fs/xfs/xfs_bmap.c

   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_rw.h"
  45 #include "xfs_quota.h"
  46 #include "xfs_trans_space.h"
  47 #include "xfs_buf_item.h"
  48 #include "xfs_filestream.h"
  49 #include "xfs_vnodeops.h"
  50 #include "xfs_trace.h"
  51
  52
  53 kmem_zone_t             *xfs_bmap_free_item_zone;
  54
  55 /*
  56  * Prototypes for internal bmap routines.
  57  */
  58
  59 #ifdef DEBUG
  60 STATIC void
  61 xfs_bmap_check_leaf_extents(
  62         struct xfs_btree_cur    *cur,
  63         struct xfs_inode        *ip,
  64         int                     whichfork);
  65 #else
  66 #define xfs_bmap_check_leaf_extents(cur, ip, whichfork)         do { } while (0)
  67 #endif
  68
  69
  70 /*
  71  * Called from xfs_bmap_add_attrfork to handle extents format files.
  72  */
  73 STATIC int                                      /* error */
  74 xfs_bmap_add_attrfork_extents(
  75         xfs_trans_t             *tp,            /* transaction pointer */
  76         xfs_inode_t             *ip,            /* incore inode pointer */
  77         xfs_fsblock_t           *firstblock,    /* first block allocated */
  78         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
  79         int                     *flags);        /* inode logging flags */
  80
  81 /*
  82  * Called from xfs_bmap_add_attrfork to handle local format files.
  83  */
  84 STATIC int                                      /* error */
  85 xfs_bmap_add_attrfork_local(
  86         xfs_trans_t             *tp,            /* transaction pointer */
  87         xfs_inode_t             *ip,            /* incore inode pointer */
  88         xfs_fsblock_t           *firstblock,    /* first block allocated */
  89         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
  90         int                     *flags);        /* inode logging flags */
  91
  92 /*
  93  * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
  94  * It figures out where to ask the underlying allocator to put the new extent.
  95  */
  96 STATIC int                              /* error */
  97 xfs_bmap_alloc(
  98         xfs_bmalloca_t          *ap);   /* bmap alloc argument struct */
  99
 100 /*
 101  * Transform a btree format file with only one leaf node, where the
 102  * extents list will fit in the inode, into an extents format file.
 103  * Since the file extents are already in-core, all we have to do is
 104  * give up the space for the btree root and pitch the leaf block.
 105  */
 106 STATIC int                              /* error */
 107 xfs_bmap_btree_to_extents(
 108         xfs_trans_t             *tp,    /* transaction pointer */
 109         xfs_inode_t             *ip,    /* incore inode pointer */
 110         xfs_btree_cur_t         *cur,   /* btree cursor */
 111         int                     *logflagsp, /* inode logging flags */
 112         int                     whichfork); /* data or attr fork */
 113
 114 /*
 115  * Remove the entry "free" from the free item list.  Prev points to the
 116  * previous entry, unless "free" is the head of the list.
 117  */
 118 STATIC void
 119 xfs_bmap_del_free(
 120         xfs_bmap_free_t         *flist, /* free item list header */
 121         xfs_bmap_free_item_t    *prev,  /* previous item on list, if any */
 122         xfs_bmap_free_item_t    *free); /* list item to be freed */
 123
 124 /*
 125  * Convert an extents-format file into a btree-format file.
 126  * The new file will have a root block (in the inode) and a single child block.
 127  */
 128 STATIC int                                      /* error */
 129 xfs_bmap_extents_to_btree(
 130         xfs_trans_t             *tp,            /* transaction pointer */
 131         xfs_inode_t             *ip,            /* incore inode pointer */
 132         xfs_fsblock_t           *firstblock,    /* first-block-allocated */
 133         xfs_bmap_free_t         *flist,         /* blocks freed in xaction */
 134         xfs_btree_cur_t         **curp,         /* cursor returned to caller */
 135         int                     wasdel,         /* converting a delayed alloc */
 136         int                     *logflagsp,     /* inode logging flags */
 137         int                     whichfork);     /* data or attr fork */
 138
 139 /*
 140  * Convert a local file to an extents file.
 141  * This code is sort of bogus, since the file data needs to get
 142  * logged so it won't be lost.  The bmap-level manipulations are ok, though.
 143  */
 144 STATIC int                              /* error */
 145 xfs_bmap_local_to_extents(
 146         xfs_trans_t     *tp,            /* transaction pointer */
 147         xfs_inode_t     *ip,            /* incore inode pointer */
 148         xfs_fsblock_t   *firstblock,    /* first block allocated in xaction */
 149         xfs_extlen_t    total,          /* total blocks needed by transaction */
 150         int             *logflagsp,     /* inode logging flags */
 151         int             whichfork);     /* data or attr fork */
 152
 153 /*
 154  * Search the extents list for the inode, for the extent containing bno.
 155  * If bno lies in a hole, point to the next entry.  If bno lies past eof,
 156  * *eofp will be set, and *prevp will contain the last entry (null if none).
 157  * Else, *lastxp will be set to the index of the found
 158  * entry; *gotp will contain the entry.
 159  */
 160 STATIC xfs_bmbt_rec_host_t *            /* pointer to found extent entry */
 161 xfs_bmap_search_extents(
 162         xfs_inode_t     *ip,            /* incore inode pointer */
 163         xfs_fileoff_t   bno,            /* block number searched for */
 164         int             whichfork,      /* data or attr fork */
 165         int             *eofp,          /* out: end of file found */
 166         xfs_extnum_t    *lastxp,        /* out: last extent index */
 167         xfs_bmbt_irec_t *gotp,          /* out: extent entry found */
 168         xfs_bmbt_irec_t *prevp);        /* out: previous extent entry found */
 169
 170 /*
 171  * Compute the worst-case number of indirect blocks that will be used
 172  * for ip's delayed extent of length "len".
 173  */
 174 STATIC xfs_filblks_t
 175 xfs_bmap_worst_indlen(
 176         xfs_inode_t             *ip,    /* incore inode pointer */
 177         xfs_filblks_t           len);   /* delayed extent length */
 178
 179 #ifdef DEBUG
 180 /*
 181  * Perform various validation checks on the values being returned
 182  * from xfs_bmapi().
 183  */
 184 STATIC void
 185 xfs_bmap_validate_ret(
 186         xfs_fileoff_t           bno,
 187         xfs_filblks_t           len,
 188         int                     flags,
 189         xfs_bmbt_irec_t         *mval,
 190         int                     nmap,
 191         int                     ret_nmap);
 192 #else
 193 #define xfs_bmap_validate_ret(bno,len,flags,mval,onmap,nmap)
 194 #endif /* DEBUG */
 195
 196 STATIC int
 197 xfs_bmap_count_tree(
 198         xfs_mount_t     *mp,
 199         xfs_trans_t     *tp,
 200         xfs_ifork_t     *ifp,
 201         xfs_fsblock_t   blockno,
 202         int             levelin,
 203         int             *count);
 204
 205 STATIC void
 206 xfs_bmap_count_leaves(
 207         xfs_ifork_t             *ifp,
 208         xfs_extnum_t            idx,
 209         int                     numrecs,
 210         int                     *count);
 211
 212 STATIC void
 213 xfs_bmap_disk_count_leaves(
 214         struct xfs_mount        *mp,
 215         struct xfs_btree_block  *block,
 216         int                     numrecs,
 217         int                     *count);
 218
 219 /*
 220  * Bmap internal routines.
 221  */
 222
 223 STATIC int                              /* error */
 224 xfs_bmbt_lookup_eq(
 225         struct xfs_btree_cur    *cur,
 226         xfs_fileoff_t           off,
 227         xfs_fsblock_t           bno,
 228         xfs_filblks_t           len,
 229         int                     *stat)  /* success/failure */
 230 {
 231         cur->bc_rec.b.br_startoff = off;
 232         cur->bc_rec.b.br_startblock = bno;
 233         cur->bc_rec.b.br_blockcount = len;
 234         return xfs_btree_lookup(cur, XFS_LOOKUP_EQ, stat);
 235 }
 236
 237 STATIC int                              /* error */
 238 xfs_bmbt_lookup_ge(
 239         struct xfs_btree_cur    *cur,
 240         xfs_fileoff_t           off,
 241         xfs_fsblock_t           bno,
 242         xfs_filblks_t           len,
 243         int                     *stat)  /* success/failure */
 244 {
 245         cur->bc_rec.b.br_startoff = off;
 246         cur->bc_rec.b.br_startblock = bno;
 247         cur->bc_rec.b.br_blockcount = len;
 248         return xfs_btree_lookup(cur, XFS_LOOKUP_GE, stat);
 249 }
 250
 251 /*
 252  * Check if the inode needs to be converted to btree format.
 253  */
 254 static inline bool xfs_bmap_needs_btree(struct xfs_inode *ip, int whichfork)
 255 {
 256         return XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
 257                 XFS_IFORK_NEXTENTS(ip, whichfork) >
 258                         XFS_IFORK_MAXEXT(ip, whichfork);
 259 }
 260
 261 /*
 262  * Check if the inode should be converted to extent format.
 263  */
 264 static inline bool xfs_bmap_wants_extents(struct xfs_inode *ip, int whichfork)
 265 {
 266         return XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE &&
 267                 XFS_IFORK_NEXTENTS(ip, whichfork) <=
 268                         XFS_IFORK_MAXEXT(ip, whichfork);
 269 }
 270
 271 /*
 272  * Update the record referred to by cur to the value given
 273  * by [off, bno, len, state].
 274  * This either works (return 0) or gets an EFSCORRUPTED error.
 275  */
 276 STATIC int
 277 xfs_bmbt_update(
 278         struct xfs_btree_cur    *cur,
 279         xfs_fileoff_t           off,
 280         xfs_fsblock_t           bno,
 281         xfs_filblks_t           len,
 282         xfs_exntst_t            state)
 283 {
 284         union xfs_btree_rec     rec;
 285
 286         xfs_bmbt_disk_set_allf(&rec.bmbt, off, bno, len, state);
 287         return xfs_btree_update(cur, &rec);
 288 }
 289
 290 /*
 291  * Called from xfs_bmap_add_attrfork to handle btree format files.
 292  */
 293 STATIC int                                      /* error */
 294 xfs_bmap_add_attrfork_btree(
 295         xfs_trans_t             *tp,            /* transaction pointer */
 296         xfs_inode_t             *ip,            /* incore inode pointer */
 297         xfs_fsblock_t           *firstblock,    /* first block allocated */
 298         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
 299         int                     *flags)         /* inode logging flags */
 300 {
 301         xfs_btree_cur_t         *cur;           /* btree cursor */
 302         int                     error;          /* error return value */
 303         xfs_mount_t             *mp;            /* file system mount struct */
 304         int                     stat;           /* newroot status */
 305
 306         mp = ip->i_mount;
 307         if (ip->i_df.if_broot_bytes <= XFS_IFORK_DSIZE(ip))
 308                 *flags |= XFS_ILOG_DBROOT;
 309         else {
 310                 cur = xfs_bmbt_init_cursor(mp, tp, ip, XFS_DATA_FORK);
 311                 cur->bc_private.b.flist = flist;
 312                 cur->bc_private.b.firstblock = *firstblock;
 313                 if ((error = xfs_bmbt_lookup_ge(cur, 0, 0, 0, &stat)))
 314                         goto error0;
 315                 /* must be at least one entry */
 316                 XFS_WANT_CORRUPTED_GOTO(stat == 1, error0);
 317                 if ((error = xfs_btree_new_iroot(cur, flags, &stat)))
 318                         goto error0;
 319                 if (stat == 0) {
 320                         xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
 321                         return XFS_ERROR(ENOSPC);
 322                 }
 323                 *firstblock = cur->bc_private.b.firstblock;
 324                 cur->bc_private.b.allocated = 0;
 325                 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
 326         }
 327         return 0;
 328 error0:
 329         xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
 330         return error;
 331 }
 332
 333 /*
 334  * Called from xfs_bmap_add_attrfork to handle extents format files.
 335  */
 336 STATIC int                                      /* error */
 337 xfs_bmap_add_attrfork_extents(
 338         xfs_trans_t             *tp,            /* transaction pointer */
 339         xfs_inode_t             *ip,            /* incore inode pointer */
 340         xfs_fsblock_t           *firstblock,    /* first block allocated */
 341         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
 342         int                     *flags)         /* inode logging flags */
 343 {
 344         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
 345         int                     error;          /* error return value */
 346
 347         if (ip->i_d.di_nextents * sizeof(xfs_bmbt_rec_t) <= XFS_IFORK_DSIZE(ip))
 348                 return 0;
 349         cur = NULL;
 350         error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist, &cur, 0,
 351                 flags, XFS_DATA_FORK);
 352         if (cur) {
 353                 cur->bc_private.b.allocated = 0;
 354                 xfs_btree_del_cursor(cur,
 355                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
 356         }
 357         return error;
 358 }
 359
 360 /*
 361  * Called from xfs_bmap_add_attrfork to handle local format files.
 362  */
 363 STATIC int                                      /* error */
 364 xfs_bmap_add_attrfork_local(
 365         xfs_trans_t             *tp,            /* transaction pointer */
 366         xfs_inode_t             *ip,            /* incore inode pointer */
 367         xfs_fsblock_t           *firstblock,    /* first block allocated */
 368         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
 369         int                     *flags)         /* inode logging flags */
 370 {
 371         xfs_da_args_t           dargs;          /* args for dir/attr code */
 372         int                     error;          /* error return value */
 373         xfs_mount_t             *mp;            /* mount structure pointer */
 374
 375         if (ip->i_df.if_bytes <= XFS_IFORK_DSIZE(ip))
 376                 return 0;
 377         if (S_ISDIR(ip->i_d.di_mode)) {
 378                 mp = ip->i_mount;
 379                 memset(&dargs, 0, sizeof(dargs));
 380                 dargs.dp = ip;
 381                 dargs.firstblock = firstblock;
 382                 dargs.flist = flist;
 383                 dargs.total = mp->m_dirblkfsbs;
 384                 dargs.whichfork = XFS_DATA_FORK;
 385                 dargs.trans = tp;
 386                 error = xfs_dir2_sf_to_block(&dargs);
 387         } else
 388                 error = xfs_bmap_local_to_extents(tp, ip, firstblock, 1, flags,
 389                         XFS_DATA_FORK);
 390         return error;
 391 }
 392
 393 /*
 394  * Convert a delayed allocation to a real allocation.
 395  */
 396 STATIC int                              /* error */
 397 xfs_bmap_add_extent_delay_real(
 398         struct xfs_bmalloca     *bma)
 399 {
 400         struct xfs_bmbt_irec    *new = &bma->got;
 401         int                     diff;   /* temp value */
 402         xfs_bmbt_rec_host_t     *ep;    /* extent entry for idx */
 403         int                     error;  /* error return value */
 404         int                     i;      /* temp state */
 405         xfs_ifork_t             *ifp;   /* inode fork pointer */
 406         xfs_fileoff_t           new_endoff;     /* end offset of new entry */
 407         xfs_bmbt_irec_t         r[3];   /* neighbor extent entries */
 408                                         /* left is 0, right is 1, prev is 2 */
 409         int                     rval=0; /* return value (logging flags) */
 410         int                     state = 0;/* state bits, accessed thru macros */
 411         xfs_filblks_t           da_new; /* new count del alloc blocks used */
 412         xfs_filblks_t           da_old; /* old count del alloc blocks used */
 413         xfs_filblks_t           temp=0; /* value for da_new calculations */
 414         xfs_filblks_t           temp2=0;/* value for da_new calculations */
 415         int                     tmp_rval;       /* partial logging flags */
 416
 417         ifp = XFS_IFORK_PTR(bma->ip, XFS_DATA_FORK);
 418
 419         ASSERT(bma->idx >= 0);
 420         ASSERT(bma->idx <= ifp->if_bytes / sizeof(struct xfs_bmbt_rec));
 421         ASSERT(!isnullstartblock(new->br_startblock));
 422         ASSERT(!bma->cur ||
 423                (bma->cur->bc_private.b.flags & XFS_BTCUR_BPRV_WASDEL));
 424
 425         XFS_STATS_INC(xs_add_exlist);
 426
 427 #define LEFT            r[0]
 428 #define RIGHT           r[1]
 429 #define PREV            r[2]
 430
 431         /*
 432          * Set up a bunch of variables to make the tests simpler.
 433          */
 434         ep = xfs_iext_get_ext(ifp, bma->idx);
 435         xfs_bmbt_get_all(ep, &PREV);
 436         new_endoff = new->br_startoff + new->br_blockcount;
 437         ASSERT(PREV.br_startoff <= new->br_startoff);
 438         ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
 439
 440         da_old = startblockval(PREV.br_startblock);
 441         da_new = 0;
 442
 443         /*
 444          * Set flags determining what part of the previous delayed allocation
 445          * extent is being replaced by a real allocation.
 446          */
 447         if (PREV.br_startoff == new->br_startoff)
 448                 state |= BMAP_LEFT_FILLING;
 449         if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
 450                 state |= BMAP_RIGHT_FILLING;
 451
 452         /*
 453          * Check and set flags if this segment has a left neighbor.
 454          * Don't set contiguous if the combined extent would be too large.
 455          */
 456         if (bma->idx > 0) {
 457                 state |= BMAP_LEFT_VALID;
 458                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx - 1), &LEFT);
 459
 460                 if (isnullstartblock(LEFT.br_startblock))
 461                         state |= BMAP_LEFT_DELAY;
 462         }
 463
 464         if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
 465             LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
 466             LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
 467             LEFT.br_state == new->br_state &&
 468             LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
 469                 state |= BMAP_LEFT_CONTIG;
 470
 471         /*
 472          * Check and set flags if this segment has a right neighbor.
 473          * Don't set contiguous if the combined extent would be too large.
 474          * Also check for all-three-contiguous being too large.
 475          */
 476         if (bma->idx < bma->ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1) {
 477                 state |= BMAP_RIGHT_VALID;
 478                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx + 1), &RIGHT);
 479
 480                 if (isnullstartblock(RIGHT.br_startblock))
 481                         state |= BMAP_RIGHT_DELAY;
 482         }
 483
 484         if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
 485             new_endoff == RIGHT.br_startoff &&
 486             new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
 487             new->br_state == RIGHT.br_state &&
 488             new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
 489             ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
 490                        BMAP_RIGHT_FILLING)) !=
 491                       (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
 492                        BMAP_RIGHT_FILLING) ||
 493              LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
 494                         <= MAXEXTLEN))
 495                 state |= BMAP_RIGHT_CONTIG;
 496
 497         error = 0;
 498         /*
 499          * Switch out based on the FILLING and CONTIG state bits.
 500          */
 501         switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
 502                          BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
 503         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
 504              BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
 505                 /*
 506                  * Filling in all of a previously delayed allocation extent.
 507                  * The left and right neighbors are both contiguous with new.
 508                  */
 509                 bma->idx--;
 510                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
 511                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
 512                         LEFT.br_blockcount + PREV.br_blockcount +
 513                         RIGHT.br_blockcount);
 514                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
 515
 516                 xfs_iext_remove(bma->ip, bma->idx + 1, 2, state);
 517                 bma->ip->i_d.di_nextents--;
 518                 if (bma->cur == NULL)
 519                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
 520                 else {
 521                         rval = XFS_ILOG_CORE;
 522                         error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff,
 523                                         RIGHT.br_startblock,
 524                                         RIGHT.br_blockcount, &i);
 525                         if (error)
 526                                 goto done;
 527                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 528                         error = xfs_btree_delete(bma->cur, &i);
 529                         if (error)
 530                                 goto done;
 531                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 532                         error = xfs_btree_decrement(bma->cur, 0, &i);
 533                         if (error)
 534                                 goto done;
 535                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 536                         error = xfs_bmbt_update(bma->cur, LEFT.br_startoff,
 537                                         LEFT.br_startblock,
 538                                         LEFT.br_blockcount +
 539                                         PREV.br_blockcount +
 540                                         RIGHT.br_blockcount, LEFT.br_state);
 541                         if (error)
 542                                 goto done;
 543                 }
 544                 break;
 545
 546         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
 547                 /*
 548                  * Filling in all of a previously delayed allocation extent.
 549                  * The left neighbor is contiguous, the right is not.
 550                  */
 551                 bma->idx--;
 552
 553                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
 554                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
 555                         LEFT.br_blockcount + PREV.br_blockcount);
 556                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
 557
 558                 xfs_iext_remove(bma->ip, bma->idx + 1, 1, state);
 559                 if (bma->cur == NULL)
 560                         rval = XFS_ILOG_DEXT;
 561                 else {
 562                         rval = 0;
 563                         error = xfs_bmbt_lookup_eq(bma->cur, LEFT.br_startoff,
 564                                         LEFT.br_startblock, LEFT.br_blockcount,
 565                                         &i);
 566                         if (error)
 567                                 goto done;
 568                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 569                         error = xfs_bmbt_update(bma->cur, LEFT.br_startoff,
 570                                         LEFT.br_startblock,
 571                                         LEFT.br_blockcount +
 572                                         PREV.br_blockcount, LEFT.br_state);
 573                         if (error)
 574                                 goto done;
 575                 }
 576                 break;
 577
 578         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
 579                 /*
 580                  * Filling in all of a previously delayed allocation extent.
 581                  * The right neighbor is contiguous, the left is not.
 582                  */
 583                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
 584                 xfs_bmbt_set_startblock(ep, new->br_startblock);
 585                 xfs_bmbt_set_blockcount(ep,
 586                         PREV.br_blockcount + RIGHT.br_blockcount);
 587                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
 588
 589                 xfs_iext_remove(bma->ip, bma->idx + 1, 1, state);
 590                 if (bma->cur == NULL)
 591                         rval = XFS_ILOG_DEXT;
 592                 else {
 593                         rval = 0;
 594                         error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff,
 595                                         RIGHT.br_startblock,
 596                                         RIGHT.br_blockcount, &i);
 597                         if (error)
 598                                 goto done;
 599                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 600                         error = xfs_bmbt_update(bma->cur, PREV.br_startoff,
 601                                         new->br_startblock,
 602                                         PREV.br_blockcount +
 603                                         RIGHT.br_blockcount, PREV.br_state);
 604                         if (error)
 605                                 goto done;
 606                 }
 607                 break;
 608
 609         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
 610                 /*
 611                  * Filling in all of a previously delayed allocation extent.
 612                  * Neither the left nor right neighbors are contiguous with
 613                  * the new one.
 614                  */
 615                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
 616                 xfs_bmbt_set_startblock(ep, new->br_startblock);
 617                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
 618
 619                 bma->ip->i_d.di_nextents++;
 620                 if (bma->cur == NULL)
 621                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
 622                 else {
 623                         rval = XFS_ILOG_CORE;
 624                         error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
 625                                         new->br_startblock, new->br_blockcount,
 626                                         &i);
 627                         if (error)
 628                                 goto done;
 629                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
 630                         bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
 631                         error = xfs_btree_insert(bma->cur, &i);
 632                         if (error)
 633                                 goto done;
 634                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 635                 }
 636                 break;
 637
 638         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
 639                 /*
 640                  * Filling in the first part of a previous delayed allocation.
 641                  * The left neighbor is contiguous.
 642                  */
 643                 trace_xfs_bmap_pre_update(bma->ip, bma->idx - 1, state, _THIS_IP_);
 644                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx - 1),
 645                         LEFT.br_blockcount + new->br_blockcount);
 646                 xfs_bmbt_set_startoff(ep,
 647                         PREV.br_startoff + new->br_blockcount);
 648                 trace_xfs_bmap_post_update(bma->ip, bma->idx - 1, state, _THIS_IP_);
 649
 650                 temp = PREV.br_blockcount - new->br_blockcount;
 651                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
 652                 xfs_bmbt_set_blockcount(ep, temp);
 653                 if (bma->cur == NULL)
 654                         rval = XFS_ILOG_DEXT;
 655                 else {
 656                         rval = 0;
 657                         error = xfs_bmbt_lookup_eq(bma->cur, LEFT.br_startoff,
 658                                         LEFT.br_startblock, LEFT.br_blockcount,
 659                                         &i);
 660                         if (error)
 661                                 goto done;
 662                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 663                         error = xfs_bmbt_update(bma->cur, LEFT.br_startoff,
 664                                         LEFT.br_startblock,
 665                                         LEFT.br_blockcount +
 666                                         new->br_blockcount,
 667                                         LEFT.br_state);
 668                         if (error)
 669                                 goto done;
 670                 }
 671                 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
 672                         startblockval(PREV.br_startblock));
 673                 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
 674                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
 675
 676                 bma->idx--;
 677                 break;
 678
 679         case BMAP_LEFT_FILLING:
 680                 /*
 681                  * Filling in the first part of a previous delayed allocation.
 682                  * The left neighbor is not contiguous.
 683                  */
 684                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
 685                 xfs_bmbt_set_startoff(ep, new_endoff);
 686                 temp = PREV.br_blockcount - new->br_blockcount;
 687                 xfs_bmbt_set_blockcount(ep, temp);
 688                 xfs_iext_insert(bma->ip, bma->idx, 1, new, state);
 689                 bma->ip->i_d.di_nextents++;
 690                 if (bma->cur == NULL)
 691                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
 692                 else {
 693                         rval = XFS_ILOG_CORE;
 694                         error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
 695                                         new->br_startblock, new->br_blockcount,
 696                                         &i);
 697                         if (error)
 698                                 goto done;
 699                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
 700                         bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
 701                         error = xfs_btree_insert(bma->cur, &i);
 702                         if (error)
 703                                 goto done;
 704                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 705                 }
 706
 707                 if (xfs_bmap_needs_btree(bma->ip, XFS_DATA_FORK)) {
 708                         error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
 709                                         bma->firstblock, bma->flist,
 710                                         &bma->cur, 1, &tmp_rval, XFS_DATA_FORK);
 711                         rval |= tmp_rval;
 712                         if (error)
 713                                 goto done;
 714                 }
 715                 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
 716                         startblockval(PREV.br_startblock) -
 717                         (bma->cur ? bma->cur->bc_private.b.allocated : 0));
 718                 ep = xfs_iext_get_ext(ifp, bma->idx + 1);
 719                 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
 720                 trace_xfs_bmap_post_update(bma->ip, bma->idx + 1, state, _THIS_IP_);
 721                 break;
 722
 723         case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
 724                 /*
 725                  * Filling in the last part of a previous delayed allocation.
 726                  * The right neighbor is contiguous with the new allocation.
 727                  */
 728                 temp = PREV.br_blockcount - new->br_blockcount;
 729                 trace_xfs_bmap_pre_update(bma->ip, bma->idx + 1, state, _THIS_IP_);
 730                 xfs_bmbt_set_blockcount(ep, temp);
 731                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, bma->idx + 1),
 732                         new->br_startoff, new->br_startblock,
 733                         new->br_blockcount + RIGHT.br_blockcount,
 734                         RIGHT.br_state);
 735                 trace_xfs_bmap_post_update(bma->ip, bma->idx + 1, state, _THIS_IP_);
 736                 if (bma->cur == NULL)
 737                         rval = XFS_ILOG_DEXT;
 738                 else {
 739                         rval = 0;
 740                         error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff,
 741                                         RIGHT.br_startblock,
 742                                         RIGHT.br_blockcount, &i);
 743                         if (error)
 744                                 goto done;
 745                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 746                         error = xfs_bmbt_update(bma->cur, new->br_startoff,
 747                                         new->br_startblock,
 748                                         new->br_blockcount +
 749                                         RIGHT.br_blockcount,
 750                                         RIGHT.br_state);
 751                         if (error)
 752                                 goto done;
 753                 }
 754
 755                 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
 756                         startblockval(PREV.br_startblock));
 757                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
 758                 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
 759                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
 760
 761                 bma->idx++;
 762                 break;
 763
 764         case BMAP_RIGHT_FILLING:
 765                 /*
 766                  * Filling in the last part of a previous delayed allocation.
 767                  * The right neighbor is not contiguous.
 768                  */
 769                 temp = PREV.br_blockcount - new->br_blockcount;
 770                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
 771                 xfs_bmbt_set_blockcount(ep, temp);
 772                 xfs_iext_insert(bma->ip, bma->idx + 1, 1, new, state);
 773                 bma->ip->i_d.di_nextents++;
 774                 if (bma->cur == NULL)
 775                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
 776                 else {
 777                         rval = XFS_ILOG_CORE;
 778                         error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
 779                                         new->br_startblock, new->br_blockcount,
 780                                         &i);
 781                         if (error)
 782                                 goto done;
 783                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
 784                         bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
 785                         error = xfs_btree_insert(bma->cur, &i);
 786                         if (error)
 787                                 goto done;
 788                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 789                 }
 790
 791                 if (xfs_bmap_needs_btree(bma->ip, XFS_DATA_FORK)) {
 792                         error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
 793                                 bma->firstblock, bma->flist, &bma->cur, 1,
 794                                 &tmp_rval, XFS_DATA_FORK);
 795                         rval |= tmp_rval;
 796                         if (error)
 797                                 goto done;
 798                 }
 799                 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
 800                         startblockval(PREV.br_startblock) -
 801                         (bma->cur ? bma->cur->bc_private.b.allocated : 0));
 802                 ep = xfs_iext_get_ext(ifp, bma->idx);
 803                 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
 804                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
 805
 806                 bma->idx++;
 807                 break;
 808
 809         case 0:
 810                 /*
 811                  * Filling in the middle part of a previous delayed allocation.
 812                  * Contiguity is impossible here.
 813                  * This case is avoided almost all the time.
 814                  *
 815                  * We start with a delayed allocation:
 816                  *
 817                  * +ddddddddddddddddddddddddddddddddddddddddddddddddddddddd+
 818                  *  PREV @ idx
 819                  *
 820                  * and we are allocating:
 821                  *                     +rrrrrrrrrrrrrrrrr+
 822                  *                            new
 823                  *
 824                  * and we set it up for insertion as:
 825                  * +ddddddddddddddddddd+rrrrrrrrrrrrrrrrr+ddddddddddddddddd+
 826                  *                            new
 827                  *  PREV @ idx          LEFT              RIGHT
 828                  *                      inserted at idx + 1
 829                  */
 830                 temp = new->br_startoff - PREV.br_startoff;
 831                 temp2 = PREV.br_startoff + PREV.br_blockcount - new_endoff;
 832                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, 0, _THIS_IP_);
 833                 xfs_bmbt_set_blockcount(ep, temp);      /* truncate PREV */
 834                 LEFT = *new;
 835                 RIGHT.br_state = PREV.br_state;
 836                 RIGHT.br_startblock = nullstartblock(
 837                                 (int)xfs_bmap_worst_indlen(bma->ip, temp2));
 838                 RIGHT.br_startoff = new_endoff;
 839                 RIGHT.br_blockcount = temp2;
 840                 /* insert LEFT (r[0]) and RIGHT (r[1]) at the same time */
 841                 xfs_iext_insert(bma->ip, bma->idx + 1, 2, &LEFT, state);
 842                 bma->ip->i_d.di_nextents++;
 843                 if (bma->cur == NULL)
 844                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
 845                 else {
 846                         rval = XFS_ILOG_CORE;
 847                         error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
 848                                         new->br_startblock, new->br_blockcount,
 849                                         &i);
 850                         if (error)
 851                                 goto done;
 852                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
 853                         bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
 854                         error = xfs_btree_insert(bma->cur, &i);
 855                         if (error)
 856                                 goto done;
 857                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 858                 }
 859
 860                 if (xfs_bmap_needs_btree(bma->ip, XFS_DATA_FORK)) {
 861                         error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
 862                                         bma->firstblock, bma->flist, &bma->cur,
 863                                         1, &tmp_rval, XFS_DATA_FORK);
 864                         rval |= tmp_rval;
 865                         if (error)
 866                                 goto done;
 867                 }
 868                 temp = xfs_bmap_worst_indlen(bma->ip, temp);
 869                 temp2 = xfs_bmap_worst_indlen(bma->ip, temp2);
 870                 diff = (int)(temp + temp2 - startblockval(PREV.br_startblock) -
 871                         (bma->cur ? bma->cur->bc_private.b.allocated : 0));
 872                 if (diff > 0) {
 873                         error = xfs_icsb_modify_counters(bma->ip->i_mount,
 874                                         XFS_SBS_FDBLOCKS,
 875                                         -((int64_t)diff), 0);
 876                         ASSERT(!error);
 877                         if (error)
 878                                 goto done;
 879                 }
 880
 881                 ep = xfs_iext_get_ext(ifp, bma->idx);
 882                 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
 883                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
 884                 trace_xfs_bmap_pre_update(bma->ip, bma->idx + 2, state, _THIS_IP_);
 885                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, bma->idx + 2),
 886                         nullstartblock((int)temp2));
 887                 trace_xfs_bmap_post_update(bma->ip, bma->idx + 2, state, _THIS_IP_);
 888
 889                 bma->idx++;
 890                 da_new = temp + temp2;
 891                 break;
 892
 893         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
 894         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
 895         case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
 896         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
 897         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
 898         case BMAP_LEFT_CONTIG:
 899         case BMAP_RIGHT_CONTIG:
 900                 /*
 901                  * These cases are all impossible.
 902                  */
 903                 ASSERT(0);
 904         }
 905
 906         /* convert to a btree if necessary */
 907         if (xfs_bmap_needs_btree(bma->ip, XFS_DATA_FORK)) {
 908                 int     tmp_logflags;   /* partial log flag return val */
 909
 910                 ASSERT(bma->cur == NULL);
 911                 error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
 912                                 bma->firstblock, bma->flist, &bma->cur,
 913                                 da_old > 0, &tmp_logflags, XFS_DATA_FORK);
 914                 bma->logflags |= tmp_logflags;
 915                 if (error)
 916                         goto done;
 917         }
 918
 919         /* adjust for changes in reserved delayed indirect blocks */
 920         if (da_old || da_new) {
 921                 temp = da_new;
 922                 if (bma->cur)
 923                         temp += bma->cur->bc_private.b.allocated;
 924                 ASSERT(temp <= da_old);
 925                 if (temp < da_old)
 926                         xfs_icsb_modify_counters(bma->ip->i_mount,
 927                                         XFS_SBS_FDBLOCKS,
 928                                         (int64_t)(da_old - temp), 0);
 929         }
 930
 931         /* clear out the allocated field, done with it now in any case. */
 932         if (bma->cur)
 933                 bma->cur->bc_private.b.allocated = 0;
 934
 935         xfs_bmap_check_leaf_extents(bma->cur, bma->ip, XFS_DATA_FORK);
 936 done:
 937         bma->logflags |= rval;
 938         return error;
 939 #undef  LEFT
 940 #undef  RIGHT
 941 #undef  PREV
 942 }
 943
 944 /*
 945  * Convert an unwritten allocation to a real allocation or vice versa.
 946  */
 947 STATIC int                              /* error */
 948 xfs_bmap_add_extent_unwritten_real(
 949         struct xfs_trans        *tp,
 950         xfs_inode_t             *ip,    /* incore inode pointer */
 951         xfs_extnum_t            *idx,   /* extent number to update/insert */
 952         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
 953         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
 954         xfs_fsblock_t           *first, /* pointer to firstblock variable */
 955         xfs_bmap_free_t         *flist, /* list of extents to be freed */
 956         int                     *logflagsp) /* inode logging flags */
 957 {
 958         xfs_btree_cur_t         *cur;   /* btree cursor */
 959         xfs_bmbt_rec_host_t     *ep;    /* extent entry for idx */
 960         int                     error;  /* error return value */
 961         int                     i;      /* temp state */
 962         xfs_ifork_t             *ifp;   /* inode fork pointer */
 963         xfs_fileoff_t           new_endoff;     /* end offset of new entry */
 964         xfs_exntst_t            newext; /* new extent state */
 965         xfs_exntst_t            oldext; /* old extent state */
 966         xfs_bmbt_irec_t         r[3];   /* neighbor extent entries */
 967                                         /* left is 0, right is 1, prev is 2 */
 968         int                     rval=0; /* return value (logging flags) */
 969         int                     state = 0;/* state bits, accessed thru macros */
 970
 971         *logflagsp = 0;
 972
 973         cur = *curp;
 974         ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
 975
 976         ASSERT(*idx >= 0);
 977         ASSERT(*idx <= ifp->if_bytes / sizeof(struct xfs_bmbt_rec));
 978         ASSERT(!isnullstartblock(new->br_startblock));
 979
 980         XFS_STATS_INC(xs_add_exlist);
 981
 982 #define LEFT            r[0]
 983 #define RIGHT           r[1]
 984 #define PREV            r[2]
 985
 986         /*
 987          * Set up a bunch of variables to make the tests simpler.
 988          */
 989         error = 0;
 990         ep = xfs_iext_get_ext(ifp, *idx);
 991         xfs_bmbt_get_all(ep, &PREV);
 992         newext = new->br_state;
 993         oldext = (newext == XFS_EXT_UNWRITTEN) ?
 994                 XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
 995         ASSERT(PREV.br_state == oldext);
 996         new_endoff = new->br_startoff + new->br_blockcount;
 997         ASSERT(PREV.br_startoff <= new->br_startoff);
 998         ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
 999
1000         /*
1001          * Set flags determining what part of the previous oldext allocation
1002          * extent is being replaced by a newext allocation.
1003          */
1004         if (PREV.br_startoff == new->br_startoff)
1005                 state |= BMAP_LEFT_FILLING;
1006         if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
1007                 state |= BMAP_RIGHT_FILLING;
1008
1009         /*
1010          * Check and set flags if this segment has a left neighbor.
1011          * Don't set contiguous if the combined extent would be too large.
1012          */
1013         if (*idx > 0) {
1014                 state |= BMAP_LEFT_VALID;
1015                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &LEFT);
1016
1017                 if (isnullstartblock(LEFT.br_startblock))
1018                         state |= BMAP_LEFT_DELAY;
1019         }
1020
1021         if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
1022             LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
1023             LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
1024             LEFT.br_state == newext &&
1025             LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
1026                 state |= BMAP_LEFT_CONTIG;
1027
1028         /*
1029          * Check and set flags if this segment has a right neighbor.
1030          * Don't set contiguous if the combined extent would be too large.
1031          * Also check for all-three-contiguous being too large.
1032          */
1033         if (*idx < ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1) {
1034                 state |= BMAP_RIGHT_VALID;
1035                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx + 1), &RIGHT);
1036                 if (isnullstartblock(RIGHT.br_startblock))
1037                         state |= BMAP_RIGHT_DELAY;
1038         }
1039
1040         if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
1041             new_endoff == RIGHT.br_startoff &&
1042             new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
1043             newext == RIGHT.br_state &&
1044             new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
1045             ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1046                        BMAP_RIGHT_FILLING)) !=
1047                       (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1048                        BMAP_RIGHT_FILLING) ||
1049              LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
1050                         <= MAXEXTLEN))
1051                 state |= BMAP_RIGHT_CONTIG;
1052
1053         /*
1054          * Switch out based on the FILLING and CONTIG state bits.
1055          */
1056         switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1057                          BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
1058         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1059              BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1060                 /*
1061                  * Setting all of a previous oldext extent to newext.
1062                  * The left and right neighbors are both contiguous with new.
1063                  */
1064                 --*idx;
1065
1066                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1067                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
1068                         LEFT.br_blockcount + PREV.br_blockcount +
1069                         RIGHT.br_blockcount);
1070                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1071
1072                 xfs_iext_remove(ip, *idx + 1, 2, state);
1073                 ip->i_d.di_nextents -= 2;
1074                 if (cur == NULL)
1075                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1076                 else {
1077                         rval = XFS_ILOG_CORE;
1078                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
1079                                         RIGHT.br_startblock,
1080                                         RIGHT.br_blockcount, &i)))
1081                                 goto done;
1082                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1083                         if ((error = xfs_btree_delete(cur, &i)))
1084                                 goto done;
1085                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1086                         if ((error = xfs_btree_decrement(cur, 0, &i)))
1087                                 goto done;
1088                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1089                         if ((error = xfs_btree_delete(cur, &i)))
1090                                 goto done;
1091                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1092                         if ((error = xfs_btree_decrement(cur, 0, &i)))
1093                                 goto done;
1094                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1095                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
1096                                 LEFT.br_startblock,
1097                                 LEFT.br_blockcount + PREV.br_blockcount +
1098                                 RIGHT.br_blockcount, LEFT.br_state)))
1099                                 goto done;
1100                 }
1101                 break;
1102
1103         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1104                 /*
1105                  * Setting all of a previous oldext extent to newext.
1106                  * The left neighbor is contiguous, the right is not.
1107                  */
1108                 --*idx;
1109
1110                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1111                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
1112                         LEFT.br_blockcount + PREV.br_blockcount);
1113                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1114
1115                 xfs_iext_remove(ip, *idx + 1, 1, state);
1116                 ip->i_d.di_nextents--;
1117                 if (cur == NULL)
1118                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1119                 else {
1120                         rval = XFS_ILOG_CORE;
1121                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1122                                         PREV.br_startblock, PREV.br_blockcount,
1123                                         &i)))
1124                                 goto done;
1125                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1126                         if ((error = xfs_btree_delete(cur, &i)))
1127                                 goto done;
1128                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1129                         if ((error = xfs_btree_decrement(cur, 0, &i)))
1130                                 goto done;
1131                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1132                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
1133                                 LEFT.br_startblock,
1134                                 LEFT.br_blockcount + PREV.br_blockcount,
1135                                 LEFT.br_state)))
1136                                 goto done;
1137                 }
1138                 break;
1139
1140         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1141                 /*
1142                  * Setting all of a previous oldext extent to newext.
1143                  * The right neighbor is contiguous, the left is not.
1144                  */
1145                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1146                 xfs_bmbt_set_blockcount(ep,
1147                         PREV.br_blockcount + RIGHT.br_blockcount);
1148                 xfs_bmbt_set_state(ep, newext);
1149                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1150                 xfs_iext_remove(ip, *idx + 1, 1, state);
1151                 ip->i_d.di_nextents--;
1152                 if (cur == NULL)
1153                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1154                 else {
1155                         rval = XFS_ILOG_CORE;
1156                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
1157                                         RIGHT.br_startblock,
1158                                         RIGHT.br_blockcount, &i)))
1159                                 goto done;
1160                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1161                         if ((error = xfs_btree_delete(cur, &i)))
1162                                 goto done;
1163                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1164                         if ((error = xfs_btree_decrement(cur, 0, &i)))
1165                                 goto done;
1166                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1167                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
1168                                 new->br_startblock,
1169                                 new->br_blockcount + RIGHT.br_blockcount,
1170                                 newext)))
1171                                 goto done;
1172                 }
1173                 break;
1174
1175         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
1176                 /*
1177                  * Setting all of a previous oldext extent to newext.
1178                  * Neither the left nor right neighbors are contiguous with
1179                  * the new one.
1180                  */
1181                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1182                 xfs_bmbt_set_state(ep, newext);
1183                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1184
1185                 if (cur == NULL)
1186                         rval = XFS_ILOG_DEXT;
1187                 else {
1188                         rval = 0;
1189                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1190                                         new->br_startblock, new->br_blockcount,
1191                                         &i)))
1192                                 goto done;
1193                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1194                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
1195                                 new->br_startblock, new->br_blockcount,
1196                                 newext)))
1197                                 goto done;
1198                 }
1199                 break;
1200
1201         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
1202                 /*
1203                  * Setting the first part of a previous oldext extent to newext.
1204                  * The left neighbor is contiguous.
1205                  */
1206                 trace_xfs_bmap_pre_update(ip, *idx - 1, state, _THIS_IP_);
1207                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx - 1),
1208                         LEFT.br_blockcount + new->br_blockcount);
1209                 xfs_bmbt_set_startoff(ep,
1210                         PREV.br_startoff + new->br_blockcount);
1211                 trace_xfs_bmap_post_update(ip, *idx - 1, state, _THIS_IP_);
1212
1213                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1214                 xfs_bmbt_set_startblock(ep,
1215                         new->br_startblock + new->br_blockcount);
1216                 xfs_bmbt_set_blockcount(ep,
1217                         PREV.br_blockcount - new->br_blockcount);
1218                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1219
1220                 --*idx;
1221
1222                 if (cur == NULL)
1223                         rval = XFS_ILOG_DEXT;
1224                 else {
1225                         rval = 0;
1226                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1227                                         PREV.br_startblock, PREV.br_blockcount,
1228                                         &i)))
1229                                 goto done;
1230                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1231                         if ((error = xfs_bmbt_update(cur,
1232                                 PREV.br_startoff + new->br_blockcount,
1233                                 PREV.br_startblock + new->br_blockcount,
1234                                 PREV.br_blockcount - new->br_blockcount,
1235                                 oldext)))
1236                                 goto done;
1237                         if ((error = xfs_btree_decrement(cur, 0, &i)))
1238                                 goto done;
1239                         error = xfs_bmbt_update(cur, LEFT.br_startoff,
1240                                 LEFT.br_startblock,
1241                                 LEFT.br_blockcount + new->br_blockcount,
1242                                 LEFT.br_state);
1243                         if (error)
1244                                 goto done;
1245                 }
1246                 break;
1247
1248         case BMAP_LEFT_FILLING:
1249                 /*
1250                  * Setting the first part of a previous oldext extent to newext.
1251                  * The left neighbor is not contiguous.
1252                  */
1253                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1254                 ASSERT(ep && xfs_bmbt_get_state(ep) == oldext);
1255                 xfs_bmbt_set_startoff(ep, new_endoff);
1256                 xfs_bmbt_set_blockcount(ep,
1257                         PREV.br_blockcount - new->br_blockcount);
1258                 xfs_bmbt_set_startblock(ep,
1259                         new->br_startblock + new->br_blockcount);
1260                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1261
1262                 xfs_iext_insert(ip, *idx, 1, new, state);
1263                 ip->i_d.di_nextents++;
1264                 if (cur == NULL)
1265                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1266                 else {
1267                         rval = XFS_ILOG_CORE;
1268                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1269                                         PREV.br_startblock, PREV.br_blockcount,
1270                                         &i)))
1271                                 goto done;
1272                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1273                         if ((error = xfs_bmbt_update(cur,
1274                                 PREV.br_startoff + new->br_blockcount,
1275                                 PREV.br_startblock + new->br_blockcount,
1276                                 PREV.br_blockcount - new->br_blockcount,
1277                                 oldext)))
1278                                 goto done;
1279                         cur->bc_rec.b = *new;
1280                         if ((error = xfs_btree_insert(cur, &i)))
1281                                 goto done;
1282                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1283                 }
1284                 break;
1285
1286         case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1287                 /*
1288                  * Setting the last part of a previous oldext extent to newext.
1289                  * The right neighbor is contiguous with the new allocation.
1290                  */
1291                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1292                 xfs_bmbt_set_blockcount(ep,
1293                         PREV.br_blockcount - new->br_blockcount);
1294                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1295
1296                 ++*idx;
1297
1298                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1299                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx),
1300                         new->br_startoff, new->br_startblock,
1301                         new->br_blockcount + RIGHT.br_blockcount, newext);
1302                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1303
1304                 if (cur == NULL)
1305                         rval = XFS_ILOG_DEXT;
1306                 else {
1307                         rval = 0;
1308                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1309                                         PREV.br_startblock,
1310                                         PREV.br_blockcount, &i)))
1311                                 goto done;
1312                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1313                         if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
1314                                 PREV.br_startblock,
1315                                 PREV.br_blockcount - new->br_blockcount,
1316                                 oldext)))
1317                                 goto done;
1318                         if ((error = xfs_btree_increment(cur, 0, &i)))
1319                                 goto done;
1320                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
1321                                 new->br_startblock,
1322                                 new->br_blockcount + RIGHT.br_blockcount,
1323                                 newext)))
1324                                 goto done;
1325                 }
1326                 break;
1327
1328         case BMAP_RIGHT_FILLING:
1329                 /*
1330                  * Setting the last part of a previous oldext extent to newext.
1331                  * The right neighbor is not contiguous.
1332                  */
1333                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1334                 xfs_bmbt_set_blockcount(ep,
1335                         PREV.br_blockcount - new->br_blockcount);
1336                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1337
1338                 ++*idx;
1339                 xfs_iext_insert(ip, *idx, 1, new, state);
1340
1341                 ip->i_d.di_nextents++;
1342                 if (cur == NULL)
1343                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1344                 else {
1345                         rval = XFS_ILOG_CORE;
1346                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1347                                         PREV.br_startblock, PREV.br_blockcount,
1348                                         &i)))
1349                                 goto done;
1350                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1351                         if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
1352                                 PREV.br_startblock,
1353                                 PREV.br_blockcount - new->br_blockcount,
1354                                 oldext)))
1355                                 goto done;
1356                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1357                                         new->br_startblock, new->br_blockcount,
1358                                         &i)))
1359                                 goto done;
1360                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1361                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
1362                         if ((error = xfs_btree_insert(cur, &i)))
1363                                 goto done;
1364                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1365                 }
1366                 break;
1367
1368         case 0:
1369                 /*
1370                  * Setting the middle part of a previous oldext extent to
1371                  * newext.  Contiguity is impossible here.
1372                  * One extent becomes three extents.
1373                  */
1374                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1375                 xfs_bmbt_set_blockcount(ep,
1376                         new->br_startoff - PREV.br_startoff);
1377                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1378
1379                 r[0] = *new;
1380                 r[1].br_startoff = new_endoff;
1381                 r[1].br_blockcount =
1382                         PREV.br_startoff + PREV.br_blockcount - new_endoff;
1383                 r[1].br_startblock = new->br_startblock + new->br_blockcount;
1384                 r[1].br_state = oldext;
1385
1386                 ++*idx;
1387                 xfs_iext_insert(ip, *idx, 2, &r[0], state);
1388
1389                 ip->i_d.di_nextents += 2;
1390                 if (cur == NULL)
1391                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1392                 else {
1393                         rval = XFS_ILOG_CORE;
1394                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1395                                         PREV.br_startblock, PREV.br_blockcount,
1396                                         &i)))
1397                                 goto done;
1398                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1399                         /* new right extent - oldext */
1400                         if ((error = xfs_bmbt_update(cur, r[1].br_startoff,
1401                                 r[1].br_startblock, r[1].br_blockcount,
1402                                 r[1].br_state)))
1403                                 goto done;
1404                         /* new left extent - oldext */
1405                         cur->bc_rec.b = PREV;
1406                         cur->bc_rec.b.br_blockcount =
1407                                 new->br_startoff - PREV.br_startoff;
1408                         if ((error = xfs_btree_insert(cur, &i)))
1409                                 goto done;
1410                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1411                         /*
1412                          * Reset the cursor to the position of the new extent
1413                          * we are about to insert as we can't trust it after
1414                          * the previous insert.
1415                          */
1416                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1417                                         new->br_startblock, new->br_blockcount,
1418                                         &i)))
1419                                 goto done;
1420                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1421                         /* new middle extent - newext */
1422                         cur->bc_rec.b.br_state = new->br_state;
1423                         if ((error = xfs_btree_insert(cur, &i)))
1424                                 goto done;
1425                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1426                 }
1427                 break;
1428
1429         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1430         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1431         case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
1432         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1433         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1434         case BMAP_LEFT_CONTIG:
1435         case BMAP_RIGHT_CONTIG:
1436                 /*
1437                  * These cases are all impossible.
1438                  */
1439                 ASSERT(0);
1440         }
1441
1442         /* convert to a btree if necessary */
1443         if (xfs_bmap_needs_btree(ip, XFS_DATA_FORK)) {
1444                 int     tmp_logflags;   /* partial log flag return val */
1445
1446                 ASSERT(cur == NULL);
1447                 error = xfs_bmap_extents_to_btree(tp, ip, first, flist, &cur,
1448                                 0, &tmp_logflags, XFS_DATA_FORK);
1449                 *logflagsp |= tmp_logflags;
1450                 if (error)
1451                         goto done;
1452         }
1453
1454         /* clear out the allocated field, done with it now in any case. */
1455         if (cur) {
1456                 cur->bc_private.b.allocated = 0;
1457                 *curp = cur;
1458         }
1459
1460         xfs_bmap_check_leaf_extents(*curp, ip, XFS_DATA_FORK);
1461 done:
1462         *logflagsp |= rval;
1463         return error;
1464 #undef  LEFT
1465 #undef  RIGHT
1466 #undef  PREV
1467 }
1468
1469 /*
1470  * Convert a hole to a delayed allocation.
1471  */
1472 STATIC void
1473 xfs_bmap_add_extent_hole_delay(
1474         xfs_inode_t             *ip,    /* incore inode pointer */
1475         xfs_extnum_t            *idx,   /* extent number to update/insert */
1476         xfs_bmbt_irec_t         *new)   /* new data to add to file extents */
1477 {
1478         xfs_ifork_t             *ifp;   /* inode fork pointer */
1479         xfs_bmbt_irec_t         left;   /* left neighbor extent entry */
1480         xfs_filblks_t           newlen=0;       /* new indirect size */
1481         xfs_filblks_t           oldlen=0;       /* old indirect size */
1482         xfs_bmbt_irec_t         right;  /* right neighbor extent entry */
1483         int                     state;  /* state bits, accessed thru macros */
1484         xfs_filblks_t           temp=0; /* temp for indirect calculations */
1485
1486         ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
1487         state = 0;
1488         ASSERT(isnullstartblock(new->br_startblock));
1489
1490         /*
1491          * Check and set flags if this segment has a left neighbor
1492          */
1493         if (*idx > 0) {
1494                 state |= BMAP_LEFT_VALID;
1495                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &left);
1496
1497                 if (isnullstartblock(left.br_startblock))
1498                         state |= BMAP_LEFT_DELAY;
1499         }
1500
1501         /*
1502          * Check and set flags if the current (right) segment exists.
1503          * If it doesn't exist, we're converting the hole at end-of-file.
1504          */
1505         if (*idx < ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t)) {
1506                 state |= BMAP_RIGHT_VALID;
1507                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx), &right);
1508
1509                 if (isnullstartblock(right.br_startblock))
1510                         state |= BMAP_RIGHT_DELAY;
1511         }
1512
1513         /*
1514          * Set contiguity flags on the left and right neighbors.
1515          * Don't let extents get too large, even if the pieces are contiguous.
1516          */
1517         if ((state & BMAP_LEFT_VALID) && (state & BMAP_LEFT_DELAY) &&
1518             left.br_startoff + left.br_blockcount == new->br_startoff &&
1519             left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
1520                 state |= BMAP_LEFT_CONTIG;
1521
1522         if ((state & BMAP_RIGHT_VALID) && (state & BMAP_RIGHT_DELAY) &&
1523             new->br_startoff + new->br_blockcount == right.br_startoff &&
1524             new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
1525             (!(state & BMAP_LEFT_CONTIG) ||
1526              (left.br_blockcount + new->br_blockcount +
1527               right.br_blockcount <= MAXEXTLEN)))
1528                 state |= BMAP_RIGHT_CONTIG;
1529
1530         /*
1531          * Switch out based on the contiguity flags.
1532          */
1533         switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
1534         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1535                 /*
1536                  * New allocation is contiguous with delayed allocations
1537                  * on the left and on the right.
1538                  * Merge all three into a single extent record.
1539                  */
1540                 --*idx;
1541                 temp = left.br_blockcount + new->br_blockcount +
1542                         right.br_blockcount;
1543
1544                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1545                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx), temp);
1546                 oldlen = startblockval(left.br_startblock) +
1547                         startblockval(new->br_startblock) +
1548                         startblockval(right.br_startblock);
1549                 newlen = xfs_bmap_worst_indlen(ip, temp);
1550                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, *idx),
1551                         nullstartblock((int)newlen));
1552                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1553
1554                 xfs_iext_remove(ip, *idx + 1, 1, state);
1555                 break;
1556
1557         case BMAP_LEFT_CONTIG:
1558                 /*
1559                  * New allocation is contiguous with a delayed allocation
1560                  * on the left.
1561                  * Merge the new allocation with the left neighbor.
1562                  */
1563                 --*idx;
1564                 temp = left.br_blockcount + new->br_blockcount;
1565
1566                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1567                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx), temp);
1568                 oldlen = startblockval(left.br_startblock) +
1569                         startblockval(new->br_startblock);
1570                 newlen = xfs_bmap_worst_indlen(ip, temp);
1571                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, *idx),
1572                         nullstartblock((int)newlen));
1573                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1574                 break;
1575
1576         case BMAP_RIGHT_CONTIG:
1577                 /*
1578                  * New allocation is contiguous with a delayed allocation
1579                  * on the right.
1580                  * Merge the new allocation with the right neighbor.
1581                  */
1582                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1583                 temp = new->br_blockcount + right.br_blockcount;
1584                 oldlen = startblockval(new->br_startblock) +
1585                         startblockval(right.br_startblock);
1586                 newlen = xfs_bmap_worst_indlen(ip, temp);
1587                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx),
1588                         new->br_startoff,
1589                         nullstartblock((int)newlen), temp, right.br_state);
1590                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1591                 break;
1592
1593         case 0:
1594                 /*
1595                  * New allocation is not contiguous with another
1596                  * delayed allocation.
1597                  * Insert a new entry.
1598                  */
1599                 oldlen = newlen = 0;
1600                 xfs_iext_insert(ip, *idx, 1, new, state);
1601                 break;
1602         }
1603         if (oldlen != newlen) {
1604                 ASSERT(oldlen > newlen);
1605                 xfs_icsb_modify_counters(ip->i_mount, XFS_SBS_FDBLOCKS,
1606                         (int64_t)(oldlen - newlen), 0);
1607                 /*
1608                  * Nothing to do for disk quota accounting here.
1609                  */
1610         }
1611 }
1612
1613 /*
1614  * Convert a hole to a real allocation.
1615  */
1616 STATIC int                              /* error */
1617 xfs_bmap_add_extent_hole_real(
1618         struct xfs_bmalloca     *bma,
1619         int                     whichfork)
1620 {
1621         struct xfs_bmbt_irec    *new = &bma->got;
1622         int                     error;  /* error return value */
1623         int                     i;      /* temp state */
1624         xfs_ifork_t             *ifp;   /* inode fork pointer */
1625         xfs_bmbt_irec_t         left;   /* left neighbor extent entry */
1626         xfs_bmbt_irec_t         right;  /* right neighbor extent entry */
1627         int                     rval=0; /* return value (logging flags) */
1628         int                     state;  /* state bits, accessed thru macros */
1629
1630         ifp = XFS_IFORK_PTR(bma->ip, whichfork);
1631
1632         ASSERT(bma->idx >= 0);
1633         ASSERT(bma->idx <= ifp->if_bytes / sizeof(struct xfs_bmbt_rec));
1634         ASSERT(!isnullstartblock(new->br_startblock));
1635         ASSERT(!bma->cur ||
1636                !(bma->cur->bc_private.b.flags & XFS_BTCUR_BPRV_WASDEL));
1637
1638         XFS_STATS_INC(xs_add_exlist);
1639
1640         state = 0;
1641         if (whichfork == XFS_ATTR_FORK)
1642                 state |= BMAP_ATTRFORK;
1643
1644         /*
1645          * Check and set flags if this segment has a left neighbor.
1646          */
1647         if (bma->idx > 0) {
1648                 state |= BMAP_LEFT_VALID;
1649                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx - 1), &left);
1650                 if (isnullstartblock(left.br_startblock))
1651                         state |= BMAP_LEFT_DELAY;
1652         }
1653
1654         /*
1655          * Check and set flags if this segment has a current value.
1656          * Not true if we're inserting into the "hole" at eof.
1657          */
1658         if (bma->idx < ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)) {
1659                 state |= BMAP_RIGHT_VALID;
1660                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx), &right);
1661                 if (isnullstartblock(right.br_startblock))
1662                         state |= BMAP_RIGHT_DELAY;
1663         }
1664
1665         /*
1666          * We're inserting a real allocation between "left" and "right".
1667          * Set the contiguity flags.  Don't let extents get too large.
1668          */
1669         if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
1670             left.br_startoff + left.br_blockcount == new->br_startoff &&
1671             left.br_startblock + left.br_blockcount == new->br_startblock &&
1672             left.br_state == new->br_state &&
1673             left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
1674                 state |= BMAP_LEFT_CONTIG;
1675
1676         if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
1677             new->br_startoff + new->br_blockcount == right.br_startoff &&
1678             new->br_startblock + new->br_blockcount == right.br_startblock &&
1679             new->br_state == right.br_state &&
1680             new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
1681             (!(state & BMAP_LEFT_CONTIG) ||
1682              left.br_blockcount + new->br_blockcount +
1683              right.br_blockcount <= MAXEXTLEN))
1684                 state |= BMAP_RIGHT_CONTIG;
1685
1686         error = 0;
1687         /*
1688          * Select which case we're in here, and implement it.
1689          */
1690         switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
1691         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1692                 /*
1693                  * New allocation is contiguous with real allocations on the
1694                  * left and on the right.
1695                  * Merge all three into a single extent record.
1696                  */
1697                 --bma->idx;
1698                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1699                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
1700                         left.br_blockcount + new->br_blockcount +
1701                         right.br_blockcount);
1702                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1703
1704                 xfs_iext_remove(bma->ip, bma->idx + 1, 1, state);
1705
1706                 XFS_IFORK_NEXT_SET(bma->ip, whichfork,
1707                         XFS_IFORK_NEXTENTS(bma->ip, whichfork) - 1);
1708                 if (bma->cur == NULL) {
1709                         rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
1710                 } else {
1711                         rval = XFS_ILOG_CORE;
1712                         error = xfs_bmbt_lookup_eq(bma->cur, right.br_startoff,
1713                                         right.br_startblock, right.br_blockcount,
1714                                         &i);
1715                         if (error)
1716                                 goto done;
1717                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1718                         error = xfs_btree_delete(bma->cur, &i);
1719                         if (error)
1720                                 goto done;
1721                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1722                         error = xfs_btree_decrement(bma->cur, 0, &i);
1723                         if (error)
1724                                 goto done;
1725                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1726                         error = xfs_bmbt_update(bma->cur, left.br_startoff,
1727                                         left.br_startblock,
1728                                         left.br_blockcount +
1729                                                 new->br_blockcount +
1730                                                 right.br_blockcount,
1731                                         left.br_state);
1732                         if (error)
1733                                 goto done;
1734                 }
1735                 break;
1736
1737         case BMAP_LEFT_CONTIG:
1738                 /*
1739                  * New allocation is contiguous with a real allocation
1740                  * on the left.
1741                  * Merge the new allocation with the left neighbor.
1742                  */
1743                 --bma->idx;
1744                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1745                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
1746                         left.br_blockcount + new->br_blockcount);
1747                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1748
1749                 if (bma->cur == NULL) {
1750                         rval = xfs_ilog_fext(whichfork);
1751                 } else {
1752                         rval = 0;
1753                         error = xfs_bmbt_lookup_eq(bma->cur, left.br_startoff,
1754                                         left.br_startblock, left.br_blockcount,
1755                                         &i);
1756                         if (error)
1757                                 goto done;
1758                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1759                         error = xfs_bmbt_update(bma->cur, left.br_startoff,
1760                                         left.br_startblock,
1761                                         left.br_blockcount +
1762                                                 new->br_blockcount,
1763                                         left.br_state);
1764                         if (error)
1765                                 goto done;
1766                 }
1767                 break;
1768
1769         case BMAP_RIGHT_CONTIG:
1770                 /*
1771                  * New allocation is contiguous with a real allocation
1772                  * on the right.
1773                  * Merge the new allocation with the right neighbor.
1774                  */
1775                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1776                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, bma->idx),
1777                         new->br_startoff, new->br_startblock,
1778                         new->br_blockcount + right.br_blockcount,
1779                         right.br_state);
1780                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1781
1782                 if (bma->cur == NULL) {
1783                         rval = xfs_ilog_fext(whichfork);
1784                 } else {
1785                         rval = 0;
1786                         error = xfs_bmbt_lookup_eq(bma->cur,
1787                                         right.br_startoff,
1788                                         right.br_startblock,
1789                                         right.br_blockcount, &i);
1790                         if (error)
1791                                 goto done;
1792                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1793                         error = xfs_bmbt_update(bma->cur, new->br_startoff,
1794                                         new->br_startblock,
1795                                         new->br_blockcount +
1796                                                 right.br_blockcount,
1797                                         right.br_state);
1798                         if (error)
1799                                 goto done;
1800                 }
1801                 break;
1802
1803         case 0:
1804                 /*
1805                  * New allocation is not contiguous with another
1806                  * real allocation.
1807                  * Insert a new entry.
1808                  */
1809                 xfs_iext_insert(bma->ip, bma->idx, 1, new, state);
1810                 XFS_IFORK_NEXT_SET(bma->ip, whichfork,
1811                         XFS_IFORK_NEXTENTS(bma->ip, whichfork) + 1);
1812                 if (bma->cur == NULL) {
1813                         rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
1814                 } else {
1815                         rval = XFS_ILOG_CORE;
1816                         error = xfs_bmbt_lookup_eq(bma->cur,
1817                                         new->br_startoff,
1818                                         new->br_startblock,
1819                                         new->br_blockcount, &i);
1820                         if (error)
1821                                 goto done;
1822                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1823                         bma->cur->bc_rec.b.br_state = new->br_state;
1824                         error = xfs_btree_insert(bma->cur, &i);
1825                         if (error)
1826                                 goto done;
1827                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1828                 }
1829                 break;
1830         }
1831
1832         /* convert to a btree if necessary */
1833         if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
1834                 int     tmp_logflags;   /* partial log flag return val */
1835
1836                 ASSERT(bma->cur == NULL);
1837                 error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
1838                                 bma->firstblock, bma->flist, &bma->cur,
1839                                 0, &tmp_logflags, whichfork);
1840                 bma->logflags |= tmp_logflags;
1841                 if (error)
1842                         goto done;
1843         }
1844
1845         /* clear out the allocated field, done with it now in any case. */
1846         if (bma->cur)
1847                 bma->cur->bc_private.b.allocated = 0;
1848
1849         xfs_bmap_check_leaf_extents(bma->cur, bma->ip, whichfork);
1850 done:
1851         bma->logflags |= rval;
1852         return error;
1853 }
1854
1855 /*
1856  * Adjust the size of the new extent based on di_extsize and rt extsize.
1857  */
1858 STATIC int
1859 xfs_bmap_extsize_align(
1860         xfs_mount_t     *mp,
1861         xfs_bmbt_irec_t *gotp,          /* next extent pointer */
1862         xfs_bmbt_irec_t *prevp,         /* previous extent pointer */
1863         xfs_extlen_t    extsz,          /* align to this extent size */
1864         int             rt,             /* is this a realtime inode? */
1865         int             eof,            /* is extent at end-of-file? */
1866         int             delay,          /* creating delalloc extent? */
1867         int             convert,        /* overwriting unwritten extent? */
1868         xfs_fileoff_t   *offp,          /* in/out: aligned offset */
1869         xfs_extlen_t    *lenp)          /* in/out: aligned length */
1870 {
1871         xfs_fileoff_t   orig_off;       /* original offset */
1872         xfs_extlen_t    orig_alen;      /* original length */
1873         xfs_fileoff_t   orig_end;       /* original off+len */
1874         xfs_fileoff_t   nexto;          /* next file offset */
1875         xfs_fileoff_t   prevo;          /* previous file offset */
1876         xfs_fileoff_t   align_off;      /* temp for offset */
1877         xfs_extlen_t    align_alen;     /* temp for length */
1878         xfs_extlen_t    temp;           /* temp for calculations */
1879
1880         if (convert)
1881                 return 0;
1882
1883         orig_off = align_off = *offp;
1884         orig_alen = align_alen = *lenp;
1885         orig_end = orig_off + orig_alen;
1886
1887         /*
1888          * If this request overlaps an existing extent, then don't
1889          * attempt to perform any additional alignment.
1890          */
1891         if (!delay && !eof &&
1892             (orig_off >= gotp->br_startoff) &&
1893             (orig_end <= gotp->br_startoff + gotp->br_blockcount)) {
1894                 return 0;
1895         }
1896
1897         /*
1898          * If the file offset is unaligned vs. the extent size
1899          * we need to align it.  This will be possible unless
1900          * the file was previously written with a kernel that didn't
1901          * perform this alignment, or if a truncate shot us in the
1902          * foot.
1903          */
1904         temp = do_mod(orig_off, extsz);
1905         if (temp) {
1906                 align_alen += temp;
1907                 align_off -= temp;
1908         }
1909         /*
1910          * Same adjustment for the end of the requested area.
1911          */
1912         if ((temp = (align_alen % extsz))) {
1913                 align_alen += extsz - temp;
1914         }
1915         /*
1916          * If the previous block overlaps with this proposed allocation
1917          * then move the start forward without adjusting the length.
1918          */
1919         if (prevp->br_startoff != NULLFILEOFF) {
1920                 if (prevp->br_startblock == HOLESTARTBLOCK)
1921                         prevo = prevp->br_startoff;
1922                 else
1923                         prevo = prevp->br_startoff + prevp->br_blockcount;
1924         } else
1925                 prevo = 0;
1926         if (align_off != orig_off && align_off < prevo)
1927                 align_off = prevo;
1928         /*
1929          * If the next block overlaps with this proposed allocation
1930          * then move the start back without adjusting the length,
1931          * but not before offset 0.
1932          * This may of course make the start overlap previous block,
1933          * and if we hit the offset 0 limit then the next block
1934          * can still overlap too.
1935          */
1936         if (!eof && gotp->br_startoff != NULLFILEOFF) {
1937                 if ((delay && gotp->br_startblock == HOLESTARTBLOCK) ||
1938                     (!delay && gotp->br_startblock == DELAYSTARTBLOCK))
1939                         nexto = gotp->br_startoff + gotp->br_blockcount;
1940                 else
1941                         nexto = gotp->br_startoff;
1942         } else
1943                 nexto = NULLFILEOFF;
1944         if (!eof &&
1945             align_off + align_alen != orig_end &&
1946             align_off + align_alen > nexto)
1947                 align_off = nexto > align_alen ? nexto - align_alen : 0;
1948         /*
1949          * If we're now overlapping the next or previous extent that
1950          * means we can't fit an extsz piece in this hole.  Just move
1951          * the start forward to the first valid spot and set
1952          * the length so we hit the end.
1953          */
1954         if (align_off != orig_off && align_off < prevo)
1955                 align_off = prevo;
1956         if (align_off + align_alen != orig_end &&
1957             align_off + align_alen > nexto &&
1958             nexto != NULLFILEOFF) {
1959                 ASSERT(nexto > prevo);
1960                 align_alen = nexto - align_off;
1961         }
1962
1963         /*
1964          * If realtime, and the result isn't a multiple of the realtime
1965          * extent size we need to remove blocks until it is.
1966          */
1967         if (rt && (temp = (align_alen % mp->m_sb.sb_rextsize))) {
1968                 /*
1969                  * We're not covering the original request, or
1970                  * we won't be able to once we fix the length.
1971                  */
1972                 if (orig_off < align_off ||
1973                     orig_end > align_off + align_alen ||
1974                     align_alen - temp < orig_alen)
1975                         return XFS_ERROR(EINVAL);
1976                 /*
1977                  * Try to fix it by moving the start up.
1978                  */
1979                 if (align_off + temp <= orig_off) {
1980                         align_alen -= temp;
1981                         align_off += temp;
1982                 }
1983                 /*
1984                  * Try to fix it by moving the end in.
1985                  */
1986                 else if (align_off + align_alen - temp >= orig_end)
1987                         align_alen -= temp;
1988                 /*
1989                  * Set the start to the minimum then trim the length.
1990                  */
1991                 else {
1992                         align_alen -= orig_off - align_off;
1993                         align_off = orig_off;
1994                         align_alen -= align_alen % mp->m_sb.sb_rextsize;
1995                 }
1996                 /*
1997                  * Result doesn't cover the request, fail it.
1998                  */
1999                 if (orig_off < align_off || orig_end > align_off + align_alen)
2000                         return XFS_ERROR(EINVAL);
2001         } else {
2002                 ASSERT(orig_off >= align_off);
2003                 ASSERT(orig_end <= align_off + align_alen);
2004         }
2005
2006 #ifdef DEBUG
2007         if (!eof && gotp->br_startoff != NULLFILEOFF)
2008                 ASSERT(align_off + align_alen <= gotp->br_startoff);
2009         if (prevp->br_startoff != NULLFILEOFF)
2010                 ASSERT(align_off >= prevp->br_startoff + prevp->br_blockcount);
2011 #endif
2012
2013         *lenp = align_alen;
2014         *offp = align_off;
2015         return 0;
2016 }
2017
2018 #define XFS_ALLOC_GAP_UNITS     4
2019
2020 STATIC void
2021 xfs_bmap_adjacent(
2022         xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
2023 {
2024         xfs_fsblock_t   adjust;         /* adjustment to block numbers */
2025         xfs_agnumber_t  fb_agno;        /* ag number of ap->firstblock */
2026         xfs_mount_t     *mp;            /* mount point structure */
2027         int             nullfb;         /* true if ap->firstblock isn't set */
2028         int             rt;             /* true if inode is realtime */
2029
2030 #define ISVALID(x,y)    \
2031         (rt ? \
2032                 (x) < mp->m_sb.sb_rblocks : \
2033                 XFS_FSB_TO_AGNO(mp, x) == XFS_FSB_TO_AGNO(mp, y) && \
2034                 XFS_FSB_TO_AGNO(mp, x) < mp->m_sb.sb_agcount && \
2035                 XFS_FSB_TO_AGBNO(mp, x) < mp->m_sb.sb_agblocks)
2036
2037         mp = ap->ip->i_mount;
2038         nullfb = *ap->firstblock == NULLFSBLOCK;
2039         rt = XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata;
2040         fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, *ap->firstblock);
2041         /*
2042          * If allocating at eof, and there's a previous real block,
2043          * try to use its last block as our starting point.
2044          */
2045         if (ap->eof && ap->prev.br_startoff != NULLFILEOFF &&
2046             !isnullstartblock(ap->prev.br_startblock) &&
2047             ISVALID(ap->prev.br_startblock + ap->prev.br_blockcount,
2048                     ap->prev.br_startblock)) {
2049                 ap->blkno = ap->prev.br_startblock + ap->prev.br_blockcount;
2050                 /*
2051                  * Adjust for the gap between prevp and us.
2052                  */
2053                 adjust = ap->offset -
2054                         (ap->prev.br_startoff + ap->prev.br_blockcount);
2055                 if (adjust &&
2056                     ISVALID(ap->blkno + adjust, ap->prev.br_startblock))
2057                         ap->blkno += adjust;
2058         }
2059         /*
2060          * If not at eof, then compare the two neighbor blocks.
2061          * Figure out whether either one gives us a good starting point,
2062          * and pick the better one.
2063          */
2064         else if (!ap->eof) {
2065                 xfs_fsblock_t   gotbno;         /* right side block number */
2066                 xfs_fsblock_t   gotdiff=0;      /* right side difference */
2067                 xfs_fsblock_t   prevbno;        /* left side block number */
2068                 xfs_fsblock_t   prevdiff=0;     /* left side difference */
2069
2070                 /*
2071                  * If there's a previous (left) block, select a requested
2072                  * start block based on it.
2073                  */
2074                 if (ap->prev.br_startoff != NULLFILEOFF &&
2075                     !isnullstartblock(ap->prev.br_startblock) &&
2076                     (prevbno = ap->prev.br_startblock +
2077                                ap->prev.br_blockcount) &&
2078                     ISVALID(prevbno, ap->prev.br_startblock)) {
2079                         /*
2080                          * Calculate gap to end of previous block.
2081                          */
2082                         adjust = prevdiff = ap->offset -
2083                                 (ap->prev.br_startoff +
2084                                  ap->prev.br_blockcount);
2085                         /*
2086                          * Figure the startblock based on the previous block's
2087                          * end and the gap size.
2088                          * Heuristic!
2089                          * If the gap is large relative to the piece we're
2090                          * allocating, or using it gives us an invalid block
2091                          * number, then just use the end of the previous block.
2092                          */
2093                         if (prevdiff <= XFS_ALLOC_GAP_UNITS * ap->length &&
2094                             ISVALID(prevbno + prevdiff,
2095                                     ap->prev.br_startblock))
2096                                 prevbno += adjust;
2097                         else
2098                                 prevdiff += adjust;
2099                         /*
2100                          * If the firstblock forbids it, can't use it,
2101                          * must use default.
2102                          */
2103                         if (!rt && !nullfb &&
2104                             XFS_FSB_TO_AGNO(mp, prevbno) != fb_agno)
2105                                 prevbno = NULLFSBLOCK;
2106                 }
2107                 /*
2108                  * No previous block or can't follow it, just default.
2109                  */
2110                 else
2111                         prevbno = NULLFSBLOCK;
2112                 /*
2113                  * If there's a following (right) block, select a requested
2114                  * start block based on it.
2115                  */
2116                 if (!isnullstartblock(ap->got.br_startblock)) {
2117                         /*
2118                          * Calculate gap to start of next block.
2119                          */
2120                         adjust = gotdiff = ap->got.br_startoff - ap->offset;
2121                         /*
2122                          * Figure the startblock based on the next block's
2123                          * start and the gap size.
2124                          */
2125                         gotbno = ap->got.br_startblock;
2126                         /*
2127                          * Heuristic!
2128                          * If the gap is large relative to the piece we're
2129                          * allocating, or using it gives us an invalid block
2130                          * number, then just use the start of the next block
2131                          * offset by our length.
2132                          */
2133                         if (gotdiff <= XFS_ALLOC_GAP_UNITS * ap->length &&
2134                             ISVALID(gotbno - gotdiff, gotbno))
2135                                 gotbno -= adjust;
2136                         else if (ISVALID(gotbno - ap->length, gotbno)) {
2137                                 gotbno -= ap->length;
2138                                 gotdiff += adjust - ap->length;
2139                         } else
2140                                 gotdiff += adjust;
2141                         /*
2142                          * If the firstblock forbids it, can't use it,
2143                          * must use default.
2144                          */
2145                         if (!rt && !nullfb &&
2146                             XFS_FSB_TO_AGNO(mp, gotbno) != fb_agno)
2147                                 gotbno = NULLFSBLOCK;
2148                 }
2149                 /*
2150                  * No next block, just default.
2151                  */
2152                 else
2153                         gotbno = NULLFSBLOCK;
2154                 /*
2155                  * If both valid, pick the better one, else the only good
2156                  * one, else ap->blkno is already set (to 0 or the inode block).
2157                  */
2158                 if (prevbno != NULLFSBLOCK && gotbno != NULLFSBLOCK)
2159                         ap->blkno = prevdiff <= gotdiff ? prevbno : gotbno;
2160                 else if (prevbno != NULLFSBLOCK)
2161                         ap->blkno = prevbno;
2162                 else if (gotbno != NULLFSBLOCK)
2163                         ap->blkno = gotbno;
2164         }
2165 #undef ISVALID
2166 }
2167
2168 STATIC int
2169 xfs_bmap_rtalloc(
2170         xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
2171 {
2172         xfs_alloctype_t atype = 0;      /* type for allocation routines */
2173         int             error;          /* error return value */
2174         xfs_mount_t     *mp;            /* mount point structure */
2175         xfs_extlen_t    prod = 0;       /* product factor for allocators */
2176         xfs_extlen_t    ralen = 0;      /* realtime allocation length */
2177         xfs_extlen_t    align;          /* minimum allocation alignment */
2178         xfs_rtblock_t   rtb;
2179
2180         mp = ap->ip->i_mount;
2181         align = xfs_get_extsz_hint(ap->ip);
2182         prod = align / mp->m_sb.sb_rextsize;
2183         error = xfs_bmap_extsize_align(mp, &ap->got, &ap->prev,
2184                                         align, 1, ap->eof, 0,
2185                                         ap->conv, &ap->offset, &ap->length);
2186         if (error)
2187                 return error;
2188         ASSERT(ap->length);
2189         ASSERT(ap->length % mp->m_sb.sb_rextsize == 0);
2190
2191         /*
2192          * If the offset & length are not perfectly aligned
2193          * then kill prod, it will just get us in trouble.
2194          */
2195         if (do_mod(ap->offset, align) || ap->length % align)
2196                 prod = 1;
2197         /*
2198          * Set ralen to be the actual requested length in rtextents.
2199          */
2200         ralen = ap->length / mp->m_sb.sb_rextsize;
2201         /*
2202          * If the old value was close enough to MAXEXTLEN that
2203          * we rounded up to it, cut it back so it's valid again.
2204          * Note that if it's a really large request (bigger than
2205          * MAXEXTLEN), we don't hear about that number, and can't
2206          * adjust the starting point to match it.
2207          */
2208         if (ralen * mp->m_sb.sb_rextsize >= MAXEXTLEN)
2209                 ralen = MAXEXTLEN / mp->m_sb.sb_rextsize;
2210
2211         /*
2212          * Lock out other modifications to the RT bitmap inode.
2213          */
2214         xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL);
2215         xfs_trans_ijoin(ap->tp, mp->m_rbmip, XFS_ILOCK_EXCL);
2216
2217         /*
2218          * If it's an allocation to an empty file at offset 0,
2219          * pick an extent that will space things out in the rt area.
2220          */
2221         if (ap->eof && ap->offset == 0) {
2222                 xfs_rtblock_t uninitialized_var(rtx); /* realtime extent no */
2223
2224                 error = xfs_rtpick_extent(mp, ap->tp, ralen, &rtx);
2225                 if (error)
2226                         return error;
2227                 ap->blkno = rtx * mp->m_sb.sb_rextsize;
2228         } else {
2229                 ap->blkno = 0;
2230         }
2231
2232         xfs_bmap_adjacent(ap);
2233
2234         /*
2235          * Realtime allocation, done through xfs_rtallocate_extent.
2236          */
2237         atype = ap->blkno == 0 ?  XFS_ALLOCTYPE_ANY_AG : XFS_ALLOCTYPE_NEAR_BNO;
2238         do_div(ap->blkno, mp->m_sb.sb_rextsize);
2239         rtb = ap->blkno;
2240         ap->length = ralen;
2241         if ((error = xfs_rtallocate_extent(ap->tp, ap->blkno, 1, ap->length,
2242                                 &ralen, atype, ap->wasdel, prod, &rtb)))
2243                 return error;
2244         if (rtb == NULLFSBLOCK && prod > 1 &&
2245             (error = xfs_rtallocate_extent(ap->tp, ap->blkno, 1,
2246                                            ap->length, &ralen, atype,
2247                                            ap->wasdel, 1, &rtb)))
2248                 return error;
2249         ap->blkno = rtb;
2250         if (ap->blkno != NULLFSBLOCK) {
2251                 ap->blkno *= mp->m_sb.sb_rextsize;
2252                 ralen *= mp->m_sb.sb_rextsize;
2253                 ap->length = ralen;
2254                 ap->ip->i_d.di_nblocks += ralen;
2255                 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
2256                 if (ap->wasdel)
2257                         ap->ip->i_delayed_blks -= ralen;
2258                 /*
2259                  * Adjust the disk quota also. This was reserved
2260                  * earlier.
2261                  */
2262                 xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
2263                         ap->wasdel ? XFS_TRANS_DQ_DELRTBCOUNT :
2264                                         XFS_TRANS_DQ_RTBCOUNT, (long) ralen);
2265         } else {
2266                 ap->length = 0;
2267         }
2268         return 0;
2269 }
2270
2271 STATIC int
2272 xfs_bmap_btalloc_nullfb(
2273         struct xfs_bmalloca     *ap,
2274         struct xfs_alloc_arg    *args,
2275         xfs_extlen_t            *blen)
2276 {
2277         struct xfs_mount        *mp = ap->ip->i_mount;
2278         struct xfs_perag        *pag;
2279         xfs_agnumber_t          ag, startag;
2280         int                     notinit = 0;
2281         int                     error;
2282
2283         if (ap->userdata && xfs_inode_is_filestream(ap->ip))
2284                 args->type = XFS_ALLOCTYPE_NEAR_BNO;
2285         else
2286                 args->type = XFS_ALLOCTYPE_START_BNO;
2287         args->total = ap->total;
2288
2289         /*
2290          * Search for an allocation group with a single extent large enough
2291          * for the request.  If one isn't found, then adjust the minimum
2292          * allocation size to the largest space found.
2293          */
2294         startag = ag = XFS_FSB_TO_AGNO(mp, args->fsbno);
2295         if (startag == NULLAGNUMBER)
2296                 startag = ag = 0;
2297
2298         pag = xfs_perag_get(mp, ag);
2299         while (*blen < args->maxlen) {
2300                 if (!pag->pagf_init) {
2301                         error = xfs_alloc_pagf_init(mp, args->tp, ag,
2302                                                     XFS_ALLOC_FLAG_TRYLOCK);
2303                         if (error) {
2304                                 xfs_perag_put(pag);
2305                                 return error;
2306                         }
2307                 }
2308
2309                 /*
2310                  * See xfs_alloc_fix_freelist...
2311                  */
2312                 if (pag->pagf_init) {
2313                         xfs_extlen_t    longest;
2314                         longest = xfs_alloc_longest_free_extent(mp, pag);
2315                         if (*blen < longest)
2316                                 *blen = longest;
2317                 } else
2318                         notinit = 1;
2319
2320                 if (xfs_inode_is_filestream(ap->ip)) {
2321                         if (*blen >= args->maxlen)
2322                                 break;
2323
2324                         if (ap->userdata) {
2325                                 /*
2326                                  * If startag is an invalid AG, we've
2327                                  * come here once before and
2328                                  * xfs_filestream_new_ag picked the
2329                                  * best currently available.
2330                                  *
2331                                  * Don't continue looping, since we
2332                                  * could loop forever.
2333                                  */
2334                                 if (startag == NULLAGNUMBER)
2335                                         break;
2336
2337                                 error = xfs_filestream_new_ag(ap, &ag);
2338                                 xfs_perag_put(pag);
2339                                 if (error)
2340                                         return error;
2341
2342                                 /* loop again to set 'blen'*/
2343                                 startag = NULLAGNUMBER;
2344                                 pag = xfs_perag_get(mp, ag);
2345                                 continue;
2346                         }
2347                 }
2348                 if (++ag == mp->m_sb.sb_agcount)
2349                         ag = 0;
2350                 if (ag == startag)
2351                         break;
2352                 xfs_perag_put(pag);
2353                 pag = xfs_perag_get(mp, ag);
2354         }
2355         xfs_perag_put(pag);
2356
2357         /*
2358          * Since the above loop did a BUF_TRYLOCK, it is
2359          * possible that there is space for this request.
2360          */
2361         if (notinit || *blen < ap->minlen)
2362                 args->minlen = ap->minlen;
2363         /*
2364          * If the best seen length is less than the request
2365          * length, use the best as the minimum.
2366          */
2367         else if (*blen < args->maxlen)
2368                 args->minlen = *blen;
2369         /*
2370          * Otherwise we've seen an extent as big as maxlen,
2371          * use that as the minimum.
2372          */
2373         else
2374                 args->minlen = args->maxlen;
2375
2376         /*
2377          * set the failure fallback case to look in the selected
2378          * AG as the stream may have moved.
2379          */
2380         if (xfs_inode_is_filestream(ap->ip))
2381                 ap->blkno = args->fsbno = XFS_AGB_TO_FSB(mp, ag, 0);
2382
2383         return 0;
2384 }
2385
2386 STATIC int
2387 xfs_bmap_btalloc(
2388         xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
2389 {
2390         xfs_mount_t     *mp;            /* mount point structure */
2391         xfs_alloctype_t atype = 0;      /* type for allocation routines */
2392         xfs_extlen_t    align;          /* minimum allocation alignment */
2393         xfs_agnumber_t  fb_agno;        /* ag number of ap->firstblock */
2394         xfs_agnumber_t  ag;
2395         xfs_alloc_arg_t args;
2396         xfs_extlen_t    blen;
2397         xfs_extlen_t    nextminlen = 0;
2398         int             nullfb;         /* true if ap->firstblock isn't set */
2399         int             isaligned;
2400         int             tryagain;
2401         int             error;
2402
2403         ASSERT(ap->length);
2404
2405         mp = ap->ip->i_mount;
2406         align = ap->userdata ? xfs_get_extsz_hint(ap->ip) : 0;
2407         if (unlikely(align)) {
2408                 error = xfs_bmap_extsize_align(mp, &ap->got, &ap->prev,
2409                                                 align, 0, ap->eof, 0, ap->conv,
2410                                                 &ap->offset, &ap->length);
2411                 ASSERT(!error);
2412                 ASSERT(ap->length);
2413         }
2414         nullfb = *ap->firstblock == NULLFSBLOCK;
2415         fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, *ap->firstblock);
2416         if (nullfb) {
2417                 if (ap->userdata && xfs_inode_is_filestream(ap->ip)) {
2418                         ag = xfs_filestream_lookup_ag(ap->ip);
2419                         ag = (ag != NULLAGNUMBER) ? ag : 0;
2420                         ap->blkno = XFS_AGB_TO_FSB(mp, ag, 0);
2421                 } else {
2422                         ap->blkno = XFS_INO_TO_FSB(mp, ap->ip->i_ino);
2423                 }
2424         } else
2425                 ap->blkno = *ap->firstblock;
2426
2427         xfs_bmap_adjacent(ap);
2428
2429         /*
2430          * If allowed, use ap->blkno; otherwise must use firstblock since
2431          * it's in the right allocation group.
2432          */
2433         if (nullfb || XFS_FSB_TO_AGNO(mp, ap->blkno) == fb_agno)
2434                 ;
2435         else
2436                 ap->blkno = *ap->firstblock;
2437         /*
2438          * Normal allocation, done through xfs_alloc_vextent.
2439          */
2440         tryagain = isaligned = 0;
2441         args.tp = ap->tp;
2442         args.mp = mp;
2443         args.fsbno = ap->blkno;
2444
2445         /* Trim the allocation back to the maximum an AG can fit. */
2446         args.maxlen = MIN(ap->length, XFS_ALLOC_AG_MAX_USABLE(mp));
2447         args.firstblock = *ap->firstblock;
2448         blen = 0;
2449         if (nullfb) {
2450                 error = xfs_bmap_btalloc_nullfb(ap, &args, &blen);
2451                 if (error)
2452                         return error;
2453         } else if (ap->flist->xbf_low) {
2454                 if (xfs_inode_is_filestream(ap->ip))
2455                         args.type = XFS_ALLOCTYPE_FIRST_AG;
2456                 else
2457                         args.type = XFS_ALLOCTYPE_START_BNO;
2458                 args.total = args.minlen = ap->minlen;
2459         } else {
2460                 args.type = XFS_ALLOCTYPE_NEAR_BNO;
2461                 args.total = ap->total;
2462                 args.minlen = ap->minlen;
2463         }
2464         /* apply extent size hints if obtained earlier */
2465         if (unlikely(align)) {
2466                 args.prod = align;
2467                 if ((args.mod = (xfs_extlen_t)do_mod(ap->offset, args.prod)))
2468                         args.mod = (xfs_extlen_t)(args.prod - args.mod);
2469         } else if (mp->m_sb.sb_blocksize >= PAGE_CACHE_SIZE) {
2470                 args.prod = 1;
2471                 args.mod = 0;
2472         } else {
2473                 args.prod = PAGE_CACHE_SIZE >> mp->m_sb.sb_blocklog;
2474                 if ((args.mod = (xfs_extlen_t)(do_mod(ap->offset, args.prod))))
2475                         args.mod = (xfs_extlen_t)(args.prod - args.mod);
2476         }
2477         /*
2478          * If we are not low on available data blocks, and the
2479          * underlying logical volume manager is a stripe, and
2480          * the file offset is zero then try to allocate data
2481          * blocks on stripe unit boundary.
2482          * NOTE: ap->aeof is only set if the allocation length
2483          * is >= the stripe unit and the allocation offset is
2484          * at the end of file.
2485          */
2486         if (!ap->flist->xbf_low && ap->aeof) {
2487                 if (!ap->offset) {
2488                         args.alignment = mp->m_dalign;
2489                         atype = args.type;
2490                         isaligned = 1;
2491                         /*
2492                          * Adjust for alignment
2493                          */
2494                         if (blen > args.alignment && blen <= args.maxlen)
2495                                 args.minlen = blen - args.alignment;
2496                         args.minalignslop = 0;
2497                 } else {
2498                         /*
2499                          * First try an exact bno allocation.
2500                          * If it fails then do a near or start bno
2501                          * allocation with alignment turned on.
2502                          */
2503                         atype = args.type;
2504                         tryagain = 1;
2505                         args.type = XFS_ALLOCTYPE_THIS_BNO;
2506                         args.alignment = 1;
2507                         /*
2508                          * Compute the minlen+alignment for the
2509                          * next case.  Set slop so that the value
2510                          * of minlen+alignment+slop doesn't go up
2511                          * between the calls.
2512                          */
2513                         if (blen > mp->m_dalign && blen <= args.maxlen)
2514                                 nextminlen = blen - mp->m_dalign;
2515                         else
2516                                 nextminlen = args.minlen;
2517                         if (nextminlen + mp->m_dalign > args.minlen + 1)
2518                                 args.minalignslop =
2519                                         nextminlen + mp->m_dalign -
2520                                         args.minlen - 1;
2521                         else
2522                                 args.minalignslop = 0;
2523                 }
2524         } else {
2525                 args.alignment = 1;
2526                 args.minalignslop = 0;
2527         }
2528         args.minleft = ap->minleft;
2529         args.wasdel = ap->wasdel;
2530         args.isfl = 0;
2531         args.userdata = ap->userdata;
2532         if ((error = xfs_alloc_vextent(&args)))
2533                 return error;
2534         if (tryagain && args.fsbno == NULLFSBLOCK) {
2535                 /*
2536                  * Exact allocation failed. Now try with alignment
2537                  * turned on.
2538                  */
2539                 args.type = atype;
2540                 args.fsbno = ap->blkno;
2541                 args.alignment = mp->m_dalign;
2542                 args.minlen = nextminlen;
2543                 args.minalignslop = 0;
2544                 isaligned = 1;
2545                 if ((error = xfs_alloc_vextent(&args)))
2546                         return error;
2547         }
2548         if (isaligned && args.fsbno == NULLFSBLOCK) {
2549                 /*
2550                  * allocation failed, so turn off alignment and
2551                  * try again.
2552                  */
2553                 args.type = atype;
2554                 args.fsbno = ap->blkno;
2555                 args.alignment = 0;
2556                 if ((error = xfs_alloc_vextent(&args)))
2557                         return error;
2558         }
2559         if (args.fsbno == NULLFSBLOCK && nullfb &&
2560             args.minlen > ap->minlen) {
2561                 args.minlen = ap->minlen;
2562                 args.type = XFS_ALLOCTYPE_START_BNO;
2563                 args.fsbno = ap->blkno;
2564                 if ((error = xfs_alloc_vextent(&args)))
2565                         return error;
2566         }
2567         if (args.fsbno == NULLFSBLOCK && nullfb) {
2568                 args.fsbno = 0;
2569                 args.type = XFS_ALLOCTYPE_FIRST_AG;
2570                 args.total = ap->minlen;
2571                 args.minleft = 0;
2572                 if ((error = xfs_alloc_vextent(&args)))
2573                         return error;
2574                 ap->flist->xbf_low = 1;
2575         }
2576         if (args.fsbno != NULLFSBLOCK) {
2577                 /*
2578                  * check the allocation happened at the same or higher AG than
2579                  * the first block that was allocated.
2580                  */
2581                 ASSERT(*ap->firstblock == NULLFSBLOCK ||
2582                        XFS_FSB_TO_AGNO(mp, *ap->firstblock) ==
2583                        XFS_FSB_TO_AGNO(mp, args.fsbno) ||
2584                        (ap->flist->xbf_low &&
2585                         XFS_FSB_TO_AGNO(mp, *ap->firstblock) <
2586                         XFS_FSB_TO_AGNO(mp, args.fsbno)));
2587
2588                 ap->blkno = args.fsbno;
2589                 if (*ap->firstblock == NULLFSBLOCK)
2590                         *ap->firstblock = args.fsbno;
2591                 ASSERT(nullfb || fb_agno == args.agno ||
2592                        (ap->flist->xbf_low && fb_agno < args.agno));
2593                 ap->length = args.len;
2594                 ap->ip->i_d.di_nblocks += args.len;
2595                 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
2596                 if (ap->wasdel)
2597                         ap->ip->i_delayed_blks -= args.len;
2598                 /*
2599                  * Adjust the disk quota also. This was reserved
2600                  * earlier.
2601                  */
2602                 xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
2603                         ap->wasdel ? XFS_TRANS_DQ_DELBCOUNT :
2604                                         XFS_TRANS_DQ_BCOUNT,
2605                         (long) args.len);
2606         } else {
2607                 ap->blkno = NULLFSBLOCK;
2608                 ap->length = 0;
2609         }
2610         return 0;
2611 }
2612
2613 /*
2614  * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
2615  * It figures out where to ask the underlying allocator to put the new extent.
2616  */
2617 STATIC int
2618 xfs_bmap_alloc(
2619         xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
2620 {
2621         if (XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata)
2622                 return xfs_bmap_rtalloc(ap);
2623         return xfs_bmap_btalloc(ap);
2624 }
2625
2626 /*
2627  * Transform a btree format file with only one leaf node, where the
2628  * extents list will fit in the inode, into an extents format file.
2629  * Since the file extents are already in-core, all we have to do is
2630  * give up the space for the btree root and pitch the leaf block.
2631  */
2632 STATIC int                              /* error */
2633 xfs_bmap_btree_to_extents(
2634         xfs_trans_t             *tp,    /* transaction pointer */
2635         xfs_inode_t             *ip,    /* incore inode pointer */
2636         xfs_btree_cur_t         *cur,   /* btree cursor */
2637         int                     *logflagsp, /* inode logging flags */
2638         int                     whichfork)  /* data or attr fork */
2639 {
2640         /* REFERENCED */
2641         struct xfs_btree_block  *cblock;/* child btree block */
2642         xfs_fsblock_t           cbno;   /* child block number */
2643         xfs_buf_t               *cbp;   /* child block's buffer */
2644         int                     error;  /* error return value */
2645         xfs_ifork_t             *ifp;   /* inode fork data */
2646         xfs_mount_t             *mp;    /* mount point structure */
2647         __be64                  *pp;    /* ptr to block address */
2648         struct xfs_btree_block  *rblock;/* root btree block */
2649
2650         mp = ip->i_mount;
2651         ifp = XFS_IFORK_PTR(ip, whichfork);
2652         ASSERT(ifp->if_flags & XFS_IFEXTENTS);
2653         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
2654         rblock = ifp->if_broot;
2655         ASSERT(be16_to_cpu(rblock->bb_level) == 1);
2656         ASSERT(be16_to_cpu(rblock->bb_numrecs) == 1);
2657         ASSERT(xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0) == 1);
2658         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, rblock, 1, ifp->if_broot_bytes);
2659         cbno = be64_to_cpu(*pp);
2660         *logflagsp = 0;
2661 #ifdef DEBUG
2662         if ((error = xfs_btree_check_lptr(cur, cbno, 1)))
2663                 return error;
2664 #endif
2665         if ((error = xfs_btree_read_bufl(mp, tp, cbno, 0, &cbp,
2666                         XFS_BMAP_BTREE_REF)))
2667                 return error;
2668         cblock = XFS_BUF_TO_BLOCK(cbp);
2669         if ((error = xfs_btree_check_block(cur, cblock, 0, cbp)))
2670                 return error;
2671         xfs_bmap_add_free(cbno, 1, cur->bc_private.b.flist, mp);
2672         ip->i_d.di_nblocks--;
2673         xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, -1L);
2674         xfs_trans_binval(tp, cbp);
2675         if (cur->bc_bufs[0] == cbp)
2676                 cur->bc_bufs[0] = NULL;
2677         xfs_iroot_realloc(ip, -1, whichfork);
2678         ASSERT(ifp->if_broot == NULL);
2679         ASSERT((ifp->if_flags & XFS_IFBROOT) == 0);
2680         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
2681         *logflagsp = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
2682         return 0;
2683 }
2684
2685 /*
2686  * Called by xfs_bmapi to update file extent records and the btree
2687  * after removing space (or undoing a delayed allocation).
2688  */
2689 STATIC int                              /* error */
2690 xfs_bmap_del_extent(
2691         xfs_inode_t             *ip,    /* incore inode pointer */
2692         xfs_trans_t             *tp,    /* current transaction pointer */
2693         xfs_extnum_t            *idx,   /* extent number to update/delete */
2694         xfs_bmap_free_t         *flist, /* list of extents to be freed */
2695         xfs_btree_cur_t         *cur,   /* if null, not a btree */
2696         xfs_bmbt_irec_t         *del,   /* data to remove from extents */
2697         int                     *logflagsp, /* inode logging flags */
2698         int                     whichfork) /* data or attr fork */
2699 {
2700         xfs_filblks_t           da_new; /* new delay-alloc indirect blocks */
2701         xfs_filblks_t           da_old; /* old delay-alloc indirect blocks */
2702         xfs_fsblock_t           del_endblock=0; /* first block past del */
2703         xfs_fileoff_t           del_endoff;     /* first offset past del */
2704         int                     delay;  /* current block is delayed allocated */
2705         int                     do_fx;  /* free extent at end of routine */
2706         xfs_bmbt_rec_host_t     *ep;    /* current extent entry pointer */
2707         int                     error;  /* error return value */
2708         int                     flags;  /* inode logging flags */
2709         xfs_bmbt_irec_t         got;    /* current extent entry */
2710         xfs_fileoff_t           got_endoff;     /* first offset past got */
2711         int                     i;      /* temp state */
2712         xfs_ifork_t             *ifp;   /* inode fork pointer */
2713         xfs_mount_t             *mp;    /* mount structure */
2714         xfs_filblks_t           nblks;  /* quota/sb block count */
2715         xfs_bmbt_irec_t         new;    /* new record to be inserted */
2716         /* REFERENCED */
2717         uint                    qfield; /* quota field to update */
2718         xfs_filblks_t           temp;   /* for indirect length calculations */
2719         xfs_filblks_t           temp2;  /* for indirect length calculations */
2720         int                     state = 0;
2721
2722         XFS_STATS_INC(xs_del_exlist);
2723
2724         if (whichfork == XFS_ATTR_FORK)
2725                 state |= BMAP_ATTRFORK;
2726
2727         mp = ip->i_mount;
2728         ifp = XFS_IFORK_PTR(ip, whichfork);
2729         ASSERT((*idx >= 0) && (*idx < ifp->if_bytes /
2730                 (uint)sizeof(xfs_bmbt_rec_t)));
2731         ASSERT(del->br_blockcount > 0);
2732         ep = xfs_iext_get_ext(ifp, *idx);
2733         xfs_bmbt_get_all(ep, &got);
2734         ASSERT(got.br_startoff <= del->br_startoff);
2735         del_endoff = del->br_startoff + del->br_blockcount;
2736         got_endoff = got.br_startoff + got.br_blockcount;
2737         ASSERT(got_endoff >= del_endoff);
2738         delay = isnullstartblock(got.br_startblock);
2739         ASSERT(isnullstartblock(del->br_startblock) == delay);
2740         flags = 0;
2741         qfield = 0;
2742         error = 0;
2743         /*
2744          * If deleting a real allocation, must free up the disk space.
2745          */
2746         if (!delay) {
2747                 flags = XFS_ILOG_CORE;
2748                 /*
2749                  * Realtime allocation.  Free it and record di_nblocks update.
2750                  */
2751                 if (whichfork == XFS_DATA_FORK && XFS_IS_REALTIME_INODE(ip)) {
2752                         xfs_fsblock_t   bno;
2753                         xfs_filblks_t   len;
2754
2755                         ASSERT(do_mod(del->br_blockcount,
2756                                       mp->m_sb.sb_rextsize) == 0);
2757                         ASSERT(do_mod(del->br_startblock,
2758                                       mp->m_sb.sb_rextsize) == 0);
2759                         bno = del->br_startblock;
2760                         len = del->br_blockcount;
2761                         do_div(bno, mp->m_sb.sb_rextsize);
2762                         do_div(len, mp->m_sb.sb_rextsize);
2763                         error = xfs_rtfree_extent(tp, bno, (xfs_extlen_t)len);
2764                         if (error)
2765                                 goto done;
2766                         do_fx = 0;
2767                         nblks = len * mp->m_sb.sb_rextsize;
2768                         qfield = XFS_TRANS_DQ_RTBCOUNT;
2769                 }
2770                 /*
2771                  * Ordinary allocation.
2772                  */
2773                 else {
2774                         do_fx = 1;
2775                         nblks = del->br_blockcount;
2776                         qfield = XFS_TRANS_DQ_BCOUNT;
2777                 }
2778                 /*
2779                  * Set up del_endblock and cur for later.
2780                  */
2781                 del_endblock = del->br_startblock + del->br_blockcount;
2782                 if (cur) {
2783                         if ((error = xfs_bmbt_lookup_eq(cur, got.br_startoff,
2784                                         got.br_startblock, got.br_blockcount,
2785                                         &i)))
2786                                 goto done;
2787                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2788                 }
2789                 da_old = da_new = 0;
2790         } else {
2791                 da_old = startblockval(got.br_startblock);
2792                 da_new = 0;
2793                 nblks = 0;
2794                 do_fx = 0;
2795         }
2796         /*
2797          * Set flag value to use in switch statement.
2798          * Left-contig is 2, right-contig is 1.
2799          */
2800         switch (((got.br_startoff == del->br_startoff) << 1) |
2801                 (got_endoff == del_endoff)) {
2802         case 3:
2803                 /*
2804                  * Matches the whole extent.  Delete the entry.
2805                  */
2806                 xfs_iext_remove(ip, *idx, 1,
2807                                 whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0);
2808                 --*idx;
2809                 if (delay)
2810                         break;
2811
2812                 XFS_IFORK_NEXT_SET(ip, whichfork,
2813                         XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
2814                 flags |= XFS_ILOG_CORE;
2815                 if (!cur) {
2816                         flags |= xfs_ilog_fext(whichfork);
2817                         break;
2818                 }
2819                 if ((error = xfs_btree_delete(cur, &i)))
2820                         goto done;
2821                 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2822                 break;
2823
2824         case 2:
2825                 /*
2826                  * Deleting the first part of the extent.
2827                  */
2828                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2829                 xfs_bmbt_set_startoff(ep, del_endoff);
2830                 temp = got.br_blockcount - del->br_blockcount;
2831                 xfs_bmbt_set_blockcount(ep, temp);
2832                 if (delay) {
2833                         temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2834                                 da_old);
2835                         xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
2836                         trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2837                         da_new = temp;
2838                         break;
2839                 }
2840                 xfs_bmbt_set_startblock(ep, del_endblock);
2841                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2842                 if (!cur) {
2843                         flags |= xfs_ilog_fext(whichfork);
2844                         break;
2845                 }
2846                 if ((error = xfs_bmbt_update(cur, del_endoff, del_endblock,
2847                                 got.br_blockcount - del->br_blockcount,
2848                                 got.br_state)))
2849                         goto done;
2850                 break;
2851
2852         case 1:
2853                 /*
2854                  * Deleting the last part of the extent.
2855                  */
2856                 temp = got.br_blockcount - del->br_blockcount;
2857                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2858                 xfs_bmbt_set_blockcount(ep, temp);
2859                 if (delay) {
2860                         temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2861                                 da_old);
2862                         xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
2863                         trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2864                         da_new = temp;
2865                         break;
2866                 }
2867                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2868                 if (!cur) {
2869                         flags |= xfs_ilog_fext(whichfork);
2870                         break;
2871                 }
2872                 if ((error = xfs_bmbt_update(cur, got.br_startoff,
2873                                 got.br_startblock,
2874                                 got.br_blockcount - del->br_blockcount,
2875                                 got.br_state)))
2876                         goto done;
2877                 break;
2878
2879         case 0:
2880                 /*
2881                  * Deleting the middle of the extent.
2882                  */
2883                 temp = del->br_startoff - got.br_startoff;
2884                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2885                 xfs_bmbt_set_blockcount(ep, temp);
2886                 new.br_startoff = del_endoff;
2887                 temp2 = got_endoff - del_endoff;
2888                 new.br_blockcount = temp2;
2889                 new.br_state = got.br_state;
2890                 if (!delay) {
2891                         new.br_startblock = del_endblock;
2892                         flags |= XFS_ILOG_CORE;
2893                         if (cur) {
2894                                 if ((error = xfs_bmbt_update(cur,
2895                                                 got.br_startoff,
2896                                                 got.br_startblock, temp,
2897                                                 got.br_state)))
2898                                         goto done;
2899                                 if ((error = xfs_btree_increment(cur, 0, &i)))
2900                                         goto done;
2901                                 cur->bc_rec.b = new;
2902                                 error = xfs_btree_insert(cur, &i);
2903                                 if (error && error != ENOSPC)
2904                                         goto done;
2905                                 /*
2906                                  * If get no-space back from btree insert,
2907                                  * it tried a split, and we have a zero
2908                                  * block reservation.
2909                                  * Fix up our state and return the error.
2910                                  */
2911                                 if (error == ENOSPC) {
2912                                         /*
2913                                          * Reset the cursor, don't trust
2914                                          * it after any insert operation.
2915                                          */
2916                                         if ((error = xfs_bmbt_lookup_eq(cur,
2917                                                         got.br_startoff,
2918                                                         got.br_startblock,
2919                                                         temp, &i)))
2920                                                 goto done;
2921                                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2922                                         /*
2923                                          * Update the btree record back
2924                                          * to the original value.
2925                                          */
2926                                         if ((error = xfs_bmbt_update(cur,
2927                                                         got.br_startoff,
2928                                                         got.br_startblock,
2929                                                         got.br_blockcount,
2930                                                         got.br_state)))
2931                                                 goto done;
2932                                         /*
2933                                          * Reset the extent record back
2934                                          * to the original value.
2935                                          */
2936                                         xfs_bmbt_set_blockcount(ep,
2937                                                 got.br_blockcount);
2938                                         flags = 0;
2939                                         error = XFS_ERROR(ENOSPC);
2940                                         goto done;
2941                                 }
2942                                 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2943                         } else
2944                                 flags |= xfs_ilog_fext(whichfork);
2945                         XFS_IFORK_NEXT_SET(ip, whichfork,
2946                                 XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
2947                 } else {
2948                         ASSERT(whichfork == XFS_DATA_FORK);
2949                         temp = xfs_bmap_worst_indlen(ip, temp);
2950                         xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
2951                         temp2 = xfs_bmap_worst_indlen(ip, temp2);
2952                         new.br_startblock = nullstartblock((int)temp2);
2953                         da_new = temp + temp2;
2954                         while (da_new > da_old) {
2955                                 if (temp) {
2956                                         temp--;
2957                                         da_new--;
2958                                         xfs_bmbt_set_startblock(ep,
2959                                                 nullstartblock((int)temp));
2960                                 }
2961                                 if (da_new == da_old)
2962                                         break;
2963                                 if (temp2) {
2964                                         temp2--;
2965                                         da_new--;
2966                                         new.br_startblock =
2967                                                 nullstartblock((int)temp2);
2968                                 }
2969                         }
2970                 }
2971                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2972                 xfs_iext_insert(ip, *idx + 1, 1, &new, state);
2973                 ++*idx;
2974                 break;
2975         }
2976         /*
2977          * If we need to, add to list of extents to delete.
2978          */
2979         if (do_fx)
2980                 xfs_bmap_add_free(del->br_startblock, del->br_blockcount, flist,
2981                         mp);
2982         /*
2983          * Adjust inode # blocks in the file.
2984          */
2985         if (nblks)
2986                 ip->i_d.di_nblocks -= nblks;
2987         /*
2988          * Adjust quota data.
2989          */
2990         if (qfield)
2991                 xfs_trans_mod_dquot_byino(tp, ip, qfield, (long)-nblks);
2992
2993         /*
2994          * Account for change in delayed indirect blocks.
2995          * Nothing to do for disk quota accounting here.
2996          */
2997         ASSERT(da_old >= da_new);
2998         if (da_old > da_new) {
2999                 xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
3000                         (int64_t)(da_old - da_new), 0);
3001         }
3002 done:
3003         *logflagsp = flags;
3004         return error;
3005 }
3006
3007 /*
3008  * Remove the entry "free" from the free item list.  Prev points to the
3009  * previous entry, unless "free" is the head of the list.
3010  */
3011 STATIC void
3012 xfs_bmap_del_free(
3013         xfs_bmap_free_t         *flist, /* free item list header */
3014         xfs_bmap_free_item_t    *prev,  /* previous item on list, if any */
3015         xfs_bmap_free_item_t    *free)  /* list item to be freed */
3016 {
3017         if (prev)
3018                 prev->xbfi_next = free->xbfi_next;
3019         else
3020                 flist->xbf_first = free->xbfi_next;
3021         flist->xbf_count--;
3022         kmem_zone_free(xfs_bmap_free_item_zone, free);
3023 }
3024
3025 /*
3026  * Convert an extents-format file into a btree-format file.
3027  * The new file will have a root block (in the inode) and a single child block.
3028  */
3029 STATIC int                                      /* error */
3030 xfs_bmap_extents_to_btree(
3031         xfs_trans_t             *tp,            /* transaction pointer */
3032         xfs_inode_t             *ip,            /* incore inode pointer */
3033         xfs_fsblock_t           *firstblock,    /* first-block-allocated */
3034         xfs_bmap_free_t         *flist,         /* blocks freed in xaction */
3035         xfs_btree_cur_t         **curp,         /* cursor returned to caller */
3036         int                     wasdel,         /* converting a delayed alloc */
3037         int                     *logflagsp,     /* inode logging flags */
3038         int                     whichfork)      /* data or attr fork */
3039 {
3040         struct xfs_btree_block  *ablock;        /* allocated (child) bt block */
3041         xfs_buf_t               *abp;           /* buffer for ablock */
3042         xfs_alloc_arg_t         args;           /* allocation arguments */
3043         xfs_bmbt_rec_t          *arp;           /* child record pointer */
3044         struct xfs_btree_block  *block;         /* btree root block */
3045         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
3046         xfs_bmbt_rec_host_t     *ep;            /* extent record pointer */
3047         int                     error;          /* error return value */
3048         xfs_extnum_t            i, cnt;         /* extent record index */
3049         xfs_ifork_t             *ifp;           /* inode fork pointer */
3050         xfs_bmbt_key_t          *kp;            /* root block key pointer */
3051         xfs_mount_t             *mp;            /* mount structure */
3052         xfs_extnum_t            nextents;       /* number of file extents */
3053         xfs_bmbt_ptr_t          *pp;            /* root block address pointer */
3054
3055         ifp = XFS_IFORK_PTR(ip, whichfork);
3056         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS);
3057
3058         /*
3059          * Make space in the inode incore.
3060          */
3061         xfs_iroot_realloc(ip, 1, whichfork);
3062         ifp->if_flags |= XFS_IFBROOT;
3063
3064         /*
3065          * Fill in the root.
3066          */
3067         block = ifp->if_broot;
3068         block->bb_magic = cpu_to_be32(XFS_BMAP_MAGIC);
3069         block->bb_level = cpu_to_be16(1);
3070         block->bb_numrecs = cpu_to_be16(1);
3071         block->bb_u.l.bb_leftsib = cpu_to_be64(NULLDFSBNO);
3072         block->bb_u.l.bb_rightsib = cpu_to_be64(NULLDFSBNO);
3073
3074         /*
3075          * Need a cursor.  Can't allocate until bb_level is filled in.
3076          */
3077         mp = ip->i_mount;
3078         cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
3079         cur->bc_private.b.firstblock = *firstblock;
3080         cur->bc_private.b.flist = flist;
3081         cur->bc_private.b.flags = wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
3082         /*
3083          * Convert to a btree with two levels, one record in root.
3084          */
3085         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_BTREE);
3086         args.tp = tp;
3087         args.mp = mp;
3088         args.firstblock = *firstblock;
3089         if (*firstblock == NULLFSBLOCK) {
3090                 args.type = XFS_ALLOCTYPE_START_BNO;
3091                 args.fsbno = XFS_INO_TO_FSB(mp, ip->i_ino);
3092         } else if (flist->xbf_low) {
3093                 args.type = XFS_ALLOCTYPE_START_BNO;
3094                 args.fsbno = *firstblock;
3095         } else {
3096                 args.type = XFS_ALLOCTYPE_NEAR_BNO;
3097                 args.fsbno = *firstblock;
3098         }
3099         args.minlen = args.maxlen = args.prod = 1;
3100         args.total = args.minleft = args.alignment = args.mod = args.isfl =
3101                 args.minalignslop = 0;
3102         args.wasdel = wasdel;
3103         *logflagsp = 0;
3104         if ((error = xfs_alloc_vextent(&args))) {
3105                 xfs_iroot_realloc(ip, -1, whichfork);
3106                 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
3107                 return error;
3108         }
3109         /*
3110          * Allocation can't fail, the space was reserved.
3111          */
3112         ASSERT(args.fsbno != NULLFSBLOCK);
3113         ASSERT(*firstblock == NULLFSBLOCK ||
3114                args.agno == XFS_FSB_TO_AGNO(mp, *firstblock) ||
3115                (flist->xbf_low &&
3116                 args.agno > XFS_FSB_TO_AGNO(mp, *firstblock)));
3117         *firstblock = cur->bc_private.b.firstblock = args.fsbno;
3118         cur->bc_private.b.allocated++;
3119         ip->i_d.di_nblocks++;
3120         xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, 1L);
3121         abp = xfs_btree_get_bufl(mp, tp, args.fsbno, 0);
3122         /*
3123          * Fill in the child block.
3124          */
3125         ablock = XFS_BUF_TO_BLOCK(abp);
3126         ablock->bb_magic = cpu_to_be32(XFS_BMAP_MAGIC);
3127         ablock->bb_level = 0;
3128         ablock->bb_u.l.bb_leftsib = cpu_to_be64(NULLDFSBNO);
3129         ablock->bb_u.l.bb_rightsib = cpu_to_be64(NULLDFSBNO);
3130         arp = XFS_BMBT_REC_ADDR(mp, ablock, 1);
3131         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3132         for (cnt = i = 0; i < nextents; i++) {
3133                 ep = xfs_iext_get_ext(ifp, i);
3134                 if (!isnullstartblock(xfs_bmbt_get_startblock(ep))) {
3135                         arp->l0 = cpu_to_be64(ep->l0);
3136                         arp->l1 = cpu_to_be64(ep->l1);
3137                         arp++; cnt++;
3138                 }
3139         }
3140         ASSERT(cnt == XFS_IFORK_NEXTENTS(ip, whichfork));
3141         xfs_btree_set_numrecs(ablock, cnt);
3142
3143         /*
3144          * Fill in the root key and pointer.
3145          */
3146         kp = XFS_BMBT_KEY_ADDR(mp, block, 1);
3147         arp = XFS_BMBT_REC_ADDR(mp, ablock, 1);
3148         kp->br_startoff = cpu_to_be64(xfs_bmbt_disk_get_startoff(arp));
3149         pp = XFS_BMBT_PTR_ADDR(mp, block, 1, xfs_bmbt_get_maxrecs(cur,
3150                                                 be16_to_cpu(block->bb_level)));
3151         *pp = cpu_to_be64(args.fsbno);
3152
3153         /*
3154          * Do all this logging at the end so that
3155          * the root is at the right level.
3156          */
3157         xfs_btree_log_block(cur, abp, XFS_BB_ALL_BITS);
3158         xfs_btree_log_recs(cur, abp, 1, be16_to_cpu(ablock->bb_numrecs));
3159         ASSERT(*curp == NULL);
3160         *curp = cur;
3161         *logflagsp = XFS_ILOG_CORE | xfs_ilog_fbroot(whichfork);
3162         return 0;
3163 }
3164
3165 /*
3166  * Calculate the default attribute fork offset for newly created inodes.
3167  */
3168 uint
3169 xfs_default_attroffset(
3170         struct xfs_inode        *ip)
3171 {
3172         struct xfs_mount        *mp = ip->i_mount;
3173         uint                    offset;
3174
3175         if (mp->m_sb.sb_inodesize == 256) {
3176                 offset = XFS_LITINO(mp) -
3177                                 XFS_BMDR_SPACE_CALC(MINABTPTRS);
3178         } else {
3179                 offset = XFS_BMDR_SPACE_CALC(6 * MINABTPTRS);
3180         }
3181
3182         ASSERT(offset < XFS_LITINO(mp));
3183         return offset;
3184 }
3185
3186 /*
3187  * Helper routine to reset inode di_forkoff field when switching
3188  * attribute fork from local to extent format - we reset it where
3189  * possible to make space available for inline data fork extents.
3190  */
3191 STATIC void
3192 xfs_bmap_forkoff_reset(
3193         xfs_mount_t     *mp,
3194         xfs_inode_t     *ip,
3195         int             whichfork)
3196 {
3197         if (whichfork == XFS_ATTR_FORK &&
3198             ip->i_d.di_format != XFS_DINODE_FMT_DEV &&
3199             ip->i_d.di_format != XFS_DINODE_FMT_UUID &&
3200             ip->i_d.di_format != XFS_DINODE_FMT_BTREE) {
3201                 uint    dfl_forkoff = xfs_default_attroffset(ip) >> 3;
3202
3203                 if (dfl_forkoff > ip->i_d.di_forkoff)
3204                         ip->i_d.di_forkoff = dfl_forkoff;
3205         }
3206 }
3207
3208 /*
3209  * Convert a local file to an extents file.
3210  * This code is out of bounds for data forks of regular files,
3211  * since the file data needs to get logged so things will stay consistent.
3212  * (The bmap-level manipulations are ok, though).
3213  */
3214 STATIC int                              /* error */
3215 xfs_bmap_local_to_extents(
3216         xfs_trans_t     *tp,            /* transaction pointer */
3217         xfs_inode_t     *ip,            /* incore inode pointer */
3218         xfs_fsblock_t   *firstblock,    /* first block allocated in xaction */
3219         xfs_extlen_t    total,          /* total blocks needed by transaction */
3220         int             *logflagsp,     /* inode logging flags */
3221         int             whichfork)      /* data or attr fork */
3222 {
3223         int             error;          /* error return value */
3224         int             flags;          /* logging flags returned */
3225         xfs_ifork_t     *ifp;           /* inode fork pointer */
3226
3227         /*
3228          * We don't want to deal with the case of keeping inode data inline yet.
3229          * So sending the data fork of a regular inode is invalid.
3230          */
3231         ASSERT(!(S_ISREG(ip->i_d.di_mode) && whichfork == XFS_DATA_FORK));
3232         ifp = XFS_IFORK_PTR(ip, whichfork);
3233         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
3234         flags = 0;
3235         error = 0;
3236         if (ifp->if_bytes) {
3237                 xfs_alloc_arg_t args;   /* allocation arguments */
3238                 xfs_buf_t       *bp;    /* buffer for extent block */
3239                 xfs_bmbt_rec_host_t *ep;/* extent record pointer */
3240
3241                 args.tp = tp;
3242                 args.mp = ip->i_mount;
3243                 args.firstblock = *firstblock;
3244                 ASSERT((ifp->if_flags &
3245                         (XFS_IFINLINE|XFS_IFEXTENTS|XFS_IFEXTIREC)) == XFS_IFINLINE);
3246                 /*
3247                  * Allocate a block.  We know we need only one, since the
3248                  * file currently fits in an inode.
3249                  */
3250                 if (*firstblock == NULLFSBLOCK) {
3251                         args.fsbno = XFS_INO_TO_FSB(args.mp, ip->i_ino);
3252                         args.type = XFS_ALLOCTYPE_START_BNO;
3253                 } else {
3254                         args.fsbno = *firstblock;
3255                         args.type = XFS_ALLOCTYPE_NEAR_BNO;
3256                 }
3257                 args.total = total;
3258                 args.mod = args.minleft = args.alignment = args.wasdel =
3259                         args.isfl = args.minalignslop = 0;
3260                 args.minlen = args.maxlen = args.prod = 1;
3261                 if ((error = xfs_alloc_vextent(&args)))
3262                         goto done;
3263                 /*
3264                  * Can't fail, the space was reserved.
3265                  */
3266                 ASSERT(args.fsbno != NULLFSBLOCK);
3267                 ASSERT(args.len == 1);
3268                 *firstblock = args.fsbno;
3269                 bp = xfs_btree_get_bufl(args.mp, tp, args.fsbno, 0);
3270                 memcpy(bp->b_addr, ifp->if_u1.if_data, ifp->if_bytes);
3271                 xfs_trans_log_buf(tp, bp, 0, ifp->if_bytes - 1);
3272                 xfs_bmap_forkoff_reset(args.mp, ip, whichfork);
3273                 xfs_idata_realloc(ip, -ifp->if_bytes, whichfork);
3274                 xfs_iext_add(ifp, 0, 1);
3275                 ep = xfs_iext_get_ext(ifp, 0);
3276                 xfs_bmbt_set_allf(ep, 0, args.fsbno, 1, XFS_EXT_NORM);
3277                 trace_xfs_bmap_post_update(ip, 0,
3278                                 whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0,
3279                                 _THIS_IP_);
3280                 XFS_IFORK_NEXT_SET(ip, whichfork, 1);
3281                 ip->i_d.di_nblocks = 1;
3282                 xfs_trans_mod_dquot_byino(tp, ip,
3283                         XFS_TRANS_DQ_BCOUNT, 1L);
3284                 flags |= xfs_ilog_fext(whichfork);
3285         } else {
3286                 ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) == 0);
3287                 xfs_bmap_forkoff_reset(ip->i_mount, ip, whichfork);
3288         }
3289         ifp->if_flags &= ~XFS_IFINLINE;
3290         ifp->if_flags |= XFS_IFEXTENTS;
3291         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
3292         flags |= XFS_ILOG_CORE;
3293 done:
3294         *logflagsp = flags;
3295         return error;
3296 }
3297
3298 /*
3299  * Search the extent records for the entry containing block bno.
3300  * If bno lies in a hole, point to the next entry.  If bno lies
3301  * past eof, *eofp will be set, and *prevp will contain the last
3302  * entry (null if none).  Else, *lastxp will be set to the index
3303  * of the found entry; *gotp will contain the entry.
3304  */
3305 STATIC xfs_bmbt_rec_host_t *            /* pointer to found extent entry */
3306 xfs_bmap_search_multi_extents(
3307         xfs_ifork_t     *ifp,           /* inode fork pointer */
3308         xfs_fileoff_t   bno,            /* block number searched for */
3309         int             *eofp,          /* out: end of file found */
3310         xfs_extnum_t    *lastxp,        /* out: last extent index */
3311         xfs_bmbt_irec_t *gotp,          /* out: extent entry found */
3312         xfs_bmbt_irec_t *prevp)         /* out: previous extent entry found */
3313 {
3314         xfs_bmbt_rec_host_t *ep;                /* extent record pointer */
3315         xfs_extnum_t    lastx;          /* last extent index */
3316
3317         /*
3318          * Initialize the extent entry structure to catch access to
3319          * uninitialized br_startblock field.
3320          */
3321         gotp->br_startoff = 0xffa5a5a5a5a5a5a5LL;
3322         gotp->br_blockcount = 0xa55a5a5a5a5a5a5aLL;
3323         gotp->br_state = XFS_EXT_INVALID;
3324 #if XFS_BIG_BLKNOS
3325         gotp->br_startblock = 0xffffa5a5a5a5a5a5LL;
3326 #else
3327         gotp->br_startblock = 0xffffa5a5;
3328 #endif
3329         prevp->br_startoff = NULLFILEOFF;
3330
3331         ep = xfs_iext_bno_to_ext(ifp, bno, &lastx);
3332         if (lastx > 0) {
3333                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx - 1), prevp);
3334         }
3335         if (lastx < (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t))) {
3336                 xfs_bmbt_get_all(ep, gotp);
3337                 *eofp = 0;
3338         } else {
3339                 if (lastx > 0) {
3340                         *gotp = *prevp;
3341                 }
3342                 *eofp = 1;
3343                 ep = NULL;
3344         }
3345         *lastxp = lastx;
3346         return ep;
3347 }
3348
3349 /*
3350  * Search the extents list for the inode, for the extent containing bno.
3351  * If bno lies in a hole, point to the next entry.  If bno lies past eof,
3352  * *eofp will be set, and *prevp will contain the last entry (null if none).
3353  * Else, *lastxp will be set to the index of the found
3354  * entry; *gotp will contain the entry.
3355  */
3356 STATIC xfs_bmbt_rec_host_t *                 /* pointer to found extent entry */
3357 xfs_bmap_search_extents(
3358         xfs_inode_t     *ip,            /* incore inode pointer */
3359         xfs_fileoff_t   bno,            /* block number searched for */
3360         int             fork,           /* data or attr fork */
3361         int             *eofp,          /* out: end of file found */
3362         xfs_extnum_t    *lastxp,        /* out: last extent index */
3363         xfs_bmbt_irec_t *gotp,          /* out: extent entry found */
3364         xfs_bmbt_irec_t *prevp)         /* out: previous extent entry found */
3365 {
3366         xfs_ifork_t     *ifp;           /* inode fork pointer */
3367         xfs_bmbt_rec_host_t  *ep;            /* extent record pointer */
3368
3369         XFS_STATS_INC(xs_look_exlist);
3370         ifp = XFS_IFORK_PTR(ip, fork);
3371
3372         ep = xfs_bmap_search_multi_extents(ifp, bno, eofp, lastxp, gotp, prevp);
3373
3374         if (unlikely(!(gotp->br_startblock) && (*lastxp != NULLEXTNUM) &&
3375                      !(XFS_IS_REALTIME_INODE(ip) && fork == XFS_DATA_FORK))) {
3376                 xfs_alert_tag(ip->i_mount, XFS_PTAG_FSBLOCK_ZERO,
3377                                 "Access to block zero in inode %llu "
3378                                 "start_block: %llx start_off: %llx "
3379                                 "blkcnt: %llx extent-state: %x lastx: %x\n",
3380                         (unsigned long long)ip->i_ino,
3381                         (unsigned long long)gotp->br_startblock,
3382                         (unsigned long long)gotp->br_startoff,
3383                         (unsigned long long)gotp->br_blockcount,
3384                         gotp->br_state, *lastxp);
3385                 *lastxp = NULLEXTNUM;
3386                 *eofp = 1;
3387                 return NULL;
3388         }
3389         return ep;
3390 }
3391
3392 /*
3393  * Compute the worst-case number of indirect blocks that will be used
3394  * for ip's delayed extent of length "len".
3395  */
3396 STATIC xfs_filblks_t
3397 xfs_bmap_worst_indlen(
3398         xfs_inode_t     *ip,            /* incore inode pointer */
3399         xfs_filblks_t   len)            /* delayed extent length */
3400 {
3401         int             level;          /* btree level number */
3402         int             maxrecs;        /* maximum record count at this level */
3403         xfs_mount_t     *mp;            /* mount structure */
3404         xfs_filblks_t   rval;           /* return value */
3405
3406         mp = ip->i_mount;
3407         maxrecs = mp->m_bmap_dmxr[0];
3408         for (level = 0, rval = 0;
3409              level < XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK);
3410              level++) {
3411                 len += maxrecs - 1;
3412                 do_div(len, maxrecs);
3413                 rval += len;
3414                 if (len == 1)
3415                         return rval + XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK) -
3416                                 level - 1;
3417                 if (level == 0)
3418                         maxrecs = mp->m_bmap_dmxr[1];
3419         }
3420         return rval;
3421 }
3422
3423 /*
3424  * Convert inode from non-attributed to attributed.
3425  * Must not be in a transaction, ip must not be locked.
3426  */
3427 int                                             /* error code */
3428 xfs_bmap_add_attrfork(
3429         xfs_inode_t             *ip,            /* incore inode pointer */
3430         int                     size,           /* space new attribute needs */
3431         int                     rsvd)           /* xact may use reserved blks */
3432 {
3433         xfs_fsblock_t           firstblock;     /* 1st block/ag allocated */
3434         xfs_bmap_free_t         flist;          /* freed extent records */
3435         xfs_mount_t             *mp;            /* mount structure */
3436         xfs_trans_t             *tp;            /* transaction pointer */
3437         int                     blks;           /* space reservation */
3438         int                     version = 1;    /* superblock attr version */
3439         int                     committed;      /* xaction was committed */
3440         int                     logflags;       /* logging flags */
3441         int                     error;          /* error return value */
3442
3443         ASSERT(XFS_IFORK_Q(ip) == 0);
3444
3445         mp = ip->i_mount;
3446         ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
3447         tp = xfs_trans_alloc(mp, XFS_TRANS_ADDAFORK);
3448         blks = XFS_ADDAFORK_SPACE_RES(mp);
3449         if (rsvd)
3450                 tp->t_flags |= XFS_TRANS_RESERVE;
3451         if ((error = xfs_trans_reserve(tp, blks, XFS_ADDAFORK_LOG_RES(mp), 0,
3452                         XFS_TRANS_PERM_LOG_RES, XFS_ADDAFORK_LOG_COUNT)))
3453                 goto error0;
3454         xfs_ilock(ip, XFS_ILOCK_EXCL);
3455         error = xfs_trans_reserve_quota_nblks(tp, ip, blks, 0, rsvd ?
3456                         XFS_QMOPT_RES_REGBLKS | XFS_QMOPT_FORCE_RES :
3457                         XFS_QMOPT_RES_REGBLKS);
3458         if (error) {
3459                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3460                 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES);
3461                 return error;
3462         }
3463         if (XFS_IFORK_Q(ip))
3464                 goto error1;
3465         if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS) {
3466                 /*
3467                  * For inodes coming from pre-6.2 filesystems.
3468                  */
3469                 ASSERT(ip->i_d.di_aformat == 0);
3470                 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
3471         }
3472         ASSERT(ip->i_d.di_anextents == 0);
3473
3474         xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
3475         xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
3476
3477         switch (ip->i_d.di_format) {
3478         case XFS_DINODE_FMT_DEV:
3479                 ip->i_d.di_forkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
3480                 break;
3481         case XFS_DINODE_FMT_UUID:
3482                 ip->i_d.di_forkoff = roundup(sizeof(uuid_t), 8) >> 3;
3483                 break;
3484         case XFS_DINODE_FMT_LOCAL:
3485         case XFS_DINODE_FMT_EXTENTS:
3486         case XFS_DINODE_FMT_BTREE:
3487                 ip->i_d.di_forkoff = xfs_attr_shortform_bytesfit(ip, size);
3488                 if (!ip->i_d.di_forkoff)
3489                         ip->i_d.di_forkoff = xfs_default_attroffset(ip) >> 3;
3490                 else if (mp->m_flags & XFS_MOUNT_ATTR2)
3491                         version = 2;
3492                 break;
3493         default:
3494                 ASSERT(0);
3495                 error = XFS_ERROR(EINVAL);
3496                 goto error1;
3497         }
3498
3499         ASSERT(ip->i_afp == NULL);
3500         ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP);
3501         ip->i_afp->if_flags = XFS_IFEXTENTS;
3502         logflags = 0;
3503         xfs_bmap_init(&flist, &firstblock);
3504         switch (ip->i_d.di_format) {
3505         case XFS_DINODE_FMT_LOCAL:
3506                 error = xfs_bmap_add_attrfork_local(tp, ip, &firstblock, &flist,
3507                         &logflags);
3508                 break;
3509         case XFS_DINODE_FMT_EXTENTS:
3510                 error = xfs_bmap_add_attrfork_extents(tp, ip, &firstblock,
3511                         &flist, &logflags);
3512                 break;
3513         case XFS_DINODE_FMT_BTREE:
3514                 error = xfs_bmap_add_attrfork_btree(tp, ip, &firstblock, &flist,
3515                         &logflags);
3516                 break;
3517         default:
3518                 error = 0;
3519                 break;
3520         }
3521         if (logflags)
3522                 xfs_trans_log_inode(tp, ip, logflags);
3523         if (error)
3524                 goto error2;
3525         if (!xfs_sb_version_hasattr(&mp->m_sb) ||
3526            (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2)) {
3527                 __int64_t sbfields = 0;
3528
3529                 spin_lock(&mp->m_sb_lock);
3530                 if (!xfs_sb_version_hasattr(&mp->m_sb)) {
3531                         xfs_sb_version_addattr(&mp->m_sb);
3532                         sbfields |= XFS_SB_VERSIONNUM;
3533                 }
3534                 if (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2) {
3535                         xfs_sb_version_addattr2(&mp->m_sb);
3536                         sbfields |= (XFS_SB_VERSIONNUM | XFS_SB_FEATURES2);
3537                 }
3538                 if (sbfields) {
3539                         spin_unlock(&mp->m_sb_lock);
3540                         xfs_mod_sb(tp, sbfields);
3541                 } else
3542                         spin_unlock(&mp->m_sb_lock);
3543         }
3544
3545         error = xfs_bmap_finish(&tp, &flist, &committed);
3546         if (error)
3547                 goto error2;
3548         return xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
3549 error2:
3550         xfs_bmap_cancel(&flist);
3551 error1:
3552         xfs_iunlock(ip, XFS_ILOCK_EXCL);
3553 error0:
3554         xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
3555         return error;
3556 }
3557
3558 /*
3559  * Add the extent to the list of extents to be free at transaction end.
3560  * The list is maintained sorted (by block number).
3561  */
3562 /* ARGSUSED */
3563 void
3564 xfs_bmap_add_free(
3565         xfs_fsblock_t           bno,            /* fs block number of extent */
3566         xfs_filblks_t           len,            /* length of extent */
3567         xfs_bmap_free_t         *flist,         /* list of extents */
3568         xfs_mount_t             *mp)            /* mount point structure */
3569 {
3570         xfs_bmap_free_item_t    *cur;           /* current (next) element */
3571         xfs_bmap_free_item_t    *new;           /* new element */
3572         xfs_bmap_free_item_t    *prev;          /* previous element */
3573 #ifdef DEBUG
3574         xfs_agnumber_t          agno;
3575         xfs_agblock_t           agbno;
3576
3577         ASSERT(bno != NULLFSBLOCK);
3578         ASSERT(len > 0);
3579         ASSERT(len <= MAXEXTLEN);
3580         ASSERT(!isnullstartblock(bno));
3581         agno = XFS_FSB_TO_AGNO(mp, bno);
3582         agbno = XFS_FSB_TO_AGBNO(mp, bno);
3583         ASSERT(agno < mp->m_sb.sb_agcount);
3584         ASSERT(agbno < mp->m_sb.sb_agblocks);
3585         ASSERT(len < mp->m_sb.sb_agblocks);
3586         ASSERT(agbno + len <= mp->m_sb.sb_agblocks);
3587 #endif
3588         ASSERT(xfs_bmap_free_item_zone != NULL);
3589         new = kmem_zone_alloc(xfs_bmap_free_item_zone, KM_SLEEP);
3590         new->xbfi_startblock = bno;
3591         new->xbfi_blockcount = (xfs_extlen_t)len;
3592         for (prev = NULL, cur = flist->xbf_first;
3593              cur != NULL;
3594              prev = cur, cur = cur->xbfi_next) {
3595                 if (cur->xbfi_startblock >= bno)
3596                         break;
3597         }
3598         if (prev)
3599                 prev->xbfi_next = new;
3600         else
3601                 flist->xbf_first = new;
3602         new->xbfi_next = cur;
3603         flist->xbf_count++;
3604 }
3605
3606 /*
3607  * Compute and fill in the value of the maximum depth of a bmap btree
3608  * in this filesystem.  Done once, during mount.
3609  */
3610 void
3611 xfs_bmap_compute_maxlevels(
3612         xfs_mount_t     *mp,            /* file system mount structure */
3613         int             whichfork)      /* data or attr fork */
3614 {
3615         int             level;          /* btree level */
3616         uint            maxblocks;      /* max blocks at this level */
3617         uint            maxleafents;    /* max leaf entries possible */
3618         int             maxrootrecs;    /* max records in root block */
3619         int             minleafrecs;    /* min records in leaf block */
3620         int             minnoderecs;    /* min records in node block */
3621         int             sz;             /* root block size */
3622
3623         /*
3624          * The maximum number of extents in a file, hence the maximum
3625          * number of leaf entries, is controlled by the type of di_nextents
3626          * (a signed 32-bit number, xfs_extnum_t), or by di_anextents
3627          * (a signed 16-bit number, xfs_aextnum_t).
3628          *
3629          * Note that we can no longer assume that if we are in ATTR1 that
3630          * the fork offset of all the inodes will be
3631          * (xfs_default_attroffset(ip) >> 3) because we could have mounted
3632          * with ATTR2 and then mounted back with ATTR1, keeping the
3633          * di_forkoff's fixed but probably at various positions. Therefore,
3634          * for both ATTR1 and ATTR2 we have to assume the worst case scenario
3635          * of a minimum size available.
3636          */
3637         if (whichfork == XFS_DATA_FORK) {
3638                 maxleafents = MAXEXTNUM;
3639                 sz = XFS_BMDR_SPACE_CALC(MINDBTPTRS);
3640         } else {
3641                 maxleafents = MAXAEXTNUM;
3642                 sz = XFS_BMDR_SPACE_CALC(MINABTPTRS);
3643         }
3644         maxrootrecs = xfs_bmdr_maxrecs(mp, sz, 0);
3645         minleafrecs = mp->m_bmap_dmnr[0];
3646         minnoderecs = mp->m_bmap_dmnr[1];
3647         maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs;
3648         for (level = 1; maxblocks > 1; level++) {
3649                 if (maxblocks <= maxrootrecs)
3650                         maxblocks = 1;
3651                 else
3652                         maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs;
3653         }
3654         mp->m_bm_maxlevels[whichfork] = level;
3655 }
3656
3657 /*
3658  * Routine to be called at transaction's end by xfs_bmapi, xfs_bunmapi
3659  * caller.  Frees all the extents that need freeing, which must be done
3660  * last due to locking considerations.  We never free any extents in
3661  * the first transaction.
3662  *
3663  * Return 1 if the given transaction was committed and a new one
3664  * started, and 0 otherwise in the committed parameter.
3665  */
3666 int                                             /* error */
3667 xfs_bmap_finish(
3668         xfs_trans_t             **tp,           /* transaction pointer addr */
3669         xfs_bmap_free_t         *flist,         /* i/o: list extents to free */
3670         int                     *committed)     /* xact committed or not */
3671 {
3672         xfs_efd_log_item_t      *efd;           /* extent free data */
3673         xfs_efi_log_item_t      *efi;           /* extent free intention */
3674         int                     error;          /* error return value */
3675         xfs_bmap_free_item_t    *free;          /* free extent item */
3676         unsigned int            logres;         /* new log reservation */
3677         unsigned int            logcount;       /* new log count */
3678         xfs_mount_t             *mp;            /* filesystem mount structure */
3679         xfs_bmap_free_item_t    *next;          /* next item on free list */
3680         xfs_trans_t             *ntp;           /* new transaction pointer */
3681
3682         ASSERT((*tp)->t_flags & XFS_TRANS_PERM_LOG_RES);
3683         if (flist->xbf_count == 0) {
3684                 *committed = 0;
3685                 return 0;
3686         }
3687         ntp = *tp;
3688         efi = xfs_trans_get_efi(ntp, flist->xbf_count);
3689         for (free = flist->xbf_first; free; free = free->xbfi_next)
3690                 xfs_trans_log_efi_extent(ntp, efi, free->xbfi_startblock,
3691                         free->xbfi_blockcount);
3692         logres = ntp->t_log_res;
3693         logcount = ntp->t_log_count;
3694         ntp = xfs_trans_dup(*tp);
3695         error = xfs_trans_commit(*tp, 0);
3696         *tp = ntp;
3697         *committed = 1;
3698         /*
3699          * We have a new transaction, so we should return committed=1,
3700          * even though we're returning an error.
3701          */
3702         if (error)
3703                 return error;
3704
3705         /*
3706          * transaction commit worked ok so we can drop the extra ticket
3707          * reference that we gained in xfs_trans_dup()
3708          */
3709         xfs_log_ticket_put(ntp->t_ticket);
3710
3711         if ((error = xfs_trans_reserve(ntp, 0, logres, 0, XFS_TRANS_PERM_LOG_RES,
3712                         logcount)))
3713                 return error;
3714         efd = xfs_trans_get_efd(ntp, efi, flist->xbf_count);
3715         for (free = flist->xbf_first; free != NULL; free = next) {
3716                 next = free->xbfi_next;
3717                 if ((error = xfs_free_extent(ntp, free->xbfi_startblock,
3718                                 free->xbfi_blockcount))) {
3719                         /*
3720                          * The bmap free list will be cleaned up at a
3721                          * higher level.  The EFI will be canceled when
3722                          * this transaction is aborted.
3723                          * Need to force shutdown here to make sure it
3724                          * happens, since this transaction may not be
3725                          * dirty yet.
3726                          */
3727                         mp = ntp->t_mountp;
3728                         if (!XFS_FORCED_SHUTDOWN(mp))
3729                                 xfs_force_shutdown(mp,
3730                                                    (error == EFSCORRUPTED) ?
3731                                                    SHUTDOWN_CORRUPT_INCORE :
3732                                                    SHUTDOWN_META_IO_ERROR);
3733                         return error;
3734                 }
3735                 xfs_trans_log_efd_extent(ntp, efd, free->xbfi_startblock,
3736                         free->xbfi_blockcount);
3737                 xfs_bmap_del_free(flist, NULL, free);
3738         }
3739         return 0;
3740 }
3741
3742 /*
3743  * Free up any items left in the list.
3744  */
3745 void
3746 xfs_bmap_cancel(
3747         xfs_bmap_free_t         *flist) /* list of bmap_free_items */
3748 {
3749         xfs_bmap_free_item_t    *free;  /* free list item */
3750         xfs_bmap_free_item_t    *next;
3751
3752         if (flist->xbf_count == 0)
3753                 return;
3754         ASSERT(flist->xbf_first != NULL);
3755         for (free = flist->xbf_first; free; free = next) {
3756                 next = free->xbfi_next;
3757                 xfs_bmap_del_free(flist, NULL, free);
3758         }
3759         ASSERT(flist->xbf_count == 0);
3760 }
3761
3762 /*
3763  * Returns the file-relative block number of the first unused block(s)
3764  * in the file with at least "len" logically contiguous blocks free.
3765  * This is the lowest-address hole if the file has holes, else the first block
3766  * past the end of file.
3767  * Return 0 if the file is currently local (in-inode).
3768  */
3769 int                                             /* error */
3770 xfs_bmap_first_unused(
3771         xfs_trans_t     *tp,                    /* transaction pointer */
3772         xfs_inode_t     *ip,                    /* incore inode */
3773         xfs_extlen_t    len,                    /* size of hole to find */
3774         xfs_fileoff_t   *first_unused,          /* unused block */
3775         int             whichfork)              /* data or attr fork */
3776 {
3777         int             error;                  /* error return value */
3778         int             idx;                    /* extent record index */
3779         xfs_ifork_t     *ifp;                   /* inode fork pointer */
3780         xfs_fileoff_t   lastaddr;               /* last block number seen */
3781         xfs_fileoff_t   lowest;                 /* lowest useful block */
3782         xfs_fileoff_t   max;                    /* starting useful block */
3783         xfs_fileoff_t   off;                    /* offset for this block */
3784         xfs_extnum_t    nextents;               /* number of extent entries */
3785
3786         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE ||
3787                XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ||
3788                XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
3789         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
3790                 *first_unused = 0;
3791                 return 0;
3792         }
3793         ifp = XFS_IFORK_PTR(ip, whichfork);
3794         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
3795             (error = xfs_iread_extents(tp, ip, whichfork)))
3796                 return error;
3797         lowest = *first_unused;
3798         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3799         for (idx = 0, lastaddr = 0, max = lowest; idx < nextents; idx++) {
3800                 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, idx);
3801                 off = xfs_bmbt_get_startoff(ep);
3802                 /*
3803                  * See if the hole before this extent will work.
3804                  */
3805                 if (off >= lowest + len && off - max >= len) {
3806                         *first_unused = max;
3807                         return 0;
3808                 }
3809                 lastaddr = off + xfs_bmbt_get_blockcount(ep);
3810                 max = XFS_FILEOFF_MAX(lastaddr, lowest);
3811         }
3812         *first_unused = max;
3813         return 0;
3814 }
3815
3816 /*
3817  * Returns the file-relative block number of the last block + 1 before
3818  * last_block (input value) in the file.
3819  * This is not based on i_size, it is based on the extent records.
3820  * Returns 0 for local files, as they do not have extent records.
3821  */
3822 int                                             /* error */
3823 xfs_bmap_last_before(
3824         xfs_trans_t     *tp,                    /* transaction pointer */
3825         xfs_inode_t     *ip,                    /* incore inode */
3826         xfs_fileoff_t   *last_block,            /* last block */
3827         int             whichfork)              /* data or attr fork */
3828 {
3829         xfs_fileoff_t   bno;                    /* input file offset */
3830         int             eof;                    /* hit end of file */
3831         xfs_bmbt_rec_host_t *ep;                /* pointer to last extent */
3832         int             error;                  /* error return value */
3833         xfs_bmbt_irec_t got;                    /* current extent value */
3834         xfs_ifork_t     *ifp;                   /* inode fork pointer */
3835         xfs_extnum_t    lastx;                  /* last extent used */
3836         xfs_bmbt_irec_t prev;                   /* previous extent value */
3837
3838         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
3839             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
3840             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL)
3841                return XFS_ERROR(EIO);
3842         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
3843                 *last_block = 0;
3844                 return 0;
3845         }
3846         ifp = XFS_IFORK_PTR(ip, whichfork);
3847         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
3848             (error = xfs_iread_extents(tp, ip, whichfork)))
3849                 return error;
3850         bno = *last_block - 1;
3851         ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
3852                 &prev);
3853         if (eof || xfs_bmbt_get_startoff(ep) > bno) {
3854                 if (prev.br_startoff == NULLFILEOFF)
3855                         *last_block = 0;
3856                 else
3857                         *last_block = prev.br_startoff + prev.br_blockcount;
3858         }
3859         /*
3860          * Otherwise *last_block is already the right answer.
3861          */
3862         return 0;
3863 }
3864
3865 STATIC int
3866 xfs_bmap_last_extent(
3867         struct xfs_trans        *tp,
3868         struct xfs_inode        *ip,
3869         int                     whichfork,
3870         struct xfs_bmbt_irec    *rec,
3871         int                     *is_empty)
3872 {
3873         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, whichfork);
3874         int                     error;
3875         int                     nextents;
3876
3877         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
3878                 error = xfs_iread_extents(tp, ip, whichfork);
3879                 if (error)
3880                         return error;
3881         }
3882
3883         nextents = ifp->if_bytes / sizeof(xfs_bmbt_rec_t);
3884         if (nextents == 0) {
3885                 *is_empty = 1;
3886                 return 0;
3887         }
3888
3889         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, nextents - 1), rec);
3890         *is_empty = 0;
3891         return 0;
3892 }
3893
3894 /*
3895  * Check the last inode extent to determine whether this allocation will result
3896  * in blocks being allocated at the end of the file. When we allocate new data
3897  * blocks at the end of the file which do not start at the previous data block,
3898  * we will try to align the new blocks at stripe unit boundaries.
3899  *
3900  * Returns 0 in bma->aeof if the file (fork) is empty as any new write will be
3901  * at, or past the EOF.
3902  */
3903 STATIC int
3904 xfs_bmap_isaeof(
3905         struct xfs_bmalloca     *bma,
3906         int                     whichfork)
3907 {
3908         struct xfs_bmbt_irec    rec;
3909         int                     is_empty;
3910         int                     error;
3911
3912         bma->aeof = 0;
3913         error = xfs_bmap_last_extent(NULL, bma->ip, whichfork, &rec,
3914                                      &is_empty);
3915         if (error || is_empty)
3916                 return error;
3917
3918         /*
3919          * Check if we are allocation or past the last extent, or at least into
3920          * the last delayed allocated extent.
3921          */
3922         bma->aeof = bma->offset >= rec.br_startoff + rec.br_blockcount ||
3923                 (bma->offset >= rec.br_startoff &&
3924                  isnullstartblock(rec.br_startblock));
3925         return 0;
3926 }
3927
3928 /*
3929  * Check if the endoff is outside the last extent. If so the caller will grow
3930  * the allocation to a stripe unit boundary.  All offsets are considered outside
3931  * the end of file for an empty fork, so 1 is returned in *eof in that case.
3932  */
3933 int
3934 xfs_bmap_eof(
3935         struct xfs_inode        *ip,
3936         xfs_fileoff_t           endoff,
3937         int                     whichfork,
3938         int                     *eof)
3939 {
3940         struct xfs_bmbt_irec    rec;
3941         int                     error;
3942
3943         error = xfs_bmap_last_extent(NULL, ip, whichfork, &rec, eof);
3944         if (error || *eof)
3945                 return error;
3946
3947         *eof = endoff >= rec.br_startoff + rec.br_blockcount;
3948         return 0;
3949 }
3950
3951 /*
3952  * Returns the file-relative block number of the first block past eof in
3953  * the file.  This is not based on i_size, it is based on the extent records.
3954  * Returns 0 for local files, as they do not have extent records.
3955  */
3956 int
3957 xfs_bmap_last_offset(
3958         struct xfs_trans        *tp,
3959         struct xfs_inode        *ip,
3960         xfs_fileoff_t           *last_block,
3961         int                     whichfork)
3962 {
3963         struct xfs_bmbt_irec    rec;
3964         int                     is_empty;
3965         int                     error;
3966
3967         *last_block = 0;
3968
3969         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL)
3970                 return 0;
3971
3972         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
3973             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
3974                return XFS_ERROR(EIO);
3975
3976         error = xfs_bmap_last_extent(NULL, ip, whichfork, &rec, &is_empty);
3977         if (error || is_empty)
3978                 return error;
3979
3980         *last_block = rec.br_startoff + rec.br_blockcount;
3981         return 0;
3982 }
3983
3984 /*
3985  * Returns whether the selected fork of the inode has exactly one
3986  * block or not.  For the data fork we check this matches di_size,
3987  * implying the file's range is 0..bsize-1.
3988  */
3989 int                                     /* 1=>1 block, 0=>otherwise */
3990 xfs_bmap_one_block(
3991         xfs_inode_t     *ip,            /* incore inode */
3992         int             whichfork)      /* data or attr fork */
3993 {
3994         xfs_bmbt_rec_host_t *ep;        /* ptr to fork's extent */
3995         xfs_ifork_t     *ifp;           /* inode fork pointer */
3996         int             rval;           /* return value */
3997         xfs_bmbt_irec_t s;              /* internal version of extent */
3998
3999 #ifndef DEBUG
4000         if (whichfork == XFS_DATA_FORK)
4001                 return XFS_ISIZE(ip) == ip->i_mount->m_sb.sb_blocksize;
4002 #endif  /* !DEBUG */
4003         if (XFS_IFORK_NEXTENTS(ip, whichfork) != 1)
4004                 return 0;
4005         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
4006                 return 0;
4007         ifp = XFS_IFORK_PTR(ip, whichfork);
4008         ASSERT(ifp->if_flags & XFS_IFEXTENTS);
4009         ep = xfs_iext_get_ext(ifp, 0);
4010         xfs_bmbt_get_all(ep, &s);
4011         rval = s.br_startoff == 0 && s.br_blockcount == 1;
4012         if (rval && whichfork == XFS_DATA_FORK)
4013                 ASSERT(XFS_ISIZE(ip) == ip->i_mount->m_sb.sb_blocksize);
4014         return rval;
4015 }
4016
4017 STATIC int
4018 xfs_bmap_sanity_check(
4019         struct xfs_mount        *mp,
4020         struct xfs_buf          *bp,
4021         int                     level)
4022 {
4023         struct xfs_btree_block  *block = XFS_BUF_TO_BLOCK(bp);
4024
4025         if (block->bb_magic != cpu_to_be32(XFS_BMAP_MAGIC) ||
4026             be16_to_cpu(block->bb_level) != level ||
4027             be16_to_cpu(block->bb_numrecs) == 0 ||
4028             be16_to_cpu(block->bb_numrecs) > mp->m_bmap_dmxr[level != 0])
4029                 return 0;
4030         return 1;
4031 }
4032
4033 /*
4034  * Read in the extents to if_extents.
4035  * All inode fields are set up by caller, we just traverse the btree
4036  * and copy the records in. If the file system cannot contain unwritten
4037  * extents, the records are checked for no "state" flags.
4038  */
4039 int                                     /* error */
4040 xfs_bmap_read_extents(
4041         xfs_trans_t             *tp,    /* transaction pointer */
4042         xfs_inode_t             *ip,    /* incore inode */
4043         int                     whichfork) /* data or attr fork */
4044 {
4045         struct xfs_btree_block  *block; /* current btree block */
4046         xfs_fsblock_t           bno;    /* block # of "block" */
4047         xfs_buf_t               *bp;    /* buffer for "block" */
4048         int                     error;  /* error return value */
4049         xfs_exntfmt_t           exntf;  /* XFS_EXTFMT_NOSTATE, if checking */
4050         xfs_extnum_t            i, j;   /* index into the extents list */
4051         xfs_ifork_t             *ifp;   /* fork structure */
4052         int                     level;  /* btree level, for checking */
4053         xfs_mount_t             *mp;    /* file system mount structure */
4054         __be64                  *pp;    /* pointer to block address */
4055         /* REFERENCED */
4056         xfs_extnum_t            room;   /* number of entries there's room for */
4057
4058         bno = NULLFSBLOCK;
4059         mp = ip->i_mount;
4060         ifp = XFS_IFORK_PTR(ip, whichfork);
4061         exntf = (whichfork != XFS_DATA_FORK) ? XFS_EXTFMT_NOSTATE :
4062                                         XFS_EXTFMT_INODE(ip);
4063         block = ifp->if_broot;
4064         /*
4065          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
4066          */
4067         level = be16_to_cpu(block->bb_level);
4068         ASSERT(level > 0);
4069         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
4070         bno = be64_to_cpu(*pp);
4071         ASSERT(bno != NULLDFSBNO);
4072         ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
4073         ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
4074         /*
4075          * Go down the tree until leaf level is reached, following the first
4076          * pointer (leftmost) at each level.
4077          */
4078         while (level-- > 0) {
4079                 if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
4080                                 XFS_BMAP_BTREE_REF)))
4081                         return error;
4082                 block = XFS_BUF_TO_BLOCK(bp);
4083                 XFS_WANT_CORRUPTED_GOTO(
4084                         xfs_bmap_sanity_check(mp, bp, level),
4085                         error0);
4086                 if (level == 0)
4087                         break;
4088                 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
4089                 bno = be64_to_cpu(*pp);
4090                 XFS_WANT_CORRUPTED_GOTO(XFS_FSB_SANITY_CHECK(mp, bno), error0);
4091                 xfs_trans_brelse(tp, bp);
4092         }
4093         /*
4094          * Here with bp and block set to the leftmost leaf node in the tree.
4095          */
4096         room = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4097         i = 0;
4098         /*
4099          * Loop over all leaf nodes.  Copy information to the extent records.
4100          */
4101         for (;;) {
4102                 xfs_bmbt_rec_t  *frp;
4103                 xfs_fsblock_t   nextbno;
4104                 xfs_extnum_t    num_recs;
4105                 xfs_extnum_t    start;
4106
4107                 num_recs = xfs_btree_get_numrecs(block);
4108                 if (unlikely(i + num_recs > room)) {
4109                         ASSERT(i + num_recs <= room);
4110                         xfs_warn(ip->i_mount,
4111                                 "corrupt dinode %Lu, (btree extents).",
4112                                 (unsigned long long) ip->i_ino);
4113                         XFS_CORRUPTION_ERROR("xfs_bmap_read_extents(1)",
4114                                 XFS_ERRLEVEL_LOW, ip->i_mount, block);
4115                         goto error0;
4116                 }
4117                 XFS_WANT_CORRUPTED_GOTO(
4118                         xfs_bmap_sanity_check(mp, bp, 0),
4119                         error0);
4120                 /*
4121                  * Read-ahead the next leaf block, if any.
4122                  */
4123                 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
4124                 if (nextbno != NULLFSBLOCK)
4125                         xfs_btree_reada_bufl(mp, nextbno, 1);
4126                 /*
4127                  * Copy records into the extent records.
4128                  */
4129                 frp = XFS_BMBT_REC_ADDR(mp, block, 1);
4130                 start = i;
4131                 for (j = 0; j < num_recs; j++, i++, frp++) {
4132                         xfs_bmbt_rec_host_t *trp = xfs_iext_get_ext(ifp, i);
4133                         trp->l0 = be64_to_cpu(frp->l0);
4134                         trp->l1 = be64_to_cpu(frp->l1);
4135                 }
4136                 if (exntf == XFS_EXTFMT_NOSTATE) {
4137                         /*
4138                          * Check all attribute bmap btree records and
4139                          * any "older" data bmap btree records for a
4140                          * set bit in the "extent flag" position.
4141                          */
4142                         if (unlikely(xfs_check_nostate_extents(ifp,
4143                                         start, num_recs))) {
4144                                 XFS_ERROR_REPORT("xfs_bmap_read_extents(2)",
4145                                                  XFS_ERRLEVEL_LOW,
4146                                                  ip->i_mount);
4147                                 goto error0;
4148                         }
4149                 }
4150                 xfs_trans_brelse(tp, bp);
4151                 bno = nextbno;
4152                 /*
4153                  * If we've reached the end, stop.
4154                  */
4155                 if (bno == NULLFSBLOCK)
4156                         break;
4157                 if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
4158                                 XFS_BMAP_BTREE_REF)))
4159                         return error;
4160                 block = XFS_BUF_TO_BLOCK(bp);
4161         }
4162         ASSERT(i == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)));
4163         ASSERT(i == XFS_IFORK_NEXTENTS(ip, whichfork));
4164         XFS_BMAP_TRACE_EXLIST(ip, i, whichfork);
4165         return 0;
4166 error0:
4167         xfs_trans_brelse(tp, bp);
4168         return XFS_ERROR(EFSCORRUPTED);
4169 }
4170
4171 #ifdef DEBUG
4172 /*
4173  * Add bmap trace insert entries for all the contents of the extent records.
4174  */
4175 void
4176 xfs_bmap_trace_exlist(
4177         xfs_inode_t     *ip,            /* incore inode pointer */
4178         xfs_extnum_t    cnt,            /* count of entries in the list */
4179         int             whichfork,      /* data or attr fork */
4180         unsigned long   caller_ip)
4181 {
4182         xfs_extnum_t    idx;            /* extent record index */
4183         xfs_ifork_t     *ifp;           /* inode fork pointer */
4184         int             state = 0;
4185
4186         if (whichfork == XFS_ATTR_FORK)
4187                 state |= BMAP_ATTRFORK;
4188
4189         ifp = XFS_IFORK_PTR(ip, whichfork);
4190         ASSERT(cnt == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)));
4191         for (idx = 0; idx < cnt; idx++)
4192                 trace_xfs_extlist(ip, idx, whichfork, caller_ip);
4193 }
4194
4195 /*
4196  * Validate that the bmbt_irecs being returned from bmapi are valid
4197  * given the callers original parameters.  Specifically check the
4198  * ranges of the returned irecs to ensure that they only extent beyond
4199  * the given parameters if the XFS_BMAPI_ENTIRE flag was set.
4200  */
4201 STATIC void
4202 xfs_bmap_validate_ret(
4203         xfs_fileoff_t           bno,
4204         xfs_filblks_t           len,
4205         int                     flags,
4206         xfs_bmbt_irec_t         *mval,
4207         int                     nmap,
4208         int                     ret_nmap)
4209 {
4210         int                     i;              /* index to map values */
4211
4212         ASSERT(ret_nmap <= nmap);
4213
4214         for (i = 0; i < ret_nmap; i++) {
4215                 ASSERT(mval[i].br_blockcount > 0);
4216                 if (!(flags & XFS_BMAPI_ENTIRE)) {
4217                         ASSERT(mval[i].br_startoff >= bno);
4218                         ASSERT(mval[i].br_blockcount <= len);
4219                         ASSERT(mval[i].br_startoff + mval[i].br_blockcount <=
4220                                bno + len);
4221                 } else {
4222                         ASSERT(mval[i].br_startoff < bno + len);
4223                         ASSERT(mval[i].br_startoff + mval[i].br_blockcount >
4224                                bno);
4225                 }
4226                 ASSERT(i == 0 ||
4227                        mval[i - 1].br_startoff + mval[i - 1].br_blockcount ==
4228                        mval[i].br_startoff);
4229                 ASSERT(mval[i].br_startblock != DELAYSTARTBLOCK &&
4230                        mval[i].br_startblock != HOLESTARTBLOCK);
4231                 ASSERT(mval[i].br_state == XFS_EXT_NORM ||
4232                        mval[i].br_state == XFS_EXT_UNWRITTEN);
4233         }
4234 }
4235 #endif /* DEBUG */
4236
4237
4238 /*
4239  * Trim the returned map to the required bounds
4240  */
4241 STATIC void
4242 xfs_bmapi_trim_map(
4243         struct xfs_bmbt_irec    *mval,
4244         struct xfs_bmbt_irec    *got,
4245         xfs_fileoff_t           *bno,
4246         xfs_filblks_t           len,
4247         xfs_fileoff_t           obno,
4248         xfs_fileoff_t           end,
4249         int                     n,
4250         int                     flags)
4251 {
4252         if ((flags & XFS_BMAPI_ENTIRE) ||
4253             got->br_startoff + got->br_blockcount <= obno) {
4254                 *mval = *got;
4255                 if (isnullstartblock(got->br_startblock))
4256                         mval->br_startblock = DELAYSTARTBLOCK;
4257                 return;
4258         }
4259
4260         if (obno > *bno)
4261                 *bno = obno;
4262         ASSERT((*bno >= obno) || (n == 0));
4263         ASSERT(*bno < end);
4264         mval->br_startoff = *bno;
4265         if (isnullstartblock(got->br_startblock))
4266                 mval->br_startblock = DELAYSTARTBLOCK;
4267         else
4268                 mval->br_startblock = got->br_startblock +
4269                                         (*bno - got->br_startoff);
4270         /*
4271          * Return the minimum of what we got and what we asked for for
4272          * the length.  We can use the len variable here because it is
4273          * modified below and we could have been there before coming
4274          * here if the first part of the allocation didn't overlap what
4275          * was asked for.
4276          */
4277         mval->br_blockcount = XFS_FILBLKS_MIN(end - *bno,
4278                         got->br_blockcount - (*bno - got->br_startoff));
4279         mval->br_state = got->br_state;
4280         ASSERT(mval->br_blockcount <= len);
4281         return;
4282 }
4283
4284 /*
4285  * Update and validate the extent map to return
4286  */
4287 STATIC void
4288 xfs_bmapi_update_map(
4289         struct xfs_bmbt_irec    **map,
4290         xfs_fileoff_t           *bno,
4291         xfs_filblks_t           *len,
4292         xfs_fileoff_t           obno,
4293         xfs_fileoff_t           end,
4294         int                     *n,
4295         int                     flags)
4296 {
4297         xfs_bmbt_irec_t *mval = *map;
4298
4299         ASSERT((flags & XFS_BMAPI_ENTIRE) ||
4300                ((mval->br_startoff + mval->br_blockcount) <= end));
4301         ASSERT((flags & XFS_BMAPI_ENTIRE) || (mval->br_blockcount <= *len) ||
4302                (mval->br_startoff < obno));
4303
4304         *bno = mval->br_startoff + mval->br_blockcount;
4305         *len = end - *bno;
4306         if (*n > 0 && mval->br_startoff == mval[-1].br_startoff) {
4307                 /* update previous map with new information */
4308                 ASSERT(mval->br_startblock == mval[-1].br_startblock);
4309                 ASSERT(mval->br_blockcount > mval[-1].br_blockcount);
4310                 ASSERT(mval->br_state == mval[-1].br_state);
4311                 mval[-1].br_blockcount = mval->br_blockcount;
4312                 mval[-1].br_state = mval->br_state;
4313         } else if (*n > 0 && mval->br_startblock != DELAYSTARTBLOCK &&
4314                    mval[-1].br_startblock != DELAYSTARTBLOCK &&
4315                    mval[-1].br_startblock != HOLESTARTBLOCK &&
4316                    mval->br_startblock == mval[-1].br_startblock +
4317                                           mval[-1].br_blockcount &&
4318                    ((flags & XFS_BMAPI_IGSTATE) ||
4319                         mval[-1].br_state == mval->br_state)) {
4320                 ASSERT(mval->br_startoff ==
4321                        mval[-1].br_startoff + mval[-1].br_blockcount);
4322                 mval[-1].br_blockcount += mval->br_blockcount;
4323         } else if (*n > 0 &&
4324                    mval->br_startblock == DELAYSTARTBLOCK &&
4325                    mval[-1].br_startblock == DELAYSTARTBLOCK &&
4326                    mval->br_startoff ==
4327                    mval[-1].br_startoff + mval[-1].br_blockcount) {
4328                 mval[-1].br_blockcount += mval->br_blockcount;
4329                 mval[-1].br_state = mval->br_state;
4330         } else if (!((*n == 0) &&
4331                      ((mval->br_startoff + mval->br_blockcount) <=
4332                       obno))) {
4333                 mval++;
4334                 (*n)++;
4335         }
4336         *map = mval;
4337 }
4338
4339 /*
4340  * Map file blocks to filesystem blocks without allocation.
4341  */
4342 int
4343 xfs_bmapi_read(
4344         struct xfs_inode        *ip,
4345         xfs_fileoff_t           bno,
4346         xfs_filblks_t           len,
4347         struct xfs_bmbt_irec    *mval,
4348         int                     *nmap,
4349         int                     flags)
4350 {
4351         struct xfs_mount        *mp = ip->i_mount;
4352         struct xfs_ifork        *ifp;
4353         struct xfs_bmbt_irec    got;
4354         struct xfs_bmbt_irec    prev;
4355         xfs_fileoff_t           obno;
4356         xfs_fileoff_t           end;
4357         xfs_extnum_t            lastx;
4358         int                     error;
4359         int                     eof;
4360         int                     n = 0;
4361         int                     whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4362                                                 XFS_ATTR_FORK : XFS_DATA_FORK;
4363
4364         ASSERT(*nmap >= 1);
4365         ASSERT(!(flags & ~(XFS_BMAPI_ATTRFORK|XFS_BMAPI_ENTIRE|
4366                            XFS_BMAPI_IGSTATE)));
4367
4368         if (unlikely(XFS_TEST_ERROR(
4369             (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4370              XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
4371              mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4372                 XFS_ERROR_REPORT("xfs_bmapi_read", XFS_ERRLEVEL_LOW, mp);
4373                 return XFS_ERROR(EFSCORRUPTED);
4374         }
4375
4376         if (XFS_FORCED_SHUTDOWN(mp))
4377                 return XFS_ERROR(EIO);
4378
4379         XFS_STATS_INC(xs_blk_mapr);
4380
4381         ifp = XFS_IFORK_PTR(ip, whichfork);
4382
4383         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4384                 error = xfs_iread_extents(NULL, ip, whichfork);
4385                 if (error)
4386                         return error;
4387         }
4388
4389         xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got, &prev);
4390         end = bno + len;
4391         obno = bno;
4392
4393         while (bno < end && n < *nmap) {
4394                 /* Reading past eof, act as though there's a hole up to end. */
4395                 if (eof)
4396                         got.br_startoff = end;
4397                 if (got.br_startoff > bno) {
4398                         /* Reading in a hole.  */
4399                         mval->br_startoff = bno;
4400                         mval->br_startblock = HOLESTARTBLOCK;
4401                         mval->br_blockcount =
4402                                 XFS_FILBLKS_MIN(len, got.br_startoff - bno);
4403                         mval->br_state = XFS_EXT_NORM;
4404                         bno += mval->br_blockcount;
4405                         len -= mval->br_blockcount;
4406                         mval++;
4407                         n++;
4408                         continue;
4409                 }
4410
4411                 /* set up the extent map to return. */
4412                 xfs_bmapi_trim_map(mval, &got, &bno, len, obno, end, n, flags);
4413                 xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4414
4415                 /* If we're done, stop now. */
4416                 if (bno >= end || n >= *nmap)
4417                         break;
4418
4419                 /* Else go on to the next record. */
4420                 if (++lastx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t))
4421                         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx), &got);
4422                 else
4423                         eof = 1;
4424         }
4425         *nmap = n;
4426         return 0;
4427 }
4428
4429 STATIC int
4430 xfs_bmapi_reserve_delalloc(
4431         struct xfs_inode        *ip,
4432         xfs_fileoff_t           aoff,
4433         xfs_filblks_t           len,
4434         struct xfs_bmbt_irec    *got,
4435         struct xfs_bmbt_irec    *prev,
4436         xfs_extnum_t            *lastx,
4437         int                     eof)
4438 {
4439         struct xfs_mount        *mp = ip->i_mount;
4440         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
4441         xfs_extlen_t            alen;
4442         xfs_extlen_t            indlen;
4443         char                    rt = XFS_IS_REALTIME_INODE(ip);
4444         xfs_extlen_t            extsz;
4445         int                     error;
4446
4447         alen = XFS_FILBLKS_MIN(len, MAXEXTLEN);
4448         if (!eof)
4449                 alen = XFS_FILBLKS_MIN(alen, got->br_startoff - aoff);
4450
4451         /* Figure out the extent size, adjust alen */
4452         extsz = xfs_get_extsz_hint(ip);
4453         if (extsz) {
4454                 /*
4455                  * Make sure we don't exceed a single extent length when we
4456                  * align the extent by reducing length we are going to
4457                  * allocate by the maximum amount extent size aligment may
4458                  * require.
4459                  */
4460                 alen = XFS_FILBLKS_MIN(len, MAXEXTLEN - (2 * extsz - 1));
4461                 error = xfs_bmap_extsize_align(mp, got, prev, extsz, rt, eof,
4462                                                1, 0, &aoff, &alen);
4463                 ASSERT(!error);
4464         }
4465
4466         if (rt)
4467                 extsz = alen / mp->m_sb.sb_rextsize;
4468
4469         /*
4470          * Make a transaction-less quota reservation for delayed allocation
4471          * blocks.  This number gets adjusted later.  We return if we haven't
4472          * allocated blocks already inside this loop.
4473          */
4474         error = xfs_trans_reserve_quota_nblks(NULL, ip, (long)alen, 0,
4475                         rt ? XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4476         if (error)
4477                 return error;
4478
4479         /*
4480          * Split changing sb for alen and indlen since they could be coming
4481          * from different places.
4482          */
4483         indlen = (xfs_extlen_t)xfs_bmap_worst_indlen(ip, alen);
4484         ASSERT(indlen > 0);
4485
4486         if (rt) {
4487                 error = xfs_mod_incore_sb(mp, XFS_SBS_FREXTENTS,
4488                                           -((int64_t)extsz), 0);
4489         } else {
4490                 error = xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
4491                                                  -((int64_t)alen), 0);
4492         }
4493
4494         if (error)
4495                 goto out_unreserve_quota;
4496
4497         error = xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
4498                                          -((int64_t)indlen), 0);
4499         if (error)
4500                 goto out_unreserve_blocks;
4501
4502
4503         ip->i_delayed_blks += alen;
4504
4505         got->br_startoff = aoff;
4506         got->br_startblock = nullstartblock(indlen);
4507         got->br_blockcount = alen;
4508         got->br_state = XFS_EXT_NORM;
4509         xfs_bmap_add_extent_hole_delay(ip, lastx, got);
4510
4511         /*
4512          * Update our extent pointer, given that xfs_bmap_add_extent_hole_delay
4513          * might have merged it into one of the neighbouring ones.
4514          */
4515         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *lastx), got);
4516
4517         ASSERT(got->br_startoff <= aoff);
4518         ASSERT(got->br_startoff + got->br_blockcount >= aoff + alen);
4519         ASSERT(isnullstartblock(got->br_startblock));
4520         ASSERT(got->br_state == XFS_EXT_NORM);
4521         return 0;
4522
4523 out_unreserve_blocks:
4524         if (rt)
4525                 xfs_mod_incore_sb(mp, XFS_SBS_FREXTENTS, extsz, 0);
4526         else
4527                 xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS, alen, 0);
4528 out_unreserve_quota:
4529         if (XFS_IS_QUOTA_ON(mp))
4530                 xfs_trans_unreserve_quota_nblks(NULL, ip, alen, 0, rt ?
4531                                 XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4532         return error;
4533 }
4534
4535 /*
4536  * Map file blocks to filesystem blocks, adding delayed allocations as needed.
4537  */
4538 int
4539 xfs_bmapi_delay(
4540         struct xfs_inode        *ip,    /* incore inode */
4541         xfs_fileoff_t           bno,    /* starting file offs. mapped */
4542         xfs_filblks_t           len,    /* length to map in file */
4543         struct xfs_bmbt_irec    *mval,  /* output: map values */
4544         int                     *nmap,  /* i/o: mval size/count */
4545         int                     flags)  /* XFS_BMAPI_... */
4546 {
4547         struct xfs_mount        *mp = ip->i_mount;
4548         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
4549         struct xfs_bmbt_irec    got;    /* current file extent record */
4550         struct xfs_bmbt_irec    prev;   /* previous file extent record */
4551         xfs_fileoff_t           obno;   /* old block number (offset) */
4552         xfs_fileoff_t           end;    /* end of mapped file region */
4553         xfs_extnum_t            lastx;  /* last useful extent number */
4554         int                     eof;    /* we've hit the end of extents */
4555         int                     n = 0;  /* current extent index */
4556         int                     error = 0;
4557
4558         ASSERT(*nmap >= 1);
4559         ASSERT(*nmap <= XFS_BMAP_MAX_NMAP);
4560         ASSERT(!(flags & ~XFS_BMAPI_ENTIRE));
4561
4562         if (unlikely(XFS_TEST_ERROR(
4563             (XFS_IFORK_FORMAT(ip, XFS_DATA_FORK) != XFS_DINODE_FMT_EXTENTS &&
4564              XFS_IFORK_FORMAT(ip, XFS_DATA_FORK) != XFS_DINODE_FMT_BTREE),
4565              mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4566                 XFS_ERROR_REPORT("xfs_bmapi_delay", XFS_ERRLEVEL_LOW, mp);
4567                 return XFS_ERROR(EFSCORRUPTED);
4568         }
4569
4570         if (XFS_FORCED_SHUTDOWN(mp))
4571                 return XFS_ERROR(EIO);
4572
4573         XFS_STATS_INC(xs_blk_mapw);
4574
4575         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4576                 error = xfs_iread_extents(NULL, ip, XFS_DATA_FORK);
4577                 if (error)
4578                         return error;
4579         }
4580
4581         xfs_bmap_search_extents(ip, bno, XFS_DATA_FORK, &eof, &lastx, &got, &prev);
4582         end = bno + len;
4583         obno = bno;
4584
4585         while (bno < end && n < *nmap) {
4586                 if (eof || got.br_startoff > bno) {
4587                         error = xfs_bmapi_reserve_delalloc(ip, bno, len, &got,
4588                                                            &prev, &lastx, eof);
4589                         if (error) {
4590                                 if (n == 0) {
4591                                         *nmap = 0;
4592                                         return error;
4593                                 }
4594                                 break;
4595                         }
4596                 }
4597
4598                 /* set up the extent map to return. */
4599                 xfs_bmapi_trim_map(mval, &got, &bno, len, obno, end, n, flags);
4600                 xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4601
4602                 /* If we're done, stop now. */
4603                 if (bno >= end || n >= *nmap)
4604                         break;
4605
4606                 /* Else go on to the next record. */
4607                 prev = got;
4608                 if (++lastx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t))
4609                         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx), &got);
4610                 else
4611                         eof = 1;
4612         }
4613
4614         *nmap = n;
4615         return 0;
4616 }
4617
4618
4619 STATIC int
4620 xfs_bmapi_allocate(
4621         struct xfs_bmalloca     *bma,
4622         int                     flags)
4623 {
4624         struct xfs_mount        *mp = bma->ip->i_mount;
4625         int                     whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4626                                                 XFS_ATTR_FORK : XFS_DATA_FORK;
4627         struct xfs_ifork        *ifp = XFS_IFORK_PTR(bma->ip, whichfork);
4628         int                     tmp_logflags = 0;
4629         int                     error;
4630         int                     rt;
4631
4632         ASSERT(bma->length > 0);
4633
4634         rt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(bma->ip);
4635
4636         /*
4637          * For the wasdelay case, we could also just allocate the stuff asked
4638          * for in this bmap call but that wouldn't be as good.
4639          */
4640         if (bma->wasdel) {
4641                 bma->length = (xfs_extlen_t)bma->got.br_blockcount;
4642                 bma->offset = bma->got.br_startoff;
4643                 if (bma->idx != NULLEXTNUM && bma->idx) {
4644                         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx - 1),
4645                                          &bma->prev);
4646                 }
4647         } else {
4648                 bma->length = XFS_FILBLKS_MIN(bma->length, MAXEXTLEN);
4649                 if (!bma->eof)
4650                         bma->length = XFS_FILBLKS_MIN(bma->length,
4651                                         bma->got.br_startoff - bma->offset);
4652         }
4653
4654         /*
4655          * Indicate if this is the first user data in the file, or just any
4656          * user data.
4657          */
4658         if (!(flags & XFS_BMAPI_METADATA)) {
4659                 bma->userdata = (bma->offset == 0) ?
4660                         XFS_ALLOC_INITIAL_USER_DATA : XFS_ALLOC_USERDATA;
4661         }
4662
4663         bma->minlen = (flags & XFS_BMAPI_CONTIG) ? bma->length : 1;
4664
4665         /*
4666          * Only want to do the alignment at the eof if it is userdata and
4667          * allocation length is larger than a stripe unit.
4668          */
4669         if (mp->m_dalign && bma->length >= mp->m_dalign &&
4670             !(flags & XFS_BMAPI_METADATA) && whichfork == XFS_DATA_FORK) {
4671                 error = xfs_bmap_isaeof(bma, whichfork);
4672                 if (error)
4673                         return error;
4674         }
4675
4676         error = xfs_bmap_alloc(bma);
4677         if (error)
4678                 return error;
4679
4680         if (bma->flist->xbf_low)
4681                 bma->minleft = 0;
4682         if (bma->cur)
4683                 bma->cur->bc_private.b.firstblock = *bma->firstblock;
4684         if (bma->blkno == NULLFSBLOCK)
4685                 return 0;
4686         if ((ifp->if_flags & XFS_IFBROOT) && !bma->cur) {
4687                 bma->cur = xfs_bmbt_init_cursor(mp, bma->tp, bma->ip, whichfork);
4688                 bma->cur->bc_private.b.firstblock = *bma->firstblock;
4689                 bma->cur->bc_private.b.flist = bma->flist;
4690         }
4691         /*
4692          * Bump the number of extents we've allocated
4693          * in this call.
4694          */
4695         bma->nallocs++;
4696
4697         if (bma->cur)
4698                 bma->cur->bc_private.b.flags =
4699                         bma->wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
4700
4701         bma->got.br_startoff = bma->offset;
4702         bma->got.br_startblock = bma->blkno;
4703         bma->got.br_blockcount = bma->length;
4704         bma->got.br_state = XFS_EXT_NORM;
4705
4706         /*
4707          * A wasdelay extent has been initialized, so shouldn't be flagged
4708          * as unwritten.
4709          */
4710         if (!bma->wasdel && (flags & XFS_BMAPI_PREALLOC) &&
4711             xfs_sb_version_hasextflgbit(&mp->m_sb))
4712                 bma->got.br_state = XFS_EXT_UNWRITTEN;
4713
4714         if (bma->wasdel)
4715                 error = xfs_bmap_add_extent_delay_real(bma);
4716         else
4717                 error = xfs_bmap_add_extent_hole_real(bma, whichfork);
4718
4719         bma->logflags |= tmp_logflags;
4720         if (error)
4721                 return error;
4722
4723         /*
4724          * Update our extent pointer, given that xfs_bmap_add_extent_delay_real
4725          * or xfs_bmap_add_extent_hole_real might have merged it into one of
4726          * the neighbouring ones.
4727          */
4728         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx), &bma->got);
4729
4730         ASSERT(bma->got.br_startoff <= bma->offset);
4731         ASSERT(bma->got.br_startoff + bma->got.br_blockcount >=
4732                bma->offset + bma->length);
4733         ASSERT(bma->got.br_state == XFS_EXT_NORM ||
4734                bma->got.br_state == XFS_EXT_UNWRITTEN);
4735         return 0;
4736 }
4737
4738 STATIC int
4739 xfs_bmapi_convert_unwritten(
4740         struct xfs_bmalloca     *bma,
4741         struct xfs_bmbt_irec    *mval,
4742         xfs_filblks_t           len,
4743         int                     flags)
4744 {
4745         int                     whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4746                                                 XFS_ATTR_FORK : XFS_DATA_FORK;
4747         struct xfs_ifork        *ifp = XFS_IFORK_PTR(bma->ip, whichfork);
4748         int                     tmp_logflags = 0;
4749         int                     error;
4750
4751         /* check if we need to do unwritten->real conversion */
4752         if (mval->br_state == XFS_EXT_UNWRITTEN &&
4753             (flags & XFS_BMAPI_PREALLOC))
4754                 return 0;
4755
4756         /* check if we need to do real->unwritten conversion */
4757         if (mval->br_state == XFS_EXT_NORM &&
4758             (flags & (XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT)) !=
4759                         (XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT))
4760                 return 0;
4761
4762         /*
4763          * Modify (by adding) the state flag, if writing.
4764          */
4765         ASSERT(mval->br_blockcount <= len);
4766         if ((ifp->if_flags & XFS_IFBROOT) && !bma->cur) {
4767                 bma->cur = xfs_bmbt_init_cursor(bma->ip->i_mount, bma->tp,
4768                                         bma->ip, whichfork);
4769                 bma->cur->bc_private.b.firstblock = *bma->firstblock;
4770                 bma->cur->bc_private.b.flist = bma->flist;
4771         }
4772         mval->br_state = (mval->br_state == XFS_EXT_UNWRITTEN)
4773                                 ? XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
4774
4775         error = xfs_bmap_add_extent_unwritten_real(bma->tp, bma->ip, &bma->idx,
4776                         &bma->cur, mval, bma->firstblock, bma->flist,
4777                         &tmp_logflags);
4778         bma->logflags |= tmp_logflags;
4779         if (error)
4780                 return error;
4781
4782         /*
4783          * Update our extent pointer, given that
4784          * xfs_bmap_add_extent_unwritten_real might have merged it into one
4785          * of the neighbouring ones.
4786          */
4787         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx), &bma->got);
4788
4789         /*
4790          * We may have combined previously unwritten space with written space,
4791          * so generate another request.
4792          */
4793         if (mval->br_blockcount < len)
4794                 return EAGAIN;
4795         return 0;
4796 }
4797
4798 /*
4799  * Map file blocks to filesystem blocks, and allocate blocks or convert the
4800  * extent state if necessary.  Details behaviour is controlled by the flags
4801  * parameter.  Only allocates blocks from a single allocation group, to avoid
4802  * locking problems.
4803  *
4804  * The returned value in "firstblock" from the first call in a transaction
4805  * must be remembered and presented to subsequent calls in "firstblock".
4806  * An upper bound for the number of blocks to be allocated is supplied to
4807  * the first call in "total"; if no allocation group has that many free
4808  * blocks then the call will fail (return NULLFSBLOCK in "firstblock").
4809  */
4810 int
4811 xfs_bmapi_write(
4812         struct xfs_trans        *tp,            /* transaction pointer */
4813         struct xfs_inode        *ip,            /* incore inode */
4814         xfs_fileoff_t           bno,            /* starting file offs. mapped */
4815         xfs_filblks_t           len,            /* length to map in file */
4816         int                     flags,          /* XFS_BMAPI_... */
4817         xfs_fsblock_t           *firstblock,    /* first allocated block
4818                                                    controls a.g. for allocs */
4819         xfs_extlen_t            total,          /* total blocks needed */
4820         struct xfs_bmbt_irec    *mval,          /* output: map values */
4821         int                     *nmap,          /* i/o: mval size/count */
4822         struct xfs_bmap_free    *flist)         /* i/o: list extents to free */
4823 {
4824         struct xfs_mount        *mp = ip->i_mount;
4825         struct xfs_ifork        *ifp;
4826         struct xfs_bmalloca     bma = { 0 };    /* args for xfs_bmap_alloc */
4827         xfs_fileoff_t           end;            /* end of mapped file region */
4828         int                     eof;            /* after the end of extents */
4829         int                     error;          /* error return */
4830         int                     n;              /* current extent index */
4831         xfs_fileoff_t           obno;           /* old block number (offset) */
4832         int                     whichfork;      /* data or attr fork */
4833         char                    inhole;         /* current location is hole in file */
4834         char                    wasdelay;       /* old extent was delayed */
4835
4836 #ifdef DEBUG
4837         xfs_fileoff_t           orig_bno;       /* original block number value */
4838         int                     orig_flags;     /* original flags arg value */
4839         xfs_filblks_t           orig_len;       /* original value of len arg */
4840         struct xfs_bmbt_irec    *orig_mval;     /* original value of mval */
4841         int                     orig_nmap;      /* original value of *nmap */
4842
4843         orig_bno = bno;
4844         orig_len = len;
4845         orig_flags = flags;
4846         orig_mval = mval;
4847         orig_nmap = *nmap;
4848 #endif
4849
4850         ASSERT(*nmap >= 1);
4851         ASSERT(*nmap <= XFS_BMAP_MAX_NMAP);
4852         ASSERT(!(flags & XFS_BMAPI_IGSTATE));
4853         ASSERT(tp != NULL);
4854         ASSERT(len > 0);
4855
4856         whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4857                 XFS_ATTR_FORK : XFS_DATA_FORK;
4858
4859         if (unlikely(XFS_TEST_ERROR(
4860             (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4861              XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
4862              XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL),
4863              mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4864                 XFS_ERROR_REPORT("xfs_bmapi_write", XFS_ERRLEVEL_LOW, mp);
4865                 return XFS_ERROR(EFSCORRUPTED);
4866         }
4867
4868         if (XFS_FORCED_SHUTDOWN(mp))
4869                 return XFS_ERROR(EIO);
4870
4871         ifp = XFS_IFORK_PTR(ip, whichfork);
4872
4873         XFS_STATS_INC(xs_blk_mapw);
4874
4875         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
4876                 error = xfs_bmap_local_to_extents(tp, ip, firstblock, total,
4877                                                   &bma.logflags, whichfork);
4878                 if (error)
4879                         goto error0;
4880         }
4881
4882         if (*firstblock == NULLFSBLOCK) {
4883                 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE)
4884                         bma.minleft = be16_to_cpu(ifp->if_broot->bb_level) + 1;
4885                 else
4886                         bma.minleft = 1;
4887         } else {
4888                 bma.minleft = 0;
4889         }
4890
4891         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4892                 error = xfs_iread_extents(tp, ip, whichfork);
4893                 if (error)
4894                         goto error0;
4895         }
4896
4897         xfs_bmap_search_extents(ip, bno, whichfork, &eof, &bma.idx, &bma.got,
4898                                 &bma.prev);
4899         n = 0;
4900         end = bno + len;
4901         obno = bno;
4902
4903         bma.tp = tp;
4904         bma.ip = ip;
4905         bma.total = total;
4906         bma.userdata = 0;
4907         bma.flist = flist;
4908         bma.firstblock = firstblock;
4909
4910         while (bno < end && n < *nmap) {
4911                 inhole = eof || bma.got.br_startoff > bno;
4912                 wasdelay = !inhole && isnullstartblock(bma.got.br_startblock);
4913
4914                 /*
4915                  * First, deal with the hole before the allocated space
4916                  * that we found, if any.
4917                  */
4918                 if (inhole || wasdelay) {
4919                         bma.eof = eof;
4920                         bma.conv = !!(flags & XFS_BMAPI_CONVERT);
4921                         bma.wasdel = wasdelay;
4922                         bma.offset = bno;
4923
4924                         /*
4925                          * There's a 32/64 bit type mismatch between the
4926                          * allocation length request (which can be 64 bits in
4927                          * length) and the bma length request, which is
4928                          * xfs_extlen_t and therefore 32 bits. Hence we have to
4929                          * check for 32-bit overflows and handle them here.
4930                          */
4931                         if (len > (xfs_filblks_t)MAXEXTLEN)
4932                                 bma.length = MAXEXTLEN;
4933                         else
4934                                 bma.length = len;
4935
4936                         ASSERT(len > 0);
4937                         ASSERT(bma.length > 0);
4938                         error = xfs_bmapi_allocate(&bma, flags);
4939                         if (error)
4940                                 goto error0;
4941                         if (bma.blkno == NULLFSBLOCK)
4942                                 break;
4943                 }
4944
4945                 /* Deal with the allocated space we found.  */
4946                 xfs_bmapi_trim_map(mval, &bma.got, &bno, len, obno,
4947                                                         end, n, flags);
4948
4949                 /* Execute unwritten extent conversion if necessary */
4950                 error = xfs_bmapi_convert_unwritten(&bma, mval, len, flags);
4951                 if (error == EAGAIN)
4952                         continue;
4953                 if (error)
4954                         goto error0;
4955
4956                 /* update the extent map to return */
4957                 xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4958
4959                 /*
4960                  * If we're done, stop now.  Stop when we've allocated
4961                  * XFS_BMAP_MAX_NMAP extents no matter what.  Otherwise
4962                  * the transaction may get too big.
4963                  */
4964                 if (bno >= end || n >= *nmap || bma.nallocs >= *nmap)
4965                         break;
4966
4967                 /* Else go on to the next record. */
4968                 bma.prev = bma.got;
4969                 if (++bma.idx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t)) {
4970                         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma.idx),
4971                                          &bma.got);
4972                 } else
4973                         eof = 1;
4974         }
4975         *nmap = n;
4976
4977         /*
4978          * Transform from btree to extents, give it cur.
4979          */
4980         if (xfs_bmap_wants_extents(ip, whichfork)) {
4981                 int             tmp_logflags = 0;
4982
4983                 ASSERT(bma.cur);
4984                 error = xfs_bmap_btree_to_extents(tp, ip, bma.cur,
4985                         &tmp_logflags, whichfork);
4986                 bma.logflags |= tmp_logflags;
4987                 if (error)
4988                         goto error0;
4989         }
4990
4991         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE ||
4992                XFS_IFORK_NEXTENTS(ip, whichfork) >
4993                 XFS_IFORK_MAXEXT(ip, whichfork));
4994         error = 0;
4995 error0:
4996         /*
4997          * Log everything.  Do this after conversion, there's no point in
4998          * logging the extent records if we've converted to btree format.
4999          */
5000         if ((bma.logflags & xfs_ilog_fext(whichfork)) &&
5001             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
5002                 bma.logflags &= ~xfs_ilog_fext(whichfork);
5003         else if ((bma.logflags & xfs_ilog_fbroot(whichfork)) &&
5004                  XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
5005                 bma.logflags &= ~xfs_ilog_fbroot(whichfork);
5006         /*
5007          * Log whatever the flags say, even if error.  Otherwise we might miss
5008          * detecting a case where the data is changed, there's an error,
5009          * and it's not logged so we don't shutdown when we should.
5010          */
5011         if (bma.logflags)
5012                 xfs_trans_log_inode(tp, ip, bma.logflags);
5013
5014         if (bma.cur) {
5015                 if (!error) {
5016                         ASSERT(*firstblock == NULLFSBLOCK ||
5017                                XFS_FSB_TO_AGNO(mp, *firstblock) ==
5018                                XFS_FSB_TO_AGNO(mp,
5019                                        bma.cur->bc_private.b.firstblock) ||
5020                                (flist->xbf_low &&
5021                                 XFS_FSB_TO_AGNO(mp, *firstblock) <
5022                                 XFS_FSB_TO_AGNO(mp,
5023                                         bma.cur->bc_private.b.firstblock)));
5024                         *firstblock = bma.cur->bc_private.b.firstblock;
5025                 }
5026                 xfs_btree_del_cursor(bma.cur,
5027                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5028         }
5029         if (!error)
5030                 xfs_bmap_validate_ret(orig_bno, orig_len, orig_flags, orig_mval,
5031                         orig_nmap, *nmap);
5032         return error;
5033 }
5034
5035 /*
5036  * Unmap (remove) blocks from a file.
5037  * If nexts is nonzero then the number of extents to remove is limited to
5038  * that value.  If not all extents in the block range can be removed then
5039  * *done is set.
5040  */
5041 int                                             /* error */
5042 xfs_bunmapi(
5043         xfs_trans_t             *tp,            /* transaction pointer */
5044         struct xfs_inode        *ip,            /* incore inode */
5045         xfs_fileoff_t           bno,            /* starting offset to unmap */
5046         xfs_filblks_t           len,            /* length to unmap in file */
5047         int                     flags,          /* misc flags */
5048         xfs_extnum_t            nexts,          /* number of extents max */
5049         xfs_fsblock_t           *firstblock,    /* first allocated block
5050                                                    controls a.g. for allocs */
5051         xfs_bmap_free_t         *flist,         /* i/o: list extents to free */
5052         int                     *done)          /* set if not done yet */
5053 {
5054         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
5055         xfs_bmbt_irec_t         del;            /* extent being deleted */
5056         int                     eof;            /* is deleting at eof */
5057         xfs_bmbt_rec_host_t     *ep;            /* extent record pointer */
5058         int                     error;          /* error return value */
5059         xfs_extnum_t            extno;          /* extent number in list */
5060         xfs_bmbt_irec_t         got;            /* current extent record */
5061         xfs_ifork_t             *ifp;           /* inode fork pointer */
5062         int                     isrt;           /* freeing in rt area */
5063         xfs_extnum_t            lastx;          /* last extent index used */
5064         int                     logflags;       /* transaction logging flags */
5065         xfs_extlen_t            mod;            /* rt extent offset */
5066         xfs_mount_t             *mp;            /* mount structure */
5067         xfs_extnum_t            nextents;       /* number of file extents */
5068         xfs_bmbt_irec_t         prev;           /* previous extent record */
5069         xfs_fileoff_t           start;          /* first file offset deleted */
5070         int                     tmp_logflags;   /* partial logging flags */
5071         int                     wasdel;         /* was a delayed alloc extent */
5072         int                     whichfork;      /* data or attribute fork */
5073         xfs_fsblock_t           sum;
5074
5075         trace_xfs_bunmap(ip, bno, len, flags, _RET_IP_);
5076
5077         whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
5078                 XFS_ATTR_FORK : XFS_DATA_FORK;
5079         ifp = XFS_IFORK_PTR(ip, whichfork);
5080         if (unlikely(
5081             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
5082             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
5083                 XFS_ERROR_REPORT("xfs_bunmapi", XFS_ERRLEVEL_LOW,
5084                                  ip->i_mount);
5085                 return XFS_ERROR(EFSCORRUPTED);
5086         }
5087         mp = ip->i_mount;
5088         if (XFS_FORCED_SHUTDOWN(mp))
5089                 return XFS_ERROR(EIO);
5090
5091         ASSERT(len > 0);
5092         ASSERT(nexts >= 0);
5093
5094         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5095             (error = xfs_iread_extents(tp, ip, whichfork)))
5096                 return error;
5097         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5098         if (nextents == 0) {
5099                 *done = 1;
5100                 return 0;
5101         }
5102         XFS_STATS_INC(xs_blk_unmap);
5103         isrt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
5104         start = bno;
5105         bno = start + len - 1;
5106         ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
5107                 &prev);
5108
5109         /*
5110          * Check to see if the given block number is past the end of the
5111          * file, back up to the last block if so...
5112          */
5113         if (eof) {
5114                 ep = xfs_iext_get_ext(ifp, --lastx);
5115                 xfs_bmbt_get_all(ep, &got);
5116                 bno = got.br_startoff + got.br_blockcount - 1;
5117         }
5118         logflags = 0;
5119         if (ifp->if_flags & XFS_IFBROOT) {
5120                 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
5121                 cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5122                 cur->bc_private.b.firstblock = *firstblock;
5123                 cur->bc_private.b.flist = flist;
5124                 cur->bc_private.b.flags = 0;
5125         } else
5126                 cur = NULL;
5127         extno = 0;
5128         while (bno != (xfs_fileoff_t)-1 && bno >= start && lastx >= 0 &&
5129                (nexts == 0 || extno < nexts)) {
5130                 /*
5131                  * Is the found extent after a hole in which bno lives?
5132                  * Just back up to the previous extent, if so.
5133                  */
5134                 if (got.br_startoff > bno) {
5135                         if (--lastx < 0)
5136                                 break;
5137                         ep = xfs_iext_get_ext(ifp, lastx);
5138                         xfs_bmbt_get_all(ep, &got);
5139                 }
5140                 /*
5141                  * Is the last block of this extent before the range
5142                  * we're supposed to delete?  If so, we're done.
5143                  */
5144                 bno = XFS_FILEOFF_MIN(bno,
5145                         got.br_startoff + got.br_blockcount - 1);
5146                 if (bno < start)
5147                         break;
5148                 /*
5149                  * Then deal with the (possibly delayed) allocated space
5150                  * we found.
5151                  */
5152                 ASSERT(ep != NULL);
5153                 del = got;
5154                 wasdel = isnullstartblock(del.br_startblock);
5155                 if (got.br_startoff < start) {
5156                         del.br_startoff = start;
5157                         del.br_blockcount -= start - got.br_startoff;
5158                         if (!wasdel)
5159                                 del.br_startblock += start - got.br_startoff;
5160                 }
5161                 if (del.br_startoff + del.br_blockcount > bno + 1)
5162                         del.br_blockcount = bno + 1 - del.br_startoff;
5163                 sum = del.br_startblock + del.br_blockcount;
5164                 if (isrt &&
5165                     (mod = do_mod(sum, mp->m_sb.sb_rextsize))) {
5166                         /*
5167                          * Realtime extent not lined up at the end.
5168                          * The extent could have been split into written
5169                          * and unwritten pieces, or we could just be
5170                          * unmapping part of it.  But we can't really
5171                          * get rid of part of a realtime extent.
5172                          */
5173                         if (del.br_state == XFS_EXT_UNWRITTEN ||
5174                             !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5175                                 /*
5176                                  * This piece is unwritten, or we're not
5177                                  * using unwritten extents.  Skip over it.
5178                                  */
5179                                 ASSERT(bno >= mod);
5180                                 bno -= mod > del.br_blockcount ?
5181                                         del.br_blockcount : mod;
5182                                 if (bno < got.br_startoff) {
5183                                         if (--lastx >= 0)
5184                                                 xfs_bmbt_get_all(xfs_iext_get_ext(
5185                                                         ifp, lastx), &got);
5186                                 }
5187                                 continue;
5188                         }
5189                         /*
5190                          * It's written, turn it unwritten.
5191                          * This is better than zeroing it.
5192                          */
5193                         ASSERT(del.br_state == XFS_EXT_NORM);
5194                         ASSERT(xfs_trans_get_block_res(tp) > 0);
5195                         /*
5196                          * If this spans a realtime extent boundary,
5197                          * chop it back to the start of the one we end at.
5198                          */
5199                         if (del.br_blockcount > mod) {
5200                                 del.br_startoff += del.br_blockcount - mod;
5201                                 del.br_startblock += del.br_blockcount - mod;
5202                                 del.br_blockcount = mod;
5203                         }
5204                         del.br_state = XFS_EXT_UNWRITTEN;
5205                         error = xfs_bmap_add_extent_unwritten_real(tp, ip,
5206                                         &lastx, &cur, &del, firstblock, flist,
5207                                         &logflags);
5208                         if (error)
5209                                 goto error0;
5210                         goto nodelete;
5211                 }
5212                 if (isrt && (mod = do_mod(del.br_startblock, mp->m_sb.sb_rextsize))) {
5213                         /*
5214                          * Realtime extent is lined up at the end but not
5215                          * at the front.  We'll get rid of full extents if
5216                          * we can.
5217                          */
5218                         mod = mp->m_sb.sb_rextsize - mod;
5219                         if (del.br_blockcount > mod) {
5220                                 del.br_blockcount -= mod;
5221                                 del.br_startoff += mod;
5222                                 del.br_startblock += mod;
5223                         } else if ((del.br_startoff == start &&
5224                                     (del.br_state == XFS_EXT_UNWRITTEN ||
5225                                      xfs_trans_get_block_res(tp) == 0)) ||
5226                                    !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5227                                 /*
5228                                  * Can't make it unwritten.  There isn't
5229                                  * a full extent here so just skip it.
5230                                  */
5231                                 ASSERT(bno >= del.br_blockcount);
5232                                 bno -= del.br_blockcount;
5233                                 if (got.br_startoff > bno) {
5234                                         if (--lastx >= 0) {
5235                                                 ep = xfs_iext_get_ext(ifp,
5236                                                                       lastx);
5237                                                 xfs_bmbt_get_all(ep, &got);
5238                                         }
5239                                 }
5240                                 continue;
5241                         } else if (del.br_state == XFS_EXT_UNWRITTEN) {
5242                                 /*
5243                                  * This one is already unwritten.
5244                                  * It must have a written left neighbor.
5245                                  * Unwrite the killed part of that one and
5246                                  * try again.
5247                                  */
5248                                 ASSERT(lastx > 0);
5249                                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp,
5250                                                 lastx - 1), &prev);
5251                                 ASSERT(prev.br_state == XFS_EXT_NORM);
5252                                 ASSERT(!isnullstartblock(prev.br_startblock));
5253                                 ASSERT(del.br_startblock ==
5254                                        prev.br_startblock + prev.br_blockcount);
5255                                 if (prev.br_startoff < start) {
5256                                         mod = start - prev.br_startoff;
5257                                         prev.br_blockcount -= mod;
5258                                         prev.br_startblock += mod;
5259                                         prev.br_startoff = start;
5260                                 }
5261                                 prev.br_state = XFS_EXT_UNWRITTEN;
5262                                 lastx--;
5263                                 error = xfs_bmap_add_extent_unwritten_real(tp,
5264                                                 ip, &lastx, &cur, &prev,
5265                                                 firstblock, flist, &logflags);
5266                                 if (error)
5267                                         goto error0;
5268                                 goto nodelete;
5269                         } else {
5270                                 ASSERT(del.br_state == XFS_EXT_NORM);
5271                                 del.br_state = XFS_EXT_UNWRITTEN;
5272                                 error = xfs_bmap_add_extent_unwritten_real(tp,
5273                                                 ip, &lastx, &cur, &del,
5274                                                 firstblock, flist, &logflags);
5275                                 if (error)
5276                                         goto error0;
5277                                 goto nodelete;
5278                         }
5279                 }
5280                 if (wasdel) {
5281                         ASSERT(startblockval(del.br_startblock) > 0);
5282                         /* Update realtime/data freespace, unreserve quota */
5283                         if (isrt) {
5284                                 xfs_filblks_t rtexts;
5285
5286                                 rtexts = XFS_FSB_TO_B(mp, del.br_blockcount);
5287                                 do_div(rtexts, mp->m_sb.sb_rextsize);
5288                                 xfs_mod_incore_sb(mp, XFS_SBS_FREXTENTS,
5289                                                 (int64_t)rtexts, 0);
5290                                 (void)xfs_trans_reserve_quota_nblks(NULL,
5291                                         ip, -((long)del.br_blockcount), 0,
5292                                         XFS_QMOPT_RES_RTBLKS);
5293                         } else {
5294                                 xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
5295                                                 (int64_t)del.br_blockcount, 0);
5296                                 (void)xfs_trans_reserve_quota_nblks(NULL,
5297                                         ip, -((long)del.br_blockcount), 0,
5298                                         XFS_QMOPT_RES_REGBLKS);
5299                         }
5300                         ip->i_delayed_blks -= del.br_blockcount;
5301                         if (cur)
5302                                 cur->bc_private.b.flags |=
5303                                         XFS_BTCUR_BPRV_WASDEL;
5304                 } else if (cur)
5305                         cur->bc_private.b.flags &= ~XFS_BTCUR_BPRV_WASDEL;
5306                 /*
5307                  * If it's the case where the directory code is running
5308                  * with no block reservation, and the deleted block is in
5309                  * the middle of its extent, and the resulting insert
5310                  * of an extent would cause transformation to btree format,
5311                  * then reject it.  The calling code will then swap
5312                  * blocks around instead.
5313                  * We have to do this now, rather than waiting for the
5314                  * conversion to btree format, since the transaction
5315                  * will be dirty.
5316                  */
5317                 if (!wasdel && xfs_trans_get_block_res(tp) == 0 &&
5318                     XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
5319                     XFS_IFORK_NEXTENTS(ip, whichfork) >= /* Note the >= */
5320                         XFS_IFORK_MAXEXT(ip, whichfork) &&
5321                     del.br_startoff > got.br_startoff &&
5322                     del.br_startoff + del.br_blockcount <
5323                     got.br_startoff + got.br_blockcount) {
5324                         error = XFS_ERROR(ENOSPC);
5325                         goto error0;
5326                 }
5327                 error = xfs_bmap_del_extent(ip, tp, &lastx, flist, cur, &del,
5328                                 &tmp_logflags, whichfork);
5329                 logflags |= tmp_logflags;
5330                 if (error)
5331                         goto error0;
5332                 bno = del.br_startoff - 1;
5333 nodelete:
5334                 /*
5335                  * If not done go on to the next (previous) record.
5336                  */
5337                 if (bno != (xfs_fileoff_t)-1 && bno >= start) {
5338                         if (lastx >= 0) {
5339                                 ep = xfs_iext_get_ext(ifp, lastx);
5340                                 if (xfs_bmbt_get_startoff(ep) > bno) {
5341                                         if (--lastx >= 0)
5342                                                 ep = xfs_iext_get_ext(ifp,
5343                                                                       lastx);
5344                                 }
5345                                 xfs_bmbt_get_all(ep, &got);
5346                         }
5347                         extno++;
5348                 }
5349         }
5350         *done = bno == (xfs_fileoff_t)-1 || bno < start || lastx < 0;
5351
5352         /*
5353          * Convert to a btree if necessary.
5354          */
5355         if (xfs_bmap_needs_btree(ip, whichfork)) {
5356                 ASSERT(cur == NULL);
5357                 error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist,
5358                         &cur, 0, &tmp_logflags, whichfork);
5359                 logflags |= tmp_logflags;
5360                 if (error)
5361                         goto error0;
5362         }
5363         /*
5364          * transform from btree to extents, give it cur
5365          */
5366         else if (xfs_bmap_wants_extents(ip, whichfork)) {
5367                 ASSERT(cur != NULL);
5368                 error = xfs_bmap_btree_to_extents(tp, ip, cur, &tmp_logflags,
5369                         whichfork);
5370                 logflags |= tmp_logflags;
5371                 if (error)
5372                         goto error0;
5373         }
5374         /*
5375          * transform from extents to local?
5376          */
5377         error = 0;
5378 error0:
5379         /*
5380          * Log everything.  Do this after conversion, there's no point in
5381          * logging the extent records if we've converted to btree format.
5382          */
5383         if ((logflags & xfs_ilog_fext(whichfork)) &&
5384             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
5385                 logflags &= ~xfs_ilog_fext(whichfork);
5386         else if ((logflags & xfs_ilog_fbroot(whichfork)) &&
5387                  XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
5388                 logflags &= ~xfs_ilog_fbroot(whichfork);
5389         /*
5390          * Log inode even in the error case, if the transaction
5391          * is dirty we'll need to shut down the filesystem.
5392          */
5393         if (logflags)
5394                 xfs_trans_log_inode(tp, ip, logflags);
5395         if (cur) {
5396                 if (!error) {
5397                         *firstblock = cur->bc_private.b.firstblock;
5398                         cur->bc_private.b.allocated = 0;
5399                 }
5400                 xfs_btree_del_cursor(cur,
5401                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5402         }
5403         return error;
5404 }
5405
5406 /*
5407  * returns 1 for success, 0 if we failed to map the extent.
5408  */
5409 STATIC int
5410 xfs_getbmapx_fix_eof_hole(
5411         xfs_inode_t             *ip,            /* xfs incore inode pointer */
5412         struct getbmapx         *out,           /* output structure */
5413         int                     prealloced,     /* this is a file with
5414                                                  * preallocated data space */
5415         __int64_t               end,            /* last block requested */
5416         xfs_fsblock_t           startblock)
5417 {
5418         __int64_t               fixlen;
5419         xfs_mount_t             *mp;            /* file system mount point */
5420         xfs_ifork_t             *ifp;           /* inode fork pointer */
5421         xfs_extnum_t            lastx;          /* last extent pointer */
5422         xfs_fileoff_t           fileblock;
5423
5424         if (startblock == HOLESTARTBLOCK) {
5425                 mp = ip->i_mount;
5426                 out->bmv_block = -1;
5427                 fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, XFS_ISIZE(ip)));
5428                 fixlen -= out->bmv_offset;
5429                 if (prealloced && out->bmv_offset + out->bmv_length == end) {
5430                         /* Came to hole at EOF. Trim it. */
5431                         if (fixlen <= 0)
5432                                 return 0;
5433                         out->bmv_length = fixlen;
5434                 }
5435         } else {
5436                 if (startblock == DELAYSTARTBLOCK)
5437                         out->bmv_block = -2;
5438                 else
5439                         out->bmv_block = xfs_fsb_to_db(ip, startblock);
5440                 fileblock = XFS_BB_TO_FSB(ip->i_mount, out->bmv_offset);
5441                 ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
5442                 if (xfs_iext_bno_to_ext(ifp, fileblock, &lastx) &&
5443                    (lastx == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t))-1))
5444                         out->bmv_oflags |= BMV_OF_LAST;
5445         }
5446
5447         return 1;
5448 }
5449
5450 /*
5451  * Get inode's extents as described in bmv, and format for output.
5452  * Calls formatter to fill the user's buffer until all extents
5453  * are mapped, until the passed-in bmv->bmv_count slots have
5454  * been filled, or until the formatter short-circuits the loop,
5455  * if it is tracking filled-in extents on its own.
5456  */
5457 int                                             /* error code */
5458 xfs_getbmap(
5459         xfs_inode_t             *ip,
5460         struct getbmapx         *bmv,           /* user bmap structure */
5461         xfs_bmap_format_t       formatter,      /* format to user */
5462         void                    *arg)           /* formatter arg */
5463 {
5464         __int64_t               bmvend;         /* last block requested */
5465         int                     error = 0;      /* return value */
5466         __int64_t               fixlen;         /* length for -1 case */
5467         int                     i;              /* extent number */
5468         int                     lock;           /* lock state */
5469         xfs_bmbt_irec_t         *map;           /* buffer for user's data */
5470         xfs_mount_t             *mp;            /* file system mount point */
5471         int                     nex;            /* # of user extents can do */
5472         int                     nexleft;        /* # of user extents left */
5473         int                     subnex;         /* # of bmapi's can do */
5474         int                     nmap;           /* number of map entries */
5475         struct getbmapx         *out;           /* output structure */
5476         int                     whichfork;      /* data or attr fork */
5477         int                     prealloced;     /* this is a file with
5478                                                  * preallocated data space */
5479         int                     iflags;         /* interface flags */
5480         int                     bmapi_flags;    /* flags for xfs_bmapi */
5481         int                     cur_ext = 0;
5482
5483         mp = ip->i_mount;
5484         iflags = bmv->bmv_iflags;
5485         whichfork = iflags & BMV_IF_ATTRFORK ? XFS_ATTR_FORK : XFS_DATA_FORK;
5486
5487         if (whichfork == XFS_ATTR_FORK) {
5488                 if (XFS_IFORK_Q(ip)) {
5489                         if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS &&
5490                             ip->i_d.di_aformat != XFS_DINODE_FMT_BTREE &&
5491                             ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)
5492                                 return XFS_ERROR(EINVAL);
5493                 } else if (unlikely(
5494                            ip->i_d.di_aformat != 0 &&
5495                            ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS)) {
5496                         XFS_ERROR_REPORT("xfs_getbmap", XFS_ERRLEVEL_LOW,
5497                                          ip->i_mount);
5498                         return XFS_ERROR(EFSCORRUPTED);
5499                 }
5500
5501                 prealloced = 0;
5502                 fixlen = 1LL << 32;
5503         } else {
5504                 if (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS &&
5505                     ip->i_d.di_format != XFS_DINODE_FMT_BTREE &&
5506                     ip->i_d.di_format != XFS_DINODE_FMT_LOCAL)
5507                         return XFS_ERROR(EINVAL);
5508
5509                 if (xfs_get_extsz_hint(ip) ||
5510                     ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC|XFS_DIFLAG_APPEND)){
5511                         prealloced = 1;
5512                         fixlen = XFS_MAXIOFFSET(mp);
5513                 } else {
5514                         prealloced = 0;
5515                         fixlen = XFS_ISIZE(ip);
5516                 }
5517         }
5518
5519         if (bmv->bmv_length == -1) {
5520                 fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, fixlen));
5521                 bmv->bmv_length =
5522                         max_t(__int64_t, fixlen - bmv->bmv_offset, 0);
5523         } else if (bmv->bmv_length == 0) {
5524                 bmv->bmv_entries = 0;
5525                 return 0;
5526         } else if (bmv->bmv_length < 0) {
5527                 return XFS_ERROR(EINVAL);
5528         }
5529
5530         nex = bmv->bmv_count - 1;
5531         if (nex <= 0)
5532                 return XFS_ERROR(EINVAL);
5533         bmvend = bmv->bmv_offset + bmv->bmv_length;
5534
5535
5536         if (bmv->bmv_count > ULONG_MAX / sizeof(struct getbmapx))
5537                 return XFS_ERROR(ENOMEM);
5538         out = kmem_zalloc(bmv->bmv_count * sizeof(struct getbmapx), KM_MAYFAIL);
5539         if (!out) {
5540                 out = kmem_zalloc_large(bmv->bmv_count *
5541                                         sizeof(struct getbmapx));
5542                 if (!out)
5543                         return XFS_ERROR(ENOMEM);
5544         }
5545
5546         xfs_ilock(ip, XFS_IOLOCK_SHARED);
5547         if (whichfork == XFS_DATA_FORK && !(iflags & BMV_IF_DELALLOC)) {
5548                 if (ip->i_delayed_blks || XFS_ISIZE(ip) > ip->i_d.di_size) {
5549                         error = xfs_flush_pages(ip, 0, -1, 0, FI_REMAPF);
5550                         if (error)
5551                                 goto out_unlock_iolock;
5552                 }
5553                 /*
5554                  * even after flushing the inode, there can still be delalloc
5555                  * blocks on the inode beyond EOF due to speculative
5556                  * preallocation. These are not removed until the release
5557                  * function is called or the inode is inactivated. Hence we
5558                  * cannot assert here that ip->i_delayed_blks == 0.
5559                  */
5560         }
5561
5562         lock = xfs_ilock_map_shared(ip);
5563
5564         /*
5565          * Don't let nex be bigger than the number of extents
5566          * we can have assuming alternating holes and real extents.
5567          */
5568         if (nex > XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1)
5569                 nex = XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1;
5570
5571         bmapi_flags = xfs_bmapi_aflag(whichfork);
5572         if (!(iflags & BMV_IF_PREALLOC))
5573                 bmapi_flags |= XFS_BMAPI_IGSTATE;
5574
5575         /*
5576          * Allocate enough space to handle "subnex" maps at a time.
5577          */
5578         error = ENOMEM;
5579         subnex = 16;
5580         map = kmem_alloc(subnex * sizeof(*map), KM_MAYFAIL | KM_NOFS);
5581         if (!map)
5582                 goto out_unlock_ilock;
5583
5584         bmv->bmv_entries = 0;
5585
5586         if (XFS_IFORK_NEXTENTS(ip, whichfork) == 0 &&
5587             (whichfork == XFS_ATTR_FORK || !(iflags & BMV_IF_DELALLOC))) {
5588                 error = 0;
5589                 goto out_free_map;
5590         }
5591
5592         nexleft = nex;
5593
5594         do {
5595                 nmap = (nexleft > subnex) ? subnex : nexleft;
5596                 error = xfs_bmapi_read(ip, XFS_BB_TO_FSBT(mp, bmv->bmv_offset),
5597                                        XFS_BB_TO_FSB(mp, bmv->bmv_length),
5598                                        map, &nmap, bmapi_flags);
5599                 if (error)
5600                         goto out_free_map;
5601                 ASSERT(nmap <= subnex);
5602
5603                 for (i = 0; i < nmap && nexleft && bmv->bmv_length; i++) {
5604                         out[cur_ext].bmv_oflags = 0;
5605                         if (map[i].br_state == XFS_EXT_UNWRITTEN)
5606                                 out[cur_ext].bmv_oflags |= BMV_OF_PREALLOC;
5607                         else if (map[i].br_startblock == DELAYSTARTBLOCK)
5608                                 out[cur_ext].bmv_oflags |= BMV_OF_DELALLOC;
5609                         out[cur_ext].bmv_offset =
5610                                 XFS_FSB_TO_BB(mp, map[i].br_startoff);
5611                         out[cur_ext].bmv_length =
5612                                 XFS_FSB_TO_BB(mp, map[i].br_blockcount);
5613                         out[cur_ext].bmv_unused1 = 0;
5614                         out[cur_ext].bmv_unused2 = 0;
5615                         ASSERT(((iflags & BMV_IF_DELALLOC) != 0) ||
5616                               (map[i].br_startblock != DELAYSTARTBLOCK));
5617                         if (map[i].br_startblock == HOLESTARTBLOCK &&
5618                             whichfork == XFS_ATTR_FORK) {
5619                                 /* came to the end of attribute fork */
5620                                 out[cur_ext].bmv_oflags |= BMV_OF_LAST;
5621                                 goto out_free_map;
5622                         }
5623
5624                         if (!xfs_getbmapx_fix_eof_hole(ip, &out[cur_ext],
5625                                         prealloced, bmvend,
5626                                         map[i].br_startblock))
5627                                 goto out_free_map;
5628
5629                         bmv->bmv_offset =
5630                                 out[cur_ext].bmv_offset +
5631                                 out[cur_ext].bmv_length;
5632                         bmv->bmv_length =
5633                                 max_t(__int64_t, 0, bmvend - bmv->bmv_offset);
5634
5635                         /*
5636                          * In case we don't want to return the hole,
5637                          * don't increase cur_ext so that we can reuse
5638                          * it in the next loop.
5639                          */
5640                         if ((iflags & BMV_IF_NO_HOLES) &&
5641                             map[i].br_startblock == HOLESTARTBLOCK) {
5642                                 memset(&out[cur_ext], 0, sizeof(out[cur_ext]));
5643                                 continue;
5644                         }
5645
5646                         nexleft--;
5647                         bmv->bmv_entries++;
5648                         cur_ext++;
5649                 }
5650         } while (nmap && nexleft && bmv->bmv_length);
5651
5652  out_free_map:
5653         kmem_free(map);
5654  out_unlock_ilock:
5655         xfs_iunlock_map_shared(ip, lock);
5656  out_unlock_iolock:
5657         xfs_iunlock(ip, XFS_IOLOCK_SHARED);
5658
5659         for (i = 0; i < cur_ext; i++) {
5660                 int full = 0;   /* user array is full */
5661
5662                 /* format results & advance arg */
5663                 error = formatter(&arg, &out[i], &full);
5664                 if (error || full)
5665                         break;
5666         }
5667
5668         if (is_vmalloc_addr(out))
5669                 kmem_free_large(out);
5670         else
5671                 kmem_free(out);
5672         return error;
5673 }
5674
5675 #ifdef DEBUG
5676 STATIC struct xfs_buf *
5677 xfs_bmap_get_bp(
5678         struct xfs_btree_cur    *cur,
5679         xfs_fsblock_t           bno)
5680 {
5681         struct xfs_log_item_desc *lidp;
5682         int                     i;
5683
5684         if (!cur)
5685                 return NULL;
5686
5687         for (i = 0; i < XFS_BTREE_MAXLEVELS; i++) {
5688                 if (!cur->bc_bufs[i])
5689                         break;
5690                 if (XFS_BUF_ADDR(cur->bc_bufs[i]) == bno)
5691                         return cur->bc_bufs[i];
5692         }
5693
5694         /* Chase down all the log items to see if the bp is there */
5695         list_for_each_entry(lidp, &cur->bc_tp->t_items, lid_trans) {
5696                 struct xfs_buf_log_item *bip;
5697                 bip = (struct xfs_buf_log_item *)lidp->lid_item;
5698                 if (bip->bli_item.li_type == XFS_LI_BUF &&
5699                     XFS_BUF_ADDR(bip->bli_buf) == bno)
5700                         return bip->bli_buf;
5701         }
5702
5703         return NULL;
5704 }
5705
5706 STATIC void
5707 xfs_check_block(
5708         struct xfs_btree_block  *block,
5709         xfs_mount_t             *mp,
5710         int                     root,
5711         short                   sz)
5712 {
5713         int                     i, j, dmxr;
5714         __be64                  *pp, *thispa;   /* pointer to block address */
5715         xfs_bmbt_key_t          *prevp, *keyp;
5716
5717         ASSERT(be16_to_cpu(block->bb_level) > 0);
5718
5719         prevp = NULL;
5720         for( i = 1; i <= xfs_btree_get_numrecs(block); i++) {
5721                 dmxr = mp->m_bmap_dmxr[0];
5722                 keyp = XFS_BMBT_KEY_ADDR(mp, block, i);
5723
5724                 if (prevp) {
5725                         ASSERT(be64_to_cpu(prevp->br_startoff) <
5726                                be64_to_cpu(keyp->br_startoff));
5727                 }
5728                 prevp = keyp;
5729
5730                 /*
5731                  * Compare the block numbers to see if there are dups.
5732                  */
5733                 if (root)
5734                         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, i, sz);
5735                 else
5736                         pp = XFS_BMBT_PTR_ADDR(mp, block, i, dmxr);
5737
5738                 for (j = i+1; j <= be16_to_cpu(block->bb_numrecs); j++) {
5739                         if (root)
5740                                 thispa = XFS_BMAP_BROOT_PTR_ADDR(mp, block, j, sz);
5741                         else
5742                                 thispa = XFS_BMBT_PTR_ADDR(mp, block, j, dmxr);
5743                         if (*thispa == *pp) {
5744                                 xfs_warn(mp, "%s: thispa(%d) == pp(%d) %Ld",
5745                                         __func__, j, i,
5746                                         (unsigned long long)be64_to_cpu(*thispa));
5747                                 panic("%s: ptrs are equal in node\n",
5748                                         __func__);
5749                         }
5750                 }
5751         }
5752 }
5753
5754 /*
5755  * Check that the extents for the inode ip are in the right order in all
5756  * btree leaves.
5757  */
5758
5759 STATIC void
5760 xfs_bmap_check_leaf_extents(
5761         xfs_btree_cur_t         *cur,   /* btree cursor or null */
5762         xfs_inode_t             *ip,            /* incore inode pointer */
5763         int                     whichfork)      /* data or attr fork */
5764 {
5765         struct xfs_btree_block  *block; /* current btree block */
5766         xfs_fsblock_t           bno;    /* block # of "block" */
5767         xfs_buf_t               *bp;    /* buffer for "block" */
5768         int                     error;  /* error return value */
5769         xfs_extnum_t            i=0, j; /* index into the extents list */
5770         xfs_ifork_t             *ifp;   /* fork structure */
5771         int                     level;  /* btree level, for checking */
5772         xfs_mount_t             *mp;    /* file system mount structure */
5773         __be64                  *pp;    /* pointer to block address */
5774         xfs_bmbt_rec_t          *ep;    /* pointer to current extent */
5775         xfs_bmbt_rec_t          last = {0, 0}; /* last extent in prev block */
5776         xfs_bmbt_rec_t          *nextp; /* pointer to next extent */
5777         int                     bp_release = 0;
5778
5779         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE) {
5780                 return;
5781         }
5782
5783         bno = NULLFSBLOCK;
5784         mp = ip->i_mount;
5785         ifp = XFS_IFORK_PTR(ip, whichfork);
5786         block = ifp->if_broot;
5787         /*
5788          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
5789          */
5790         level = be16_to_cpu(block->bb_level);
5791         ASSERT(level > 0);
5792         xfs_check_block(block, mp, 1, ifp->if_broot_bytes);
5793         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
5794         bno = be64_to_cpu(*pp);
5795
5796         ASSERT(bno != NULLDFSBNO);
5797         ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
5798         ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
5799
5800         /*
5801          * Go down the tree until leaf level is reached, following the first
5802          * pointer (leftmost) at each level.
5803          */
5804         while (level-- > 0) {
5805                 /* See if buf is in cur first */
5806                 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
5807                 if (bp) {
5808                         bp_release = 0;
5809                 } else {
5810                         bp_release = 1;
5811                 }
5812                 if (!bp && (error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
5813                                 XFS_BMAP_BTREE_REF)))
5814                         goto error_norelse;
5815                 block = XFS_BUF_TO_BLOCK(bp);
5816                 XFS_WANT_CORRUPTED_GOTO(
5817                         xfs_bmap_sanity_check(mp, bp, level),
5818                         error0);
5819                 if (level == 0)
5820                         break;
5821
5822                 /*
5823                  * Check this block for basic sanity (increasing keys and
5824                  * no duplicate blocks).
5825                  */
5826
5827                 xfs_check_block(block, mp, 0, 0);
5828                 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
5829                 bno = be64_to_cpu(*pp);
5830                 XFS_WANT_CORRUPTED_GOTO(XFS_FSB_SANITY_CHECK(mp, bno), error0);
5831                 if (bp_release) {
5832                         bp_release = 0;
5833                         xfs_trans_brelse(NULL, bp);
5834                 }
5835         }
5836
5837         /*
5838          * Here with bp and block set to the leftmost leaf node in the tree.
5839          */
5840         i = 0;
5841
5842         /*
5843          * Loop over all leaf nodes checking that all extents are in the right order.
5844          */
5845         for (;;) {
5846                 xfs_fsblock_t   nextbno;
5847                 xfs_extnum_t    num_recs;
5848
5849
5850                 num_recs = xfs_btree_get_numrecs(block);
5851
5852                 /*
5853                  * Read-ahead the next leaf block, if any.
5854                  */
5855
5856                 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
5857
5858                 /*
5859                  * Check all the extents to make sure they are OK.
5860                  * If we had a previous block, the last entry should
5861                  * conform with the first entry in this one.
5862                  */
5863
5864                 ep = XFS_BMBT_REC_ADDR(mp, block, 1);
5865                 if (i) {
5866                         ASSERT(xfs_bmbt_disk_get_startoff(&last) +
5867                                xfs_bmbt_disk_get_blockcount(&last) <=
5868                                xfs_bmbt_disk_get_startoff(ep));
5869                 }
5870                 for (j = 1; j < num_recs; j++) {
5871                         nextp = XFS_BMBT_REC_ADDR(mp, block, j + 1);
5872                         ASSERT(xfs_bmbt_disk_get_startoff(ep) +
5873                                xfs_bmbt_disk_get_blockcount(ep) <=
5874                                xfs_bmbt_disk_get_startoff(nextp));
5875                         ep = nextp;
5876                 }
5877
5878                 last = *ep;
5879                 i += num_recs;
5880                 if (bp_release) {
5881                         bp_release = 0;
5882                         xfs_trans_brelse(NULL, bp);
5883                 }
5884                 bno = nextbno;
5885                 /*
5886                  * If we've reached the end, stop.
5887                  */
5888                 if (bno == NULLFSBLOCK)
5889                         break;
5890
5891                 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
5892                 if (bp) {
5893                         bp_release = 0;
5894                 } else {
5895                         bp_release = 1;
5896                 }
5897                 if (!bp && (error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
5898                                 XFS_BMAP_BTREE_REF)))
5899                         goto error_norelse;
5900                 block = XFS_BUF_TO_BLOCK(bp);
5901         }
5902         if (bp_release) {
5903                 bp_release = 0;
5904                 xfs_trans_brelse(NULL, bp);
5905         }
5906         return;
5907
5908 error0:
5909         xfs_warn(mp, "%s: at error0", __func__);
5910         if (bp_release)
5911                 xfs_trans_brelse(NULL, bp);
5912 error_norelse:
5913         xfs_warn(mp, "%s: BAD after btree leaves for %d extents",
5914                 __func__, i);
5915         panic("%s: CORRUPTED BTREE OR SOMETHING", __func__);
5916         return;
5917 }
5918 #endif
5919
5920 /*
5921  * Count fsblocks of the given fork.
5922  */
5923 int                                             /* error */
5924 xfs_bmap_count_blocks(
5925         xfs_trans_t             *tp,            /* transaction pointer */
5926         xfs_inode_t             *ip,            /* incore inode */
5927         int                     whichfork,      /* data or attr fork */
5928         int                     *count)         /* out: count of blocks */
5929 {
5930         struct xfs_btree_block  *block; /* current btree block */
5931         xfs_fsblock_t           bno;    /* block # of "block" */
5932         xfs_ifork_t             *ifp;   /* fork structure */
5933         int                     level;  /* btree level, for checking */
5934         xfs_mount_t             *mp;    /* file system mount structure */
5935         __be64                  *pp;    /* pointer to block address */
5936
5937         bno = NULLFSBLOCK;
5938         mp = ip->i_mount;
5939         ifp = XFS_IFORK_PTR(ip, whichfork);
5940         if ( XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ) {
5941                 xfs_bmap_count_leaves(ifp, 0,
5942                         ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t),
5943                         count);
5944                 return 0;
5945         }
5946
5947         /*
5948          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
5949          */
5950         block = ifp->if_broot;
5951         level = be16_to_cpu(block->bb_level);
5952         ASSERT(level > 0);
5953         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
5954         bno = be64_to_cpu(*pp);
5955         ASSERT(bno != NULLDFSBNO);
5956         ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
5957         ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
5958
5959         if (unlikely(xfs_bmap_count_tree(mp, tp, ifp, bno, level, count) < 0)) {
5960                 XFS_ERROR_REPORT("xfs_bmap_count_blocks(2)", XFS_ERRLEVEL_LOW,
5961                                  mp);
5962                 return XFS_ERROR(EFSCORRUPTED);
5963         }
5964
5965         return 0;
5966 }
5967
5968 /*
5969  * Recursively walks each level of a btree
5970  * to count total fsblocks is use.
5971  */
5972 STATIC int                                     /* error */
5973 xfs_bmap_count_tree(
5974         xfs_mount_t     *mp,            /* file system mount point */
5975         xfs_trans_t     *tp,            /* transaction pointer */
5976         xfs_ifork_t     *ifp,           /* inode fork pointer */
5977         xfs_fsblock_t   blockno,        /* file system block number */
5978         int             levelin,        /* level in btree */
5979         int             *count)         /* Count of blocks */
5980 {
5981         int                     error;
5982         xfs_buf_t               *bp, *nbp;
5983         int                     level = levelin;
5984         __be64                  *pp;
5985         xfs_fsblock_t           bno = blockno;
5986         xfs_fsblock_t           nextbno;
5987         struct xfs_btree_block  *block, *nextblock;
5988         int                     numrecs;
5989
5990         if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp, XFS_BMAP_BTREE_REF)))
5991                 return error;
5992         *count += 1;
5993         block = XFS_BUF_TO_BLOCK(bp);
5994
5995         if (--level) {
5996                 /* Not at node above leaves, count this level of nodes */
5997                 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
5998                 while (nextbno != NULLFSBLOCK) {
5999                         if ((error = xfs_btree_read_bufl(mp, tp, nextbno,
6000                                 0, &nbp, XFS_BMAP_BTREE_REF)))
6001                                 return error;
6002                         *count += 1;
6003                         nextblock = XFS_BUF_TO_BLOCK(nbp);
6004                         nextbno = be64_to_cpu(nextblock->bb_u.l.bb_rightsib);
6005                         xfs_trans_brelse(tp, nbp);
6006                 }
6007
6008                 /* Dive to the next level */
6009                 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
6010                 bno = be64_to_cpu(*pp);
6011                 if (unlikely((error =
6012                      xfs_bmap_count_tree(mp, tp, ifp, bno, level, count)) < 0)) {
6013                         xfs_trans_brelse(tp, bp);
6014                         XFS_ERROR_REPORT("xfs_bmap_count_tree(1)",
6015                                          XFS_ERRLEVEL_LOW, mp);
6016                         return XFS_ERROR(EFSCORRUPTED);
6017                 }
6018                 xfs_trans_brelse(tp, bp);
6019         } else {
6020                 /* count all level 1 nodes and their leaves */
6021                 for (;;) {
6022                         nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
6023                         numrecs = be16_to_cpu(block->bb_numrecs);
6024                         xfs_bmap_disk_count_leaves(mp, block, numrecs, count);
6025                         xfs_trans_brelse(tp, bp);
6026                         if (nextbno == NULLFSBLOCK)
6027                                 break;
6028                         bno = nextbno;
6029                         if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
6030                                 XFS_BMAP_BTREE_REF)))
6031                                 return error;
6032                         *count += 1;
6033                         block = XFS_BUF_TO_BLOCK(bp);
6034                 }
6035         }
6036         return 0;
6037 }
6038
6039 /*
6040  * Count leaf blocks given a range of extent records.
6041  */
6042 STATIC void
6043 xfs_bmap_count_leaves(
6044         xfs_ifork_t             *ifp,
6045         xfs_extnum_t            idx,
6046         int                     numrecs,
6047         int                     *count)
6048 {
6049         int             b;
6050
6051         for (b = 0; b < numrecs; b++) {
6052                 xfs_bmbt_rec_host_t *frp = xfs_iext_get_ext(ifp, idx + b);
6053                 *count += xfs_bmbt_get_blockcount(frp);
6054         }
6055 }
6056
6057 /*
6058  * Count leaf blocks given a range of extent records originally
6059  * in btree format.
6060  */
6061 STATIC void
6062 xfs_bmap_disk_count_leaves(
6063         struct xfs_mount        *mp,
6064         struct xfs_btree_block  *block,
6065         int                     numrecs,
6066         int                     *count)
6067 {
6068         int             b;
6069         xfs_bmbt_rec_t  *frp;
6070
6071         for (b = 1; b <= numrecs; b++) {
6072                 frp = XFS_BMBT_REC_ADDR(mp, block, b);
6073                 *count += xfs_bmbt_disk_get_blockcount(frp);
6074         }
6075 }
6076
6077 /*
6078  * dead simple method of punching delalyed allocation blocks from a range in
6079  * the inode. Walks a block at a time so will be slow, but is only executed in
6080  * rare error cases so the overhead is not critical. This will alays punch out
6081  * both the start and end blocks, even if the ranges only partially overlap
6082  * them, so it is up to the caller to ensure that partial blocks are not
6083  * passed in.
6084  */
6085 int
6086 xfs_bmap_punch_delalloc_range(
6087         struct xfs_inode        *ip,
6088         xfs_fileoff_t           start_fsb,
6089         xfs_fileoff_t           length)
6090 {
6091         xfs_fileoff_t           remaining = length;
6092         int                     error = 0;
6093
6094         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
6095
6096         do {
6097                 int             done;
6098                 xfs_bmbt_irec_t imap;
6099                 int             nimaps = 1;
6100                 xfs_fsblock_t   firstblock;
6101                 xfs_bmap_free_t flist;
6102
6103                 /*
6104                  * Map the range first and check that it is a delalloc extent
6105                  * before trying to unmap the range. Otherwise we will be
6106                  * trying to remove a real extent (which requires a
6107                  * transaction) or a hole, which is probably a bad idea...
6108                  */
6109                 error = xfs_bmapi_read(ip, start_fsb, 1, &imap, &nimaps,
6110                                        XFS_BMAPI_ENTIRE);
6111
6112                 if (error) {
6113                         /* something screwed, just bail */
6114                         if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
6115                                 xfs_alert(ip->i_mount,
6116                         "Failed delalloc mapping lookup ino %lld fsb %lld.",
6117                                                 ip->i_ino, start_fsb);
6118                         }
6119                         break;
6120                 }
6121                 if (!nimaps) {
6122                         /* nothing there */
6123                         goto next_block;
6124                 }
6125                 if (imap.br_startblock != DELAYSTARTBLOCK) {
6126                         /* been converted, ignore */
6127                         goto next_block;
6128                 }
6129                 WARN_ON(imap.br_blockcount == 0);
6130
6131                 /*
6132                  * Note: while we initialise the firstblock/flist pair, they
6133                  * should never be used because blocks should never be
6134                  * allocated or freed for a delalloc extent and hence we need
6135                  * don't cancel or finish them after the xfs_bunmapi() call.
6136                  */
6137                 xfs_bmap_init(&flist, &firstblock);
6138                 error = xfs_bunmapi(NULL, ip, start_fsb, 1, 0, 1, &firstblock,
6139                                         &flist, &done);
6140                 if (error)
6141                         break;
6142
6143                 ASSERT(!flist.xbf_count && !flist.xbf_first);
6144 next_block:
6145                 start_fsb++;
6146                 remaining--;
6147         } while(remaining > 0);
6148
6149         return error;
6150 }