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 * Update the record referred to by cur to the value given
 253  * by [off, bno, len, state].
 254  * This either works (return 0) or gets an EFSCORRUPTED error.
 255  */
 256 STATIC int
 257 xfs_bmbt_update(
 258         struct xfs_btree_cur    *cur,
 259         xfs_fileoff_t           off,
 260         xfs_fsblock_t           bno,
 261         xfs_filblks_t           len,
 262         xfs_exntst_t            state)
 263 {
 264         union xfs_btree_rec     rec;
 265
 266         xfs_bmbt_disk_set_allf(&rec.bmbt, off, bno, len, state);
 267         return xfs_btree_update(cur, &rec);
 268 }
 269
 270 /*
 271  * Called from xfs_bmap_add_attrfork to handle btree format files.
 272  */
 273 STATIC int                                      /* error */
 274 xfs_bmap_add_attrfork_btree(
 275         xfs_trans_t             *tp,            /* transaction pointer */
 276         xfs_inode_t             *ip,            /* incore inode pointer */
 277         xfs_fsblock_t           *firstblock,    /* first block allocated */
 278         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
 279         int                     *flags)         /* inode logging flags */
 280 {
 281         xfs_btree_cur_t         *cur;           /* btree cursor */
 282         int                     error;          /* error return value */
 283         xfs_mount_t             *mp;            /* file system mount struct */
 284         int                     stat;           /* newroot status */
 285
 286         mp = ip->i_mount;
 287         if (ip->i_df.if_broot_bytes <= XFS_IFORK_DSIZE(ip))
 288                 *flags |= XFS_ILOG_DBROOT;
 289         else {
 290                 cur = xfs_bmbt_init_cursor(mp, tp, ip, XFS_DATA_FORK);
 291                 cur->bc_private.b.flist = flist;
 292                 cur->bc_private.b.firstblock = *firstblock;
 293                 if ((error = xfs_bmbt_lookup_ge(cur, 0, 0, 0, &stat)))
 294                         goto error0;
 295                 /* must be at least one entry */
 296                 XFS_WANT_CORRUPTED_GOTO(stat == 1, error0);
 297                 if ((error = xfs_btree_new_iroot(cur, flags, &stat)))
 298                         goto error0;
 299                 if (stat == 0) {
 300                         xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
 301                         return XFS_ERROR(ENOSPC);
 302                 }
 303                 *firstblock = cur->bc_private.b.firstblock;
 304                 cur->bc_private.b.allocated = 0;
 305                 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
 306         }
 307         return 0;
 308 error0:
 309         xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
 310         return error;
 311 }
 312
 313 /*
 314  * Called from xfs_bmap_add_attrfork to handle extents format files.
 315  */
 316 STATIC int                                      /* error */
 317 xfs_bmap_add_attrfork_extents(
 318         xfs_trans_t             *tp,            /* transaction pointer */
 319         xfs_inode_t             *ip,            /* incore inode pointer */
 320         xfs_fsblock_t           *firstblock,    /* first block allocated */
 321         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
 322         int                     *flags)         /* inode logging flags */
 323 {
 324         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
 325         int                     error;          /* error return value */
 326
 327         if (ip->i_d.di_nextents * sizeof(xfs_bmbt_rec_t) <= XFS_IFORK_DSIZE(ip))
 328                 return 0;
 329         cur = NULL;
 330         error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist, &cur, 0,
 331                 flags, XFS_DATA_FORK);
 332         if (cur) {
 333                 cur->bc_private.b.allocated = 0;
 334                 xfs_btree_del_cursor(cur,
 335                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
 336         }
 337         return error;
 338 }
 339
 340 /*
 341  * Called from xfs_bmap_add_attrfork to handle local format files.
 342  */
 343 STATIC int                                      /* error */
 344 xfs_bmap_add_attrfork_local(
 345         xfs_trans_t             *tp,            /* transaction pointer */
 346         xfs_inode_t             *ip,            /* incore inode pointer */
 347         xfs_fsblock_t           *firstblock,    /* first block allocated */
 348         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
 349         int                     *flags)         /* inode logging flags */
 350 {
 351         xfs_da_args_t           dargs;          /* args for dir/attr code */
 352         int                     error;          /* error return value */
 353         xfs_mount_t             *mp;            /* mount structure pointer */
 354
 355         if (ip->i_df.if_bytes <= XFS_IFORK_DSIZE(ip))
 356                 return 0;
 357         if (S_ISDIR(ip->i_d.di_mode)) {
 358                 mp = ip->i_mount;
 359                 memset(&dargs, 0, sizeof(dargs));
 360                 dargs.dp = ip;
 361                 dargs.firstblock = firstblock;
 362                 dargs.flist = flist;
 363                 dargs.total = mp->m_dirblkfsbs;
 364                 dargs.whichfork = XFS_DATA_FORK;
 365                 dargs.trans = tp;
 366                 error = xfs_dir2_sf_to_block(&dargs);
 367         } else
 368                 error = xfs_bmap_local_to_extents(tp, ip, firstblock, 1, flags,
 369                         XFS_DATA_FORK);
 370         return error;
 371 }
 372
 373 /*
 374  * Convert a delayed allocation to a real allocation.
 375  */
 376 STATIC int                              /* error */
 377 xfs_bmap_add_extent_delay_real(
 378         struct xfs_trans        *tp,    /* transaction pointer */
 379         xfs_inode_t             *ip,    /* incore inode pointer */
 380         xfs_extnum_t            *idx,   /* extent number to update/insert */
 381         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
 382         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
 383         xfs_fsblock_t           *first, /* pointer to firstblock variable */
 384         xfs_bmap_free_t         *flist, /* list of extents to be freed */
 385         int                     *logflagsp) /* inode logging flags */
 386 {
 387         xfs_btree_cur_t         *cur;   /* btree cursor */
 388         int                     diff;   /* temp value */
 389         xfs_bmbt_rec_host_t     *ep;    /* extent entry for idx */
 390         int                     error;  /* error return value */
 391         int                     i;      /* temp state */
 392         xfs_ifork_t             *ifp;   /* inode fork pointer */
 393         xfs_fileoff_t           new_endoff;     /* end offset of new entry */
 394         xfs_bmbt_irec_t         r[3];   /* neighbor extent entries */
 395                                         /* left is 0, right is 1, prev is 2 */
 396         int                     rval=0; /* return value (logging flags) */
 397         int                     state = 0;/* state bits, accessed thru macros */
 398         xfs_filblks_t           da_new; /* new count del alloc blocks used */
 399         xfs_filblks_t           da_old; /* old count del alloc blocks used */
 400         xfs_filblks_t           temp=0; /* value for da_new calculations */
 401         xfs_filblks_t           temp2=0;/* value for da_new calculations */
 402         int                     tmp_rval;       /* partial logging flags */
 403
 404         ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
 405         cur = *curp;
 406
 407         ASSERT(*idx >= 0);
 408         ASSERT(*idx <= ifp->if_bytes / sizeof(struct xfs_bmbt_rec));
 409         ASSERT(!isnullstartblock(new->br_startblock));
 410         ASSERT(!cur || (cur->bc_private.b.flags & XFS_BTCUR_BPRV_WASDEL));
 411
 412         XFS_STATS_INC(xs_add_exlist);
 413
 414         *logflagsp = 0;
 415
 416 #define LEFT            r[0]
 417 #define RIGHT           r[1]
 418 #define PREV            r[2]
 419
 420         /*
 421          * Set up a bunch of variables to make the tests simpler.
 422          */
 423         ep = xfs_iext_get_ext(ifp, *idx);
 424         xfs_bmbt_get_all(ep, &PREV);
 425         new_endoff = new->br_startoff + new->br_blockcount;
 426         ASSERT(PREV.br_startoff <= new->br_startoff);
 427         ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
 428
 429         da_old = startblockval(PREV.br_startblock);
 430         da_new = 0;
 431
 432         /*
 433          * Set flags determining what part of the previous delayed allocation
 434          * extent is being replaced by a real allocation.
 435          */
 436         if (PREV.br_startoff == new->br_startoff)
 437                 state |= BMAP_LEFT_FILLING;
 438         if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
 439                 state |= BMAP_RIGHT_FILLING;
 440
 441         /*
 442          * Check and set flags if this segment has a left neighbor.
 443          * Don't set contiguous if the combined extent would be too large.
 444          */
 445         if (*idx > 0) {
 446                 state |= BMAP_LEFT_VALID;
 447                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &LEFT);
 448
 449                 if (isnullstartblock(LEFT.br_startblock))
 450                         state |= BMAP_LEFT_DELAY;
 451         }
 452
 453         if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
 454             LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
 455             LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
 456             LEFT.br_state == new->br_state &&
 457             LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
 458                 state |= BMAP_LEFT_CONTIG;
 459
 460         /*
 461          * Check and set flags if this segment has a right neighbor.
 462          * Don't set contiguous if the combined extent would be too large.
 463          * Also check for all-three-contiguous being too large.
 464          */
 465         if (*idx < ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1) {
 466                 state |= BMAP_RIGHT_VALID;
 467                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx + 1), &RIGHT);
 468
 469                 if (isnullstartblock(RIGHT.br_startblock))
 470                         state |= BMAP_RIGHT_DELAY;
 471         }
 472
 473         if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
 474             new_endoff == RIGHT.br_startoff &&
 475             new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
 476             new->br_state == RIGHT.br_state &&
 477             new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
 478             ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
 479                        BMAP_RIGHT_FILLING)) !=
 480                       (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
 481                        BMAP_RIGHT_FILLING) ||
 482              LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
 483                         <= MAXEXTLEN))
 484                 state |= BMAP_RIGHT_CONTIG;
 485
 486         error = 0;
 487         /*
 488          * Switch out based on the FILLING and CONTIG state bits.
 489          */
 490         switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
 491                          BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
 492         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
 493              BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
 494                 /*
 495                  * Filling in all of a previously delayed allocation extent.
 496                  * The left and right neighbors are both contiguous with new.
 497                  */
 498                 --*idx;
 499                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
 500                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
 501                         LEFT.br_blockcount + PREV.br_blockcount +
 502                         RIGHT.br_blockcount);
 503                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
 504
 505                 xfs_iext_remove(ip, *idx + 1, 2, state);
 506                 ip->i_d.di_nextents--;
 507                 if (cur == NULL)
 508                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
 509                 else {
 510                         rval = XFS_ILOG_CORE;
 511                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
 512                                         RIGHT.br_startblock,
 513                                         RIGHT.br_blockcount, &i)))
 514                                 goto done;
 515                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 516                         if ((error = xfs_btree_delete(cur, &i)))
 517                                 goto done;
 518                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 519                         if ((error = xfs_btree_decrement(cur, 0, &i)))
 520                                 goto done;
 521                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 522                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
 523                                         LEFT.br_startblock,
 524                                         LEFT.br_blockcount +
 525                                         PREV.br_blockcount +
 526                                         RIGHT.br_blockcount, LEFT.br_state)))
 527                                 goto done;
 528                 }
 529                 break;
 530
 531         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
 532                 /*
 533                  * Filling in all of a previously delayed allocation extent.
 534                  * The left neighbor is contiguous, the right is not.
 535                  */
 536                 --*idx;
 537
 538                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
 539                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
 540                         LEFT.br_blockcount + PREV.br_blockcount);
 541                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
 542
 543                 xfs_iext_remove(ip, *idx + 1, 1, state);
 544                 if (cur == NULL)
 545                         rval = XFS_ILOG_DEXT;
 546                 else {
 547                         rval = 0;
 548                         if ((error = xfs_bmbt_lookup_eq(cur, LEFT.br_startoff,
 549                                         LEFT.br_startblock, LEFT.br_blockcount,
 550                                         &i)))
 551                                 goto done;
 552                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 553                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
 554                                         LEFT.br_startblock,
 555                                         LEFT.br_blockcount +
 556                                         PREV.br_blockcount, LEFT.br_state)))
 557                                 goto done;
 558                 }
 559                 break;
 560
 561         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
 562                 /*
 563                  * Filling in all of a previously delayed allocation extent.
 564                  * The right neighbor is contiguous, the left is not.
 565                  */
 566                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
 567                 xfs_bmbt_set_startblock(ep, new->br_startblock);
 568                 xfs_bmbt_set_blockcount(ep,
 569                         PREV.br_blockcount + RIGHT.br_blockcount);
 570                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
 571
 572                 xfs_iext_remove(ip, *idx + 1, 1, state);
 573                 if (cur == NULL)
 574                         rval = XFS_ILOG_DEXT;
 575                 else {
 576                         rval = 0;
 577                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
 578                                         RIGHT.br_startblock,
 579                                         RIGHT.br_blockcount, &i)))
 580                                 goto done;
 581                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 582                         if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
 583                                         new->br_startblock,
 584                                         PREV.br_blockcount +
 585                                         RIGHT.br_blockcount, PREV.br_state)))
 586                                 goto done;
 587                 }
 588                 break;
 589
 590         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
 591                 /*
 592                  * Filling in all of a previously delayed allocation extent.
 593                  * Neither the left nor right neighbors are contiguous with
 594                  * the new one.
 595                  */
 596                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
 597                 xfs_bmbt_set_startblock(ep, new->br_startblock);
 598                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
 599
 600                 ip->i_d.di_nextents++;
 601                 if (cur == NULL)
 602                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
 603                 else {
 604                         rval = XFS_ILOG_CORE;
 605                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
 606                                         new->br_startblock, new->br_blockcount,
 607                                         &i)))
 608                                 goto done;
 609                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
 610                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
 611                         if ((error = xfs_btree_insert(cur, &i)))
 612                                 goto done;
 613                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 614                 }
 615                 break;
 616
 617         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
 618                 /*
 619                  * Filling in the first part of a previous delayed allocation.
 620                  * The left neighbor is contiguous.
 621                  */
 622                 trace_xfs_bmap_pre_update(ip, *idx - 1, state, _THIS_IP_);
 623                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx - 1),
 624                         LEFT.br_blockcount + new->br_blockcount);
 625                 xfs_bmbt_set_startoff(ep,
 626                         PREV.br_startoff + new->br_blockcount);
 627                 trace_xfs_bmap_post_update(ip, *idx - 1, state, _THIS_IP_);
 628
 629                 temp = PREV.br_blockcount - new->br_blockcount;
 630                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
 631                 xfs_bmbt_set_blockcount(ep, temp);
 632                 if (cur == NULL)
 633                         rval = XFS_ILOG_DEXT;
 634                 else {
 635                         rval = 0;
 636                         if ((error = xfs_bmbt_lookup_eq(cur, LEFT.br_startoff,
 637                                         LEFT.br_startblock, LEFT.br_blockcount,
 638                                         &i)))
 639                                 goto done;
 640                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 641                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
 642                                         LEFT.br_startblock,
 643                                         LEFT.br_blockcount +
 644                                         new->br_blockcount,
 645                                         LEFT.br_state)))
 646                                 goto done;
 647                 }
 648                 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
 649                         startblockval(PREV.br_startblock));
 650                 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
 651                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
 652
 653                 --*idx;
 654                 break;
 655
 656         case BMAP_LEFT_FILLING:
 657                 /*
 658                  * Filling in the first part of a previous delayed allocation.
 659                  * The left neighbor is not contiguous.
 660                  */
 661                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
 662                 xfs_bmbt_set_startoff(ep, new_endoff);
 663                 temp = PREV.br_blockcount - new->br_blockcount;
 664                 xfs_bmbt_set_blockcount(ep, temp);
 665                 xfs_iext_insert(ip, *idx, 1, new, state);
 666                 ip->i_d.di_nextents++;
 667                 if (cur == NULL)
 668                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
 669                 else {
 670                         rval = XFS_ILOG_CORE;
 671                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
 672                                         new->br_startblock, new->br_blockcount,
 673                                         &i)))
 674                                 goto done;
 675                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
 676                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
 677                         if ((error = xfs_btree_insert(cur, &i)))
 678                                 goto done;
 679                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 680                 }
 681                 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
 682                     ip->i_d.di_nextents > ip->i_df.if_ext_max) {
 683                         error = xfs_bmap_extents_to_btree(tp, ip,
 684                                         first, flist, &cur, 1, &tmp_rval,
 685                                         XFS_DATA_FORK);
 686                         rval |= tmp_rval;
 687                         if (error)
 688                                 goto done;
 689                 }
 690                 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
 691                         startblockval(PREV.br_startblock) -
 692                         (cur ? cur->bc_private.b.allocated : 0));
 693                 ep = xfs_iext_get_ext(ifp, *idx + 1);
 694                 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
 695                 trace_xfs_bmap_post_update(ip, *idx + 1, state, _THIS_IP_);
 696                 break;
 697
 698         case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
 699                 /*
 700                  * Filling in the last part of a previous delayed allocation.
 701                  * The right neighbor is contiguous with the new allocation.
 702                  */
 703                 temp = PREV.br_blockcount - new->br_blockcount;
 704                 trace_xfs_bmap_pre_update(ip, *idx + 1, state, _THIS_IP_);
 705                 xfs_bmbt_set_blockcount(ep, temp);
 706                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx + 1),
 707                         new->br_startoff, new->br_startblock,
 708                         new->br_blockcount + RIGHT.br_blockcount,
 709                         RIGHT.br_state);
 710                 trace_xfs_bmap_post_update(ip, *idx + 1, state, _THIS_IP_);
 711                 if (cur == NULL)
 712                         rval = XFS_ILOG_DEXT;
 713                 else {
 714                         rval = 0;
 715                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
 716                                         RIGHT.br_startblock,
 717                                         RIGHT.br_blockcount, &i)))
 718                                 goto done;
 719                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 720                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
 721                                         new->br_startblock,
 722                                         new->br_blockcount +
 723                                         RIGHT.br_blockcount,
 724                                         RIGHT.br_state)))
 725                                 goto done;
 726                 }
 727
 728                 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
 729                         startblockval(PREV.br_startblock));
 730                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
 731                 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
 732                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
 733
 734                 ++*idx;
 735                 break;
 736
 737         case BMAP_RIGHT_FILLING:
 738                 /*
 739                  * Filling in the last part of a previous delayed allocation.
 740                  * The right neighbor is not contiguous.
 741                  */
 742                 temp = PREV.br_blockcount - new->br_blockcount;
 743                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
 744                 xfs_bmbt_set_blockcount(ep, temp);
 745                 xfs_iext_insert(ip, *idx + 1, 1, new, state);
 746                 ip->i_d.di_nextents++;
 747                 if (cur == NULL)
 748                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
 749                 else {
 750                         rval = XFS_ILOG_CORE;
 751                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
 752                                         new->br_startblock, new->br_blockcount,
 753                                         &i)))
 754                                 goto done;
 755                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
 756                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
 757                         if ((error = xfs_btree_insert(cur, &i)))
 758                                 goto done;
 759                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 760                 }
 761                 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
 762                     ip->i_d.di_nextents > ip->i_df.if_ext_max) {
 763                         error = xfs_bmap_extents_to_btree(tp, ip,
 764                                 first, flist, &cur, 1, &tmp_rval,
 765                                 XFS_DATA_FORK);
 766                         rval |= tmp_rval;
 767                         if (error)
 768                                 goto done;
 769                 }
 770                 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
 771                         startblockval(PREV.br_startblock) -
 772                         (cur ? cur->bc_private.b.allocated : 0));
 773                 ep = xfs_iext_get_ext(ifp, *idx);
 774                 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
 775                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
 776
 777                 ++*idx;
 778                 break;
 779
 780         case 0:
 781                 /*
 782                  * Filling in the middle part of a previous delayed allocation.
 783                  * Contiguity is impossible here.
 784                  * This case is avoided almost all the time.
 785                  *
 786                  * We start with a delayed allocation:
 787                  *
 788                  * +ddddddddddddddddddddddddddddddddddddddddddddddddddddddd+
 789                  *  PREV @ idx
 790                  *
 791                  * and we are allocating:
 792                  *                     +rrrrrrrrrrrrrrrrr+
 793                  *                            new
 794                  *
 795                  * and we set it up for insertion as:
 796                  * +ddddddddddddddddddd+rrrrrrrrrrrrrrrrr+ddddddddddddddddd+
 797                  *                            new
 798                  *  PREV @ idx          LEFT              RIGHT
 799                  *                      inserted at idx + 1
 800                  */
 801                 temp = new->br_startoff - PREV.br_startoff;
 802                 temp2 = PREV.br_startoff + PREV.br_blockcount - new_endoff;
 803                 trace_xfs_bmap_pre_update(ip, *idx, 0, _THIS_IP_);
 804                 xfs_bmbt_set_blockcount(ep, temp);      /* truncate PREV */
 805                 LEFT = *new;
 806                 RIGHT.br_state = PREV.br_state;
 807                 RIGHT.br_startblock = nullstartblock(
 808                                 (int)xfs_bmap_worst_indlen(ip, temp2));
 809                 RIGHT.br_startoff = new_endoff;
 810                 RIGHT.br_blockcount = temp2;
 811                 /* insert LEFT (r[0]) and RIGHT (r[1]) at the same time */
 812                 xfs_iext_insert(ip, *idx + 1, 2, &LEFT, state);
 813                 ip->i_d.di_nextents++;
 814                 if (cur == NULL)
 815                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
 816                 else {
 817                         rval = XFS_ILOG_CORE;
 818                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
 819                                         new->br_startblock, new->br_blockcount,
 820                                         &i)))
 821                                 goto done;
 822                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
 823                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
 824                         if ((error = xfs_btree_insert(cur, &i)))
 825                                 goto done;
 826                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 827                 }
 828                 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
 829                     ip->i_d.di_nextents > ip->i_df.if_ext_max) {
 830                         error = xfs_bmap_extents_to_btree(tp, ip,
 831                                         first, flist, &cur, 1, &tmp_rval,
 832                                         XFS_DATA_FORK);
 833                         rval |= tmp_rval;
 834                         if (error)
 835                                 goto done;
 836                 }
 837                 temp = xfs_bmap_worst_indlen(ip, temp);
 838                 temp2 = xfs_bmap_worst_indlen(ip, temp2);
 839                 diff = (int)(temp + temp2 - startblockval(PREV.br_startblock) -
 840                         (cur ? cur->bc_private.b.allocated : 0));
 841                 if (diff > 0) {
 842                         error = xfs_icsb_modify_counters(ip->i_mount,
 843                                         XFS_SBS_FDBLOCKS,
 844                                         -((int64_t)diff), 0);
 845                         ASSERT(!error);
 846                         if (error)
 847                                 goto done;
 848                 }
 849
 850                 ep = xfs_iext_get_ext(ifp, *idx);
 851                 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
 852                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
 853                 trace_xfs_bmap_pre_update(ip, *idx + 2, state, _THIS_IP_);
 854                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, *idx + 2),
 855                         nullstartblock((int)temp2));
 856                 trace_xfs_bmap_post_update(ip, *idx + 2, state, _THIS_IP_);
 857
 858                 ++*idx;
 859                 da_new = temp + temp2;
 860                 break;
 861
 862         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
 863         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
 864         case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
 865         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
 866         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
 867         case BMAP_LEFT_CONTIG:
 868         case BMAP_RIGHT_CONTIG:
 869                 /*
 870                  * These cases are all impossible.
 871                  */
 872                 ASSERT(0);
 873         }
 874
 875         /* convert to a btree if necessary */
 876         if (XFS_IFORK_FORMAT(ip, XFS_DATA_FORK) == XFS_DINODE_FMT_EXTENTS &&
 877             XFS_IFORK_NEXTENTS(ip, XFS_DATA_FORK) > ifp->if_ext_max) {
 878                 int     tmp_logflags;   /* partial log flag return val */
 879
 880                 ASSERT(cur == NULL);
 881                 error = xfs_bmap_extents_to_btree(tp, ip, first, flist, &cur,
 882                                 da_old > 0, &tmp_logflags, XFS_DATA_FORK);
 883                 *logflagsp |= tmp_logflags;
 884                 if (error)
 885                         goto done;
 886         }
 887
 888         /* adjust for changes in reserved delayed indirect blocks */
 889         if (da_old || da_new) {
 890                 temp = da_new;
 891                 if (cur)
 892                         temp += cur->bc_private.b.allocated;
 893                 ASSERT(temp <= da_old);
 894                 if (temp < da_old)
 895                         xfs_icsb_modify_counters(ip->i_mount, XFS_SBS_FDBLOCKS,
 896                                 (int64_t)(da_old - temp), 0);
 897         }
 898
 899         /* clear out the allocated field, done with it now in any case. */
 900         if (cur) {
 901                 cur->bc_private.b.allocated = 0;
 902                 *curp = cur;
 903         }
 904         xfs_bmap_check_leaf_extents(cur, ip, XFS_DATA_FORK);
 905 done:
 906         *logflagsp |= rval;
 907         return error;
 908 #undef  LEFT
 909 #undef  RIGHT
 910 #undef  PREV
 911 }
 912
 913 /*
 914  * Convert an unwritten allocation to a real allocation or vice versa.
 915  */
 916 STATIC int                              /* error */
 917 xfs_bmap_add_extent_unwritten_real(
 918         struct xfs_trans        *tp,
 919         xfs_inode_t             *ip,    /* incore inode pointer */
 920         xfs_extnum_t            *idx,   /* extent number to update/insert */
 921         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
 922         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
 923         xfs_fsblock_t           *first, /* pointer to firstblock variable */
 924         xfs_bmap_free_t         *flist, /* list of extents to be freed */
 925         int                     *logflagsp) /* inode logging flags */
 926 {
 927         xfs_btree_cur_t         *cur;   /* btree cursor */
 928         xfs_bmbt_rec_host_t     *ep;    /* extent entry for idx */
 929         int                     error;  /* error return value */
 930         int                     i;      /* temp state */
 931         xfs_ifork_t             *ifp;   /* inode fork pointer */
 932         xfs_fileoff_t           new_endoff;     /* end offset of new entry */
 933         xfs_exntst_t            newext; /* new extent state */
 934         xfs_exntst_t            oldext; /* old extent state */
 935         xfs_bmbt_irec_t         r[3];   /* neighbor extent entries */
 936                                         /* left is 0, right is 1, prev is 2 */
 937         int                     rval=0; /* return value (logging flags) */
 938         int                     state = 0;/* state bits, accessed thru macros */
 939
 940         *logflagsp = 0;
 941
 942         cur = *curp;
 943         ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
 944
 945         ASSERT(*idx >= 0);
 946         ASSERT(*idx <= ifp->if_bytes / sizeof(struct xfs_bmbt_rec));
 947         ASSERT(!isnullstartblock(new->br_startblock));
 948
 949         XFS_STATS_INC(xs_add_exlist);
 950
 951 #define LEFT            r[0]
 952 #define RIGHT           r[1]
 953 #define PREV            r[2]
 954
 955         /*
 956          * Set up a bunch of variables to make the tests simpler.
 957          */
 958         error = 0;
 959         ep = xfs_iext_get_ext(ifp, *idx);
 960         xfs_bmbt_get_all(ep, &PREV);
 961         newext = new->br_state;
 962         oldext = (newext == XFS_EXT_UNWRITTEN) ?
 963                 XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
 964         ASSERT(PREV.br_state == oldext);
 965         new_endoff = new->br_startoff + new->br_blockcount;
 966         ASSERT(PREV.br_startoff <= new->br_startoff);
 967         ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
 968
 969         /*
 970          * Set flags determining what part of the previous oldext allocation
 971          * extent is being replaced by a newext allocation.
 972          */
 973         if (PREV.br_startoff == new->br_startoff)
 974                 state |= BMAP_LEFT_FILLING;
 975         if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
 976                 state |= BMAP_RIGHT_FILLING;
 977
 978         /*
 979          * Check and set flags if this segment has a left neighbor.
 980          * Don't set contiguous if the combined extent would be too large.
 981          */
 982         if (*idx > 0) {
 983                 state |= BMAP_LEFT_VALID;
 984                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &LEFT);
 985
 986                 if (isnullstartblock(LEFT.br_startblock))
 987                         state |= BMAP_LEFT_DELAY;
 988         }
 989
 990         if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
 991             LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
 992             LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
 993             LEFT.br_state == newext &&
 994             LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
 995                 state |= BMAP_LEFT_CONTIG;
 996
 997         /*
 998          * Check and set flags if this segment has a right neighbor.
 999          * Don't set contiguous if the combined extent would be too large.
1000          * Also check for all-three-contiguous being too large.
1001          */
1002         if (*idx < ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1) {
1003                 state |= BMAP_RIGHT_VALID;
1004                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx + 1), &RIGHT);
1005                 if (isnullstartblock(RIGHT.br_startblock))
1006                         state |= BMAP_RIGHT_DELAY;
1007         }
1008
1009         if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
1010             new_endoff == RIGHT.br_startoff &&
1011             new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
1012             newext == RIGHT.br_state &&
1013             new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
1014             ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1015                        BMAP_RIGHT_FILLING)) !=
1016                       (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1017                        BMAP_RIGHT_FILLING) ||
1018              LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
1019                         <= MAXEXTLEN))
1020                 state |= BMAP_RIGHT_CONTIG;
1021
1022         /*
1023          * Switch out based on the FILLING and CONTIG state bits.
1024          */
1025         switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1026                          BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
1027         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1028              BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1029                 /*
1030                  * Setting all of a previous oldext extent to newext.
1031                  * The left and right neighbors are both contiguous with new.
1032                  */
1033                 --*idx;
1034
1035                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1036                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
1037                         LEFT.br_blockcount + PREV.br_blockcount +
1038                         RIGHT.br_blockcount);
1039                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1040
1041                 xfs_iext_remove(ip, *idx + 1, 2, state);
1042                 ip->i_d.di_nextents -= 2;
1043                 if (cur == NULL)
1044                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1045                 else {
1046                         rval = XFS_ILOG_CORE;
1047                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
1048                                         RIGHT.br_startblock,
1049                                         RIGHT.br_blockcount, &i)))
1050                                 goto done;
1051                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1052                         if ((error = xfs_btree_delete(cur, &i)))
1053                                 goto done;
1054                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1055                         if ((error = xfs_btree_decrement(cur, 0, &i)))
1056                                 goto done;
1057                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1058                         if ((error = xfs_btree_delete(cur, &i)))
1059                                 goto done;
1060                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1061                         if ((error = xfs_btree_decrement(cur, 0, &i)))
1062                                 goto done;
1063                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1064                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
1065                                 LEFT.br_startblock,
1066                                 LEFT.br_blockcount + PREV.br_blockcount +
1067                                 RIGHT.br_blockcount, LEFT.br_state)))
1068                                 goto done;
1069                 }
1070                 break;
1071
1072         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1073                 /*
1074                  * Setting all of a previous oldext extent to newext.
1075                  * The left neighbor is contiguous, the right is not.
1076                  */
1077                 --*idx;
1078
1079                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1080                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
1081                         LEFT.br_blockcount + PREV.br_blockcount);
1082                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1083
1084                 xfs_iext_remove(ip, *idx + 1, 1, state);
1085                 ip->i_d.di_nextents--;
1086                 if (cur == NULL)
1087                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1088                 else {
1089                         rval = XFS_ILOG_CORE;
1090                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1091                                         PREV.br_startblock, PREV.br_blockcount,
1092                                         &i)))
1093                                 goto done;
1094                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1095                         if ((error = xfs_btree_delete(cur, &i)))
1096                                 goto done;
1097                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1098                         if ((error = xfs_btree_decrement(cur, 0, &i)))
1099                                 goto done;
1100                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1101                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
1102                                 LEFT.br_startblock,
1103                                 LEFT.br_blockcount + PREV.br_blockcount,
1104                                 LEFT.br_state)))
1105                                 goto done;
1106                 }
1107                 break;
1108
1109         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1110                 /*
1111                  * Setting all of a previous oldext extent to newext.
1112                  * The right neighbor is contiguous, the left is not.
1113                  */
1114                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1115                 xfs_bmbt_set_blockcount(ep,
1116                         PREV.br_blockcount + RIGHT.br_blockcount);
1117                 xfs_bmbt_set_state(ep, newext);
1118                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1119                 xfs_iext_remove(ip, *idx + 1, 1, state);
1120                 ip->i_d.di_nextents--;
1121                 if (cur == NULL)
1122                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1123                 else {
1124                         rval = XFS_ILOG_CORE;
1125                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
1126                                         RIGHT.br_startblock,
1127                                         RIGHT.br_blockcount, &i)))
1128                                 goto done;
1129                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1130                         if ((error = xfs_btree_delete(cur, &i)))
1131                                 goto done;
1132                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1133                         if ((error = xfs_btree_decrement(cur, 0, &i)))
1134                                 goto done;
1135                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1136                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
1137                                 new->br_startblock,
1138                                 new->br_blockcount + RIGHT.br_blockcount,
1139                                 newext)))
1140                                 goto done;
1141                 }
1142                 break;
1143
1144         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
1145                 /*
1146                  * Setting all of a previous oldext extent to newext.
1147                  * Neither the left nor right neighbors are contiguous with
1148                  * the new one.
1149                  */
1150                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1151                 xfs_bmbt_set_state(ep, newext);
1152                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1153
1154                 if (cur == NULL)
1155                         rval = XFS_ILOG_DEXT;
1156                 else {
1157                         rval = 0;
1158                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1159                                         new->br_startblock, new->br_blockcount,
1160                                         &i)))
1161                                 goto done;
1162                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1163                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
1164                                 new->br_startblock, new->br_blockcount,
1165                                 newext)))
1166                                 goto done;
1167                 }
1168                 break;
1169
1170         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
1171                 /*
1172                  * Setting the first part of a previous oldext extent to newext.
1173                  * The left neighbor is contiguous.
1174                  */
1175                 trace_xfs_bmap_pre_update(ip, *idx - 1, state, _THIS_IP_);
1176                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx - 1),
1177                         LEFT.br_blockcount + new->br_blockcount);
1178                 xfs_bmbt_set_startoff(ep,
1179                         PREV.br_startoff + new->br_blockcount);
1180                 trace_xfs_bmap_post_update(ip, *idx - 1, state, _THIS_IP_);
1181
1182                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1183                 xfs_bmbt_set_startblock(ep,
1184                         new->br_startblock + new->br_blockcount);
1185                 xfs_bmbt_set_blockcount(ep,
1186                         PREV.br_blockcount - new->br_blockcount);
1187                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1188
1189                 --*idx;
1190
1191                 if (cur == NULL)
1192                         rval = XFS_ILOG_DEXT;
1193                 else {
1194                         rval = 0;
1195                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1196                                         PREV.br_startblock, PREV.br_blockcount,
1197                                         &i)))
1198                                 goto done;
1199                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1200                         if ((error = xfs_bmbt_update(cur,
1201                                 PREV.br_startoff + new->br_blockcount,
1202                                 PREV.br_startblock + new->br_blockcount,
1203                                 PREV.br_blockcount - new->br_blockcount,
1204                                 oldext)))
1205                                 goto done;
1206                         if ((error = xfs_btree_decrement(cur, 0, &i)))
1207                                 goto done;
1208                         if (xfs_bmbt_update(cur, LEFT.br_startoff,
1209                                 LEFT.br_startblock,
1210                                 LEFT.br_blockcount + new->br_blockcount,
1211                                 LEFT.br_state))
1212                                 goto done;
1213                 }
1214                 break;
1215
1216         case BMAP_LEFT_FILLING:
1217                 /*
1218                  * Setting the first part of a previous oldext extent to newext.
1219                  * The left neighbor is not contiguous.
1220                  */
1221                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1222                 ASSERT(ep && xfs_bmbt_get_state(ep) == oldext);
1223                 xfs_bmbt_set_startoff(ep, new_endoff);
1224                 xfs_bmbt_set_blockcount(ep,
1225                         PREV.br_blockcount - new->br_blockcount);
1226                 xfs_bmbt_set_startblock(ep,
1227                         new->br_startblock + new->br_blockcount);
1228                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1229
1230                 xfs_iext_insert(ip, *idx, 1, new, state);
1231                 ip->i_d.di_nextents++;
1232                 if (cur == NULL)
1233                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1234                 else {
1235                         rval = XFS_ILOG_CORE;
1236                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1237                                         PREV.br_startblock, PREV.br_blockcount,
1238                                         &i)))
1239                                 goto done;
1240                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1241                         if ((error = xfs_bmbt_update(cur,
1242                                 PREV.br_startoff + new->br_blockcount,
1243                                 PREV.br_startblock + new->br_blockcount,
1244                                 PREV.br_blockcount - new->br_blockcount,
1245                                 oldext)))
1246                                 goto done;
1247                         cur->bc_rec.b = *new;
1248                         if ((error = xfs_btree_insert(cur, &i)))
1249                                 goto done;
1250                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1251                 }
1252                 break;
1253
1254         case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1255                 /*
1256                  * Setting the last part of a previous oldext extent to newext.
1257                  * The right neighbor is contiguous with the new allocation.
1258                  */
1259                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1260                 xfs_bmbt_set_blockcount(ep,
1261                         PREV.br_blockcount - new->br_blockcount);
1262                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1263
1264                 ++*idx;
1265
1266                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1267                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx),
1268                         new->br_startoff, new->br_startblock,
1269                         new->br_blockcount + RIGHT.br_blockcount, newext);
1270                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1271
1272                 if (cur == NULL)
1273                         rval = XFS_ILOG_DEXT;
1274                 else {
1275                         rval = 0;
1276                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1277                                         PREV.br_startblock,
1278                                         PREV.br_blockcount, &i)))
1279                                 goto done;
1280                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1281                         if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
1282                                 PREV.br_startblock,
1283                                 PREV.br_blockcount - new->br_blockcount,
1284                                 oldext)))
1285                                 goto done;
1286                         if ((error = xfs_btree_increment(cur, 0, &i)))
1287                                 goto done;
1288                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
1289                                 new->br_startblock,
1290                                 new->br_blockcount + RIGHT.br_blockcount,
1291                                 newext)))
1292                                 goto done;
1293                 }
1294                 break;
1295
1296         case BMAP_RIGHT_FILLING:
1297                 /*
1298                  * Setting the last part of a previous oldext extent to newext.
1299                  * The right neighbor is not contiguous.
1300                  */
1301                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1302                 xfs_bmbt_set_blockcount(ep,
1303                         PREV.br_blockcount - new->br_blockcount);
1304                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1305
1306                 ++*idx;
1307                 xfs_iext_insert(ip, *idx, 1, new, state);
1308
1309                 ip->i_d.di_nextents++;
1310                 if (cur == NULL)
1311                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1312                 else {
1313                         rval = XFS_ILOG_CORE;
1314                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1315                                         PREV.br_startblock, PREV.br_blockcount,
1316                                         &i)))
1317                                 goto done;
1318                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1319                         if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
1320                                 PREV.br_startblock,
1321                                 PREV.br_blockcount - new->br_blockcount,
1322                                 oldext)))
1323                                 goto done;
1324                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1325                                         new->br_startblock, new->br_blockcount,
1326                                         &i)))
1327                                 goto done;
1328                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1329                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
1330                         if ((error = xfs_btree_insert(cur, &i)))
1331                                 goto done;
1332                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1333                 }
1334                 break;
1335
1336         case 0:
1337                 /*
1338                  * Setting the middle part of a previous oldext extent to
1339                  * newext.  Contiguity is impossible here.
1340                  * One extent becomes three extents.
1341                  */
1342                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1343                 xfs_bmbt_set_blockcount(ep,
1344                         new->br_startoff - PREV.br_startoff);
1345                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1346
1347                 r[0] = *new;
1348                 r[1].br_startoff = new_endoff;
1349                 r[1].br_blockcount =
1350                         PREV.br_startoff + PREV.br_blockcount - new_endoff;
1351                 r[1].br_startblock = new->br_startblock + new->br_blockcount;
1352                 r[1].br_state = oldext;
1353
1354                 ++*idx;
1355                 xfs_iext_insert(ip, *idx, 2, &r[0], state);
1356
1357                 ip->i_d.di_nextents += 2;
1358                 if (cur == NULL)
1359                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1360                 else {
1361                         rval = XFS_ILOG_CORE;
1362                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1363                                         PREV.br_startblock, PREV.br_blockcount,
1364                                         &i)))
1365                                 goto done;
1366                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1367                         /* new right extent - oldext */
1368                         if ((error = xfs_bmbt_update(cur, r[1].br_startoff,
1369                                 r[1].br_startblock, r[1].br_blockcount,
1370                                 r[1].br_state)))
1371                                 goto done;
1372                         /* new left extent - oldext */
1373                         cur->bc_rec.b = PREV;
1374                         cur->bc_rec.b.br_blockcount =
1375                                 new->br_startoff - PREV.br_startoff;
1376                         if ((error = xfs_btree_insert(cur, &i)))
1377                                 goto done;
1378                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1379                         /*
1380                          * Reset the cursor to the position of the new extent
1381                          * we are about to insert as we can't trust it after
1382                          * the previous insert.
1383                          */
1384                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1385                                         new->br_startblock, new->br_blockcount,
1386                                         &i)))
1387                                 goto done;
1388                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1389                         /* new middle extent - newext */
1390                         cur->bc_rec.b.br_state = new->br_state;
1391                         if ((error = xfs_btree_insert(cur, &i)))
1392                                 goto done;
1393                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1394                 }
1395                 break;
1396
1397         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1398         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1399         case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
1400         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1401         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1402         case BMAP_LEFT_CONTIG:
1403         case BMAP_RIGHT_CONTIG:
1404                 /*
1405                  * These cases are all impossible.
1406                  */
1407                 ASSERT(0);
1408         }
1409
1410         /* convert to a btree if necessary */
1411         if (XFS_IFORK_FORMAT(ip, XFS_DATA_FORK) == XFS_DINODE_FMT_EXTENTS &&
1412             XFS_IFORK_NEXTENTS(ip, XFS_DATA_FORK) > ifp->if_ext_max) {
1413                 int     tmp_logflags;   /* partial log flag return val */
1414
1415                 ASSERT(cur == NULL);
1416                 error = xfs_bmap_extents_to_btree(tp, ip, first, flist, &cur,
1417                                 0, &tmp_logflags, XFS_DATA_FORK);
1418                 *logflagsp |= tmp_logflags;
1419                 if (error)
1420                         goto done;
1421         }
1422
1423         /* clear out the allocated field, done with it now in any case. */
1424         if (cur) {
1425                 cur->bc_private.b.allocated = 0;
1426                 *curp = cur;
1427         }
1428
1429         xfs_bmap_check_leaf_extents(*curp, ip, XFS_DATA_FORK);
1430 done:
1431         *logflagsp |= rval;
1432         return error;
1433 #undef  LEFT
1434 #undef  RIGHT
1435 #undef  PREV
1436 }
1437
1438 /*
1439  * Convert a hole to a delayed allocation.
1440  */
1441 STATIC void
1442 xfs_bmap_add_extent_hole_delay(
1443         xfs_inode_t             *ip,    /* incore inode pointer */
1444         xfs_extnum_t            *idx,   /* extent number to update/insert */
1445         xfs_bmbt_irec_t         *new)   /* new data to add to file extents */
1446 {
1447         xfs_ifork_t             *ifp;   /* inode fork pointer */
1448         xfs_bmbt_irec_t         left;   /* left neighbor extent entry */
1449         xfs_filblks_t           newlen=0;       /* new indirect size */
1450         xfs_filblks_t           oldlen=0;       /* old indirect size */
1451         xfs_bmbt_irec_t         right;  /* right neighbor extent entry */
1452         int                     state;  /* state bits, accessed thru macros */
1453         xfs_filblks_t           temp=0; /* temp for indirect calculations */
1454
1455         ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
1456         state = 0;
1457         ASSERT(isnullstartblock(new->br_startblock));
1458
1459         /*
1460          * Check and set flags if this segment has a left neighbor
1461          */
1462         if (*idx > 0) {
1463                 state |= BMAP_LEFT_VALID;
1464                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &left);
1465
1466                 if (isnullstartblock(left.br_startblock))
1467                         state |= BMAP_LEFT_DELAY;
1468         }
1469
1470         /*
1471          * Check and set flags if the current (right) segment exists.
1472          * If it doesn't exist, we're converting the hole at end-of-file.
1473          */
1474         if (*idx < ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t)) {
1475                 state |= BMAP_RIGHT_VALID;
1476                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx), &right);
1477
1478                 if (isnullstartblock(right.br_startblock))
1479                         state |= BMAP_RIGHT_DELAY;
1480         }
1481
1482         /*
1483          * Set contiguity flags on the left and right neighbors.
1484          * Don't let extents get too large, even if the pieces are contiguous.
1485          */
1486         if ((state & BMAP_LEFT_VALID) && (state & BMAP_LEFT_DELAY) &&
1487             left.br_startoff + left.br_blockcount == new->br_startoff &&
1488             left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
1489                 state |= BMAP_LEFT_CONTIG;
1490
1491         if ((state & BMAP_RIGHT_VALID) && (state & BMAP_RIGHT_DELAY) &&
1492             new->br_startoff + new->br_blockcount == right.br_startoff &&
1493             new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
1494             (!(state & BMAP_LEFT_CONTIG) ||
1495              (left.br_blockcount + new->br_blockcount +
1496               right.br_blockcount <= MAXEXTLEN)))
1497                 state |= BMAP_RIGHT_CONTIG;
1498
1499         /*
1500          * Switch out based on the contiguity flags.
1501          */
1502         switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
1503         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1504                 /*
1505                  * New allocation is contiguous with delayed allocations
1506                  * on the left and on the right.
1507                  * Merge all three into a single extent record.
1508                  */
1509                 --*idx;
1510                 temp = left.br_blockcount + new->br_blockcount +
1511                         right.br_blockcount;
1512
1513                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1514                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx), temp);
1515                 oldlen = startblockval(left.br_startblock) +
1516                         startblockval(new->br_startblock) +
1517                         startblockval(right.br_startblock);
1518                 newlen = xfs_bmap_worst_indlen(ip, temp);
1519                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, *idx),
1520                         nullstartblock((int)newlen));
1521                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1522
1523                 xfs_iext_remove(ip, *idx + 1, 1, state);
1524                 break;
1525
1526         case BMAP_LEFT_CONTIG:
1527                 /*
1528                  * New allocation is contiguous with a delayed allocation
1529                  * on the left.
1530                  * Merge the new allocation with the left neighbor.
1531                  */
1532                 --*idx;
1533                 temp = left.br_blockcount + new->br_blockcount;
1534
1535                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1536                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx), temp);
1537                 oldlen = startblockval(left.br_startblock) +
1538                         startblockval(new->br_startblock);
1539                 newlen = xfs_bmap_worst_indlen(ip, temp);
1540                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, *idx),
1541                         nullstartblock((int)newlen));
1542                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1543                 break;
1544
1545         case BMAP_RIGHT_CONTIG:
1546                 /*
1547                  * New allocation is contiguous with a delayed allocation
1548                  * on the right.
1549                  * Merge the new allocation with the right neighbor.
1550                  */
1551                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1552                 temp = new->br_blockcount + right.br_blockcount;
1553                 oldlen = startblockval(new->br_startblock) +
1554                         startblockval(right.br_startblock);
1555                 newlen = xfs_bmap_worst_indlen(ip, temp);
1556                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx),
1557                         new->br_startoff,
1558                         nullstartblock((int)newlen), temp, right.br_state);
1559                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1560                 break;
1561
1562         case 0:
1563                 /*
1564                  * New allocation is not contiguous with another
1565                  * delayed allocation.
1566                  * Insert a new entry.
1567                  */
1568                 oldlen = newlen = 0;
1569                 xfs_iext_insert(ip, *idx, 1, new, state);
1570                 break;
1571         }
1572         if (oldlen != newlen) {
1573                 ASSERT(oldlen > newlen);
1574                 xfs_icsb_modify_counters(ip->i_mount, XFS_SBS_FDBLOCKS,
1575                         (int64_t)(oldlen - newlen), 0);
1576                 /*
1577                  * Nothing to do for disk quota accounting here.
1578                  */
1579         }
1580 }
1581
1582 /*
1583  * Convert a hole to a real allocation.
1584  */
1585 STATIC int                              /* error */
1586 xfs_bmap_add_extent_hole_real(
1587         struct xfs_trans        *tp,
1588         xfs_inode_t             *ip,    /* incore inode pointer */
1589         xfs_extnum_t            *idx,   /* extent number to update/insert */
1590         xfs_btree_cur_t         **curp, /* if null, not a btree */
1591         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
1592         xfs_fsblock_t           *first, /* pointer to firstblock variable */
1593         xfs_bmap_free_t         *flist, /* list of extents to be freed */
1594         int                     *logflagsp, /* inode logging flags */
1595         int                     whichfork) /* data or attr fork */
1596 {
1597         int                     error;  /* error return value */
1598         int                     i;      /* temp state */
1599         xfs_btree_cur_t         *cur;   /* if null, not a btree */
1600         xfs_ifork_t             *ifp;   /* inode fork pointer */
1601         xfs_bmbt_irec_t         left;   /* left neighbor extent entry */
1602         xfs_bmbt_irec_t         right;  /* right neighbor extent entry */
1603         int                     rval=0; /* return value (logging flags) */
1604         int                     state;  /* state bits, accessed thru macros */
1605
1606         *logflagsp = 0;
1607
1608         ifp = XFS_IFORK_PTR(ip, whichfork);
1609         cur = *curp;
1610
1611         ASSERT(*idx >= 0);
1612         ASSERT(*idx <= ifp->if_bytes / sizeof(struct xfs_bmbt_rec));
1613         ASSERT(!isnullstartblock(new->br_startblock));
1614         ASSERT(!cur || !(cur->bc_private.b.flags & XFS_BTCUR_BPRV_WASDEL));
1615
1616         XFS_STATS_INC(xs_add_exlist);
1617
1618         state = 0;
1619         if (whichfork == XFS_ATTR_FORK)
1620                 state |= BMAP_ATTRFORK;
1621
1622         /*
1623          * Check and set flags if this segment has a left neighbor.
1624          */
1625         if (*idx > 0) {
1626                 state |= BMAP_LEFT_VALID;
1627                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &left);
1628                 if (isnullstartblock(left.br_startblock))
1629                         state |= BMAP_LEFT_DELAY;
1630         }
1631
1632         /*
1633          * Check and set flags if this segment has a current value.
1634          * Not true if we're inserting into the "hole" at eof.
1635          */
1636         if (*idx < ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)) {
1637                 state |= BMAP_RIGHT_VALID;
1638                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx), &right);
1639                 if (isnullstartblock(right.br_startblock))
1640                         state |= BMAP_RIGHT_DELAY;
1641         }
1642
1643         /*
1644          * We're inserting a real allocation between "left" and "right".
1645          * Set the contiguity flags.  Don't let extents get too large.
1646          */
1647         if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
1648             left.br_startoff + left.br_blockcount == new->br_startoff &&
1649             left.br_startblock + left.br_blockcount == new->br_startblock &&
1650             left.br_state == new->br_state &&
1651             left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
1652                 state |= BMAP_LEFT_CONTIG;
1653
1654         if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
1655             new->br_startoff + new->br_blockcount == right.br_startoff &&
1656             new->br_startblock + new->br_blockcount == right.br_startblock &&
1657             new->br_state == right.br_state &&
1658             new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
1659             (!(state & BMAP_LEFT_CONTIG) ||
1660              left.br_blockcount + new->br_blockcount +
1661              right.br_blockcount <= MAXEXTLEN))
1662                 state |= BMAP_RIGHT_CONTIG;
1663
1664         error = 0;
1665         /*
1666          * Select which case we're in here, and implement it.
1667          */
1668         switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
1669         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1670                 /*
1671                  * New allocation is contiguous with real allocations on the
1672                  * left and on the right.
1673                  * Merge all three into a single extent record.
1674                  */
1675                 --*idx;
1676                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1677                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
1678                         left.br_blockcount + new->br_blockcount +
1679                         right.br_blockcount);
1680                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1681
1682                 xfs_iext_remove(ip, *idx + 1, 1, state);
1683
1684                 XFS_IFORK_NEXT_SET(ip, whichfork,
1685                         XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
1686                 if (cur == NULL) {
1687                         rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
1688                 } else {
1689                         rval = XFS_ILOG_CORE;
1690                         if ((error = xfs_bmbt_lookup_eq(cur,
1691                                         right.br_startoff,
1692                                         right.br_startblock,
1693                                         right.br_blockcount, &i)))
1694                                 goto done;
1695                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1696                         if ((error = xfs_btree_delete(cur, &i)))
1697                                 goto done;
1698                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1699                         if ((error = xfs_btree_decrement(cur, 0, &i)))
1700                                 goto done;
1701                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1702                         if ((error = xfs_bmbt_update(cur, left.br_startoff,
1703                                         left.br_startblock,
1704                                         left.br_blockcount +
1705                                                 new->br_blockcount +
1706                                                 right.br_blockcount,
1707                                         left.br_state)))
1708                                 goto done;
1709                 }
1710                 break;
1711
1712         case BMAP_LEFT_CONTIG:
1713                 /*
1714                  * New allocation is contiguous with a real allocation
1715                  * on the left.
1716                  * Merge the new allocation with the left neighbor.
1717                  */
1718                 --*idx;
1719                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1720                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
1721                         left.br_blockcount + new->br_blockcount);
1722                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1723
1724                 if (cur == NULL) {
1725                         rval = xfs_ilog_fext(whichfork);
1726                 } else {
1727                         rval = 0;
1728                         if ((error = xfs_bmbt_lookup_eq(cur,
1729                                         left.br_startoff,
1730                                         left.br_startblock,
1731                                         left.br_blockcount, &i)))
1732                                 goto done;
1733                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1734                         if ((error = xfs_bmbt_update(cur, left.br_startoff,
1735                                         left.br_startblock,
1736                                         left.br_blockcount +
1737                                                 new->br_blockcount,
1738                                         left.br_state)))
1739                                 goto done;
1740                 }
1741                 break;
1742
1743         case BMAP_RIGHT_CONTIG:
1744                 /*
1745                  * New allocation is contiguous with a real allocation
1746                  * on the right.
1747                  * Merge the new allocation with the right neighbor.
1748                  */
1749                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1750                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx),
1751                         new->br_startoff, new->br_startblock,
1752                         new->br_blockcount + right.br_blockcount,
1753                         right.br_state);
1754                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1755
1756                 if (cur == NULL) {
1757                         rval = xfs_ilog_fext(whichfork);
1758                 } else {
1759                         rval = 0;
1760                         if ((error = xfs_bmbt_lookup_eq(cur,
1761                                         right.br_startoff,
1762                                         right.br_startblock,
1763                                         right.br_blockcount, &i)))
1764                                 goto done;
1765                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1766                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
1767                                         new->br_startblock,
1768                                         new->br_blockcount +
1769                                                 right.br_blockcount,
1770                                         right.br_state)))
1771                                 goto done;
1772                 }
1773                 break;
1774
1775         case 0:
1776                 /*
1777                  * New allocation is not contiguous with another
1778                  * real allocation.
1779                  * Insert a new entry.
1780                  */
1781                 xfs_iext_insert(ip, *idx, 1, new, state);
1782                 XFS_IFORK_NEXT_SET(ip, whichfork,
1783                         XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
1784                 if (cur == NULL) {
1785                         rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
1786                 } else {
1787                         rval = XFS_ILOG_CORE;
1788                         if ((error = xfs_bmbt_lookup_eq(cur,
1789                                         new->br_startoff,
1790                                         new->br_startblock,
1791                                         new->br_blockcount, &i)))
1792                                 goto done;
1793                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1794                         cur->bc_rec.b.br_state = new->br_state;
1795                         if ((error = xfs_btree_insert(cur, &i)))
1796                                 goto done;
1797                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1798                 }
1799                 break;
1800         }
1801
1802         /* convert to a btree if necessary */
1803         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
1804             XFS_IFORK_NEXTENTS(ip, whichfork) > ifp->if_ext_max) {
1805                 int     tmp_logflags;   /* partial log flag return val */
1806
1807                 ASSERT(cur == NULL);
1808                 error = xfs_bmap_extents_to_btree(tp, ip, first,
1809                         flist, &cur, 0, &tmp_logflags, whichfork);
1810                 *logflagsp |= tmp_logflags;
1811                 if (error)
1812                         goto done;
1813         }
1814
1815         /* clear out the allocated field, done with it now in any case. */
1816         if (cur) {
1817                 cur->bc_private.b.allocated = 0;
1818                 *curp = cur;
1819         }
1820         xfs_bmap_check_leaf_extents(cur, ip, whichfork);
1821 done:
1822         *logflagsp |= rval;
1823         return error;
1824 }
1825
1826 /*
1827  * Adjust the size of the new extent based on di_extsize and rt extsize.
1828  */
1829 STATIC int
1830 xfs_bmap_extsize_align(
1831         xfs_mount_t     *mp,
1832         xfs_bmbt_irec_t *gotp,          /* next extent pointer */
1833         xfs_bmbt_irec_t *prevp,         /* previous extent pointer */
1834         xfs_extlen_t    extsz,          /* align to this extent size */
1835         int             rt,             /* is this a realtime inode? */
1836         int             eof,            /* is extent at end-of-file? */
1837         int             delay,          /* creating delalloc extent? */
1838         int             convert,        /* overwriting unwritten extent? */
1839         xfs_fileoff_t   *offp,          /* in/out: aligned offset */
1840         xfs_extlen_t    *lenp)          /* in/out: aligned length */
1841 {
1842         xfs_fileoff_t   orig_off;       /* original offset */
1843         xfs_extlen_t    orig_alen;      /* original length */
1844         xfs_fileoff_t   orig_end;       /* original off+len */
1845         xfs_fileoff_t   nexto;          /* next file offset */
1846         xfs_fileoff_t   prevo;          /* previous file offset */
1847         xfs_fileoff_t   align_off;      /* temp for offset */
1848         xfs_extlen_t    align_alen;     /* temp for length */
1849         xfs_extlen_t    temp;           /* temp for calculations */
1850
1851         if (convert)
1852                 return 0;
1853
1854         orig_off = align_off = *offp;
1855         orig_alen = align_alen = *lenp;
1856         orig_end = orig_off + orig_alen;
1857
1858         /*
1859          * If this request overlaps an existing extent, then don't
1860          * attempt to perform any additional alignment.
1861          */
1862         if (!delay && !eof &&
1863             (orig_off >= gotp->br_startoff) &&
1864             (orig_end <= gotp->br_startoff + gotp->br_blockcount)) {
1865                 return 0;
1866         }
1867
1868         /*
1869          * If the file offset is unaligned vs. the extent size
1870          * we need to align it.  This will be possible unless
1871          * the file was previously written with a kernel that didn't
1872          * perform this alignment, or if a truncate shot us in the
1873          * foot.
1874          */
1875         temp = do_mod(orig_off, extsz);
1876         if (temp) {
1877                 align_alen += temp;
1878                 align_off -= temp;
1879         }
1880         /*
1881          * Same adjustment for the end of the requested area.
1882          */
1883         if ((temp = (align_alen % extsz))) {
1884                 align_alen += extsz - temp;
1885         }
1886         /*
1887          * If the previous block overlaps with this proposed allocation
1888          * then move the start forward without adjusting the length.
1889          */
1890         if (prevp->br_startoff != NULLFILEOFF) {
1891                 if (prevp->br_startblock == HOLESTARTBLOCK)
1892                         prevo = prevp->br_startoff;
1893                 else
1894                         prevo = prevp->br_startoff + prevp->br_blockcount;
1895         } else
1896                 prevo = 0;
1897         if (align_off != orig_off && align_off < prevo)
1898                 align_off = prevo;
1899         /*
1900          * If the next block overlaps with this proposed allocation
1901          * then move the start back without adjusting the length,
1902          * but not before offset 0.
1903          * This may of course make the start overlap previous block,
1904          * and if we hit the offset 0 limit then the next block
1905          * can still overlap too.
1906          */
1907         if (!eof && gotp->br_startoff != NULLFILEOFF) {
1908                 if ((delay && gotp->br_startblock == HOLESTARTBLOCK) ||
1909                     (!delay && gotp->br_startblock == DELAYSTARTBLOCK))
1910                         nexto = gotp->br_startoff + gotp->br_blockcount;
1911                 else
1912                         nexto = gotp->br_startoff;
1913         } else
1914                 nexto = NULLFILEOFF;
1915         if (!eof &&
1916             align_off + align_alen != orig_end &&
1917             align_off + align_alen > nexto)
1918                 align_off = nexto > align_alen ? nexto - align_alen : 0;
1919         /*
1920          * If we're now overlapping the next or previous extent that
1921          * means we can't fit an extsz piece in this hole.  Just move
1922          * the start forward to the first valid spot and set
1923          * the length so we hit the end.
1924          */
1925         if (align_off != orig_off && align_off < prevo)
1926                 align_off = prevo;
1927         if (align_off + align_alen != orig_end &&
1928             align_off + align_alen > nexto &&
1929             nexto != NULLFILEOFF) {
1930                 ASSERT(nexto > prevo);
1931                 align_alen = nexto - align_off;
1932         }
1933
1934         /*
1935          * If realtime, and the result isn't a multiple of the realtime
1936          * extent size we need to remove blocks until it is.
1937          */
1938         if (rt && (temp = (align_alen % mp->m_sb.sb_rextsize))) {
1939                 /*
1940                  * We're not covering the original request, or
1941                  * we won't be able to once we fix the length.
1942                  */
1943                 if (orig_off < align_off ||
1944                     orig_end > align_off + align_alen ||
1945                     align_alen - temp < orig_alen)
1946                         return XFS_ERROR(EINVAL);
1947                 /*
1948                  * Try to fix it by moving the start up.
1949                  */
1950                 if (align_off + temp <= orig_off) {
1951                         align_alen -= temp;
1952                         align_off += temp;
1953                 }
1954                 /*
1955                  * Try to fix it by moving the end in.
1956                  */
1957                 else if (align_off + align_alen - temp >= orig_end)
1958                         align_alen -= temp;
1959                 /*
1960                  * Set the start to the minimum then trim the length.
1961                  */
1962                 else {
1963                         align_alen -= orig_off - align_off;
1964                         align_off = orig_off;
1965                         align_alen -= align_alen % mp->m_sb.sb_rextsize;
1966                 }
1967                 /*
1968                  * Result doesn't cover the request, fail it.
1969                  */
1970                 if (orig_off < align_off || orig_end > align_off + align_alen)
1971                         return XFS_ERROR(EINVAL);
1972         } else {
1973                 ASSERT(orig_off >= align_off);
1974                 ASSERT(orig_end <= align_off + align_alen);
1975         }
1976
1977 #ifdef DEBUG
1978         if (!eof && gotp->br_startoff != NULLFILEOFF)
1979                 ASSERT(align_off + align_alen <= gotp->br_startoff);
1980         if (prevp->br_startoff != NULLFILEOFF)
1981                 ASSERT(align_off >= prevp->br_startoff + prevp->br_blockcount);
1982 #endif
1983
1984         *lenp = align_alen;
1985         *offp = align_off;
1986         return 0;
1987 }
1988
1989 #define XFS_ALLOC_GAP_UNITS     4
1990
1991 STATIC void
1992 xfs_bmap_adjacent(
1993         xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
1994 {
1995         xfs_fsblock_t   adjust;         /* adjustment to block numbers */
1996         xfs_agnumber_t  fb_agno;        /* ag number of ap->firstblock */
1997         xfs_mount_t     *mp;            /* mount point structure */
1998         int             nullfb;         /* true if ap->firstblock isn't set */
1999         int             rt;             /* true if inode is realtime */
2000
2001 #define ISVALID(x,y)    \
2002         (rt ? \
2003                 (x) < mp->m_sb.sb_rblocks : \
2004                 XFS_FSB_TO_AGNO(mp, x) == XFS_FSB_TO_AGNO(mp, y) && \
2005                 XFS_FSB_TO_AGNO(mp, x) < mp->m_sb.sb_agcount && \
2006                 XFS_FSB_TO_AGBNO(mp, x) < mp->m_sb.sb_agblocks)
2007
2008         mp = ap->ip->i_mount;
2009         nullfb = *ap->firstblock == NULLFSBLOCK;
2010         rt = XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata;
2011         fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, *ap->firstblock);
2012         /*
2013          * If allocating at eof, and there's a previous real block,
2014          * try to use its last block as our starting point.
2015          */
2016         if (ap->eof && ap->prev.br_startoff != NULLFILEOFF &&
2017             !isnullstartblock(ap->prev.br_startblock) &&
2018             ISVALID(ap->prev.br_startblock + ap->prev.br_blockcount,
2019                     ap->prev.br_startblock)) {
2020                 ap->rval = ap->prev.br_startblock + ap->prev.br_blockcount;
2021                 /*
2022                  * Adjust for the gap between prevp and us.
2023                  */
2024                 adjust = ap->off -
2025                         (ap->prev.br_startoff + ap->prev.br_blockcount);
2026                 if (adjust &&
2027                     ISVALID(ap->rval + adjust, ap->prev.br_startblock))
2028                         ap->rval += adjust;
2029         }
2030         /*
2031          * If not at eof, then compare the two neighbor blocks.
2032          * Figure out whether either one gives us a good starting point,
2033          * and pick the better one.
2034          */
2035         else if (!ap->eof) {
2036                 xfs_fsblock_t   gotbno;         /* right side block number */
2037                 xfs_fsblock_t   gotdiff=0;      /* right side difference */
2038                 xfs_fsblock_t   prevbno;        /* left side block number */
2039                 xfs_fsblock_t   prevdiff=0;     /* left side difference */
2040
2041                 /*
2042                  * If there's a previous (left) block, select a requested
2043                  * start block based on it.
2044                  */
2045                 if (ap->prev.br_startoff != NULLFILEOFF &&
2046                     !isnullstartblock(ap->prev.br_startblock) &&
2047                     (prevbno = ap->prev.br_startblock +
2048                                ap->prev.br_blockcount) &&
2049                     ISVALID(prevbno, ap->prev.br_startblock)) {
2050                         /*
2051                          * Calculate gap to end of previous block.
2052                          */
2053                         adjust = prevdiff = ap->off -
2054                                 (ap->prev.br_startoff +
2055                                  ap->prev.br_blockcount);
2056                         /*
2057                          * Figure the startblock based on the previous block's
2058                          * end and the gap size.
2059                          * Heuristic!
2060                          * If the gap is large relative to the piece we're
2061                          * allocating, or using it gives us an invalid block
2062                          * number, then just use the end of the previous block.
2063                          */
2064                         if (prevdiff <= XFS_ALLOC_GAP_UNITS * ap->alen &&
2065                             ISVALID(prevbno + prevdiff,
2066                                     ap->prev.br_startblock))
2067                                 prevbno += adjust;
2068                         else
2069                                 prevdiff += adjust;
2070                         /*
2071                          * If the firstblock forbids it, can't use it,
2072                          * must use default.
2073                          */
2074                         if (!rt && !nullfb &&
2075                             XFS_FSB_TO_AGNO(mp, prevbno) != fb_agno)
2076                                 prevbno = NULLFSBLOCK;
2077                 }
2078                 /*
2079                  * No previous block or can't follow it, just default.
2080                  */
2081                 else
2082                         prevbno = NULLFSBLOCK;
2083                 /*
2084                  * If there's a following (right) block, select a requested
2085                  * start block based on it.
2086                  */
2087                 if (!isnullstartblock(ap->got.br_startblock)) {
2088                         /*
2089                          * Calculate gap to start of next block.
2090                          */
2091                         adjust = gotdiff = ap->got.br_startoff - ap->off;
2092                         /*
2093                          * Figure the startblock based on the next block's
2094                          * start and the gap size.
2095                          */
2096                         gotbno = ap->got.br_startblock;
2097                         /*
2098                          * Heuristic!
2099                          * If the gap is large relative to the piece we're
2100                          * allocating, or using it gives us an invalid block
2101                          * number, then just use the start of the next block
2102                          * offset by our length.
2103                          */
2104                         if (gotdiff <= XFS_ALLOC_GAP_UNITS * ap->alen &&
2105                             ISVALID(gotbno - gotdiff, gotbno))
2106                                 gotbno -= adjust;
2107                         else if (ISVALID(gotbno - ap->alen, gotbno)) {
2108                                 gotbno -= ap->alen;
2109                                 gotdiff += adjust - ap->alen;
2110                         } else
2111                                 gotdiff += adjust;
2112                         /*
2113                          * If the firstblock forbids it, can't use it,
2114                          * must use default.
2115                          */
2116                         if (!rt && !nullfb &&
2117                             XFS_FSB_TO_AGNO(mp, gotbno) != fb_agno)
2118                                 gotbno = NULLFSBLOCK;
2119                 }
2120                 /*
2121                  * No next block, just default.
2122                  */
2123                 else
2124                         gotbno = NULLFSBLOCK;
2125                 /*
2126                  * If both valid, pick the better one, else the only good
2127                  * one, else ap->rval is already set (to 0 or the inode block).
2128                  */
2129                 if (prevbno != NULLFSBLOCK && gotbno != NULLFSBLOCK)
2130                         ap->rval = prevdiff <= gotdiff ? prevbno : gotbno;
2131                 else if (prevbno != NULLFSBLOCK)
2132                         ap->rval = prevbno;
2133                 else if (gotbno != NULLFSBLOCK)
2134                         ap->rval = gotbno;
2135         }
2136 #undef ISVALID
2137 }
2138
2139 STATIC int
2140 xfs_bmap_rtalloc(
2141         xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
2142 {
2143         xfs_alloctype_t atype = 0;      /* type for allocation routines */
2144         int             error;          /* error return value */
2145         xfs_mount_t     *mp;            /* mount point structure */
2146         xfs_extlen_t    prod = 0;       /* product factor for allocators */
2147         xfs_extlen_t    ralen = 0;      /* realtime allocation length */
2148         xfs_extlen_t    align;          /* minimum allocation alignment */
2149         xfs_rtblock_t   rtb;
2150
2151         mp = ap->ip->i_mount;
2152         align = xfs_get_extsz_hint(ap->ip);
2153         prod = align / mp->m_sb.sb_rextsize;
2154         error = xfs_bmap_extsize_align(mp, &ap->got, &ap->prev,
2155                                         align, 1, ap->eof, 0,
2156                                         ap->conv, &ap->off, &ap->alen);
2157         if (error)
2158                 return error;
2159         ASSERT(ap->alen);
2160         ASSERT(ap->alen % mp->m_sb.sb_rextsize == 0);
2161
2162         /*
2163          * If the offset & length are not perfectly aligned
2164          * then kill prod, it will just get us in trouble.
2165          */
2166         if (do_mod(ap->off, align) || ap->alen % align)
2167                 prod = 1;
2168         /*
2169          * Set ralen to be the actual requested length in rtextents.
2170          */
2171         ralen = ap->alen / mp->m_sb.sb_rextsize;
2172         /*
2173          * If the old value was close enough to MAXEXTLEN that
2174          * we rounded up to it, cut it back so it's valid again.
2175          * Note that if it's a really large request (bigger than
2176          * MAXEXTLEN), we don't hear about that number, and can't
2177          * adjust the starting point to match it.
2178          */
2179         if (ralen * mp->m_sb.sb_rextsize >= MAXEXTLEN)
2180                 ralen = MAXEXTLEN / mp->m_sb.sb_rextsize;
2181
2182         /*
2183          * Lock out other modifications to the RT bitmap inode.
2184          */
2185         xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL);
2186         xfs_trans_ijoin_ref(ap->tp, mp->m_rbmip, XFS_ILOCK_EXCL);
2187
2188         /*
2189          * If it's an allocation to an empty file at offset 0,
2190          * pick an extent that will space things out in the rt area.
2191          */
2192         if (ap->eof && ap->off == 0) {
2193                 xfs_rtblock_t uninitialized_var(rtx); /* realtime extent no */
2194
2195                 error = xfs_rtpick_extent(mp, ap->tp, ralen, &rtx);
2196                 if (error)
2197                         return error;
2198                 ap->rval = rtx * mp->m_sb.sb_rextsize;
2199         } else {
2200                 ap->rval = 0;
2201         }
2202
2203         xfs_bmap_adjacent(ap);
2204
2205         /*
2206          * Realtime allocation, done through xfs_rtallocate_extent.
2207          */
2208         atype = ap->rval == 0 ?  XFS_ALLOCTYPE_ANY_AG : XFS_ALLOCTYPE_NEAR_BNO;
2209         do_div(ap->rval, mp->m_sb.sb_rextsize);
2210         rtb = ap->rval;
2211         ap->alen = ralen;
2212         if ((error = xfs_rtallocate_extent(ap->tp, ap->rval, 1, ap->alen,
2213                                 &ralen, atype, ap->wasdel, prod, &rtb)))
2214                 return error;
2215         if (rtb == NULLFSBLOCK && prod > 1 &&
2216             (error = xfs_rtallocate_extent(ap->tp, ap->rval, 1,
2217                                            ap->alen, &ralen, atype,
2218                                            ap->wasdel, 1, &rtb)))
2219                 return error;
2220         ap->rval = rtb;
2221         if (ap->rval != NULLFSBLOCK) {
2222                 ap->rval *= mp->m_sb.sb_rextsize;
2223                 ralen *= mp->m_sb.sb_rextsize;
2224                 ap->alen = ralen;
2225                 ap->ip->i_d.di_nblocks += ralen;
2226                 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
2227                 if (ap->wasdel)
2228                         ap->ip->i_delayed_blks -= ralen;
2229                 /*
2230                  * Adjust the disk quota also. This was reserved
2231                  * earlier.
2232                  */
2233                 xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
2234                         ap->wasdel ? XFS_TRANS_DQ_DELRTBCOUNT :
2235                                         XFS_TRANS_DQ_RTBCOUNT, (long) ralen);
2236         } else {
2237                 ap->alen = 0;
2238         }
2239         return 0;
2240 }
2241
2242 STATIC int
2243 xfs_bmap_btalloc_nullfb(
2244         struct xfs_bmalloca     *ap,
2245         struct xfs_alloc_arg    *args,
2246         xfs_extlen_t            *blen)
2247 {
2248         struct xfs_mount        *mp = ap->ip->i_mount;
2249         struct xfs_perag        *pag;
2250         xfs_agnumber_t          ag, startag;
2251         int                     notinit = 0;
2252         int                     error;
2253
2254         if (ap->userdata && xfs_inode_is_filestream(ap->ip))
2255                 args->type = XFS_ALLOCTYPE_NEAR_BNO;
2256         else
2257                 args->type = XFS_ALLOCTYPE_START_BNO;
2258         args->total = ap->total;
2259
2260         /*
2261          * Search for an allocation group with a single extent large enough
2262          * for the request.  If one isn't found, then adjust the minimum
2263          * allocation size to the largest space found.
2264          */
2265         startag = ag = XFS_FSB_TO_AGNO(mp, args->fsbno);
2266         if (startag == NULLAGNUMBER)
2267                 startag = ag = 0;
2268
2269         pag = xfs_perag_get(mp, ag);
2270         while (*blen < args->maxlen) {
2271                 if (!pag->pagf_init) {
2272                         error = xfs_alloc_pagf_init(mp, args->tp, ag,
2273                                                     XFS_ALLOC_FLAG_TRYLOCK);
2274                         if (error) {
2275                                 xfs_perag_put(pag);
2276                                 return error;
2277                         }
2278                 }
2279
2280                 /*
2281                  * See xfs_alloc_fix_freelist...
2282                  */
2283                 if (pag->pagf_init) {
2284                         xfs_extlen_t    longest;
2285                         longest = xfs_alloc_longest_free_extent(mp, pag);
2286                         if (*blen < longest)
2287                                 *blen = longest;
2288                 } else
2289                         notinit = 1;
2290
2291                 if (xfs_inode_is_filestream(ap->ip)) {
2292                         if (*blen >= args->maxlen)
2293                                 break;
2294
2295                         if (ap->userdata) {
2296                                 /*
2297                                  * If startag is an invalid AG, we've
2298                                  * come here once before and
2299                                  * xfs_filestream_new_ag picked the
2300                                  * best currently available.
2301                                  *
2302                                  * Don't continue looping, since we
2303                                  * could loop forever.
2304                                  */
2305                                 if (startag == NULLAGNUMBER)
2306                                         break;
2307
2308                                 error = xfs_filestream_new_ag(ap, &ag);
2309                                 xfs_perag_put(pag);
2310                                 if (error)
2311                                         return error;
2312
2313                                 /* loop again to set 'blen'*/
2314                                 startag = NULLAGNUMBER;
2315                                 pag = xfs_perag_get(mp, ag);
2316                                 continue;
2317                         }
2318                 }
2319                 if (++ag == mp->m_sb.sb_agcount)
2320                         ag = 0;
2321                 if (ag == startag)
2322                         break;
2323                 xfs_perag_put(pag);
2324                 pag = xfs_perag_get(mp, ag);
2325         }
2326         xfs_perag_put(pag);
2327
2328         /*
2329          * Since the above loop did a BUF_TRYLOCK, it is
2330          * possible that there is space for this request.
2331          */
2332         if (notinit || *blen < ap->minlen)
2333                 args->minlen = ap->minlen;
2334         /*
2335          * If the best seen length is less than the request
2336          * length, use the best as the minimum.
2337          */
2338         else if (*blen < args->maxlen)
2339                 args->minlen = *blen;
2340         /*
2341          * Otherwise we've seen an extent as big as maxlen,
2342          * use that as the minimum.
2343          */
2344         else
2345                 args->minlen = args->maxlen;
2346
2347         /*
2348          * set the failure fallback case to look in the selected
2349          * AG as the stream may have moved.
2350          */
2351         if (xfs_inode_is_filestream(ap->ip))
2352                 ap->rval = args->fsbno = XFS_AGB_TO_FSB(mp, ag, 0);
2353
2354         return 0;
2355 }
2356
2357 STATIC int
2358 xfs_bmap_btalloc(
2359         xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
2360 {
2361         xfs_mount_t     *mp;            /* mount point structure */
2362         xfs_alloctype_t atype = 0;      /* type for allocation routines */
2363         xfs_extlen_t    align;          /* minimum allocation alignment */
2364         xfs_agnumber_t  fb_agno;        /* ag number of ap->firstblock */
2365         xfs_agnumber_t  ag;
2366         xfs_alloc_arg_t args;
2367         xfs_extlen_t    blen;
2368         xfs_extlen_t    nextminlen = 0;
2369         int             nullfb;         /* true if ap->firstblock isn't set */
2370         int             isaligned;
2371         int             tryagain;
2372         int             error;
2373
2374         mp = ap->ip->i_mount;
2375         align = ap->userdata ? xfs_get_extsz_hint(ap->ip) : 0;
2376         if (unlikely(align)) {
2377                 error = xfs_bmap_extsize_align(mp, &ap->got, &ap->prev,
2378                                                 align, 0, ap->eof, 0, ap->conv,
2379                                                 &ap->off, &ap->alen);
2380                 ASSERT(!error);
2381                 ASSERT(ap->alen);
2382         }
2383         nullfb = *ap->firstblock == NULLFSBLOCK;
2384         fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, *ap->firstblock);
2385         if (nullfb) {
2386                 if (ap->userdata && xfs_inode_is_filestream(ap->ip)) {
2387                         ag = xfs_filestream_lookup_ag(ap->ip);
2388                         ag = (ag != NULLAGNUMBER) ? ag : 0;
2389                         ap->rval = XFS_AGB_TO_FSB(mp, ag, 0);
2390                 } else {
2391                         ap->rval = XFS_INO_TO_FSB(mp, ap->ip->i_ino);
2392                 }
2393         } else
2394                 ap->rval = *ap->firstblock;
2395
2396         xfs_bmap_adjacent(ap);
2397
2398         /*
2399          * If allowed, use ap->rval; otherwise must use firstblock since
2400          * it's in the right allocation group.
2401          */
2402         if (nullfb || XFS_FSB_TO_AGNO(mp, ap->rval) == fb_agno)
2403                 ;
2404         else
2405                 ap->rval = *ap->firstblock;
2406         /*
2407          * Normal allocation, done through xfs_alloc_vextent.
2408          */
2409         tryagain = isaligned = 0;
2410         args.tp = ap->tp;
2411         args.mp = mp;
2412         args.fsbno = ap->rval;
2413
2414         /* Trim the allocation back to the maximum an AG can fit. */
2415         args.maxlen = MIN(ap->alen, XFS_ALLOC_AG_MAX_USABLE(mp));
2416         args.firstblock = *ap->firstblock;
2417         blen = 0;
2418         if (nullfb) {
2419                 error = xfs_bmap_btalloc_nullfb(ap, &args, &blen);
2420                 if (error)
2421                         return error;
2422         } else if (ap->flist->xbf_low) {
2423                 if (xfs_inode_is_filestream(ap->ip))
2424                         args.type = XFS_ALLOCTYPE_FIRST_AG;
2425                 else
2426                         args.type = XFS_ALLOCTYPE_START_BNO;
2427                 args.total = args.minlen = ap->minlen;
2428         } else {
2429                 args.type = XFS_ALLOCTYPE_NEAR_BNO;
2430                 args.total = ap->total;
2431                 args.minlen = ap->minlen;
2432         }
2433         /* apply extent size hints if obtained earlier */
2434         if (unlikely(align)) {
2435                 args.prod = align;
2436                 if ((args.mod = (xfs_extlen_t)do_mod(ap->off, args.prod)))
2437                         args.mod = (xfs_extlen_t)(args.prod - args.mod);
2438         } else if (mp->m_sb.sb_blocksize >= PAGE_CACHE_SIZE) {
2439                 args.prod = 1;
2440                 args.mod = 0;
2441         } else {
2442                 args.prod = PAGE_CACHE_SIZE >> mp->m_sb.sb_blocklog;
2443                 if ((args.mod = (xfs_extlen_t)(do_mod(ap->off, args.prod))))
2444                         args.mod = (xfs_extlen_t)(args.prod - args.mod);
2445         }
2446         /*
2447          * If we are not low on available data blocks, and the
2448          * underlying logical volume manager is a stripe, and
2449          * the file offset is zero then try to allocate data
2450          * blocks on stripe unit boundary.
2451          * NOTE: ap->aeof is only set if the allocation length
2452          * is >= the stripe unit and the allocation offset is
2453          * at the end of file.
2454          */
2455         if (!ap->flist->xbf_low && ap->aeof) {
2456                 if (!ap->off) {
2457                         args.alignment = mp->m_dalign;
2458                         atype = args.type;
2459                         isaligned = 1;
2460                         /*
2461                          * Adjust for alignment
2462                          */
2463                         if (blen > args.alignment && blen <= args.maxlen)
2464                                 args.minlen = blen - args.alignment;
2465                         args.minalignslop = 0;
2466                 } else {
2467                         /*
2468                          * First try an exact bno allocation.
2469                          * If it fails then do a near or start bno
2470                          * allocation with alignment turned on.
2471                          */
2472                         atype = args.type;
2473                         tryagain = 1;
2474                         args.type = XFS_ALLOCTYPE_THIS_BNO;
2475                         args.alignment = 1;
2476                         /*
2477                          * Compute the minlen+alignment for the
2478                          * next case.  Set slop so that the value
2479                          * of minlen+alignment+slop doesn't go up
2480                          * between the calls.
2481                          */
2482                         if (blen > mp->m_dalign && blen <= args.maxlen)
2483                                 nextminlen = blen - mp->m_dalign;
2484                         else
2485                                 nextminlen = args.minlen;
2486                         if (nextminlen + mp->m_dalign > args.minlen + 1)
2487                                 args.minalignslop =
2488                                         nextminlen + mp->m_dalign -
2489                                         args.minlen - 1;
2490                         else
2491                                 args.minalignslop = 0;
2492                 }
2493         } else {
2494                 args.alignment = 1;
2495                 args.minalignslop = 0;
2496         }
2497         args.minleft = ap->minleft;
2498         args.wasdel = ap->wasdel;
2499         args.isfl = 0;
2500         args.userdata = ap->userdata;
2501         if ((error = xfs_alloc_vextent(&args)))
2502                 return error;
2503         if (tryagain && args.fsbno == NULLFSBLOCK) {
2504                 /*
2505                  * Exact allocation failed. Now try with alignment
2506                  * turned on.
2507                  */
2508                 args.type = atype;
2509                 args.fsbno = ap->rval;
2510                 args.alignment = mp->m_dalign;
2511                 args.minlen = nextminlen;
2512                 args.minalignslop = 0;
2513                 isaligned = 1;
2514                 if ((error = xfs_alloc_vextent(&args)))
2515                         return error;
2516         }
2517         if (isaligned && args.fsbno == NULLFSBLOCK) {
2518                 /*
2519                  * allocation failed, so turn off alignment and
2520                  * try again.
2521                  */
2522                 args.type = atype;
2523                 args.fsbno = ap->rval;
2524                 args.alignment = 0;
2525                 if ((error = xfs_alloc_vextent(&args)))
2526                         return error;
2527         }
2528         if (args.fsbno == NULLFSBLOCK && nullfb &&
2529             args.minlen > ap->minlen) {
2530                 args.minlen = ap->minlen;
2531                 args.type = XFS_ALLOCTYPE_START_BNO;
2532                 args.fsbno = ap->rval;
2533                 if ((error = xfs_alloc_vextent(&args)))
2534                         return error;
2535         }
2536         if (args.fsbno == NULLFSBLOCK && nullfb) {
2537                 args.fsbno = 0;
2538                 args.type = XFS_ALLOCTYPE_FIRST_AG;
2539                 args.total = ap->minlen;
2540                 args.minleft = 0;
2541                 if ((error = xfs_alloc_vextent(&args)))
2542                         return error;
2543                 ap->flist->xbf_low = 1;
2544         }
2545         if (args.fsbno != NULLFSBLOCK) {
2546                 /*
2547                  * check the allocation happened at the same or higher AG than
2548                  * the first block that was allocated.
2549                  */
2550                 ASSERT(*ap->firstblock == NULLFSBLOCK ||
2551                        XFS_FSB_TO_AGNO(mp, *ap->firstblock) ==
2552                        XFS_FSB_TO_AGNO(mp, args.fsbno) ||
2553                        (ap->flist->xbf_low &&
2554                         XFS_FSB_TO_AGNO(mp, *ap->firstblock) <
2555                         XFS_FSB_TO_AGNO(mp, args.fsbno)));
2556
2557                 ap->rval = args.fsbno;
2558                 if (*ap->firstblock == NULLFSBLOCK)
2559                         *ap->firstblock = args.fsbno;
2560                 ASSERT(nullfb || fb_agno == args.agno ||
2561                        (ap->flist->xbf_low && fb_agno < args.agno));
2562                 ap->alen = args.len;
2563                 ap->ip->i_d.di_nblocks += args.len;
2564                 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
2565                 if (ap->wasdel)
2566                         ap->ip->i_delayed_blks -= args.len;
2567                 /*
2568                  * Adjust the disk quota also. This was reserved
2569                  * earlier.
2570                  */
2571                 xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
2572                         ap->wasdel ? XFS_TRANS_DQ_DELBCOUNT :
2573                                         XFS_TRANS_DQ_BCOUNT,
2574                         (long) args.len);
2575         } else {
2576                 ap->rval = NULLFSBLOCK;
2577                 ap->alen = 0;
2578         }
2579         return 0;
2580 }
2581
2582 /*
2583  * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
2584  * It figures out where to ask the underlying allocator to put the new extent.
2585  */
2586 STATIC int
2587 xfs_bmap_alloc(
2588         xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
2589 {
2590         if (XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata)
2591                 return xfs_bmap_rtalloc(ap);
2592         return xfs_bmap_btalloc(ap);
2593 }
2594
2595 /*
2596  * Transform a btree format file with only one leaf node, where the
2597  * extents list will fit in the inode, into an extents format file.
2598  * Since the file extents are already in-core, all we have to do is
2599  * give up the space for the btree root and pitch the leaf block.
2600  */
2601 STATIC int                              /* error */
2602 xfs_bmap_btree_to_extents(
2603         xfs_trans_t             *tp,    /* transaction pointer */
2604         xfs_inode_t             *ip,    /* incore inode pointer */
2605         xfs_btree_cur_t         *cur,   /* btree cursor */
2606         int                     *logflagsp, /* inode logging flags */
2607         int                     whichfork)  /* data or attr fork */
2608 {
2609         /* REFERENCED */
2610         struct xfs_btree_block  *cblock;/* child btree block */
2611         xfs_fsblock_t           cbno;   /* child block number */
2612         xfs_buf_t               *cbp;   /* child block's buffer */
2613         int                     error;  /* error return value */
2614         xfs_ifork_t             *ifp;   /* inode fork data */
2615         xfs_mount_t             *mp;    /* mount point structure */
2616         __be64                  *pp;    /* ptr to block address */
2617         struct xfs_btree_block  *rblock;/* root btree block */
2618
2619         mp = ip->i_mount;
2620         ifp = XFS_IFORK_PTR(ip, whichfork);
2621         ASSERT(ifp->if_flags & XFS_IFEXTENTS);
2622         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
2623         rblock = ifp->if_broot;
2624         ASSERT(be16_to_cpu(rblock->bb_level) == 1);
2625         ASSERT(be16_to_cpu(rblock->bb_numrecs) == 1);
2626         ASSERT(xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0) == 1);
2627         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, rblock, 1, ifp->if_broot_bytes);
2628         cbno = be64_to_cpu(*pp);
2629         *logflagsp = 0;
2630 #ifdef DEBUG
2631         if ((error = xfs_btree_check_lptr(cur, cbno, 1)))
2632                 return error;
2633 #endif
2634         if ((error = xfs_btree_read_bufl(mp, tp, cbno, 0, &cbp,
2635                         XFS_BMAP_BTREE_REF)))
2636                 return error;
2637         cblock = XFS_BUF_TO_BLOCK(cbp);
2638         if ((error = xfs_btree_check_block(cur, cblock, 0, cbp)))
2639                 return error;
2640         xfs_bmap_add_free(cbno, 1, cur->bc_private.b.flist, mp);
2641         ip->i_d.di_nblocks--;
2642         xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, -1L);
2643         xfs_trans_binval(tp, cbp);
2644         if (cur->bc_bufs[0] == cbp)
2645                 cur->bc_bufs[0] = NULL;
2646         xfs_iroot_realloc(ip, -1, whichfork);
2647         ASSERT(ifp->if_broot == NULL);
2648         ASSERT((ifp->if_flags & XFS_IFBROOT) == 0);
2649         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
2650         *logflagsp = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
2651         return 0;
2652 }
2653
2654 /*
2655  * Called by xfs_bmapi to update file extent records and the btree
2656  * after removing space (or undoing a delayed allocation).
2657  */
2658 STATIC int                              /* error */
2659 xfs_bmap_del_extent(
2660         xfs_inode_t             *ip,    /* incore inode pointer */
2661         xfs_trans_t             *tp,    /* current transaction pointer */
2662         xfs_extnum_t            *idx,   /* extent number to update/delete */
2663         xfs_bmap_free_t         *flist, /* list of extents to be freed */
2664         xfs_btree_cur_t         *cur,   /* if null, not a btree */
2665         xfs_bmbt_irec_t         *del,   /* data to remove from extents */
2666         int                     *logflagsp, /* inode logging flags */
2667         int                     whichfork) /* data or attr fork */
2668 {
2669         xfs_filblks_t           da_new; /* new delay-alloc indirect blocks */
2670         xfs_filblks_t           da_old; /* old delay-alloc indirect blocks */
2671         xfs_fsblock_t           del_endblock=0; /* first block past del */
2672         xfs_fileoff_t           del_endoff;     /* first offset past del */
2673         int                     delay;  /* current block is delayed allocated */
2674         int                     do_fx;  /* free extent at end of routine */
2675         xfs_bmbt_rec_host_t     *ep;    /* current extent entry pointer */
2676         int                     error;  /* error return value */
2677         int                     flags;  /* inode logging flags */
2678         xfs_bmbt_irec_t         got;    /* current extent entry */
2679         xfs_fileoff_t           got_endoff;     /* first offset past got */
2680         int                     i;      /* temp state */
2681         xfs_ifork_t             *ifp;   /* inode fork pointer */
2682         xfs_mount_t             *mp;    /* mount structure */
2683         xfs_filblks_t           nblks;  /* quota/sb block count */
2684         xfs_bmbt_irec_t         new;    /* new record to be inserted */
2685         /* REFERENCED */
2686         uint                    qfield; /* quota field to update */
2687         xfs_filblks_t           temp;   /* for indirect length calculations */
2688         xfs_filblks_t           temp2;  /* for indirect length calculations */
2689         int                     state = 0;
2690
2691         XFS_STATS_INC(xs_del_exlist);
2692
2693         if (whichfork == XFS_ATTR_FORK)
2694                 state |= BMAP_ATTRFORK;
2695
2696         mp = ip->i_mount;
2697         ifp = XFS_IFORK_PTR(ip, whichfork);
2698         ASSERT((*idx >= 0) && (*idx < ifp->if_bytes /
2699                 (uint)sizeof(xfs_bmbt_rec_t)));
2700         ASSERT(del->br_blockcount > 0);
2701         ep = xfs_iext_get_ext(ifp, *idx);
2702         xfs_bmbt_get_all(ep, &got);
2703         ASSERT(got.br_startoff <= del->br_startoff);
2704         del_endoff = del->br_startoff + del->br_blockcount;
2705         got_endoff = got.br_startoff + got.br_blockcount;
2706         ASSERT(got_endoff >= del_endoff);
2707         delay = isnullstartblock(got.br_startblock);
2708         ASSERT(isnullstartblock(del->br_startblock) == delay);
2709         flags = 0;
2710         qfield = 0;
2711         error = 0;
2712         /*
2713          * If deleting a real allocation, must free up the disk space.
2714          */
2715         if (!delay) {
2716                 flags = XFS_ILOG_CORE;
2717                 /*
2718                  * Realtime allocation.  Free it and record di_nblocks update.
2719                  */
2720                 if (whichfork == XFS_DATA_FORK && XFS_IS_REALTIME_INODE(ip)) {
2721                         xfs_fsblock_t   bno;
2722                         xfs_filblks_t   len;
2723
2724                         ASSERT(do_mod(del->br_blockcount,
2725                                       mp->m_sb.sb_rextsize) == 0);
2726                         ASSERT(do_mod(del->br_startblock,
2727                                       mp->m_sb.sb_rextsize) == 0);
2728                         bno = del->br_startblock;
2729                         len = del->br_blockcount;
2730                         do_div(bno, mp->m_sb.sb_rextsize);
2731                         do_div(len, mp->m_sb.sb_rextsize);
2732                         error = xfs_rtfree_extent(tp, bno, (xfs_extlen_t)len);
2733                         if (error)
2734                                 goto done;
2735                         do_fx = 0;
2736                         nblks = len * mp->m_sb.sb_rextsize;
2737                         qfield = XFS_TRANS_DQ_RTBCOUNT;
2738                 }
2739                 /*
2740                  * Ordinary allocation.
2741                  */
2742                 else {
2743                         do_fx = 1;
2744                         nblks = del->br_blockcount;
2745                         qfield = XFS_TRANS_DQ_BCOUNT;
2746                 }
2747                 /*
2748                  * Set up del_endblock and cur for later.
2749                  */
2750                 del_endblock = del->br_startblock + del->br_blockcount;
2751                 if (cur) {
2752                         if ((error = xfs_bmbt_lookup_eq(cur, got.br_startoff,
2753                                         got.br_startblock, got.br_blockcount,
2754                                         &i)))
2755                                 goto done;
2756                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2757                 }
2758                 da_old = da_new = 0;
2759         } else {
2760                 da_old = startblockval(got.br_startblock);
2761                 da_new = 0;
2762                 nblks = 0;
2763                 do_fx = 0;
2764         }
2765         /*
2766          * Set flag value to use in switch statement.
2767          * Left-contig is 2, right-contig is 1.
2768          */
2769         switch (((got.br_startoff == del->br_startoff) << 1) |
2770                 (got_endoff == del_endoff)) {
2771         case 3:
2772                 /*
2773                  * Matches the whole extent.  Delete the entry.
2774                  */
2775                 xfs_iext_remove(ip, *idx, 1,
2776                                 whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0);
2777                 --*idx;
2778                 if (delay)
2779                         break;
2780
2781                 XFS_IFORK_NEXT_SET(ip, whichfork,
2782                         XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
2783                 flags |= XFS_ILOG_CORE;
2784                 if (!cur) {
2785                         flags |= xfs_ilog_fext(whichfork);
2786                         break;
2787                 }
2788                 if ((error = xfs_btree_delete(cur, &i)))
2789                         goto done;
2790                 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2791                 break;
2792
2793         case 2:
2794                 /*
2795                  * Deleting the first part of the extent.
2796                  */
2797                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2798                 xfs_bmbt_set_startoff(ep, del_endoff);
2799                 temp = got.br_blockcount - del->br_blockcount;
2800                 xfs_bmbt_set_blockcount(ep, temp);
2801                 if (delay) {
2802                         temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2803                                 da_old);
2804                         xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
2805                         trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2806                         da_new = temp;
2807                         break;
2808                 }
2809                 xfs_bmbt_set_startblock(ep, del_endblock);
2810                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2811                 if (!cur) {
2812                         flags |= xfs_ilog_fext(whichfork);
2813                         break;
2814                 }
2815                 if ((error = xfs_bmbt_update(cur, del_endoff, del_endblock,
2816                                 got.br_blockcount - del->br_blockcount,
2817                                 got.br_state)))
2818                         goto done;
2819                 break;
2820
2821         case 1:
2822                 /*
2823                  * Deleting the last part of the extent.
2824                  */
2825                 temp = got.br_blockcount - del->br_blockcount;
2826                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2827                 xfs_bmbt_set_blockcount(ep, temp);
2828                 if (delay) {
2829                         temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2830                                 da_old);
2831                         xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
2832                         trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2833                         da_new = temp;
2834                         break;
2835                 }
2836                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2837                 if (!cur) {
2838                         flags |= xfs_ilog_fext(whichfork);
2839                         break;
2840                 }
2841                 if ((error = xfs_bmbt_update(cur, got.br_startoff,
2842                                 got.br_startblock,
2843                                 got.br_blockcount - del->br_blockcount,
2844                                 got.br_state)))
2845                         goto done;
2846                 break;
2847
2848         case 0:
2849                 /*
2850                  * Deleting the middle of the extent.
2851                  */
2852                 temp = del->br_startoff - got.br_startoff;
2853                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2854                 xfs_bmbt_set_blockcount(ep, temp);
2855                 new.br_startoff = del_endoff;
2856                 temp2 = got_endoff - del_endoff;
2857                 new.br_blockcount = temp2;
2858                 new.br_state = got.br_state;
2859                 if (!delay) {
2860                         new.br_startblock = del_endblock;
2861                         flags |= XFS_ILOG_CORE;
2862                         if (cur) {
2863                                 if ((error = xfs_bmbt_update(cur,
2864                                                 got.br_startoff,
2865                                                 got.br_startblock, temp,
2866                                                 got.br_state)))
2867                                         goto done;
2868                                 if ((error = xfs_btree_increment(cur, 0, &i)))
2869                                         goto done;
2870                                 cur->bc_rec.b = new;
2871                                 error = xfs_btree_insert(cur, &i);
2872                                 if (error && error != ENOSPC)
2873                                         goto done;
2874                                 /*
2875                                  * If get no-space back from btree insert,
2876                                  * it tried a split, and we have a zero
2877                                  * block reservation.
2878                                  * Fix up our state and return the error.
2879                                  */
2880                                 if (error == ENOSPC) {
2881                                         /*
2882                                          * Reset the cursor, don't trust
2883                                          * it after any insert operation.
2884                                          */
2885                                         if ((error = xfs_bmbt_lookup_eq(cur,
2886                                                         got.br_startoff,
2887                                                         got.br_startblock,
2888                                                         temp, &i)))
2889                                                 goto done;
2890                                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2891                                         /*
2892                                          * Update the btree record back
2893                                          * to the original value.
2894                                          */
2895                                         if ((error = xfs_bmbt_update(cur,
2896                                                         got.br_startoff,
2897                                                         got.br_startblock,
2898                                                         got.br_blockcount,
2899                                                         got.br_state)))
2900                                                 goto done;
2901                                         /*
2902                                          * Reset the extent record back
2903                                          * to the original value.
2904                                          */
2905                                         xfs_bmbt_set_blockcount(ep,
2906                                                 got.br_blockcount);
2907                                         flags = 0;
2908                                         error = XFS_ERROR(ENOSPC);
2909                                         goto done;
2910                                 }
2911                                 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2912                         } else
2913                                 flags |= xfs_ilog_fext(whichfork);
2914                         XFS_IFORK_NEXT_SET(ip, whichfork,
2915                                 XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
2916                 } else {
2917                         ASSERT(whichfork == XFS_DATA_FORK);
2918                         temp = xfs_bmap_worst_indlen(ip, temp);
2919                         xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
2920                         temp2 = xfs_bmap_worst_indlen(ip, temp2);
2921                         new.br_startblock = nullstartblock((int)temp2);
2922                         da_new = temp + temp2;
2923                         while (da_new > da_old) {
2924                                 if (temp) {
2925                                         temp--;
2926                                         da_new--;
2927                                         xfs_bmbt_set_startblock(ep,
2928                                                 nullstartblock((int)temp));
2929                                 }
2930                                 if (da_new == da_old)
2931                                         break;
2932                                 if (temp2) {
2933                                         temp2--;
2934                                         da_new--;
2935                                         new.br_startblock =
2936                                                 nullstartblock((int)temp2);
2937                                 }
2938                         }
2939                 }
2940                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2941                 xfs_iext_insert(ip, *idx + 1, 1, &new, state);
2942                 ++*idx;
2943                 break;
2944         }
2945         /*
2946          * If we need to, add to list of extents to delete.
2947          */
2948         if (do_fx)
2949                 xfs_bmap_add_free(del->br_startblock, del->br_blockcount, flist,
2950                         mp);
2951         /*
2952          * Adjust inode # blocks in the file.
2953          */
2954         if (nblks)
2955                 ip->i_d.di_nblocks -= nblks;
2956         /*
2957          * Adjust quota data.
2958          */
2959         if (qfield)
2960                 xfs_trans_mod_dquot_byino(tp, ip, qfield, (long)-nblks);
2961
2962         /*
2963          * Account for change in delayed indirect blocks.
2964          * Nothing to do for disk quota accounting here.
2965          */
2966         ASSERT(da_old >= da_new);
2967         if (da_old > da_new) {
2968                 xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
2969                         (int64_t)(da_old - da_new), 0);
2970         }
2971 done:
2972         *logflagsp = flags;
2973         return error;
2974 }
2975
2976 /*
2977  * Remove the entry "free" from the free item list.  Prev points to the
2978  * previous entry, unless "free" is the head of the list.
2979  */
2980 STATIC void
2981 xfs_bmap_del_free(
2982         xfs_bmap_free_t         *flist, /* free item list header */
2983         xfs_bmap_free_item_t    *prev,  /* previous item on list, if any */
2984         xfs_bmap_free_item_t    *free)  /* list item to be freed */
2985 {
2986         if (prev)
2987                 prev->xbfi_next = free->xbfi_next;
2988         else
2989                 flist->xbf_first = free->xbfi_next;
2990         flist->xbf_count--;
2991         kmem_zone_free(xfs_bmap_free_item_zone, free);
2992 }
2993
2994 /*
2995  * Convert an extents-format file into a btree-format file.
2996  * The new file will have a root block (in the inode) and a single child block.
2997  */
2998 STATIC int                                      /* error */
2999 xfs_bmap_extents_to_btree(
3000         xfs_trans_t             *tp,            /* transaction pointer */
3001         xfs_inode_t             *ip,            /* incore inode pointer */
3002         xfs_fsblock_t           *firstblock,    /* first-block-allocated */
3003         xfs_bmap_free_t         *flist,         /* blocks freed in xaction */
3004         xfs_btree_cur_t         **curp,         /* cursor returned to caller */
3005         int                     wasdel,         /* converting a delayed alloc */
3006         int                     *logflagsp,     /* inode logging flags */
3007         int                     whichfork)      /* data or attr fork */
3008 {
3009         struct xfs_btree_block  *ablock;        /* allocated (child) bt block */
3010         xfs_buf_t               *abp;           /* buffer for ablock */
3011         xfs_alloc_arg_t         args;           /* allocation arguments */
3012         xfs_bmbt_rec_t          *arp;           /* child record pointer */
3013         struct xfs_btree_block  *block;         /* btree root block */
3014         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
3015         xfs_bmbt_rec_host_t     *ep;            /* extent record pointer */
3016         int                     error;          /* error return value */
3017         xfs_extnum_t            i, cnt;         /* extent record index */
3018         xfs_ifork_t             *ifp;           /* inode fork pointer */
3019         xfs_bmbt_key_t          *kp;            /* root block key pointer */
3020         xfs_mount_t             *mp;            /* mount structure */
3021         xfs_extnum_t            nextents;       /* number of file extents */
3022         xfs_bmbt_ptr_t          *pp;            /* root block address pointer */
3023
3024         ifp = XFS_IFORK_PTR(ip, whichfork);
3025         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS);
3026         ASSERT(ifp->if_ext_max ==
3027                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
3028         /*
3029          * Make space in the inode incore.
3030          */
3031         xfs_iroot_realloc(ip, 1, whichfork);
3032         ifp->if_flags |= XFS_IFBROOT;
3033
3034         /*
3035          * Fill in the root.
3036          */
3037         block = ifp->if_broot;
3038         block->bb_magic = cpu_to_be32(XFS_BMAP_MAGIC);
3039         block->bb_level = cpu_to_be16(1);
3040         block->bb_numrecs = cpu_to_be16(1);
3041         block->bb_u.l.bb_leftsib = cpu_to_be64(NULLDFSBNO);
3042         block->bb_u.l.bb_rightsib = cpu_to_be64(NULLDFSBNO);
3043
3044         /*
3045          * Need a cursor.  Can't allocate until bb_level is filled in.
3046          */
3047         mp = ip->i_mount;
3048         cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
3049         cur->bc_private.b.firstblock = *firstblock;
3050         cur->bc_private.b.flist = flist;
3051         cur->bc_private.b.flags = wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
3052         /*
3053          * Convert to a btree with two levels, one record in root.
3054          */
3055         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_BTREE);
3056         args.tp = tp;
3057         args.mp = mp;
3058         args.firstblock = *firstblock;
3059         if (*firstblock == NULLFSBLOCK) {
3060                 args.type = XFS_ALLOCTYPE_START_BNO;
3061                 args.fsbno = XFS_INO_TO_FSB(mp, ip->i_ino);
3062         } else if (flist->xbf_low) {
3063                 args.type = XFS_ALLOCTYPE_START_BNO;
3064                 args.fsbno = *firstblock;
3065         } else {
3066                 args.type = XFS_ALLOCTYPE_NEAR_BNO;
3067                 args.fsbno = *firstblock;
3068         }
3069         args.minlen = args.maxlen = args.prod = 1;
3070         args.total = args.minleft = args.alignment = args.mod = args.isfl =
3071                 args.minalignslop = 0;
3072         args.wasdel = wasdel;
3073         *logflagsp = 0;
3074         if ((error = xfs_alloc_vextent(&args))) {
3075                 xfs_iroot_realloc(ip, -1, whichfork);
3076                 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
3077                 return error;
3078         }
3079         /*
3080          * Allocation can't fail, the space was reserved.
3081          */
3082         ASSERT(args.fsbno != NULLFSBLOCK);
3083         ASSERT(*firstblock == NULLFSBLOCK ||
3084                args.agno == XFS_FSB_TO_AGNO(mp, *firstblock) ||
3085                (flist->xbf_low &&
3086                 args.agno > XFS_FSB_TO_AGNO(mp, *firstblock)));
3087         *firstblock = cur->bc_private.b.firstblock = args.fsbno;
3088         cur->bc_private.b.allocated++;
3089         ip->i_d.di_nblocks++;
3090         xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, 1L);
3091         abp = xfs_btree_get_bufl(mp, tp, args.fsbno, 0);
3092         /*
3093          * Fill in the child block.
3094          */
3095         ablock = XFS_BUF_TO_BLOCK(abp);
3096         ablock->bb_magic = cpu_to_be32(XFS_BMAP_MAGIC);
3097         ablock->bb_level = 0;
3098         ablock->bb_u.l.bb_leftsib = cpu_to_be64(NULLDFSBNO);
3099         ablock->bb_u.l.bb_rightsib = cpu_to_be64(NULLDFSBNO);
3100         arp = XFS_BMBT_REC_ADDR(mp, ablock, 1);
3101         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3102         for (cnt = i = 0; i < nextents; i++) {
3103                 ep = xfs_iext_get_ext(ifp, i);
3104                 if (!isnullstartblock(xfs_bmbt_get_startblock(ep))) {
3105                         arp->l0 = cpu_to_be64(ep->l0);
3106                         arp->l1 = cpu_to_be64(ep->l1);
3107                         arp++; cnt++;
3108                 }
3109         }
3110         ASSERT(cnt == XFS_IFORK_NEXTENTS(ip, whichfork));
3111         xfs_btree_set_numrecs(ablock, cnt);
3112
3113         /*
3114          * Fill in the root key and pointer.
3115          */
3116         kp = XFS_BMBT_KEY_ADDR(mp, block, 1);
3117         arp = XFS_BMBT_REC_ADDR(mp, ablock, 1);
3118         kp->br_startoff = cpu_to_be64(xfs_bmbt_disk_get_startoff(arp));
3119         pp = XFS_BMBT_PTR_ADDR(mp, block, 1, xfs_bmbt_get_maxrecs(cur,
3120                                                 be16_to_cpu(block->bb_level)));
3121         *pp = cpu_to_be64(args.fsbno);
3122
3123         /*
3124          * Do all this logging at the end so that
3125          * the root is at the right level.
3126          */
3127         xfs_btree_log_block(cur, abp, XFS_BB_ALL_BITS);
3128         xfs_btree_log_recs(cur, abp, 1, be16_to_cpu(ablock->bb_numrecs));
3129         ASSERT(*curp == NULL);
3130         *curp = cur;
3131         *logflagsp = XFS_ILOG_CORE | xfs_ilog_fbroot(whichfork);
3132         return 0;
3133 }
3134
3135 /*
3136  * Calculate the default attribute fork offset for newly created inodes.
3137  */
3138 uint
3139 xfs_default_attroffset(
3140         struct xfs_inode        *ip)
3141 {
3142         struct xfs_mount        *mp = ip->i_mount;
3143         uint                    offset;
3144
3145         if (mp->m_sb.sb_inodesize == 256) {
3146                 offset = XFS_LITINO(mp) -
3147                                 XFS_BMDR_SPACE_CALC(MINABTPTRS);
3148         } else {
3149                 offset = XFS_BMDR_SPACE_CALC(6 * MINABTPTRS);
3150         }
3151
3152         ASSERT(offset < XFS_LITINO(mp));
3153         return offset;
3154 }
3155
3156 /*
3157  * Helper routine to reset inode di_forkoff field when switching
3158  * attribute fork from local to extent format - we reset it where
3159  * possible to make space available for inline data fork extents.
3160  */
3161 STATIC void
3162 xfs_bmap_forkoff_reset(
3163         xfs_mount_t     *mp,
3164         xfs_inode_t     *ip,
3165         int             whichfork)
3166 {
3167         if (whichfork == XFS_ATTR_FORK &&
3168             ip->i_d.di_format != XFS_DINODE_FMT_DEV &&
3169             ip->i_d.di_format != XFS_DINODE_FMT_UUID &&
3170             ip->i_d.di_format != XFS_DINODE_FMT_BTREE) {
3171                 uint    dfl_forkoff = xfs_default_attroffset(ip) >> 3;
3172
3173                 if (dfl_forkoff > ip->i_d.di_forkoff) {
3174                         ip->i_d.di_forkoff = dfl_forkoff;
3175                         ip->i_df.if_ext_max =
3176                                 XFS_IFORK_DSIZE(ip) / sizeof(xfs_bmbt_rec_t);
3177                         ip->i_afp->if_ext_max =
3178                                 XFS_IFORK_ASIZE(ip) / sizeof(xfs_bmbt_rec_t);
3179                 }
3180         }
3181 }
3182
3183 /*
3184  * Convert a local file to an extents file.
3185  * This code is out of bounds for data forks of regular files,
3186  * since the file data needs to get logged so things will stay consistent.
3187  * (The bmap-level manipulations are ok, though).
3188  */
3189 STATIC int                              /* error */
3190 xfs_bmap_local_to_extents(
3191         xfs_trans_t     *tp,            /* transaction pointer */
3192         xfs_inode_t     *ip,            /* incore inode pointer */
3193         xfs_fsblock_t   *firstblock,    /* first block allocated in xaction */
3194         xfs_extlen_t    total,          /* total blocks needed by transaction */
3195         int             *logflagsp,     /* inode logging flags */
3196         int             whichfork)      /* data or attr fork */
3197 {
3198         int             error;          /* error return value */
3199         int             flags;          /* logging flags returned */
3200         xfs_ifork_t     *ifp;           /* inode fork pointer */
3201
3202         /*
3203          * We don't want to deal with the case of keeping inode data inline yet.
3204          * So sending the data fork of a regular inode is invalid.
3205          */
3206         ASSERT(!(S_ISREG(ip->i_d.di_mode) && whichfork == XFS_DATA_FORK));
3207         ifp = XFS_IFORK_PTR(ip, whichfork);
3208         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
3209         flags = 0;
3210         error = 0;
3211         if (ifp->if_bytes) {
3212                 xfs_alloc_arg_t args;   /* allocation arguments */
3213                 xfs_buf_t       *bp;    /* buffer for extent block */
3214                 xfs_bmbt_rec_host_t *ep;/* extent record pointer */
3215
3216                 args.tp = tp;
3217                 args.mp = ip->i_mount;
3218                 args.firstblock = *firstblock;
3219                 ASSERT((ifp->if_flags &
3220                         (XFS_IFINLINE|XFS_IFEXTENTS|XFS_IFEXTIREC)) == XFS_IFINLINE);
3221                 /*
3222                  * Allocate a block.  We know we need only one, since the
3223                  * file currently fits in an inode.
3224                  */
3225                 if (*firstblock == NULLFSBLOCK) {
3226                         args.fsbno = XFS_INO_TO_FSB(args.mp, ip->i_ino);
3227                         args.type = XFS_ALLOCTYPE_START_BNO;
3228                 } else {
3229                         args.fsbno = *firstblock;
3230                         args.type = XFS_ALLOCTYPE_NEAR_BNO;
3231                 }
3232                 args.total = total;
3233                 args.mod = args.minleft = args.alignment = args.wasdel =
3234                         args.isfl = args.minalignslop = 0;
3235                 args.minlen = args.maxlen = args.prod = 1;
3236                 if ((error = xfs_alloc_vextent(&args)))
3237                         goto done;
3238                 /*
3239                  * Can't fail, the space was reserved.
3240                  */
3241                 ASSERT(args.fsbno != NULLFSBLOCK);
3242                 ASSERT(args.len == 1);
3243                 *firstblock = args.fsbno;
3244                 bp = xfs_btree_get_bufl(args.mp, tp, args.fsbno, 0);
3245                 memcpy(bp->b_addr, ifp->if_u1.if_data, ifp->if_bytes);
3246                 xfs_trans_log_buf(tp, bp, 0, ifp->if_bytes - 1);
3247                 xfs_bmap_forkoff_reset(args.mp, ip, whichfork);
3248                 xfs_idata_realloc(ip, -ifp->if_bytes, whichfork);
3249                 xfs_iext_add(ifp, 0, 1);
3250                 ep = xfs_iext_get_ext(ifp, 0);
3251                 xfs_bmbt_set_allf(ep, 0, args.fsbno, 1, XFS_EXT_NORM);
3252                 trace_xfs_bmap_post_update(ip, 0,
3253                                 whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0,
3254                                 _THIS_IP_);
3255                 XFS_IFORK_NEXT_SET(ip, whichfork, 1);
3256                 ip->i_d.di_nblocks = 1;
3257                 xfs_trans_mod_dquot_byino(tp, ip,
3258                         XFS_TRANS_DQ_BCOUNT, 1L);
3259                 flags |= xfs_ilog_fext(whichfork);
3260         } else {
3261                 ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) == 0);
3262                 xfs_bmap_forkoff_reset(ip->i_mount, ip, whichfork);
3263         }
3264         ifp->if_flags &= ~XFS_IFINLINE;
3265         ifp->if_flags |= XFS_IFEXTENTS;
3266         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
3267         flags |= XFS_ILOG_CORE;
3268 done:
3269         *logflagsp = flags;
3270         return error;
3271 }
3272
3273 /*
3274  * Search the extent records for the entry containing block bno.
3275  * If bno lies in a hole, point to the next entry.  If bno lies
3276  * past eof, *eofp will be set, and *prevp will contain the last
3277  * entry (null if none).  Else, *lastxp will be set to the index
3278  * of the found entry; *gotp will contain the entry.
3279  */
3280 STATIC xfs_bmbt_rec_host_t *            /* pointer to found extent entry */
3281 xfs_bmap_search_multi_extents(
3282         xfs_ifork_t     *ifp,           /* inode fork pointer */
3283         xfs_fileoff_t   bno,            /* block number searched for */
3284         int             *eofp,          /* out: end of file found */
3285         xfs_extnum_t    *lastxp,        /* out: last extent index */
3286         xfs_bmbt_irec_t *gotp,          /* out: extent entry found */
3287         xfs_bmbt_irec_t *prevp)         /* out: previous extent entry found */
3288 {
3289         xfs_bmbt_rec_host_t *ep;                /* extent record pointer */
3290         xfs_extnum_t    lastx;          /* last extent index */
3291
3292         /*
3293          * Initialize the extent entry structure to catch access to
3294          * uninitialized br_startblock field.
3295          */
3296         gotp->br_startoff = 0xffa5a5a5a5a5a5a5LL;
3297         gotp->br_blockcount = 0xa55a5a5a5a5a5a5aLL;
3298         gotp->br_state = XFS_EXT_INVALID;
3299 #if XFS_BIG_BLKNOS
3300         gotp->br_startblock = 0xffffa5a5a5a5a5a5LL;
3301 #else
3302         gotp->br_startblock = 0xffffa5a5;
3303 #endif
3304         prevp->br_startoff = NULLFILEOFF;
3305
3306         ep = xfs_iext_bno_to_ext(ifp, bno, &lastx);
3307         if (lastx > 0) {
3308                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx - 1), prevp);
3309         }
3310         if (lastx < (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t))) {
3311                 xfs_bmbt_get_all(ep, gotp);
3312                 *eofp = 0;
3313         } else {
3314                 if (lastx > 0) {
3315                         *gotp = *prevp;
3316                 }
3317                 *eofp = 1;
3318                 ep = NULL;
3319         }
3320         *lastxp = lastx;
3321         return ep;
3322 }
3323
3324 /*
3325  * Search the extents list for the inode, for the extent containing bno.
3326  * If bno lies in a hole, point to the next entry.  If bno lies past eof,
3327  * *eofp will be set, and *prevp will contain the last entry (null if none).
3328  * Else, *lastxp will be set to the index of the found
3329  * entry; *gotp will contain the entry.
3330  */
3331 STATIC xfs_bmbt_rec_host_t *                 /* pointer to found extent entry */
3332 xfs_bmap_search_extents(
3333         xfs_inode_t     *ip,            /* incore inode pointer */
3334         xfs_fileoff_t   bno,            /* block number searched for */
3335         int             fork,           /* data or attr fork */
3336         int             *eofp,          /* out: end of file found */
3337         xfs_extnum_t    *lastxp,        /* out: last extent index */
3338         xfs_bmbt_irec_t *gotp,          /* out: extent entry found */
3339         xfs_bmbt_irec_t *prevp)         /* out: previous extent entry found */
3340 {
3341         xfs_ifork_t     *ifp;           /* inode fork pointer */
3342         xfs_bmbt_rec_host_t  *ep;            /* extent record pointer */
3343
3344         XFS_STATS_INC(xs_look_exlist);
3345         ifp = XFS_IFORK_PTR(ip, fork);
3346
3347         ep = xfs_bmap_search_multi_extents(ifp, bno, eofp, lastxp, gotp, prevp);
3348
3349         if (unlikely(!(gotp->br_startblock) && (*lastxp != NULLEXTNUM) &&
3350                      !(XFS_IS_REALTIME_INODE(ip) && fork == XFS_DATA_FORK))) {
3351                 xfs_alert_tag(ip->i_mount, XFS_PTAG_FSBLOCK_ZERO,
3352                                 "Access to block zero in inode %llu "
3353                                 "start_block: %llx start_off: %llx "
3354                                 "blkcnt: %llx extent-state: %x lastx: %x\n",
3355                         (unsigned long long)ip->i_ino,
3356                         (unsigned long long)gotp->br_startblock,
3357                         (unsigned long long)gotp->br_startoff,
3358                         (unsigned long long)gotp->br_blockcount,
3359                         gotp->br_state, *lastxp);
3360                 *lastxp = NULLEXTNUM;
3361                 *eofp = 1;
3362                 return NULL;
3363         }
3364         return ep;
3365 }
3366
3367 /*
3368  * Compute the worst-case number of indirect blocks that will be used
3369  * for ip's delayed extent of length "len".
3370  */
3371 STATIC xfs_filblks_t
3372 xfs_bmap_worst_indlen(
3373         xfs_inode_t     *ip,            /* incore inode pointer */
3374         xfs_filblks_t   len)            /* delayed extent length */
3375 {
3376         int             level;          /* btree level number */
3377         int             maxrecs;        /* maximum record count at this level */
3378         xfs_mount_t     *mp;            /* mount structure */
3379         xfs_filblks_t   rval;           /* return value */
3380
3381         mp = ip->i_mount;
3382         maxrecs = mp->m_bmap_dmxr[0];
3383         for (level = 0, rval = 0;
3384              level < XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK);
3385              level++) {
3386                 len += maxrecs - 1;
3387                 do_div(len, maxrecs);
3388                 rval += len;
3389                 if (len == 1)
3390                         return rval + XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK) -
3391                                 level - 1;
3392                 if (level == 0)
3393                         maxrecs = mp->m_bmap_dmxr[1];
3394         }
3395         return rval;
3396 }
3397
3398 /*
3399  * Convert inode from non-attributed to attributed.
3400  * Must not be in a transaction, ip must not be locked.
3401  */
3402 int                                             /* error code */
3403 xfs_bmap_add_attrfork(
3404         xfs_inode_t             *ip,            /* incore inode pointer */
3405         int                     size,           /* space new attribute needs */
3406         int                     rsvd)           /* xact may use reserved blks */
3407 {
3408         xfs_fsblock_t           firstblock;     /* 1st block/ag allocated */
3409         xfs_bmap_free_t         flist;          /* freed extent records */
3410         xfs_mount_t             *mp;            /* mount structure */
3411         xfs_trans_t             *tp;            /* transaction pointer */
3412         int                     blks;           /* space reservation */
3413         int                     version = 1;    /* superblock attr version */
3414         int                     committed;      /* xaction was committed */
3415         int                     logflags;       /* logging flags */
3416         int                     error;          /* error return value */
3417
3418         ASSERT(XFS_IFORK_Q(ip) == 0);
3419         ASSERT(ip->i_df.if_ext_max ==
3420                XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t));
3421
3422         mp = ip->i_mount;
3423         ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
3424         tp = xfs_trans_alloc(mp, XFS_TRANS_ADDAFORK);
3425         blks = XFS_ADDAFORK_SPACE_RES(mp);
3426         if (rsvd)
3427                 tp->t_flags |= XFS_TRANS_RESERVE;
3428         if ((error = xfs_trans_reserve(tp, blks, XFS_ADDAFORK_LOG_RES(mp), 0,
3429                         XFS_TRANS_PERM_LOG_RES, XFS_ADDAFORK_LOG_COUNT)))
3430                 goto error0;
3431         xfs_ilock(ip, XFS_ILOCK_EXCL);
3432         error = xfs_trans_reserve_quota_nblks(tp, ip, blks, 0, rsvd ?
3433                         XFS_QMOPT_RES_REGBLKS | XFS_QMOPT_FORCE_RES :
3434                         XFS_QMOPT_RES_REGBLKS);
3435         if (error) {
3436                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3437                 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES);
3438                 return error;
3439         }
3440         if (XFS_IFORK_Q(ip))
3441                 goto error1;
3442         if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS) {
3443                 /*
3444                  * For inodes coming from pre-6.2 filesystems.
3445                  */
3446                 ASSERT(ip->i_d.di_aformat == 0);
3447                 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
3448         }
3449         ASSERT(ip->i_d.di_anextents == 0);
3450
3451         xfs_trans_ijoin_ref(tp, ip, XFS_ILOCK_EXCL);
3452         xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
3453
3454         switch (ip->i_d.di_format) {
3455         case XFS_DINODE_FMT_DEV:
3456                 ip->i_d.di_forkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
3457                 break;
3458         case XFS_DINODE_FMT_UUID:
3459                 ip->i_d.di_forkoff = roundup(sizeof(uuid_t), 8) >> 3;
3460                 break;
3461         case XFS_DINODE_FMT_LOCAL:
3462         case XFS_DINODE_FMT_EXTENTS:
3463         case XFS_DINODE_FMT_BTREE:
3464                 ip->i_d.di_forkoff = xfs_attr_shortform_bytesfit(ip, size);
3465                 if (!ip->i_d.di_forkoff)
3466                         ip->i_d.di_forkoff = xfs_default_attroffset(ip) >> 3;
3467                 else if (mp->m_flags & XFS_MOUNT_ATTR2)
3468                         version = 2;
3469                 break;
3470         default:
3471                 ASSERT(0);
3472                 error = XFS_ERROR(EINVAL);
3473                 goto error1;
3474         }
3475         ip->i_df.if_ext_max =
3476                 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
3477         ASSERT(ip->i_afp == NULL);
3478         ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP);
3479         ip->i_afp->if_ext_max =
3480                 XFS_IFORK_ASIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
3481         ip->i_afp->if_flags = XFS_IFEXTENTS;
3482         logflags = 0;
3483         xfs_bmap_init(&flist, &firstblock);
3484         switch (ip->i_d.di_format) {
3485         case XFS_DINODE_FMT_LOCAL:
3486                 error = xfs_bmap_add_attrfork_local(tp, ip, &firstblock, &flist,
3487                         &logflags);
3488                 break;
3489         case XFS_DINODE_FMT_EXTENTS:
3490                 error = xfs_bmap_add_attrfork_extents(tp, ip, &firstblock,
3491                         &flist, &logflags);
3492                 break;
3493         case XFS_DINODE_FMT_BTREE:
3494                 error = xfs_bmap_add_attrfork_btree(tp, ip, &firstblock, &flist,
3495                         &logflags);
3496                 break;
3497         default:
3498                 error = 0;
3499                 break;
3500         }
3501         if (logflags)
3502                 xfs_trans_log_inode(tp, ip, logflags);
3503         if (error)
3504                 goto error2;
3505         if (!xfs_sb_version_hasattr(&mp->m_sb) ||
3506            (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2)) {
3507                 __int64_t sbfields = 0;
3508
3509                 spin_lock(&mp->m_sb_lock);
3510                 if (!xfs_sb_version_hasattr(&mp->m_sb)) {
3511                         xfs_sb_version_addattr(&mp->m_sb);
3512                         sbfields |= XFS_SB_VERSIONNUM;
3513                 }
3514                 if (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2) {
3515                         xfs_sb_version_addattr2(&mp->m_sb);
3516                         sbfields |= (XFS_SB_VERSIONNUM | XFS_SB_FEATURES2);
3517                 }
3518                 if (sbfields) {
3519                         spin_unlock(&mp->m_sb_lock);
3520                         xfs_mod_sb(tp, sbfields);
3521                 } else
3522                         spin_unlock(&mp->m_sb_lock);
3523         }
3524         if ((error = xfs_bmap_finish(&tp, &flist, &committed)))
3525                 goto error2;
3526         error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
3527         ASSERT(ip->i_df.if_ext_max ==
3528                XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t));
3529         return error;
3530 error2:
3531         xfs_bmap_cancel(&flist);
3532 error1:
3533         xfs_iunlock(ip, XFS_ILOCK_EXCL);
3534 error0:
3535         xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
3536         ASSERT(ip->i_df.if_ext_max ==
3537                XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t));
3538         return error;
3539 }
3540
3541 /*
3542  * Add the extent to the list of extents to be free at transaction end.
3543  * The list is maintained sorted (by block number).
3544  */
3545 /* ARGSUSED */
3546 void
3547 xfs_bmap_add_free(
3548         xfs_fsblock_t           bno,            /* fs block number of extent */
3549         xfs_filblks_t           len,            /* length of extent */
3550         xfs_bmap_free_t         *flist,         /* list of extents */
3551         xfs_mount_t             *mp)            /* mount point structure */
3552 {
3553         xfs_bmap_free_item_t    *cur;           /* current (next) element */
3554         xfs_bmap_free_item_t    *new;           /* new element */
3555         xfs_bmap_free_item_t    *prev;          /* previous element */
3556 #ifdef DEBUG
3557         xfs_agnumber_t          agno;
3558         xfs_agblock_t           agbno;
3559
3560         ASSERT(bno != NULLFSBLOCK);
3561         ASSERT(len > 0);
3562         ASSERT(len <= MAXEXTLEN);
3563         ASSERT(!isnullstartblock(bno));
3564         agno = XFS_FSB_TO_AGNO(mp, bno);
3565         agbno = XFS_FSB_TO_AGBNO(mp, bno);
3566         ASSERT(agno < mp->m_sb.sb_agcount);
3567         ASSERT(agbno < mp->m_sb.sb_agblocks);
3568         ASSERT(len < mp->m_sb.sb_agblocks);
3569         ASSERT(agbno + len <= mp->m_sb.sb_agblocks);
3570 #endif
3571         ASSERT(xfs_bmap_free_item_zone != NULL);
3572         new = kmem_zone_alloc(xfs_bmap_free_item_zone, KM_SLEEP);
3573         new->xbfi_startblock = bno;
3574         new->xbfi_blockcount = (xfs_extlen_t)len;
3575         for (prev = NULL, cur = flist->xbf_first;
3576              cur != NULL;
3577              prev = cur, cur = cur->xbfi_next) {
3578                 if (cur->xbfi_startblock >= bno)
3579                         break;
3580         }
3581         if (prev)
3582                 prev->xbfi_next = new;
3583         else
3584                 flist->xbf_first = new;
3585         new->xbfi_next = cur;
3586         flist->xbf_count++;
3587 }
3588
3589 /*
3590  * Compute and fill in the value of the maximum depth of a bmap btree
3591  * in this filesystem.  Done once, during mount.
3592  */
3593 void
3594 xfs_bmap_compute_maxlevels(
3595         xfs_mount_t     *mp,            /* file system mount structure */
3596         int             whichfork)      /* data or attr fork */
3597 {
3598         int             level;          /* btree level */
3599         uint            maxblocks;      /* max blocks at this level */
3600         uint            maxleafents;    /* max leaf entries possible */
3601         int             maxrootrecs;    /* max records in root block */
3602         int             minleafrecs;    /* min records in leaf block */
3603         int             minnoderecs;    /* min records in node block */
3604         int             sz;             /* root block size */
3605
3606         /*
3607          * The maximum number of extents in a file, hence the maximum
3608          * number of leaf entries, is controlled by the type of di_nextents
3609          * (a signed 32-bit number, xfs_extnum_t), or by di_anextents
3610          * (a signed 16-bit number, xfs_aextnum_t).
3611          *
3612          * Note that we can no longer assume that if we are in ATTR1 that
3613          * the fork offset of all the inodes will be
3614          * (xfs_default_attroffset(ip) >> 3) because we could have mounted
3615          * with ATTR2 and then mounted back with ATTR1, keeping the
3616          * di_forkoff's fixed but probably at various positions. Therefore,
3617          * for both ATTR1 and ATTR2 we have to assume the worst case scenario
3618          * of a minimum size available.
3619          */
3620         if (whichfork == XFS_DATA_FORK) {
3621                 maxleafents = MAXEXTNUM;
3622                 sz = XFS_BMDR_SPACE_CALC(MINDBTPTRS);
3623         } else {
3624                 maxleafents = MAXAEXTNUM;
3625                 sz = XFS_BMDR_SPACE_CALC(MINABTPTRS);
3626         }
3627         maxrootrecs = xfs_bmdr_maxrecs(mp, sz, 0);
3628         minleafrecs = mp->m_bmap_dmnr[0];
3629         minnoderecs = mp->m_bmap_dmnr[1];
3630         maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs;
3631         for (level = 1; maxblocks > 1; level++) {
3632                 if (maxblocks <= maxrootrecs)
3633                         maxblocks = 1;
3634                 else
3635                         maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs;
3636         }
3637         mp->m_bm_maxlevels[whichfork] = level;
3638 }
3639
3640 /*
3641  * Routine to be called at transaction's end by xfs_bmapi, xfs_bunmapi
3642  * caller.  Frees all the extents that need freeing, which must be done
3643  * last due to locking considerations.  We never free any extents in
3644  * the first transaction.
3645  *
3646  * Return 1 if the given transaction was committed and a new one
3647  * started, and 0 otherwise in the committed parameter.
3648  */
3649 int                                             /* error */
3650 xfs_bmap_finish(
3651         xfs_trans_t             **tp,           /* transaction pointer addr */
3652         xfs_bmap_free_t         *flist,         /* i/o: list extents to free */
3653         int                     *committed)     /* xact committed or not */
3654 {
3655         xfs_efd_log_item_t      *efd;           /* extent free data */
3656         xfs_efi_log_item_t      *efi;           /* extent free intention */
3657         int                     error;          /* error return value */
3658         xfs_bmap_free_item_t    *free;          /* free extent item */
3659         unsigned int            logres;         /* new log reservation */
3660         unsigned int            logcount;       /* new log count */
3661         xfs_mount_t             *mp;            /* filesystem mount structure */
3662         xfs_bmap_free_item_t    *next;          /* next item on free list */
3663         xfs_trans_t             *ntp;           /* new transaction pointer */
3664
3665         ASSERT((*tp)->t_flags & XFS_TRANS_PERM_LOG_RES);
3666         if (flist->xbf_count == 0) {
3667                 *committed = 0;
3668                 return 0;
3669         }
3670         ntp = *tp;
3671         efi = xfs_trans_get_efi(ntp, flist->xbf_count);
3672         for (free = flist->xbf_first; free; free = free->xbfi_next)
3673                 xfs_trans_log_efi_extent(ntp, efi, free->xbfi_startblock,
3674                         free->xbfi_blockcount);
3675         logres = ntp->t_log_res;
3676         logcount = ntp->t_log_count;
3677         ntp = xfs_trans_dup(*tp);
3678         error = xfs_trans_commit(*tp, 0);
3679         *tp = ntp;
3680         *committed = 1;
3681         /*
3682          * We have a new transaction, so we should return committed=1,
3683          * even though we're returning an error.
3684          */
3685         if (error)
3686                 return error;
3687
3688         /*
3689          * transaction commit worked ok so we can drop the extra ticket
3690          * reference that we gained in xfs_trans_dup()
3691          */
3692         xfs_log_ticket_put(ntp->t_ticket);
3693
3694         if ((error = xfs_trans_reserve(ntp, 0, logres, 0, XFS_TRANS_PERM_LOG_RES,
3695                         logcount)))
3696                 return error;
3697         efd = xfs_trans_get_efd(ntp, efi, flist->xbf_count);
3698         for (free = flist->xbf_first; free != NULL; free = next) {
3699                 next = free->xbfi_next;
3700                 if ((error = xfs_free_extent(ntp, free->xbfi_startblock,
3701                                 free->xbfi_blockcount))) {
3702                         /*
3703                          * The bmap free list will be cleaned up at a
3704                          * higher level.  The EFI will be canceled when
3705                          * this transaction is aborted.
3706                          * Need to force shutdown here to make sure it
3707                          * happens, since this transaction may not be
3708                          * dirty yet.
3709                          */
3710                         mp = ntp->t_mountp;
3711                         if (!XFS_FORCED_SHUTDOWN(mp))
3712                                 xfs_force_shutdown(mp,
3713                                                    (error == EFSCORRUPTED) ?
3714                                                    SHUTDOWN_CORRUPT_INCORE :
3715                                                    SHUTDOWN_META_IO_ERROR);
3716                         return error;
3717                 }
3718                 xfs_trans_log_efd_extent(ntp, efd, free->xbfi_startblock,
3719                         free->xbfi_blockcount);
3720                 xfs_bmap_del_free(flist, NULL, free);
3721         }
3722         return 0;
3723 }
3724
3725 /*
3726  * Free up any items left in the list.
3727  */
3728 void
3729 xfs_bmap_cancel(
3730         xfs_bmap_free_t         *flist) /* list of bmap_free_items */
3731 {
3732         xfs_bmap_free_item_t    *free;  /* free list item */
3733         xfs_bmap_free_item_t    *next;
3734
3735         if (flist->xbf_count == 0)
3736                 return;
3737         ASSERT(flist->xbf_first != NULL);
3738         for (free = flist->xbf_first; free; free = next) {
3739                 next = free->xbfi_next;
3740                 xfs_bmap_del_free(flist, NULL, free);
3741         }
3742         ASSERT(flist->xbf_count == 0);
3743 }
3744
3745 /*
3746  * Returns the file-relative block number of the first unused block(s)
3747  * in the file with at least "len" logically contiguous blocks free.
3748  * This is the lowest-address hole if the file has holes, else the first block
3749  * past the end of file.
3750  * Return 0 if the file is currently local (in-inode).
3751  */
3752 int                                             /* error */
3753 xfs_bmap_first_unused(
3754         xfs_trans_t     *tp,                    /* transaction pointer */
3755         xfs_inode_t     *ip,                    /* incore inode */
3756         xfs_extlen_t    len,                    /* size of hole to find */
3757         xfs_fileoff_t   *first_unused,          /* unused block */
3758         int             whichfork)              /* data or attr fork */
3759 {
3760         int             error;                  /* error return value */
3761         int             idx;                    /* extent record index */
3762         xfs_ifork_t     *ifp;                   /* inode fork pointer */
3763         xfs_fileoff_t   lastaddr;               /* last block number seen */
3764         xfs_fileoff_t   lowest;                 /* lowest useful block */
3765         xfs_fileoff_t   max;                    /* starting useful block */
3766         xfs_fileoff_t   off;                    /* offset for this block */
3767         xfs_extnum_t    nextents;               /* number of extent entries */
3768
3769         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE ||
3770                XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ||
3771                XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
3772         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
3773                 *first_unused = 0;
3774                 return 0;
3775         }
3776         ifp = XFS_IFORK_PTR(ip, whichfork);
3777         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
3778             (error = xfs_iread_extents(tp, ip, whichfork)))
3779                 return error;
3780         lowest = *first_unused;
3781         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3782         for (idx = 0, lastaddr = 0, max = lowest; idx < nextents; idx++) {
3783                 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, idx);
3784                 off = xfs_bmbt_get_startoff(ep);
3785                 /*
3786                  * See if the hole before this extent will work.
3787                  */
3788                 if (off >= lowest + len && off - max >= len) {
3789                         *first_unused = max;
3790                         return 0;
3791                 }
3792                 lastaddr = off + xfs_bmbt_get_blockcount(ep);
3793                 max = XFS_FILEOFF_MAX(lastaddr, lowest);
3794         }
3795         *first_unused = max;
3796         return 0;
3797 }
3798
3799 /*
3800  * Returns the file-relative block number of the last block + 1 before
3801  * last_block (input value) in the file.
3802  * This is not based on i_size, it is based on the extent records.
3803  * Returns 0 for local files, as they do not have extent records.
3804  */
3805 int                                             /* error */
3806 xfs_bmap_last_before(
3807         xfs_trans_t     *tp,                    /* transaction pointer */
3808         xfs_inode_t     *ip,                    /* incore inode */
3809         xfs_fileoff_t   *last_block,            /* last block */
3810         int             whichfork)              /* data or attr fork */
3811 {
3812         xfs_fileoff_t   bno;                    /* input file offset */
3813         int             eof;                    /* hit end of file */
3814         xfs_bmbt_rec_host_t *ep;                /* pointer to last extent */
3815         int             error;                  /* error return value */
3816         xfs_bmbt_irec_t got;                    /* current extent value */
3817         xfs_ifork_t     *ifp;                   /* inode fork pointer */
3818         xfs_extnum_t    lastx;                  /* last extent used */
3819         xfs_bmbt_irec_t prev;                   /* previous extent value */
3820
3821         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
3822             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
3823             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL)
3824                return XFS_ERROR(EIO);
3825         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
3826                 *last_block = 0;
3827                 return 0;
3828         }
3829         ifp = XFS_IFORK_PTR(ip, whichfork);
3830         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
3831             (error = xfs_iread_extents(tp, ip, whichfork)))
3832                 return error;
3833         bno = *last_block - 1;
3834         ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
3835                 &prev);
3836         if (eof || xfs_bmbt_get_startoff(ep) > bno) {
3837                 if (prev.br_startoff == NULLFILEOFF)
3838                         *last_block = 0;
3839                 else
3840                         *last_block = prev.br_startoff + prev.br_blockcount;
3841         }
3842         /*
3843          * Otherwise *last_block is already the right answer.
3844          */
3845         return 0;
3846 }
3847
3848 STATIC int
3849 xfs_bmap_last_extent(
3850         struct xfs_trans        *tp,
3851         struct xfs_inode        *ip,
3852         int                     whichfork,
3853         struct xfs_bmbt_irec    *rec,
3854         int                     *is_empty)
3855 {
3856         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, whichfork);
3857         int                     error;
3858         int                     nextents;
3859
3860         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
3861                 error = xfs_iread_extents(tp, ip, whichfork);
3862                 if (error)
3863                         return error;
3864         }
3865
3866         nextents = ifp->if_bytes / sizeof(xfs_bmbt_rec_t);
3867         if (nextents == 0) {
3868                 *is_empty = 1;
3869                 return 0;
3870         }
3871
3872         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, nextents - 1), rec);
3873         *is_empty = 0;
3874         return 0;
3875 }
3876
3877 /*
3878  * Check the last inode extent to determine whether this allocation will result
3879  * in blocks being allocated at the end of the file. When we allocate new data
3880  * blocks at the end of the file which do not start at the previous data block,
3881  * we will try to align the new blocks at stripe unit boundaries.
3882  *
3883  * Returns 0 in bma->aeof if the file (fork) is empty as any new write will be
3884  * at, or past the EOF.
3885  */
3886 STATIC int
3887 xfs_bmap_isaeof(
3888         struct xfs_bmalloca     *bma,
3889         int                     whichfork)
3890 {
3891         struct xfs_bmbt_irec    rec;
3892         int                     is_empty;
3893         int                     error;
3894
3895         bma->aeof = 0;
3896         error = xfs_bmap_last_extent(NULL, bma->ip, whichfork, &rec,
3897                                      &is_empty);
3898         if (error || is_empty)
3899                 return error;
3900
3901         /*
3902          * Check if we are allocation or past the last extent, or at least into
3903          * the last delayed allocated extent.
3904          */
3905         bma->aeof = bma->off >= rec.br_startoff + rec.br_blockcount ||
3906                 (bma->off >= rec.br_startoff &&
3907                  isnullstartblock(rec.br_startblock));
3908         return 0;
3909 }
3910
3911 /*
3912  * Check if the endoff is outside the last extent. If so the caller will grow
3913  * the allocation to a stripe unit boundary.  All offsets are considered outside
3914  * the end of file for an empty fork, so 1 is returned in *eof in that case.
3915  */
3916 int
3917 xfs_bmap_eof(
3918         struct xfs_inode        *ip,
3919         xfs_fileoff_t           endoff,
3920         int                     whichfork,
3921         int                     *eof)
3922 {
3923         struct xfs_bmbt_irec    rec;
3924         int                     error;
3925
3926         error = xfs_bmap_last_extent(NULL, ip, whichfork, &rec, eof);
3927         if (error || *eof)
3928                 return error;
3929
3930         *eof = endoff >= rec.br_startoff + rec.br_blockcount;
3931         return 0;
3932 }
3933
3934 /*
3935  * Returns the file-relative block number of the first block past eof in
3936  * the file.  This is not based on i_size, it is based on the extent records.
3937  * Returns 0 for local files, as they do not have extent records.
3938  */
3939 int
3940 xfs_bmap_last_offset(
3941         struct xfs_trans        *tp,
3942         struct xfs_inode        *ip,
3943         xfs_fileoff_t           *last_block,
3944         int                     whichfork)
3945 {
3946         struct xfs_bmbt_irec    rec;
3947         int                     is_empty;
3948         int                     error;
3949
3950         *last_block = 0;
3951
3952         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL)
3953                 return 0;
3954
3955         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
3956             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
3957                return XFS_ERROR(EIO);
3958
3959         error = xfs_bmap_last_extent(NULL, ip, whichfork, &rec, &is_empty);
3960         if (error || is_empty)
3961                 return error;
3962
3963         *last_block = rec.br_startoff + rec.br_blockcount;
3964         return 0;
3965 }
3966
3967 /*
3968  * Returns whether the selected fork of the inode has exactly one
3969  * block or not.  For the data fork we check this matches di_size,
3970  * implying the file's range is 0..bsize-1.
3971  */
3972 int                                     /* 1=>1 block, 0=>otherwise */
3973 xfs_bmap_one_block(
3974         xfs_inode_t     *ip,            /* incore inode */
3975         int             whichfork)      /* data or attr fork */
3976 {
3977         xfs_bmbt_rec_host_t *ep;        /* ptr to fork's extent */
3978         xfs_ifork_t     *ifp;           /* inode fork pointer */
3979         int             rval;           /* return value */
3980         xfs_bmbt_irec_t s;              /* internal version of extent */
3981
3982 #ifndef DEBUG
3983         if (whichfork == XFS_DATA_FORK) {
3984                 return S_ISREG(ip->i_d.di_mode) ?
3985                         (ip->i_size == ip->i_mount->m_sb.sb_blocksize) :
3986                         (ip->i_d.di_size == ip->i_mount->m_sb.sb_blocksize);
3987         }
3988 #endif  /* !DEBUG */
3989         if (XFS_IFORK_NEXTENTS(ip, whichfork) != 1)
3990                 return 0;
3991         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
3992                 return 0;
3993         ifp = XFS_IFORK_PTR(ip, whichfork);
3994         ASSERT(ifp->if_flags & XFS_IFEXTENTS);
3995         ep = xfs_iext_get_ext(ifp, 0);
3996         xfs_bmbt_get_all(ep, &s);
3997         rval = s.br_startoff == 0 && s.br_blockcount == 1;
3998         if (rval && whichfork == XFS_DATA_FORK)
3999                 ASSERT(ip->i_size == ip->i_mount->m_sb.sb_blocksize);
4000         return rval;
4001 }
4002
4003 STATIC int
4004 xfs_bmap_sanity_check(
4005         struct xfs_mount        *mp,
4006         struct xfs_buf          *bp,
4007         int                     level)
4008 {
4009         struct xfs_btree_block  *block = XFS_BUF_TO_BLOCK(bp);
4010
4011         if (block->bb_magic != cpu_to_be32(XFS_BMAP_MAGIC) ||
4012             be16_to_cpu(block->bb_level) != level ||
4013             be16_to_cpu(block->bb_numrecs) == 0 ||
4014             be16_to_cpu(block->bb_numrecs) > mp->m_bmap_dmxr[level != 0])
4015                 return 0;
4016         return 1;
4017 }
4018
4019 /*
4020  * Read in the extents to if_extents.
4021  * All inode fields are set up by caller, we just traverse the btree
4022  * and copy the records in. If the file system cannot contain unwritten
4023  * extents, the records are checked for no "state" flags.
4024  */
4025 int                                     /* error */
4026 xfs_bmap_read_extents(
4027         xfs_trans_t             *tp,    /* transaction pointer */
4028         xfs_inode_t             *ip,    /* incore inode */
4029         int                     whichfork) /* data or attr fork */
4030 {
4031         struct xfs_btree_block  *block; /* current btree block */
4032         xfs_fsblock_t           bno;    /* block # of "block" */
4033         xfs_buf_t               *bp;    /* buffer for "block" */
4034         int                     error;  /* error return value */
4035         xfs_exntfmt_t           exntf;  /* XFS_EXTFMT_NOSTATE, if checking */
4036         xfs_extnum_t            i, j;   /* index into the extents list */
4037         xfs_ifork_t             *ifp;   /* fork structure */
4038         int                     level;  /* btree level, for checking */
4039         xfs_mount_t             *mp;    /* file system mount structure */
4040         __be64                  *pp;    /* pointer to block address */
4041         /* REFERENCED */
4042         xfs_extnum_t            room;   /* number of entries there's room for */
4043
4044         bno = NULLFSBLOCK;
4045         mp = ip->i_mount;
4046         ifp = XFS_IFORK_PTR(ip, whichfork);
4047         exntf = (whichfork != XFS_DATA_FORK) ? XFS_EXTFMT_NOSTATE :
4048                                         XFS_EXTFMT_INODE(ip);
4049         block = ifp->if_broot;
4050         /*
4051          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
4052          */
4053         level = be16_to_cpu(block->bb_level);
4054         ASSERT(level > 0);
4055         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
4056         bno = be64_to_cpu(*pp);
4057         ASSERT(bno != NULLDFSBNO);
4058         ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
4059         ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
4060         /*
4061          * Go down the tree until leaf level is reached, following the first
4062          * pointer (leftmost) at each level.
4063          */
4064         while (level-- > 0) {
4065                 if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
4066                                 XFS_BMAP_BTREE_REF)))
4067                         return error;
4068                 block = XFS_BUF_TO_BLOCK(bp);
4069                 XFS_WANT_CORRUPTED_GOTO(
4070                         xfs_bmap_sanity_check(mp, bp, level),
4071                         error0);
4072                 if (level == 0)
4073                         break;
4074                 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
4075                 bno = be64_to_cpu(*pp);
4076                 XFS_WANT_CORRUPTED_GOTO(XFS_FSB_SANITY_CHECK(mp, bno), error0);
4077                 xfs_trans_brelse(tp, bp);
4078         }
4079         /*
4080          * Here with bp and block set to the leftmost leaf node in the tree.
4081          */
4082         room = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4083         i = 0;
4084         /*
4085          * Loop over all leaf nodes.  Copy information to the extent records.
4086          */
4087         for (;;) {
4088                 xfs_bmbt_rec_t  *frp;
4089                 xfs_fsblock_t   nextbno;
4090                 xfs_extnum_t    num_recs;
4091                 xfs_extnum_t    start;
4092
4093                 num_recs = xfs_btree_get_numrecs(block);
4094                 if (unlikely(i + num_recs > room)) {
4095                         ASSERT(i + num_recs <= room);
4096                         xfs_warn(ip->i_mount,
4097                                 "corrupt dinode %Lu, (btree extents).",
4098                                 (unsigned long long) ip->i_ino);
4099                         XFS_CORRUPTION_ERROR("xfs_bmap_read_extents(1)",
4100                                 XFS_ERRLEVEL_LOW, ip->i_mount, block);
4101                         goto error0;
4102                 }
4103                 XFS_WANT_CORRUPTED_GOTO(
4104                         xfs_bmap_sanity_check(mp, bp, 0),
4105                         error0);
4106                 /*
4107                  * Read-ahead the next leaf block, if any.
4108                  */
4109                 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
4110                 if (nextbno != NULLFSBLOCK)
4111                         xfs_btree_reada_bufl(mp, nextbno, 1);
4112                 /*
4113                  * Copy records into the extent records.
4114                  */
4115                 frp = XFS_BMBT_REC_ADDR(mp, block, 1);
4116                 start = i;
4117                 for (j = 0; j < num_recs; j++, i++, frp++) {
4118                         xfs_bmbt_rec_host_t *trp = xfs_iext_get_ext(ifp, i);
4119                         trp->l0 = be64_to_cpu(frp->l0);
4120                         trp->l1 = be64_to_cpu(frp->l1);
4121                 }
4122                 if (exntf == XFS_EXTFMT_NOSTATE) {
4123                         /*
4124                          * Check all attribute bmap btree records and
4125                          * any "older" data bmap btree records for a
4126                          * set bit in the "extent flag" position.
4127                          */
4128                         if (unlikely(xfs_check_nostate_extents(ifp,
4129                                         start, num_recs))) {
4130                                 XFS_ERROR_REPORT("xfs_bmap_read_extents(2)",
4131                                                  XFS_ERRLEVEL_LOW,
4132                                                  ip->i_mount);
4133                                 goto error0;
4134                         }
4135                 }
4136                 xfs_trans_brelse(tp, bp);
4137                 bno = nextbno;
4138                 /*
4139                  * If we've reached the end, stop.
4140                  */
4141                 if (bno == NULLFSBLOCK)
4142                         break;
4143                 if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
4144                                 XFS_BMAP_BTREE_REF)))
4145                         return error;
4146                 block = XFS_BUF_TO_BLOCK(bp);
4147         }
4148         ASSERT(i == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)));
4149         ASSERT(i == XFS_IFORK_NEXTENTS(ip, whichfork));
4150         XFS_BMAP_TRACE_EXLIST(ip, i, whichfork);
4151         return 0;
4152 error0:
4153         xfs_trans_brelse(tp, bp);
4154         return XFS_ERROR(EFSCORRUPTED);
4155 }
4156
4157 #ifdef DEBUG
4158 /*
4159  * Add bmap trace insert entries for all the contents of the extent records.
4160  */
4161 void
4162 xfs_bmap_trace_exlist(
4163         xfs_inode_t     *ip,            /* incore inode pointer */
4164         xfs_extnum_t    cnt,            /* count of entries in the list */
4165         int             whichfork,      /* data or attr fork */
4166         unsigned long   caller_ip)
4167 {
4168         xfs_extnum_t    idx;            /* extent record index */
4169         xfs_ifork_t     *ifp;           /* inode fork pointer */
4170         int             state = 0;
4171
4172         if (whichfork == XFS_ATTR_FORK)
4173                 state |= BMAP_ATTRFORK;
4174
4175         ifp = XFS_IFORK_PTR(ip, whichfork);
4176         ASSERT(cnt == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)));
4177         for (idx = 0; idx < cnt; idx++)
4178                 trace_xfs_extlist(ip, idx, whichfork, caller_ip);
4179 }
4180
4181 /*
4182  * Validate that the bmbt_irecs being returned from bmapi are valid
4183  * given the callers original parameters.  Specifically check the
4184  * ranges of the returned irecs to ensure that they only extent beyond
4185  * the given parameters if the XFS_BMAPI_ENTIRE flag was set.
4186  */
4187 STATIC void
4188 xfs_bmap_validate_ret(
4189         xfs_fileoff_t           bno,
4190         xfs_filblks_t           len,
4191         int                     flags,
4192         xfs_bmbt_irec_t         *mval,
4193         int                     nmap,
4194         int                     ret_nmap)
4195 {
4196         int                     i;              /* index to map values */
4197
4198         ASSERT(ret_nmap <= nmap);
4199
4200         for (i = 0; i < ret_nmap; i++) {
4201                 ASSERT(mval[i].br_blockcount > 0);
4202                 if (!(flags & XFS_BMAPI_ENTIRE)) {
4203                         ASSERT(mval[i].br_startoff >= bno);
4204                         ASSERT(mval[i].br_blockcount <= len);
4205                         ASSERT(mval[i].br_startoff + mval[i].br_blockcount <=
4206                                bno + len);
4207                 } else {
4208                         ASSERT(mval[i].br_startoff < bno + len);
4209                         ASSERT(mval[i].br_startoff + mval[i].br_blockcount >
4210                                bno);
4211                 }
4212                 ASSERT(i == 0 ||
4213                        mval[i - 1].br_startoff + mval[i - 1].br_blockcount ==
4214                        mval[i].br_startoff);
4215                 ASSERT(mval[i].br_startblock != DELAYSTARTBLOCK &&
4216                        mval[i].br_startblock != HOLESTARTBLOCK);
4217                 ASSERT(mval[i].br_state == XFS_EXT_NORM ||
4218                        mval[i].br_state == XFS_EXT_UNWRITTEN);
4219         }
4220 }
4221 #endif /* DEBUG */
4222
4223
4224 /*
4225  * Trim the returned map to the required bounds
4226  */
4227 STATIC void
4228 xfs_bmapi_trim_map(
4229         struct xfs_bmbt_irec    *mval,
4230         struct xfs_bmbt_irec    *got,
4231         xfs_fileoff_t           *bno,
4232         xfs_filblks_t           len,
4233         xfs_fileoff_t           obno,
4234         xfs_fileoff_t           end,
4235         int                     n,
4236         int                     flags)
4237 {
4238         if ((flags & XFS_BMAPI_ENTIRE) ||
4239             got->br_startoff + got->br_blockcount <= obno) {
4240                 *mval = *got;
4241                 if (isnullstartblock(got->br_startblock))
4242                         mval->br_startblock = DELAYSTARTBLOCK;
4243                 return;
4244         }
4245
4246         if (obno > *bno)
4247                 *bno = obno;
4248         ASSERT((*bno >= obno) || (n == 0));
4249         ASSERT(*bno < end);
4250         mval->br_startoff = *bno;
4251         if (isnullstartblock(got->br_startblock))
4252                 mval->br_startblock = DELAYSTARTBLOCK;
4253         else
4254                 mval->br_startblock = got->br_startblock +
4255                                         (*bno - got->br_startoff);
4256         /*
4257          * Return the minimum of what we got and what we asked for for
4258          * the length.  We can use the len variable here because it is
4259          * modified below and we could have been there before coming
4260          * here if the first part of the allocation didn't overlap what
4261          * was asked for.
4262          */
4263         mval->br_blockcount = XFS_FILBLKS_MIN(end - *bno,
4264                         got->br_blockcount - (*bno - got->br_startoff));
4265         mval->br_state = got->br_state;
4266         ASSERT(mval->br_blockcount <= len);
4267         return;
4268 }
4269
4270 /*
4271  * Update and validate the extent map to return
4272  */
4273 STATIC void
4274 xfs_bmapi_update_map(
4275         struct xfs_bmbt_irec    **map,
4276         xfs_fileoff_t           *bno,
4277         xfs_filblks_t           *len,
4278         xfs_fileoff_t           obno,
4279         xfs_fileoff_t           end,
4280         int                     *n,
4281         int                     flags)
4282 {
4283         xfs_bmbt_irec_t *mval = *map;
4284
4285         ASSERT((flags & XFS_BMAPI_ENTIRE) ||
4286                ((mval->br_startoff + mval->br_blockcount) <= end));
4287         ASSERT((flags & XFS_BMAPI_ENTIRE) || (mval->br_blockcount <= *len) ||
4288                (mval->br_startoff < obno));
4289
4290         *bno = mval->br_startoff + mval->br_blockcount;
4291         *len = end - *bno;
4292         if (*n > 0 && mval->br_startoff == mval[-1].br_startoff) {
4293                 /* update previous map with new information */
4294                 ASSERT(mval->br_startblock == mval[-1].br_startblock);
4295                 ASSERT(mval->br_blockcount > mval[-1].br_blockcount);
4296                 ASSERT(mval->br_state == mval[-1].br_state);
4297                 mval[-1].br_blockcount = mval->br_blockcount;
4298                 mval[-1].br_state = mval->br_state;
4299         } else if (*n > 0 && mval->br_startblock != DELAYSTARTBLOCK &&
4300                    mval[-1].br_startblock != DELAYSTARTBLOCK &&
4301                    mval[-1].br_startblock != HOLESTARTBLOCK &&
4302                    mval->br_startblock == mval[-1].br_startblock +
4303                                           mval[-1].br_blockcount &&
4304                    ((flags & XFS_BMAPI_IGSTATE) ||
4305                         mval[-1].br_state == mval->br_state)) {
4306                 ASSERT(mval->br_startoff ==
4307                        mval[-1].br_startoff + mval[-1].br_blockcount);
4308                 mval[-1].br_blockcount += mval->br_blockcount;
4309         } else if (*n > 0 &&
4310                    mval->br_startblock == DELAYSTARTBLOCK &&
4311                    mval[-1].br_startblock == DELAYSTARTBLOCK &&
4312                    mval->br_startoff ==
4313                    mval[-1].br_startoff + mval[-1].br_blockcount) {
4314                 mval[-1].br_blockcount += mval->br_blockcount;
4315                 mval[-1].br_state = mval->br_state;
4316         } else if (!((*n == 0) &&
4317                      ((mval->br_startoff + mval->br_blockcount) <=
4318                       obno))) {
4319                 mval++;
4320                 (*n)++;
4321         }
4322         *map = mval;
4323 }
4324
4325 /*
4326  * Map file blocks to filesystem blocks without allocation.
4327  */
4328 int
4329 xfs_bmapi_read(
4330         struct xfs_inode        *ip,
4331         xfs_fileoff_t           bno,
4332         xfs_filblks_t           len,
4333         struct xfs_bmbt_irec    *mval,
4334         int                     *nmap,
4335         int                     flags)
4336 {
4337         struct xfs_mount        *mp = ip->i_mount;
4338         struct xfs_ifork        *ifp;
4339         struct xfs_bmbt_irec    got;
4340         struct xfs_bmbt_irec    prev;
4341         xfs_fileoff_t           obno;
4342         xfs_fileoff_t           end;
4343         xfs_extnum_t            lastx;
4344         int                     error;
4345         int                     eof;
4346         int                     n = 0;
4347         int                     whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4348                                                 XFS_ATTR_FORK : XFS_DATA_FORK;
4349
4350         ASSERT(*nmap >= 1);
4351         ASSERT(!(flags & ~(XFS_BMAPI_ATTRFORK|XFS_BMAPI_ENTIRE|
4352                            XFS_BMAPI_IGSTATE)));
4353
4354         if (unlikely(XFS_TEST_ERROR(
4355             (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4356              XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
4357              mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4358                 XFS_ERROR_REPORT("xfs_bmapi_read", XFS_ERRLEVEL_LOW, mp);
4359                 return XFS_ERROR(EFSCORRUPTED);
4360         }
4361
4362         if (XFS_FORCED_SHUTDOWN(mp))
4363                 return XFS_ERROR(EIO);
4364
4365         XFS_STATS_INC(xs_blk_mapr);
4366
4367         ifp = XFS_IFORK_PTR(ip, whichfork);
4368         ASSERT(ifp->if_ext_max ==
4369                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
4370
4371         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4372                 error = xfs_iread_extents(NULL, ip, whichfork);
4373                 if (error)
4374                         return error;
4375         }
4376
4377         xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got, &prev);
4378         end = bno + len;
4379         obno = bno;
4380
4381         while (bno < end && n < *nmap) {
4382                 /* Reading past eof, act as though there's a hole up to end. */
4383                 if (eof)
4384                         got.br_startoff = end;
4385                 if (got.br_startoff > bno) {
4386                         /* Reading in a hole.  */
4387                         mval->br_startoff = bno;
4388                         mval->br_startblock = HOLESTARTBLOCK;
4389                         mval->br_blockcount =
4390                                 XFS_FILBLKS_MIN(len, got.br_startoff - bno);
4391                         mval->br_state = XFS_EXT_NORM;
4392                         bno += mval->br_blockcount;
4393                         len -= mval->br_blockcount;
4394                         mval++;
4395                         n++;
4396                         continue;
4397                 }
4398
4399                 /* set up the extent map to return. */
4400                 xfs_bmapi_trim_map(mval, &got, &bno, len, obno, end, n, flags);
4401                 xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4402
4403                 /* If we're done, stop now. */
4404                 if (bno >= end || n >= *nmap)
4405                         break;
4406
4407                 /* Else go on to the next record. */
4408                 if (++lastx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t))
4409                         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx), &got);
4410                 else
4411                         eof = 1;
4412         }
4413         *nmap = n;
4414         return 0;
4415 }
4416
4417 STATIC int
4418 xfs_bmapi_reserve_delalloc(
4419         struct xfs_inode        *ip,
4420         xfs_fileoff_t           aoff,
4421         xfs_filblks_t           len,
4422         struct xfs_bmbt_irec    *got,
4423         struct xfs_bmbt_irec    *prev,
4424         xfs_extnum_t            *lastx,
4425         int                     eof)
4426 {
4427         struct xfs_mount        *mp = ip->i_mount;
4428         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
4429         xfs_extlen_t            alen;
4430         xfs_extlen_t            indlen;
4431         char                    rt = XFS_IS_REALTIME_INODE(ip);
4432         xfs_extlen_t            extsz;
4433         int                     error;
4434
4435         alen = XFS_FILBLKS_MIN(len, MAXEXTLEN);
4436         if (!eof)
4437                 alen = XFS_FILBLKS_MIN(alen, got->br_startoff - aoff);
4438
4439         /* Figure out the extent size, adjust alen */
4440         extsz = xfs_get_extsz_hint(ip);
4441         if (extsz) {
4442                 /*
4443                  * Make sure we don't exceed a single extent length when we
4444                  * align the extent by reducing length we are going to
4445                  * allocate by the maximum amount extent size aligment may
4446                  * require.
4447                  */
4448                 alen = XFS_FILBLKS_MIN(len, MAXEXTLEN - (2 * extsz - 1));
4449                 error = xfs_bmap_extsize_align(mp, got, prev, extsz, rt, eof,
4450                                                1, 0, &aoff, &alen);
4451                 ASSERT(!error);
4452         }
4453
4454         if (rt)
4455                 extsz = alen / mp->m_sb.sb_rextsize;
4456
4457         /*
4458          * Make a transaction-less quota reservation for delayed allocation
4459          * blocks.  This number gets adjusted later.  We return if we haven't
4460          * allocated blocks already inside this loop.
4461          */
4462         error = xfs_trans_reserve_quota_nblks(NULL, ip, (long)alen, 0,
4463                         rt ? XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4464         if (error)
4465                 return error;
4466
4467         /*
4468          * Split changing sb for alen and indlen since they could be coming
4469          * from different places.
4470          */
4471         indlen = (xfs_extlen_t)xfs_bmap_worst_indlen(ip, alen);
4472         ASSERT(indlen > 0);
4473
4474         if (rt) {
4475                 error = xfs_mod_incore_sb(mp, XFS_SBS_FREXTENTS,
4476                                           -((int64_t)extsz), 0);
4477         } else {
4478                 error = xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
4479                                                  -((int64_t)alen), 0);
4480         }
4481
4482         if (error)
4483                 goto out_unreserve_quota;
4484
4485         error = xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
4486                                          -((int64_t)indlen), 0);
4487         if (error)
4488                 goto out_unreserve_blocks;
4489
4490
4491         ip->i_delayed_blks += alen;
4492
4493         got->br_startoff = aoff;
4494         got->br_startblock = nullstartblock(indlen);
4495         got->br_blockcount = alen;
4496         got->br_state = XFS_EXT_NORM;
4497         xfs_bmap_add_extent_hole_delay(ip, lastx, got);
4498
4499         /*
4500          * Update our extent pointer, given that xfs_bmap_add_extent_hole_delay
4501          * might have merged it into one of the neighbouring ones.
4502          */
4503         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *lastx), got);
4504
4505         ASSERT(got->br_startoff <= aoff);
4506         ASSERT(got->br_startoff + got->br_blockcount >= aoff + alen);
4507         ASSERT(isnullstartblock(got->br_startblock));
4508         ASSERT(got->br_state == XFS_EXT_NORM);
4509         return 0;
4510
4511 out_unreserve_blocks:
4512         if (rt)
4513                 xfs_mod_incore_sb(mp, XFS_SBS_FREXTENTS, extsz, 0);
4514         else
4515                 xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS, alen, 0);
4516 out_unreserve_quota:
4517         if (XFS_IS_QUOTA_ON(mp))
4518                 xfs_trans_unreserve_quota_nblks(NULL, ip, alen, 0, rt ?
4519                                 XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4520         return error;
4521 }
4522
4523 /*
4524  * Map file blocks to filesystem blocks, adding delayed allocations as needed.
4525  */
4526 int
4527 xfs_bmapi_delay(
4528         struct xfs_inode        *ip,    /* incore inode */
4529         xfs_fileoff_t           bno,    /* starting file offs. mapped */
4530         xfs_filblks_t           len,    /* length to map in file */
4531         struct xfs_bmbt_irec    *mval,  /* output: map values */
4532         int                     *nmap,  /* i/o: mval size/count */
4533         int                     flags)  /* XFS_BMAPI_... */
4534 {
4535         struct xfs_mount        *mp = ip->i_mount;
4536         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
4537         struct xfs_bmbt_irec    got;    /* current file extent record */
4538         struct xfs_bmbt_irec    prev;   /* previous file extent record */
4539         xfs_fileoff_t           obno;   /* old block number (offset) */
4540         xfs_fileoff_t           end;    /* end of mapped file region */
4541         xfs_extnum_t            lastx;  /* last useful extent number */
4542         int                     eof;    /* we've hit the end of extents */
4543         int                     n = 0;  /* current extent index */
4544         int                     error = 0;
4545
4546         ASSERT(*nmap >= 1);
4547         ASSERT(*nmap <= XFS_BMAP_MAX_NMAP);
4548         ASSERT(!(flags & ~XFS_BMAPI_ENTIRE));
4549
4550         if (unlikely(XFS_TEST_ERROR(
4551             (XFS_IFORK_FORMAT(ip, XFS_DATA_FORK) != XFS_DINODE_FMT_EXTENTS &&
4552              XFS_IFORK_FORMAT(ip, XFS_DATA_FORK) != XFS_DINODE_FMT_BTREE),
4553              mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4554                 XFS_ERROR_REPORT("xfs_bmapi_delay", XFS_ERRLEVEL_LOW, mp);
4555                 return XFS_ERROR(EFSCORRUPTED);
4556         }
4557
4558         if (XFS_FORCED_SHUTDOWN(mp))
4559                 return XFS_ERROR(EIO);
4560
4561         XFS_STATS_INC(xs_blk_mapw);
4562
4563         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4564                 error = xfs_iread_extents(NULL, ip, XFS_DATA_FORK);
4565                 if (error)
4566                         return error;
4567         }
4568
4569         xfs_bmap_search_extents(ip, bno, XFS_DATA_FORK, &eof, &lastx, &got, &prev);
4570         end = bno + len;
4571         obno = bno;
4572
4573         while (bno < end && n < *nmap) {
4574                 if (eof || got.br_startoff > bno) {
4575                         error = xfs_bmapi_reserve_delalloc(ip, bno, len, &got,
4576                                                            &prev, &lastx, eof);
4577                         if (error) {
4578                                 if (n == 0) {
4579                                         *nmap = 0;
4580                                         return error;
4581                                 }
4582                                 break;
4583                         }
4584                 }
4585
4586                 /* set up the extent map to return. */
4587                 xfs_bmapi_trim_map(mval, &got, &bno, len, obno, end, n, flags);
4588                 xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4589
4590                 /* If we're done, stop now. */
4591                 if (bno >= end || n >= *nmap)
4592                         break;
4593
4594                 /* Else go on to the next record. */
4595                 prev = got;
4596                 if (++lastx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t))
4597                         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx), &got);
4598                 else
4599                         eof = 1;
4600         }
4601
4602         *nmap = n;
4603         return 0;
4604 }
4605
4606
4607 STATIC int
4608 xfs_bmapi_allocate(
4609         struct xfs_bmalloca     *bma,
4610         xfs_extnum_t            *lastx,
4611         struct xfs_btree_cur    **cur,
4612         int                     flags,
4613         int                     *nallocs,
4614         int                     *logflags)
4615 {
4616         struct xfs_mount        *mp = bma->ip->i_mount;
4617         int                     whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4618                                                 XFS_ATTR_FORK : XFS_DATA_FORK;
4619         struct xfs_ifork        *ifp = XFS_IFORK_PTR(bma->ip, whichfork);
4620         xfs_fsblock_t           abno;
4621         xfs_extlen_t            alen;
4622         xfs_fileoff_t           aoff;
4623         int                     error;
4624         int                     rt;
4625
4626         rt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(bma->ip);
4627
4628         /*
4629          * For the wasdelay case, we could also just allocate the stuff asked
4630          * for in this bmap call but that wouldn't be as good.
4631          */
4632         if (bma->wasdel) {
4633                 alen = (xfs_extlen_t)bma->got.br_blockcount;
4634                 aoff = bma->got.br_startoff;
4635                 if (*lastx != NULLEXTNUM && *lastx) {
4636                         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *lastx - 1),
4637                                          &bma->prev);
4638                 }
4639         } else {
4640                 alen = (xfs_extlen_t)XFS_FILBLKS_MIN(bma->alen, MAXEXTLEN);
4641                 if (!bma->eof)
4642                         alen = (xfs_extlen_t)XFS_FILBLKS_MIN(alen,
4643                                         bma->got.br_startoff - bma->off);
4644                 aoff = bma->off;
4645         }
4646
4647         /*
4648          * Indicate if this is the first user data in the file, or just any
4649          * user data.
4650          */
4651         if (!(flags & XFS_BMAPI_METADATA)) {
4652                 bma->userdata = (aoff == 0) ?
4653                         XFS_ALLOC_INITIAL_USER_DATA : XFS_ALLOC_USERDATA;
4654         }
4655
4656         /*
4657          * Fill in changeable bma fields.
4658          */
4659         bma->alen = alen;
4660         bma->off = aoff;
4661         bma->minlen = (flags & XFS_BMAPI_CONTIG) ? alen : 1;
4662         bma->aeof = 0;
4663
4664         /*
4665          * Only want to do the alignment at the eof if it is userdata and
4666          * allocation length is larger than a stripe unit.
4667          */
4668         if (mp->m_dalign && alen >= mp->m_dalign &&
4669             !(flags & XFS_BMAPI_METADATA) && whichfork == XFS_DATA_FORK) {
4670                 error = xfs_bmap_isaeof(bma, whichfork);
4671                 if (error)
4672                         return error;
4673         }
4674
4675         error = xfs_bmap_alloc(bma);
4676         if (error)
4677                 return error;
4678
4679         /*
4680          * Copy out result fields.
4681          */
4682         abno = bma->rval;
4683         alen = bma->alen;
4684         aoff = bma->off;
4685         if (bma->flist->xbf_low)
4686                 bma->minleft = 0;
4687         if (*cur)
4688                 (*cur)->bc_private.b.firstblock = *bma->firstblock;
4689         if (abno == NULLFSBLOCK)
4690                 return 0;
4691         if ((ifp->if_flags & XFS_IFBROOT) && !*cur) {
4692                 (*cur) = xfs_bmbt_init_cursor(mp, bma->tp, bma->ip, whichfork);
4693                 (*cur)->bc_private.b.firstblock = *bma->firstblock;
4694                 (*cur)->bc_private.b.flist = bma->flist;
4695         }
4696         /*
4697          * Bump the number of extents we've allocated
4698          * in this call.
4699          */
4700         (*nallocs)++;
4701
4702         if (*cur)
4703                 (*cur)->bc_private.b.flags =
4704                         bma->wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
4705
4706         bma->got.br_startoff = aoff;
4707         bma->got.br_startblock = abno;
4708         bma->got.br_blockcount = alen;
4709         bma->got.br_state = XFS_EXT_NORM;
4710
4711         /*
4712          * A wasdelay extent has been initialized, so shouldn't be flagged
4713          * as unwritten.
4714          */
4715         if (!bma->wasdel && (flags & XFS_BMAPI_PREALLOC) &&
4716             xfs_sb_version_hasextflgbit(&mp->m_sb))
4717                 bma->got.br_state = XFS_EXT_UNWRITTEN;
4718
4719         if (bma->wasdel) {
4720                 error = xfs_bmap_add_extent_delay_real(bma->tp, bma->ip, lastx,
4721                                 cur, &bma->got, bma->firstblock, bma->flist,
4722                                 logflags);
4723         } else {
4724                 error = xfs_bmap_add_extent_hole_real(bma->tp, bma->ip, lastx,
4725                                 cur, &bma->got, bma->firstblock, bma->flist,
4726                                 logflags, whichfork);
4727         }
4728
4729         if (error)
4730                 return error;
4731
4732         /*
4733          * Update our extent pointer, given that xfs_bmap_add_extent_delay_real
4734          * or xfs_bmap_add_extent_hole_real might have merged it into one of
4735          * the neighbouring ones.
4736          */
4737         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *lastx), &bma->got);
4738
4739         ASSERT(bma->got.br_startoff <= aoff);
4740         ASSERT(bma->got.br_startoff + bma->got.br_blockcount >= aoff + alen);
4741         ASSERT(bma->got.br_state == XFS_EXT_NORM ||
4742                bma->got.br_state == XFS_EXT_UNWRITTEN);
4743         return 0;
4744 }
4745
4746 STATIC int
4747 xfs_bmapi_convert_unwritten(
4748         struct xfs_bmalloca     *bma,
4749         struct xfs_bmbt_irec    *mval,
4750         xfs_filblks_t           len,
4751         xfs_extnum_t            *lastx,
4752         struct xfs_btree_cur    **cur,
4753         int                     flags,
4754         int                     *logflags)
4755 {
4756         int                     whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4757                                                 XFS_ATTR_FORK : XFS_DATA_FORK;
4758         struct xfs_ifork        *ifp = XFS_IFORK_PTR(bma->ip, whichfork);
4759         int                     error;
4760
4761         *logflags = 0;
4762
4763         /* check if we need to do unwritten->real conversion */
4764         if (mval->br_state == XFS_EXT_UNWRITTEN &&
4765             (flags & XFS_BMAPI_PREALLOC))
4766                 return 0;
4767
4768         /* check if we need to do real->unwritten conversion */
4769         if (mval->br_state == XFS_EXT_NORM &&
4770             (flags & (XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT)) !=
4771                         (XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT))
4772                 return 0;
4773
4774         /*
4775          * Modify (by adding) the state flag, if writing.
4776          */
4777         ASSERT(mval->br_blockcount <= len);
4778         if ((ifp->if_flags & XFS_IFBROOT) && !*cur) {
4779                 *cur = xfs_bmbt_init_cursor(bma->ip->i_mount, bma->tp,
4780                                         bma->ip, whichfork);
4781                 (*cur)->bc_private.b.firstblock = *bma->firstblock;
4782                 (*cur)->bc_private.b.flist = bma->flist;
4783         }
4784         mval->br_state = (mval->br_state == XFS_EXT_UNWRITTEN)
4785                                 ? XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
4786
4787         error = xfs_bmap_add_extent_unwritten_real(bma->tp, bma->ip, lastx,
4788                         cur, mval, bma->firstblock, bma->flist, logflags);
4789         if (error)
4790                 return error;
4791
4792         /*
4793          * Update our extent pointer, given that
4794          * xfs_bmap_add_extent_unwritten_real might have merged it into one
4795          * of the neighbouring ones.
4796          */
4797         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *lastx), &bma->got);
4798
4799         /*
4800          * We may have combined previously unwritten space with written space,
4801          * so generate another request.
4802          */
4803         if (mval->br_blockcount < len)
4804                 return EAGAIN;
4805         return 0;
4806 }
4807
4808 /*
4809  * Map file blocks to filesystem blocks, and allocate blocks or convert the
4810  * extent state if necessary.  Details behaviour is controlled by the flags
4811  * parameter.  Only allocates blocks from a single allocation group, to avoid
4812  * locking problems.
4813  *
4814  * The returned value in "firstblock" from the first call in a transaction
4815  * must be remembered and presented to subsequent calls in "firstblock".
4816  * An upper bound for the number of blocks to be allocated is supplied to
4817  * the first call in "total"; if no allocation group has that many free
4818  * blocks then the call will fail (return NULLFSBLOCK in "firstblock").
4819  */
4820 int
4821 xfs_bmapi_write(
4822         struct xfs_trans        *tp,            /* transaction pointer */
4823         struct xfs_inode        *ip,            /* incore inode */
4824         xfs_fileoff_t           bno,            /* starting file offs. mapped */
4825         xfs_filblks_t           len,            /* length to map in file */
4826         int                     flags,          /* XFS_BMAPI_... */
4827         xfs_fsblock_t           *firstblock,    /* first allocated block
4828                                                    controls a.g. for allocs */
4829         xfs_extlen_t            total,          /* total blocks needed */
4830         struct xfs_bmbt_irec    *mval,          /* output: map values */
4831         int                     *nmap,          /* i/o: mval size/count */
4832         struct xfs_bmap_free    *flist)         /* i/o: list extents to free */
4833 {
4834         struct xfs_mount        *mp = ip->i_mount;
4835         struct xfs_ifork        *ifp;
4836         struct xfs_bmalloca     bma = { 0 };    /* args for xfs_bmap_alloc */
4837         struct xfs_btree_cur    *cur;           /* bmap btree cursor */
4838         xfs_fileoff_t           end;            /* end of mapped file region */
4839         int                     eof;            /* after the end of extents */
4840         int                     error;          /* error return */
4841         xfs_extnum_t            lastx;          /* last useful extent number */
4842         int                     logflags;       /* flags for transaction logging */
4843         int                     n;              /* current extent index */
4844         int                     nallocs;        /* number of extents alloc'd */
4845         xfs_fileoff_t           obno;           /* old block number (offset) */
4846         int                     tmp_logflags;   /* temp flags holder */
4847         int                     whichfork;      /* data or attr fork */
4848         char                    inhole;         /* current location is hole in file */
4849         char                    wasdelay;       /* old extent was delayed */
4850
4851 #ifdef DEBUG
4852         xfs_fileoff_t           orig_bno;       /* original block number value */
4853         int                     orig_flags;     /* original flags arg value */
4854         xfs_filblks_t           orig_len;       /* original value of len arg */
4855         struct xfs_bmbt_irec    *orig_mval;     /* original value of mval */
4856         int                     orig_nmap;      /* original value of *nmap */
4857
4858         orig_bno = bno;
4859         orig_len = len;
4860         orig_flags = flags;
4861         orig_mval = mval;
4862         orig_nmap = *nmap;
4863 #endif
4864
4865         ASSERT(*nmap >= 1);
4866         ASSERT(*nmap <= XFS_BMAP_MAX_NMAP);
4867         ASSERT(!(flags & XFS_BMAPI_IGSTATE));
4868         ASSERT(tp != NULL);
4869
4870         whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4871                 XFS_ATTR_FORK : XFS_DATA_FORK;
4872
4873         if (unlikely(XFS_TEST_ERROR(
4874             (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4875              XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
4876              XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL),
4877              mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4878                 XFS_ERROR_REPORT("xfs_bmapi_write", XFS_ERRLEVEL_LOW, mp);
4879                 return XFS_ERROR(EFSCORRUPTED);
4880         }
4881
4882         if (XFS_FORCED_SHUTDOWN(mp))
4883                 return XFS_ERROR(EIO);
4884
4885         ifp = XFS_IFORK_PTR(ip, whichfork);
4886         ASSERT(ifp->if_ext_max ==
4887                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
4888
4889         XFS_STATS_INC(xs_blk_mapw);
4890
4891         logflags = 0;
4892         nallocs = 0;
4893         cur = NULL;
4894
4895         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
4896                 error = xfs_bmap_local_to_extents(tp, ip, firstblock, total,
4897                                                   &logflags, whichfork);
4898                 if (error)
4899                         goto error0;
4900         }
4901
4902         if (*firstblock == NULLFSBLOCK) {
4903                 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE)
4904                         bma.minleft = be16_to_cpu(ifp->if_broot->bb_level) + 1;
4905                 else
4906                         bma.minleft = 1;
4907         } else {
4908                 bma.minleft = 0;
4909         }
4910
4911         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4912                 error = xfs_iread_extents(tp, ip, whichfork);
4913                 if (error)
4914                         goto error0;
4915         }
4916
4917         xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &bma.got,
4918                                 &bma.prev);
4919         n = 0;
4920         end = bno + len;
4921         obno = bno;
4922
4923         bma.tp = tp;
4924         bma.ip = ip;
4925         bma.total = total;
4926         bma.userdata = 0;
4927         bma.flist = flist;
4928         bma.firstblock = firstblock;
4929
4930         while (bno < end && n < *nmap) {
4931                 inhole = eof || bma.got.br_startoff > bno;
4932                 wasdelay = !inhole && isnullstartblock(bma.got.br_startblock);
4933
4934                 /*
4935                  * First, deal with the hole before the allocated space
4936                  * that we found, if any.
4937                  */
4938                 if (inhole || wasdelay) {
4939                         bma.eof = eof;
4940                         bma.conv = !!(flags & XFS_BMAPI_CONVERT);
4941                         bma.wasdel = wasdelay;
4942                         bma.alen = len;
4943                         bma.off = bno;
4944
4945                         error = xfs_bmapi_allocate(&bma, &lastx, &cur, flags,
4946                                         &nallocs, &tmp_logflags);
4947                         logflags |= tmp_logflags;
4948                         if (error)
4949                                 goto error0;
4950                         if (bma.rval == NULLFSBLOCK)
4951                                 break;
4952                 }
4953
4954                 /* Deal with the allocated space we found.  */
4955                 xfs_bmapi_trim_map(mval, &bma.got, &bno, len, obno,
4956                                                         end, n, flags);
4957
4958                 /* Execute unwritten extent conversion if necessary */
4959                 error = xfs_bmapi_convert_unwritten(&bma, mval, len, &lastx,
4960                                                     &cur, flags, &tmp_logflags);
4961                 logflags |= tmp_logflags;
4962                 if (error == EAGAIN)
4963                         continue;
4964                 if (error)
4965                         goto error0;
4966
4967                 /* update the extent map to return */
4968                 xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4969
4970                 /*
4971                  * If we're done, stop now.  Stop when we've allocated
4972                  * XFS_BMAP_MAX_NMAP extents no matter what.  Otherwise
4973                  * the transaction may get too big.
4974                  */
4975                 if (bno >= end || n >= *nmap || nallocs >= *nmap)
4976                         break;
4977
4978                 /* Else go on to the next record. */
4979                 bma.prev = bma.got;
4980                 if (++lastx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t))
4981                         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx), &bma.got);
4982                 else
4983                         eof = 1;
4984         }
4985         *nmap = n;
4986
4987         /*
4988          * Transform from btree to extents, give it cur.
4989          */
4990         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE &&
4991             XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max) {
4992                 ASSERT(cur);
4993                 error = xfs_bmap_btree_to_extents(tp, ip, cur,
4994                         &tmp_logflags, whichfork);
4995                 logflags |= tmp_logflags;
4996                 if (error)
4997                         goto error0;
4998         }
4999         ASSERT(ifp->if_ext_max ==
5000                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5001         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE ||
5002                XFS_IFORK_NEXTENTS(ip, whichfork) > ifp->if_ext_max);
5003         error = 0;
5004 error0:
5005         /*
5006          * Log everything.  Do this after conversion, there's no point in
5007          * logging the extent records if we've converted to btree format.
5008          */
5009         if ((logflags & xfs_ilog_fext(whichfork)) &&
5010             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
5011                 logflags &= ~xfs_ilog_fext(whichfork);
5012         else if ((logflags & xfs_ilog_fbroot(whichfork)) &&
5013                  XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
5014                 logflags &= ~xfs_ilog_fbroot(whichfork);
5015         /*
5016          * Log whatever the flags say, even if error.  Otherwise we might miss
5017          * detecting a case where the data is changed, there's an error,
5018          * and it's not logged so we don't shutdown when we should.
5019          */
5020         if (logflags)
5021                 xfs_trans_log_inode(tp, ip, logflags);
5022
5023         if (cur) {
5024                 if (!error) {
5025                         ASSERT(*firstblock == NULLFSBLOCK ||
5026                                XFS_FSB_TO_AGNO(mp, *firstblock) ==
5027                                XFS_FSB_TO_AGNO(mp,
5028                                        cur->bc_private.b.firstblock) ||
5029                                (flist->xbf_low &&
5030                                 XFS_FSB_TO_AGNO(mp, *firstblock) <
5031                                 XFS_FSB_TO_AGNO(mp,
5032                                         cur->bc_private.b.firstblock)));
5033                         *firstblock = cur->bc_private.b.firstblock;
5034                 }
5035                 xfs_btree_del_cursor(cur,
5036                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5037         }
5038         if (!error)
5039                 xfs_bmap_validate_ret(orig_bno, orig_len, orig_flags, orig_mval,
5040                         orig_nmap, *nmap);
5041         return error;
5042 }
5043
5044 /*
5045  * Unmap (remove) blocks from a file.
5046  * If nexts is nonzero then the number of extents to remove is limited to
5047  * that value.  If not all extents in the block range can be removed then
5048  * *done is set.
5049  */
5050 int                                             /* error */
5051 xfs_bunmapi(
5052         xfs_trans_t             *tp,            /* transaction pointer */
5053         struct xfs_inode        *ip,            /* incore inode */
5054         xfs_fileoff_t           bno,            /* starting offset to unmap */
5055         xfs_filblks_t           len,            /* length to unmap in file */
5056         int                     flags,          /* misc flags */
5057         xfs_extnum_t            nexts,          /* number of extents max */
5058         xfs_fsblock_t           *firstblock,    /* first allocated block
5059                                                    controls a.g. for allocs */
5060         xfs_bmap_free_t         *flist,         /* i/o: list extents to free */
5061         int                     *done)          /* set if not done yet */
5062 {
5063         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
5064         xfs_bmbt_irec_t         del;            /* extent being deleted */
5065         int                     eof;            /* is deleting at eof */
5066         xfs_bmbt_rec_host_t     *ep;            /* extent record pointer */
5067         int                     error;          /* error return value */
5068         xfs_extnum_t            extno;          /* extent number in list */
5069         xfs_bmbt_irec_t         got;            /* current extent record */
5070         xfs_ifork_t             *ifp;           /* inode fork pointer */
5071         int                     isrt;           /* freeing in rt area */
5072         xfs_extnum_t            lastx;          /* last extent index used */
5073         int                     logflags;       /* transaction logging flags */
5074         xfs_extlen_t            mod;            /* rt extent offset */
5075         xfs_mount_t             *mp;            /* mount structure */
5076         xfs_extnum_t            nextents;       /* number of file extents */
5077         xfs_bmbt_irec_t         prev;           /* previous extent record */
5078         xfs_fileoff_t           start;          /* first file offset deleted */
5079         int                     tmp_logflags;   /* partial logging flags */
5080         int                     wasdel;         /* was a delayed alloc extent */
5081         int                     whichfork;      /* data or attribute fork */
5082         xfs_fsblock_t           sum;
5083
5084         trace_xfs_bunmap(ip, bno, len, flags, _RET_IP_);
5085
5086         whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
5087                 XFS_ATTR_FORK : XFS_DATA_FORK;
5088         ifp = XFS_IFORK_PTR(ip, whichfork);
5089         if (unlikely(
5090             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
5091             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
5092                 XFS_ERROR_REPORT("xfs_bunmapi", XFS_ERRLEVEL_LOW,
5093                                  ip->i_mount);
5094                 return XFS_ERROR(EFSCORRUPTED);
5095         }
5096         mp = ip->i_mount;
5097         if (XFS_FORCED_SHUTDOWN(mp))
5098                 return XFS_ERROR(EIO);
5099
5100         ASSERT(len > 0);
5101         ASSERT(nexts >= 0);
5102         ASSERT(ifp->if_ext_max ==
5103                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5104         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5105             (error = xfs_iread_extents(tp, ip, whichfork)))
5106                 return error;
5107         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5108         if (nextents == 0) {
5109                 *done = 1;
5110                 return 0;
5111         }
5112         XFS_STATS_INC(xs_blk_unmap);
5113         isrt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
5114         start = bno;
5115         bno = start + len - 1;
5116         ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
5117                 &prev);
5118
5119         /*
5120          * Check to see if the given block number is past the end of the
5121          * file, back up to the last block if so...
5122          */
5123         if (eof) {
5124                 ep = xfs_iext_get_ext(ifp, --lastx);
5125                 xfs_bmbt_get_all(ep, &got);
5126                 bno = got.br_startoff + got.br_blockcount - 1;
5127         }
5128         logflags = 0;
5129         if (ifp->if_flags & XFS_IFBROOT) {
5130                 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
5131                 cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5132                 cur->bc_private.b.firstblock = *firstblock;
5133                 cur->bc_private.b.flist = flist;
5134                 cur->bc_private.b.flags = 0;
5135         } else
5136                 cur = NULL;
5137         extno = 0;
5138         while (bno != (xfs_fileoff_t)-1 && bno >= start && lastx >= 0 &&
5139                (nexts == 0 || extno < nexts)) {
5140                 /*
5141                  * Is the found extent after a hole in which bno lives?
5142                  * Just back up to the previous extent, if so.
5143                  */
5144                 if (got.br_startoff > bno) {
5145                         if (--lastx < 0)
5146                                 break;
5147                         ep = xfs_iext_get_ext(ifp, lastx);
5148                         xfs_bmbt_get_all(ep, &got);
5149                 }
5150                 /*
5151                  * Is the last block of this extent before the range
5152                  * we're supposed to delete?  If so, we're done.
5153                  */
5154                 bno = XFS_FILEOFF_MIN(bno,
5155                         got.br_startoff + got.br_blockcount - 1);
5156                 if (bno < start)
5157                         break;
5158                 /*
5159                  * Then deal with the (possibly delayed) allocated space
5160                  * we found.
5161                  */
5162                 ASSERT(ep != NULL);
5163                 del = got;
5164                 wasdel = isnullstartblock(del.br_startblock);
5165                 if (got.br_startoff < start) {
5166                         del.br_startoff = start;
5167                         del.br_blockcount -= start - got.br_startoff;
5168                         if (!wasdel)
5169                                 del.br_startblock += start - got.br_startoff;
5170                 }
5171                 if (del.br_startoff + del.br_blockcount > bno + 1)
5172                         del.br_blockcount = bno + 1 - del.br_startoff;
5173                 sum = del.br_startblock + del.br_blockcount;
5174                 if (isrt &&
5175                     (mod = do_mod(sum, mp->m_sb.sb_rextsize))) {
5176                         /*
5177                          * Realtime extent not lined up at the end.
5178                          * The extent could have been split into written
5179                          * and unwritten pieces, or we could just be
5180                          * unmapping part of it.  But we can't really
5181                          * get rid of part of a realtime extent.
5182                          */
5183                         if (del.br_state == XFS_EXT_UNWRITTEN ||
5184                             !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5185                                 /*
5186                                  * This piece is unwritten, or we're not
5187                                  * using unwritten extents.  Skip over it.
5188                                  */
5189                                 ASSERT(bno >= mod);
5190                                 bno -= mod > del.br_blockcount ?
5191                                         del.br_blockcount : mod;
5192                                 if (bno < got.br_startoff) {
5193                                         if (--lastx >= 0)
5194                                                 xfs_bmbt_get_all(xfs_iext_get_ext(
5195                                                         ifp, lastx), &got);
5196                                 }
5197                                 continue;
5198                         }
5199                         /*
5200                          * It's written, turn it unwritten.
5201                          * This is better than zeroing it.
5202                          */
5203                         ASSERT(del.br_state == XFS_EXT_NORM);
5204                         ASSERT(xfs_trans_get_block_res(tp) > 0);
5205                         /*
5206                          * If this spans a realtime extent boundary,
5207                          * chop it back to the start of the one we end at.
5208                          */
5209                         if (del.br_blockcount > mod) {
5210                                 del.br_startoff += del.br_blockcount - mod;
5211                                 del.br_startblock += del.br_blockcount - mod;
5212                                 del.br_blockcount = mod;
5213                         }
5214                         del.br_state = XFS_EXT_UNWRITTEN;
5215                         error = xfs_bmap_add_extent_unwritten_real(tp, ip,
5216                                         &lastx, &cur, &del, firstblock, flist,
5217                                         &logflags);
5218                         if (error)
5219                                 goto error0;
5220                         goto nodelete;
5221                 }
5222                 if (isrt && (mod = do_mod(del.br_startblock, mp->m_sb.sb_rextsize))) {
5223                         /*
5224                          * Realtime extent is lined up at the end but not
5225                          * at the front.  We'll get rid of full extents if
5226                          * we can.
5227                          */
5228                         mod = mp->m_sb.sb_rextsize - mod;
5229                         if (del.br_blockcount > mod) {
5230                                 del.br_blockcount -= mod;
5231                                 del.br_startoff += mod;
5232                                 del.br_startblock += mod;
5233                         } else if ((del.br_startoff == start &&
5234                                     (del.br_state == XFS_EXT_UNWRITTEN ||
5235                                      xfs_trans_get_block_res(tp) == 0)) ||
5236                                    !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5237                                 /*
5238                                  * Can't make it unwritten.  There isn't
5239                                  * a full extent here so just skip it.
5240                                  */
5241                                 ASSERT(bno >= del.br_blockcount);
5242                                 bno -= del.br_blockcount;
5243                                 if (got.br_startoff > bno) {
5244                                         if (--lastx >= 0) {
5245                                                 ep = xfs_iext_get_ext(ifp,
5246                                                                       lastx);
5247                                                 xfs_bmbt_get_all(ep, &got);
5248                                         }
5249                                 }
5250                                 continue;
5251                         } else if (del.br_state == XFS_EXT_UNWRITTEN) {
5252                                 /*
5253                                  * This one is already unwritten.
5254                                  * It must have a written left neighbor.
5255                                  * Unwrite the killed part of that one and
5256                                  * try again.
5257                                  */
5258                                 ASSERT(lastx > 0);
5259                                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp,
5260                                                 lastx - 1), &prev);
5261                                 ASSERT(prev.br_state == XFS_EXT_NORM);
5262                                 ASSERT(!isnullstartblock(prev.br_startblock));
5263                                 ASSERT(del.br_startblock ==
5264                                        prev.br_startblock + prev.br_blockcount);
5265                                 if (prev.br_startoff < start) {
5266                                         mod = start - prev.br_startoff;
5267                                         prev.br_blockcount -= mod;
5268                                         prev.br_startblock += mod;
5269                                         prev.br_startoff = start;
5270                                 }
5271                                 prev.br_state = XFS_EXT_UNWRITTEN;
5272                                 lastx--;
5273                                 error = xfs_bmap_add_extent_unwritten_real(tp,
5274                                                 ip, &lastx, &cur, &prev,
5275                                                 firstblock, flist, &logflags);
5276                                 if (error)
5277                                         goto error0;
5278                                 goto nodelete;
5279                         } else {
5280                                 ASSERT(del.br_state == XFS_EXT_NORM);
5281                                 del.br_state = XFS_EXT_UNWRITTEN;
5282                                 error = xfs_bmap_add_extent_unwritten_real(tp,
5283                                                 ip, &lastx, &cur, &del,
5284                                                 firstblock, flist, &logflags);
5285                                 if (error)
5286                                         goto error0;
5287                                 goto nodelete;
5288                         }
5289                 }
5290                 if (wasdel) {
5291                         ASSERT(startblockval(del.br_startblock) > 0);
5292                         /* Update realtime/data freespace, unreserve quota */
5293                         if (isrt) {
5294                                 xfs_filblks_t rtexts;
5295
5296                                 rtexts = XFS_FSB_TO_B(mp, del.br_blockcount);
5297                                 do_div(rtexts, mp->m_sb.sb_rextsize);
5298                                 xfs_mod_incore_sb(mp, XFS_SBS_FREXTENTS,
5299                                                 (int64_t)rtexts, 0);
5300                                 (void)xfs_trans_reserve_quota_nblks(NULL,
5301                                         ip, -((long)del.br_blockcount), 0,
5302                                         XFS_QMOPT_RES_RTBLKS);
5303                         } else {
5304                                 xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
5305                                                 (int64_t)del.br_blockcount, 0);
5306                                 (void)xfs_trans_reserve_quota_nblks(NULL,
5307                                         ip, -((long)del.br_blockcount), 0,
5308                                         XFS_QMOPT_RES_REGBLKS);
5309                         }
5310                         ip->i_delayed_blks -= del.br_blockcount;
5311                         if (cur)
5312                                 cur->bc_private.b.flags |=
5313                                         XFS_BTCUR_BPRV_WASDEL;
5314                 } else if (cur)
5315                         cur->bc_private.b.flags &= ~XFS_BTCUR_BPRV_WASDEL;
5316                 /*
5317                  * If it's the case where the directory code is running
5318                  * with no block reservation, and the deleted block is in
5319                  * the middle of its extent, and the resulting insert
5320                  * of an extent would cause transformation to btree format,
5321                  * then reject it.  The calling code will then swap
5322                  * blocks around instead.
5323                  * We have to do this now, rather than waiting for the
5324                  * conversion to btree format, since the transaction
5325                  * will be dirty.
5326                  */
5327                 if (!wasdel && xfs_trans_get_block_res(tp) == 0 &&
5328                     XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
5329                     XFS_IFORK_NEXTENTS(ip, whichfork) >= ifp->if_ext_max &&
5330                     del.br_startoff > got.br_startoff &&
5331                     del.br_startoff + del.br_blockcount <
5332                     got.br_startoff + got.br_blockcount) {
5333                         error = XFS_ERROR(ENOSPC);
5334                         goto error0;
5335                 }
5336                 error = xfs_bmap_del_extent(ip, tp, &lastx, flist, cur, &del,
5337                                 &tmp_logflags, whichfork);
5338                 logflags |= tmp_logflags;
5339                 if (error)
5340                         goto error0;
5341                 bno = del.br_startoff - 1;
5342 nodelete:
5343                 /*
5344                  * If not done go on to the next (previous) record.
5345                  */
5346                 if (bno != (xfs_fileoff_t)-1 && bno >= start) {
5347                         if (lastx >= 0) {
5348                                 ep = xfs_iext_get_ext(ifp, lastx);
5349                                 if (xfs_bmbt_get_startoff(ep) > bno) {
5350                                         if (--lastx >= 0)
5351                                                 ep = xfs_iext_get_ext(ifp,
5352                                                                       lastx);
5353                                 }
5354                                 xfs_bmbt_get_all(ep, &got);
5355                         }
5356                         extno++;
5357                 }
5358         }
5359         *done = bno == (xfs_fileoff_t)-1 || bno < start || lastx < 0;
5360         ASSERT(ifp->if_ext_max ==
5361                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5362         /*
5363          * Convert to a btree if necessary.
5364          */
5365         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
5366             XFS_IFORK_NEXTENTS(ip, whichfork) > ifp->if_ext_max) {
5367                 ASSERT(cur == NULL);
5368                 error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist,
5369                         &cur, 0, &tmp_logflags, whichfork);
5370                 logflags |= tmp_logflags;
5371                 if (error)
5372                         goto error0;
5373         }
5374         /*
5375          * transform from btree to extents, give it cur
5376          */
5377         else if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE &&
5378                  XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max) {
5379                 ASSERT(cur != NULL);
5380                 error = xfs_bmap_btree_to_extents(tp, ip, cur, &tmp_logflags,
5381                         whichfork);
5382                 logflags |= tmp_logflags;
5383                 if (error)
5384                         goto error0;
5385         }
5386         /*
5387          * transform from extents to local?
5388          */
5389         ASSERT(ifp->if_ext_max ==
5390                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5391         error = 0;
5392 error0:
5393         /*
5394          * Log everything.  Do this after conversion, there's no point in
5395          * logging the extent records if we've converted to btree format.
5396          */
5397         if ((logflags & xfs_ilog_fext(whichfork)) &&
5398             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
5399                 logflags &= ~xfs_ilog_fext(whichfork);
5400         else if ((logflags & xfs_ilog_fbroot(whichfork)) &&
5401                  XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
5402                 logflags &= ~xfs_ilog_fbroot(whichfork);
5403         /*
5404          * Log inode even in the error case, if the transaction
5405          * is dirty we'll need to shut down the filesystem.
5406          */
5407         if (logflags)
5408                 xfs_trans_log_inode(tp, ip, logflags);
5409         if (cur) {
5410                 if (!error) {
5411                         *firstblock = cur->bc_private.b.firstblock;
5412                         cur->bc_private.b.allocated = 0;
5413                 }
5414                 xfs_btree_del_cursor(cur,
5415                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5416         }
5417         return error;
5418 }
5419
5420 /*
5421  * returns 1 for success, 0 if we failed to map the extent.
5422  */
5423 STATIC int
5424 xfs_getbmapx_fix_eof_hole(
5425         xfs_inode_t             *ip,            /* xfs incore inode pointer */
5426         struct getbmapx         *out,           /* output structure */
5427         int                     prealloced,     /* this is a file with
5428                                                  * preallocated data space */
5429         __int64_t               end,            /* last block requested */
5430         xfs_fsblock_t           startblock)
5431 {
5432         __int64_t               fixlen;
5433         xfs_mount_t             *mp;            /* file system mount point */
5434         xfs_ifork_t             *ifp;           /* inode fork pointer */
5435         xfs_extnum_t            lastx;          /* last extent pointer */
5436         xfs_fileoff_t           fileblock;
5437
5438         if (startblock == HOLESTARTBLOCK) {
5439                 mp = ip->i_mount;
5440                 out->bmv_block = -1;
5441                 fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, ip->i_size));
5442                 fixlen -= out->bmv_offset;
5443                 if (prealloced && out->bmv_offset + out->bmv_length == end) {
5444                         /* Came to hole at EOF. Trim it. */
5445                         if (fixlen <= 0)
5446                                 return 0;
5447                         out->bmv_length = fixlen;
5448                 }
5449         } else {
5450                 if (startblock == DELAYSTARTBLOCK)
5451                         out->bmv_block = -2;
5452                 else
5453                         out->bmv_block = xfs_fsb_to_db(ip, startblock);
5454                 fileblock = XFS_BB_TO_FSB(ip->i_mount, out->bmv_offset);
5455                 ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
5456                 if (xfs_iext_bno_to_ext(ifp, fileblock, &lastx) &&
5457                    (lastx == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t))-1))
5458                         out->bmv_oflags |= BMV_OF_LAST;
5459         }
5460
5461         return 1;
5462 }
5463
5464 /*
5465  * Get inode's extents as described in bmv, and format for output.
5466  * Calls formatter to fill the user's buffer until all extents
5467  * are mapped, until the passed-in bmv->bmv_count slots have
5468  * been filled, or until the formatter short-circuits the loop,
5469  * if it is tracking filled-in extents on its own.
5470  */
5471 int                                             /* error code */
5472 xfs_getbmap(
5473         xfs_inode_t             *ip,
5474         struct getbmapx         *bmv,           /* user bmap structure */
5475         xfs_bmap_format_t       formatter,      /* format to user */
5476         void                    *arg)           /* formatter arg */
5477 {
5478         __int64_t               bmvend;         /* last block requested */
5479         int                     error = 0;      /* return value */
5480         __int64_t               fixlen;         /* length for -1 case */
5481         int                     i;              /* extent number */
5482         int                     lock;           /* lock state */
5483         xfs_bmbt_irec_t         *map;           /* buffer for user's data */
5484         xfs_mount_t             *mp;            /* file system mount point */
5485         int                     nex;            /* # of user extents can do */
5486         int                     nexleft;        /* # of user extents left */
5487         int                     subnex;         /* # of bmapi's can do */
5488         int                     nmap;           /* number of map entries */
5489         struct getbmapx         *out;           /* output structure */
5490         int                     whichfork;      /* data or attr fork */
5491         int                     prealloced;     /* this is a file with
5492                                                  * preallocated data space */
5493         int                     iflags;         /* interface flags */
5494         int                     bmapi_flags;    /* flags for xfs_bmapi */
5495         int                     cur_ext = 0;
5496
5497         mp = ip->i_mount;
5498         iflags = bmv->bmv_iflags;
5499         whichfork = iflags & BMV_IF_ATTRFORK ? XFS_ATTR_FORK : XFS_DATA_FORK;
5500
5501         if (whichfork == XFS_ATTR_FORK) {
5502                 if (XFS_IFORK_Q(ip)) {
5503                         if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS &&
5504                             ip->i_d.di_aformat != XFS_DINODE_FMT_BTREE &&
5505                             ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)
5506                                 return XFS_ERROR(EINVAL);
5507                 } else if (unlikely(
5508                            ip->i_d.di_aformat != 0 &&
5509                            ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS)) {
5510                         XFS_ERROR_REPORT("xfs_getbmap", XFS_ERRLEVEL_LOW,
5511                                          ip->i_mount);
5512                         return XFS_ERROR(EFSCORRUPTED);
5513                 }
5514
5515                 prealloced = 0;
5516                 fixlen = 1LL << 32;
5517         } else {
5518                 if (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS &&
5519                     ip->i_d.di_format != XFS_DINODE_FMT_BTREE &&
5520                     ip->i_d.di_format != XFS_DINODE_FMT_LOCAL)
5521                         return XFS_ERROR(EINVAL);
5522
5523                 if (xfs_get_extsz_hint(ip) ||
5524                     ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC|XFS_DIFLAG_APPEND)){
5525                         prealloced = 1;
5526                         fixlen = XFS_MAXIOFFSET(mp);
5527                 } else {
5528                         prealloced = 0;
5529                         fixlen = ip->i_size;
5530                 }
5531         }
5532
5533         if (bmv->bmv_length == -1) {
5534                 fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, fixlen));
5535                 bmv->bmv_length =
5536                         max_t(__int64_t, fixlen - bmv->bmv_offset, 0);
5537         } else if (bmv->bmv_length == 0) {
5538                 bmv->bmv_entries = 0;
5539                 return 0;
5540         } else if (bmv->bmv_length < 0) {
5541                 return XFS_ERROR(EINVAL);
5542         }
5543
5544         nex = bmv->bmv_count - 1;
5545         if (nex <= 0)
5546                 return XFS_ERROR(EINVAL);
5547         bmvend = bmv->bmv_offset + bmv->bmv_length;
5548
5549
5550         if (bmv->bmv_count > ULONG_MAX / sizeof(struct getbmapx))
5551                 return XFS_ERROR(ENOMEM);
5552         out = kmem_zalloc(bmv->bmv_count * sizeof(struct getbmapx), KM_MAYFAIL);
5553         if (!out)
5554                 return XFS_ERROR(ENOMEM);
5555
5556         xfs_ilock(ip, XFS_IOLOCK_SHARED);
5557         if (whichfork == XFS_DATA_FORK && !(iflags & BMV_IF_DELALLOC)) {
5558                 if (ip->i_delayed_blks || ip->i_size > ip->i_d.di_size) {
5559                         error = xfs_flush_pages(ip, 0, -1, 0, FI_REMAPF);
5560                         if (error)
5561                                 goto out_unlock_iolock;
5562                 }
5563                 /*
5564                  * even after flushing the inode, there can still be delalloc
5565                  * blocks on the inode beyond EOF due to speculative
5566                  * preallocation. These are not removed until the release
5567                  * function is called or the inode is inactivated. Hence we
5568                  * cannot assert here that ip->i_delayed_blks == 0.
5569                  */
5570         }
5571
5572         lock = xfs_ilock_map_shared(ip);
5573
5574         /*
5575          * Don't let nex be bigger than the number of extents
5576          * we can have assuming alternating holes and real extents.
5577          */
5578         if (nex > XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1)
5579                 nex = XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1;
5580
5581         bmapi_flags = xfs_bmapi_aflag(whichfork);
5582         if (!(iflags & BMV_IF_PREALLOC))
5583                 bmapi_flags |= XFS_BMAPI_IGSTATE;
5584
5585         /*
5586          * Allocate enough space to handle "subnex" maps at a time.
5587          */
5588         error = ENOMEM;
5589         subnex = 16;
5590         map = kmem_alloc(subnex * sizeof(*map), KM_MAYFAIL | KM_NOFS);
5591         if (!map)
5592                 goto out_unlock_ilock;
5593
5594         bmv->bmv_entries = 0;
5595
5596         if (XFS_IFORK_NEXTENTS(ip, whichfork) == 0 &&
5597             (whichfork == XFS_ATTR_FORK || !(iflags & BMV_IF_DELALLOC))) {
5598                 error = 0;
5599                 goto out_free_map;
5600         }
5601
5602         nexleft = nex;
5603
5604         do {
5605                 nmap = (nexleft > subnex) ? subnex : nexleft;
5606                 error = xfs_bmapi_read(ip, XFS_BB_TO_FSBT(mp, bmv->bmv_offset),
5607                                        XFS_BB_TO_FSB(mp, bmv->bmv_length),
5608                                        map, &nmap, bmapi_flags);
5609                 if (error)
5610                         goto out_free_map;
5611                 ASSERT(nmap <= subnex);
5612
5613                 for (i = 0; i < nmap && nexleft && bmv->bmv_length; i++) {
5614                         out[cur_ext].bmv_oflags = 0;
5615                         if (map[i].br_state == XFS_EXT_UNWRITTEN)
5616                                 out[cur_ext].bmv_oflags |= BMV_OF_PREALLOC;
5617                         else if (map[i].br_startblock == DELAYSTARTBLOCK)
5618                                 out[cur_ext].bmv_oflags |= BMV_OF_DELALLOC;
5619                         out[cur_ext].bmv_offset =
5620                                 XFS_FSB_TO_BB(mp, map[i].br_startoff);
5621                         out[cur_ext].bmv_length =
5622                                 XFS_FSB_TO_BB(mp, map[i].br_blockcount);
5623                         out[cur_ext].bmv_unused1 = 0;
5624                         out[cur_ext].bmv_unused2 = 0;
5625                         ASSERT(((iflags & BMV_IF_DELALLOC) != 0) ||
5626                               (map[i].br_startblock != DELAYSTARTBLOCK));
5627                         if (map[i].br_startblock == HOLESTARTBLOCK &&
5628                             whichfork == XFS_ATTR_FORK) {
5629                                 /* came to the end of attribute fork */
5630                                 out[cur_ext].bmv_oflags |= BMV_OF_LAST;
5631                                 goto out_free_map;
5632                         }
5633
5634                         if (!xfs_getbmapx_fix_eof_hole(ip, &out[cur_ext],
5635                                         prealloced, bmvend,
5636                                         map[i].br_startblock))
5637                                 goto out_free_map;
5638
5639                         bmv->bmv_offset =
5640                                 out[cur_ext].bmv_offset +
5641                                 out[cur_ext].bmv_length;
5642                         bmv->bmv_length =
5643                                 max_t(__int64_t, 0, bmvend - bmv->bmv_offset);
5644
5645                         /*
5646                          * In case we don't want to return the hole,
5647                          * don't increase cur_ext so that we can reuse
5648                          * it in the next loop.
5649                          */
5650                         if ((iflags & BMV_IF_NO_HOLES) &&
5651                             map[i].br_startblock == HOLESTARTBLOCK) {
5652                                 memset(&out[cur_ext], 0, sizeof(out[cur_ext]));
5653                                 continue;
5654                         }
5655
5656                         nexleft--;
5657                         bmv->bmv_entries++;
5658                         cur_ext++;
5659                 }
5660         } while (nmap && nexleft && bmv->bmv_length);
5661
5662  out_free_map:
5663         kmem_free(map);
5664  out_unlock_ilock:
5665         xfs_iunlock_map_shared(ip, lock);
5666  out_unlock_iolock:
5667         xfs_iunlock(ip, XFS_IOLOCK_SHARED);
5668
5669         for (i = 0; i < cur_ext; i++) {
5670                 int full = 0;   /* user array is full */
5671
5672                 /* format results & advance arg */
5673                 error = formatter(&arg, &out[i], &full);
5674                 if (error || full)
5675                         break;
5676         }
5677
5678         kmem_free(out);
5679         return error;
5680 }
5681
5682 #ifdef DEBUG
5683 STATIC struct xfs_buf *
5684 xfs_bmap_get_bp(
5685         struct xfs_btree_cur    *cur,
5686         xfs_fsblock_t           bno)
5687 {
5688         struct xfs_log_item_desc *lidp;
5689         int                     i;
5690
5691         if (!cur)
5692                 return NULL;
5693
5694         for (i = 0; i < XFS_BTREE_MAXLEVELS; i++) {
5695                 if (!cur->bc_bufs[i])
5696                         break;
5697                 if (XFS_BUF_ADDR(cur->bc_bufs[i]) == bno)
5698                         return cur->bc_bufs[i];
5699         }
5700
5701         /* Chase down all the log items to see if the bp is there */
5702         list_for_each_entry(lidp, &cur->bc_tp->t_items, lid_trans) {
5703                 struct xfs_buf_log_item *bip;
5704                 bip = (struct xfs_buf_log_item *)lidp->lid_item;
5705                 if (bip->bli_item.li_type == XFS_LI_BUF &&
5706                     XFS_BUF_ADDR(bip->bli_buf) == bno)
5707                         return bip->bli_buf;
5708         }
5709
5710         return NULL;
5711 }
5712
5713 STATIC void
5714 xfs_check_block(
5715         struct xfs_btree_block  *block,
5716         xfs_mount_t             *mp,
5717         int                     root,
5718         short                   sz)
5719 {
5720         int                     i, j, dmxr;
5721         __be64                  *pp, *thispa;   /* pointer to block address */
5722         xfs_bmbt_key_t          *prevp, *keyp;
5723
5724         ASSERT(be16_to_cpu(block->bb_level) > 0);
5725
5726         prevp = NULL;
5727         for( i = 1; i <= xfs_btree_get_numrecs(block); i++) {
5728                 dmxr = mp->m_bmap_dmxr[0];
5729                 keyp = XFS_BMBT_KEY_ADDR(mp, block, i);
5730
5731                 if (prevp) {
5732                         ASSERT(be64_to_cpu(prevp->br_startoff) <
5733                                be64_to_cpu(keyp->br_startoff));
5734                 }
5735                 prevp = keyp;
5736
5737                 /*
5738                  * Compare the block numbers to see if there are dups.
5739                  */
5740                 if (root)
5741                         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, i, sz);
5742                 else
5743                         pp = XFS_BMBT_PTR_ADDR(mp, block, i, dmxr);
5744
5745                 for (j = i+1; j <= be16_to_cpu(block->bb_numrecs); j++) {
5746                         if (root)
5747                                 thispa = XFS_BMAP_BROOT_PTR_ADDR(mp, block, j, sz);
5748                         else
5749                                 thispa = XFS_BMBT_PTR_ADDR(mp, block, j, dmxr);
5750                         if (*thispa == *pp) {
5751                                 xfs_warn(mp, "%s: thispa(%d) == pp(%d) %Ld",
5752                                         __func__, j, i,
5753                                         (unsigned long long)be64_to_cpu(*thispa));
5754                                 panic("%s: ptrs are equal in node\n",
5755                                         __func__);
5756                         }
5757                 }
5758         }
5759 }
5760
5761 /*
5762  * Check that the extents for the inode ip are in the right order in all
5763  * btree leaves.
5764  */
5765
5766 STATIC void
5767 xfs_bmap_check_leaf_extents(
5768         xfs_btree_cur_t         *cur,   /* btree cursor or null */
5769         xfs_inode_t             *ip,            /* incore inode pointer */
5770         int                     whichfork)      /* data or attr fork */
5771 {
5772         struct xfs_btree_block  *block; /* current btree block */
5773         xfs_fsblock_t           bno;    /* block # of "block" */
5774         xfs_buf_t               *bp;    /* buffer for "block" */
5775         int                     error;  /* error return value */
5776         xfs_extnum_t            i=0, j; /* index into the extents list */
5777         xfs_ifork_t             *ifp;   /* fork structure */
5778         int                     level;  /* btree level, for checking */
5779         xfs_mount_t             *mp;    /* file system mount structure */
5780         __be64                  *pp;    /* pointer to block address */
5781         xfs_bmbt_rec_t          *ep;    /* pointer to current extent */
5782         xfs_bmbt_rec_t          last = {0, 0}; /* last extent in prev block */
5783         xfs_bmbt_rec_t          *nextp; /* pointer to next extent */
5784         int                     bp_release = 0;
5785
5786         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE) {
5787                 return;
5788         }
5789
5790         bno = NULLFSBLOCK;
5791         mp = ip->i_mount;
5792         ifp = XFS_IFORK_PTR(ip, whichfork);
5793         block = ifp->if_broot;
5794         /*
5795          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
5796          */
5797         level = be16_to_cpu(block->bb_level);
5798         ASSERT(level > 0);
5799         xfs_check_block(block, mp, 1, ifp->if_broot_bytes);
5800         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
5801         bno = be64_to_cpu(*pp);
5802
5803         ASSERT(bno != NULLDFSBNO);
5804         ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
5805         ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
5806
5807         /*
5808          * Go down the tree until leaf level is reached, following the first
5809          * pointer (leftmost) at each level.
5810          */
5811         while (level-- > 0) {
5812                 /* See if buf is in cur first */
5813                 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
5814                 if (bp) {
5815                         bp_release = 0;
5816                 } else {
5817                         bp_release = 1;
5818                 }
5819                 if (!bp && (error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
5820                                 XFS_BMAP_BTREE_REF)))
5821                         goto error_norelse;
5822                 block = XFS_BUF_TO_BLOCK(bp);
5823                 XFS_WANT_CORRUPTED_GOTO(
5824                         xfs_bmap_sanity_check(mp, bp, level),
5825                         error0);
5826                 if (level == 0)
5827                         break;
5828
5829                 /*
5830                  * Check this block for basic sanity (increasing keys and
5831                  * no duplicate blocks).
5832                  */
5833
5834                 xfs_check_block(block, mp, 0, 0);
5835                 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
5836                 bno = be64_to_cpu(*pp);
5837                 XFS_WANT_CORRUPTED_GOTO(XFS_FSB_SANITY_CHECK(mp, bno), error0);
5838                 if (bp_release) {
5839                         bp_release = 0;
5840                         xfs_trans_brelse(NULL, bp);
5841                 }
5842         }
5843
5844         /*
5845          * Here with bp and block set to the leftmost leaf node in the tree.
5846          */
5847         i = 0;
5848
5849         /*
5850          * Loop over all leaf nodes checking that all extents are in the right order.
5851          */
5852         for (;;) {
5853                 xfs_fsblock_t   nextbno;
5854                 xfs_extnum_t    num_recs;
5855
5856
5857                 num_recs = xfs_btree_get_numrecs(block);
5858
5859                 /*
5860                  * Read-ahead the next leaf block, if any.
5861                  */
5862
5863                 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
5864
5865                 /*
5866                  * Check all the extents to make sure they are OK.
5867                  * If we had a previous block, the last entry should
5868                  * conform with the first entry in this one.
5869                  */
5870
5871                 ep = XFS_BMBT_REC_ADDR(mp, block, 1);
5872                 if (i) {
5873                         ASSERT(xfs_bmbt_disk_get_startoff(&last) +
5874                                xfs_bmbt_disk_get_blockcount(&last) <=
5875                                xfs_bmbt_disk_get_startoff(ep));
5876                 }
5877                 for (j = 1; j < num_recs; j++) {
5878                         nextp = XFS_BMBT_REC_ADDR(mp, block, j + 1);
5879                         ASSERT(xfs_bmbt_disk_get_startoff(ep) +
5880                                xfs_bmbt_disk_get_blockcount(ep) <=
5881                                xfs_bmbt_disk_get_startoff(nextp));
5882                         ep = nextp;
5883                 }
5884
5885                 last = *ep;
5886                 i += num_recs;
5887                 if (bp_release) {
5888                         bp_release = 0;
5889                         xfs_trans_brelse(NULL, bp);
5890                 }
5891                 bno = nextbno;
5892                 /*
5893                  * If we've reached the end, stop.
5894                  */
5895                 if (bno == NULLFSBLOCK)
5896                         break;
5897
5898                 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
5899                 if (bp) {
5900                         bp_release = 0;
5901                 } else {
5902                         bp_release = 1;
5903                 }
5904                 if (!bp && (error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
5905                                 XFS_BMAP_BTREE_REF)))
5906                         goto error_norelse;
5907                 block = XFS_BUF_TO_BLOCK(bp);
5908         }
5909         if (bp_release) {
5910                 bp_release = 0;
5911                 xfs_trans_brelse(NULL, bp);
5912         }
5913         return;
5914
5915 error0:
5916         xfs_warn(mp, "%s: at error0", __func__);
5917         if (bp_release)
5918                 xfs_trans_brelse(NULL, bp);
5919 error_norelse:
5920         xfs_warn(mp, "%s: BAD after btree leaves for %d extents",
5921                 __func__, i);
5922         panic("%s: CORRUPTED BTREE OR SOMETHING", __func__);
5923         return;
5924 }
5925 #endif
5926
5927 /*
5928  * Count fsblocks of the given fork.
5929  */
5930 int                                             /* error */
5931 xfs_bmap_count_blocks(
5932         xfs_trans_t             *tp,            /* transaction pointer */
5933         xfs_inode_t             *ip,            /* incore inode */
5934         int                     whichfork,      /* data or attr fork */
5935         int                     *count)         /* out: count of blocks */
5936 {
5937         struct xfs_btree_block  *block; /* current btree block */
5938         xfs_fsblock_t           bno;    /* block # of "block" */
5939         xfs_ifork_t             *ifp;   /* fork structure */
5940         int                     level;  /* btree level, for checking */
5941         xfs_mount_t             *mp;    /* file system mount structure */
5942         __be64                  *pp;    /* pointer to block address */
5943
5944         bno = NULLFSBLOCK;
5945         mp = ip->i_mount;
5946         ifp = XFS_IFORK_PTR(ip, whichfork);
5947         if ( XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ) {
5948                 xfs_bmap_count_leaves(ifp, 0,
5949                         ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t),
5950                         count);
5951                 return 0;
5952         }
5953
5954         /*
5955          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
5956          */
5957         block = ifp->if_broot;
5958         level = be16_to_cpu(block->bb_level);
5959         ASSERT(level > 0);
5960         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
5961         bno = be64_to_cpu(*pp);
5962         ASSERT(bno != NULLDFSBNO);
5963         ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
5964         ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
5965
5966         if (unlikely(xfs_bmap_count_tree(mp, tp, ifp, bno, level, count) < 0)) {
5967                 XFS_ERROR_REPORT("xfs_bmap_count_blocks(2)", XFS_ERRLEVEL_LOW,
5968                                  mp);
5969                 return XFS_ERROR(EFSCORRUPTED);
5970         }
5971
5972         return 0;
5973 }
5974
5975 /*
5976  * Recursively walks each level of a btree
5977  * to count total fsblocks is use.
5978  */
5979 STATIC int                                     /* error */
5980 xfs_bmap_count_tree(
5981         xfs_mount_t     *mp,            /* file system mount point */
5982         xfs_trans_t     *tp,            /* transaction pointer */
5983         xfs_ifork_t     *ifp,           /* inode fork pointer */
5984         xfs_fsblock_t   blockno,        /* file system block number */
5985         int             levelin,        /* level in btree */
5986         int             *count)         /* Count of blocks */
5987 {
5988         int                     error;
5989         xfs_buf_t               *bp, *nbp;
5990         int                     level = levelin;
5991         __be64                  *pp;
5992         xfs_fsblock_t           bno = blockno;
5993         xfs_fsblock_t           nextbno;
5994         struct xfs_btree_block  *block, *nextblock;
5995         int                     numrecs;
5996
5997         if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp, XFS_BMAP_BTREE_REF)))
5998                 return error;
5999         *count += 1;
6000         block = XFS_BUF_TO_BLOCK(bp);
6001
6002         if (--level) {
6003                 /* Not at node above leaves, count this level of nodes */
6004                 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
6005                 while (nextbno != NULLFSBLOCK) {
6006                         if ((error = xfs_btree_read_bufl(mp, tp, nextbno,
6007                                 0, &nbp, XFS_BMAP_BTREE_REF)))
6008                                 return error;
6009                         *count += 1;
6010                         nextblock = XFS_BUF_TO_BLOCK(nbp);
6011                         nextbno = be64_to_cpu(nextblock->bb_u.l.bb_rightsib);
6012                         xfs_trans_brelse(tp, nbp);
6013                 }
6014
6015                 /* Dive to the next level */
6016                 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
6017                 bno = be64_to_cpu(*pp);
6018                 if (unlikely((error =
6019                      xfs_bmap_count_tree(mp, tp, ifp, bno, level, count)) < 0)) {
6020                         xfs_trans_brelse(tp, bp);
6021                         XFS_ERROR_REPORT("xfs_bmap_count_tree(1)",
6022                                          XFS_ERRLEVEL_LOW, mp);
6023                         return XFS_ERROR(EFSCORRUPTED);
6024                 }
6025                 xfs_trans_brelse(tp, bp);
6026         } else {
6027                 /* count all level 1 nodes and their leaves */
6028                 for (;;) {
6029                         nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
6030                         numrecs = be16_to_cpu(block->bb_numrecs);
6031                         xfs_bmap_disk_count_leaves(mp, block, numrecs, count);
6032                         xfs_trans_brelse(tp, bp);
6033                         if (nextbno == NULLFSBLOCK)
6034                                 break;
6035                         bno = nextbno;
6036                         if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
6037                                 XFS_BMAP_BTREE_REF)))
6038                                 return error;
6039                         *count += 1;
6040                         block = XFS_BUF_TO_BLOCK(bp);
6041                 }
6042         }
6043         return 0;
6044 }
6045
6046 /*
6047  * Count leaf blocks given a range of extent records.
6048  */
6049 STATIC void
6050 xfs_bmap_count_leaves(
6051         xfs_ifork_t             *ifp,
6052         xfs_extnum_t            idx,
6053         int                     numrecs,
6054         int                     *count)
6055 {
6056         int             b;
6057
6058         for (b = 0; b < numrecs; b++) {
6059                 xfs_bmbt_rec_host_t *frp = xfs_iext_get_ext(ifp, idx + b);
6060                 *count += xfs_bmbt_get_blockcount(frp);
6061         }
6062 }
6063
6064 /*
6065  * Count leaf blocks given a range of extent records originally
6066  * in btree format.
6067  */
6068 STATIC void
6069 xfs_bmap_disk_count_leaves(
6070         struct xfs_mount        *mp,
6071         struct xfs_btree_block  *block,
6072         int                     numrecs,
6073         int                     *count)
6074 {
6075         int             b;
6076         xfs_bmbt_rec_t  *frp;
6077
6078         for (b = 1; b <= numrecs; b++) {
6079                 frp = XFS_BMBT_REC_ADDR(mp, block, b);
6080                 *count += xfs_bmbt_disk_get_blockcount(frp);
6081         }
6082 }
6083
6084 /*
6085  * dead simple method of punching delalyed allocation blocks from a range in
6086  * the inode. Walks a block at a time so will be slow, but is only executed in
6087  * rare error cases so the overhead is not critical. This will alays punch out
6088  * both the start and end blocks, even if the ranges only partially overlap
6089  * them, so it is up to the caller to ensure that partial blocks are not
6090  * passed in.
6091  */
6092 int
6093 xfs_bmap_punch_delalloc_range(
6094         struct xfs_inode        *ip,
6095         xfs_fileoff_t           start_fsb,
6096         xfs_fileoff_t           length)
6097 {
6098         xfs_fileoff_t           remaining = length;
6099         int                     error = 0;
6100
6101         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
6102
6103         do {
6104                 int             done;
6105                 xfs_bmbt_irec_t imap;
6106                 int             nimaps = 1;
6107                 xfs_fsblock_t   firstblock;
6108                 xfs_bmap_free_t flist;
6109
6110                 /*
6111                  * Map the range first and check that it is a delalloc extent
6112                  * before trying to unmap the range. Otherwise we will be
6113                  * trying to remove a real extent (which requires a
6114                  * transaction) or a hole, which is probably a bad idea...
6115                  */
6116                 error = xfs_bmapi_read(ip, start_fsb, 1, &imap, &nimaps,
6117                                        XFS_BMAPI_ENTIRE);
6118
6119                 if (error) {
6120                         /* something screwed, just bail */
6121                         if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
6122                                 xfs_alert(ip->i_mount,
6123                         "Failed delalloc mapping lookup ino %lld fsb %lld.",
6124                                                 ip->i_ino, start_fsb);
6125                         }
6126                         break;
6127                 }
6128                 if (!nimaps) {
6129                         /* nothing there */
6130                         goto next_block;
6131                 }
6132                 if (imap.br_startblock != DELAYSTARTBLOCK) {
6133                         /* been converted, ignore */
6134                         goto next_block;
6135                 }
6136                 WARN_ON(imap.br_blockcount == 0);
6137
6138                 /*
6139                  * Note: while we initialise the firstblock/flist pair, they
6140                  * should never be used because blocks should never be
6141                  * allocated or freed for a delalloc extent and hence we need
6142                  * don't cancel or finish them after the xfs_bunmapi() call.
6143                  */
6144                 xfs_bmap_init(&flist, &firstblock);
6145                 error = xfs_bunmapi(NULL, ip, start_fsb, 1, 0, 1, &firstblock,
6146                                         &flist, &done);
6147                 if (error)
6148                         break;
6149
6150                 ASSERT(!flist.xbf_count && !flist.xbf_first);
6151 next_block:
6152                 start_fsb++;
6153                 remaining--;
6154         } while(remaining > 0);
6155
6156         return error;
6157 }