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_bmalloca     *bma)
 379 {
 380         struct xfs_bmbt_irec    *new = &bma->got;
 381         int                     diff;   /* temp value */
 382         xfs_bmbt_rec_host_t     *ep;    /* extent entry for idx */
 383         int                     error;  /* error return value */
 384         int                     i;      /* temp state */
 385         xfs_ifork_t             *ifp;   /* inode fork pointer */
 386         xfs_fileoff_t           new_endoff;     /* end offset of new entry */
 387         xfs_bmbt_irec_t         r[3];   /* neighbor extent entries */
 388                                         /* left is 0, right is 1, prev is 2 */
 389         int                     rval=0; /* return value (logging flags) */
 390         int                     state = 0;/* state bits, accessed thru macros */
 391         xfs_filblks_t           da_new; /* new count del alloc blocks used */
 392         xfs_filblks_t           da_old; /* old count del alloc blocks used */
 393         xfs_filblks_t           temp=0; /* value for da_new calculations */
 394         xfs_filblks_t           temp2=0;/* value for da_new calculations */
 395         int                     tmp_rval;       /* partial logging flags */
 396
 397         ifp = XFS_IFORK_PTR(bma->ip, XFS_DATA_FORK);
 398
 399         ASSERT(bma->idx >= 0);
 400         ASSERT(bma->idx <= ifp->if_bytes / sizeof(struct xfs_bmbt_rec));
 401         ASSERT(!isnullstartblock(new->br_startblock));
 402         ASSERT(!bma->cur ||
 403                (bma->cur->bc_private.b.flags & XFS_BTCUR_BPRV_WASDEL));
 404
 405         XFS_STATS_INC(xs_add_exlist);
 406
 407 #define LEFT            r[0]
 408 #define RIGHT           r[1]
 409 #define PREV            r[2]
 410
 411         /*
 412          * Set up a bunch of variables to make the tests simpler.
 413          */
 414         ep = xfs_iext_get_ext(ifp, bma->idx);
 415         xfs_bmbt_get_all(ep, &PREV);
 416         new_endoff = new->br_startoff + new->br_blockcount;
 417         ASSERT(PREV.br_startoff <= new->br_startoff);
 418         ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
 419
 420         da_old = startblockval(PREV.br_startblock);
 421         da_new = 0;
 422
 423         /*
 424          * Set flags determining what part of the previous delayed allocation
 425          * extent is being replaced by a real allocation.
 426          */
 427         if (PREV.br_startoff == new->br_startoff)
 428                 state |= BMAP_LEFT_FILLING;
 429         if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
 430                 state |= BMAP_RIGHT_FILLING;
 431
 432         /*
 433          * Check and set flags if this segment has a left neighbor.
 434          * Don't set contiguous if the combined extent would be too large.
 435          */
 436         if (bma->idx > 0) {
 437                 state |= BMAP_LEFT_VALID;
 438                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx - 1), &LEFT);
 439
 440                 if (isnullstartblock(LEFT.br_startblock))
 441                         state |= BMAP_LEFT_DELAY;
 442         }
 443
 444         if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
 445             LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
 446             LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
 447             LEFT.br_state == new->br_state &&
 448             LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
 449                 state |= BMAP_LEFT_CONTIG;
 450
 451         /*
 452          * Check and set flags if this segment has a right neighbor.
 453          * Don't set contiguous if the combined extent would be too large.
 454          * Also check for all-three-contiguous being too large.
 455          */
 456         if (bma->idx < bma->ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1) {
 457                 state |= BMAP_RIGHT_VALID;
 458                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx + 1), &RIGHT);
 459
 460                 if (isnullstartblock(RIGHT.br_startblock))
 461                         state |= BMAP_RIGHT_DELAY;
 462         }
 463
 464         if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
 465             new_endoff == RIGHT.br_startoff &&
 466             new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
 467             new->br_state == RIGHT.br_state &&
 468             new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
 469             ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
 470                        BMAP_RIGHT_FILLING)) !=
 471                       (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
 472                        BMAP_RIGHT_FILLING) ||
 473              LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
 474                         <= MAXEXTLEN))
 475                 state |= BMAP_RIGHT_CONTIG;
 476
 477         error = 0;
 478         /*
 479          * Switch out based on the FILLING and CONTIG state bits.
 480          */
 481         switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
 482                          BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
 483         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
 484              BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
 485                 /*
 486                  * Filling in all of a previously delayed allocation extent.
 487                  * The left and right neighbors are both contiguous with new.
 488                  */
 489                 bma->idx--;
 490                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
 491                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
 492                         LEFT.br_blockcount + PREV.br_blockcount +
 493                         RIGHT.br_blockcount);
 494                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
 495
 496                 xfs_iext_remove(bma->ip, bma->idx + 1, 2, state);
 497                 bma->ip->i_d.di_nextents--;
 498                 if (bma->cur == NULL)
 499                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
 500                 else {
 501                         rval = XFS_ILOG_CORE;
 502                         error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff,
 503                                         RIGHT.br_startblock,
 504                                         RIGHT.br_blockcount, &i);
 505                         if (error)
 506                                 goto done;
 507                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 508                         error = xfs_btree_delete(bma->cur, &i);
 509                         if (error)
 510                                 goto done;
 511                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 512                         error = xfs_btree_decrement(bma->cur, 0, &i);
 513                         if (error)
 514                                 goto done;
 515                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 516                         error = xfs_bmbt_update(bma->cur, LEFT.br_startoff,
 517                                         LEFT.br_startblock,
 518                                         LEFT.br_blockcount +
 519                                         PREV.br_blockcount +
 520                                         RIGHT.br_blockcount, LEFT.br_state);
 521                         if (error)
 522                                 goto done;
 523                 }
 524                 break;
 525
 526         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
 527                 /*
 528                  * Filling in all of a previously delayed allocation extent.
 529                  * The left neighbor is contiguous, the right is not.
 530                  */
 531                 bma->idx--;
 532
 533                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
 534                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
 535                         LEFT.br_blockcount + PREV.br_blockcount);
 536                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
 537
 538                 xfs_iext_remove(bma->ip, bma->idx + 1, 1, state);
 539                 if (bma->cur == NULL)
 540                         rval = XFS_ILOG_DEXT;
 541                 else {
 542                         rval = 0;
 543                         error = xfs_bmbt_lookup_eq(bma->cur, LEFT.br_startoff,
 544                                         LEFT.br_startblock, LEFT.br_blockcount,
 545                                         &i);
 546                         if (error)
 547                                 goto done;
 548                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 549                         error = xfs_bmbt_update(bma->cur, LEFT.br_startoff,
 550                                         LEFT.br_startblock,
 551                                         LEFT.br_blockcount +
 552                                         PREV.br_blockcount, LEFT.br_state);
 553                         if (error)
 554                                 goto done;
 555                 }
 556                 break;
 557
 558         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
 559                 /*
 560                  * Filling in all of a previously delayed allocation extent.
 561                  * The right neighbor is contiguous, the left is not.
 562                  */
 563                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
 564                 xfs_bmbt_set_startblock(ep, new->br_startblock);
 565                 xfs_bmbt_set_blockcount(ep,
 566                         PREV.br_blockcount + RIGHT.br_blockcount);
 567                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
 568
 569                 xfs_iext_remove(bma->ip, bma->idx + 1, 1, state);
 570                 if (bma->cur == NULL)
 571                         rval = XFS_ILOG_DEXT;
 572                 else {
 573                         rval = 0;
 574                         error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff,
 575                                         RIGHT.br_startblock,
 576                                         RIGHT.br_blockcount, &i);
 577                         if (error)
 578                                 goto done;
 579                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 580                         error = xfs_bmbt_update(bma->cur, PREV.br_startoff,
 581                                         new->br_startblock,
 582                                         PREV.br_blockcount +
 583                                         RIGHT.br_blockcount, PREV.br_state);
 584                         if (error)
 585                                 goto done;
 586                 }
 587                 break;
 588
 589         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
 590                 /*
 591                  * Filling in all of a previously delayed allocation extent.
 592                  * Neither the left nor right neighbors are contiguous with
 593                  * the new one.
 594                  */
 595                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
 596                 xfs_bmbt_set_startblock(ep, new->br_startblock);
 597                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
 598
 599                 bma->ip->i_d.di_nextents++;
 600                 if (bma->cur == NULL)
 601                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
 602                 else {
 603                         rval = XFS_ILOG_CORE;
 604                         error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
 605                                         new->br_startblock, new->br_blockcount,
 606                                         &i);
 607                         if (error)
 608                                 goto done;
 609                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
 610                         bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
 611                         error = xfs_btree_insert(bma->cur, &i);
 612                         if (error)
 613                                 goto done;
 614                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 615                 }
 616                 break;
 617
 618         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
 619                 /*
 620                  * Filling in the first part of a previous delayed allocation.
 621                  * The left neighbor is contiguous.
 622                  */
 623                 trace_xfs_bmap_pre_update(bma->ip, bma->idx - 1, state, _THIS_IP_);
 624                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx - 1),
 625                         LEFT.br_blockcount + new->br_blockcount);
 626                 xfs_bmbt_set_startoff(ep,
 627                         PREV.br_startoff + new->br_blockcount);
 628                 trace_xfs_bmap_post_update(bma->ip, bma->idx - 1, state, _THIS_IP_);
 629
 630                 temp = PREV.br_blockcount - new->br_blockcount;
 631                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
 632                 xfs_bmbt_set_blockcount(ep, temp);
 633                 if (bma->cur == NULL)
 634                         rval = XFS_ILOG_DEXT;
 635                 else {
 636                         rval = 0;
 637                         error = xfs_bmbt_lookup_eq(bma->cur, LEFT.br_startoff,
 638                                         LEFT.br_startblock, LEFT.br_blockcount,
 639                                         &i);
 640                         if (error)
 641                                 goto done;
 642                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 643                         error = xfs_bmbt_update(bma->cur, LEFT.br_startoff,
 644                                         LEFT.br_startblock,
 645                                         LEFT.br_blockcount +
 646                                         new->br_blockcount,
 647                                         LEFT.br_state);
 648                         if (error)
 649                                 goto done;
 650                 }
 651                 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
 652                         startblockval(PREV.br_startblock));
 653                 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
 654                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
 655
 656                 bma->idx--;
 657                 break;
 658
 659         case BMAP_LEFT_FILLING:
 660                 /*
 661                  * Filling in the first part of a previous delayed allocation.
 662                  * The left neighbor is not contiguous.
 663                  */
 664                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
 665                 xfs_bmbt_set_startoff(ep, new_endoff);
 666                 temp = PREV.br_blockcount - new->br_blockcount;
 667                 xfs_bmbt_set_blockcount(ep, temp);
 668                 xfs_iext_insert(bma->ip, bma->idx, 1, new, state);
 669                 bma->ip->i_d.di_nextents++;
 670                 if (bma->cur == NULL)
 671                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
 672                 else {
 673                         rval = XFS_ILOG_CORE;
 674                         error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
 675                                         new->br_startblock, new->br_blockcount,
 676                                         &i);
 677                         if (error)
 678                                 goto done;
 679                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
 680                         bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
 681                         error = xfs_btree_insert(bma->cur, &i);
 682                         if (error)
 683                                 goto done;
 684                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 685                 }
 686                 if (bma->ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
 687                     bma->ip->i_d.di_nextents > bma->ip->i_df.if_ext_max) {
 688                         error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
 689                                         bma->firstblock, bma->flist,
 690                                         &bma->cur, 1, &tmp_rval, XFS_DATA_FORK);
 691                         rval |= tmp_rval;
 692                         if (error)
 693                                 goto done;
 694                 }
 695                 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
 696                         startblockval(PREV.br_startblock) -
 697                         (bma->cur ? bma->cur->bc_private.b.allocated : 0));
 698                 ep = xfs_iext_get_ext(ifp, bma->idx + 1);
 699                 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
 700                 trace_xfs_bmap_post_update(bma->ip, bma->idx + 1, state, _THIS_IP_);
 701                 break;
 702
 703         case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
 704                 /*
 705                  * Filling in the last part of a previous delayed allocation.
 706                  * The right neighbor is contiguous with the new allocation.
 707                  */
 708                 temp = PREV.br_blockcount - new->br_blockcount;
 709                 trace_xfs_bmap_pre_update(bma->ip, bma->idx + 1, state, _THIS_IP_);
 710                 xfs_bmbt_set_blockcount(ep, temp);
 711                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, bma->idx + 1),
 712                         new->br_startoff, new->br_startblock,
 713                         new->br_blockcount + RIGHT.br_blockcount,
 714                         RIGHT.br_state);
 715                 trace_xfs_bmap_post_update(bma->ip, bma->idx + 1, state, _THIS_IP_);
 716                 if (bma->cur == NULL)
 717                         rval = XFS_ILOG_DEXT;
 718                 else {
 719                         rval = 0;
 720                         error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff,
 721                                         RIGHT.br_startblock,
 722                                         RIGHT.br_blockcount, &i);
 723                         if (error)
 724                                 goto done;
 725                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 726                         error = xfs_bmbt_update(bma->cur, new->br_startoff,
 727                                         new->br_startblock,
 728                                         new->br_blockcount +
 729                                         RIGHT.br_blockcount,
 730                                         RIGHT.br_state);
 731                         if (error)
 732                                 goto done;
 733                 }
 734
 735                 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
 736                         startblockval(PREV.br_startblock));
 737                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
 738                 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
 739                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
 740
 741                 bma->idx++;
 742                 break;
 743
 744         case BMAP_RIGHT_FILLING:
 745                 /*
 746                  * Filling in the last part of a previous delayed allocation.
 747                  * The right neighbor is not contiguous.
 748                  */
 749                 temp = PREV.br_blockcount - new->br_blockcount;
 750                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
 751                 xfs_bmbt_set_blockcount(ep, temp);
 752                 xfs_iext_insert(bma->ip, bma->idx + 1, 1, new, state);
 753                 bma->ip->i_d.di_nextents++;
 754                 if (bma->cur == NULL)
 755                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
 756                 else {
 757                         rval = XFS_ILOG_CORE;
 758                         error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
 759                                         new->br_startblock, new->br_blockcount,
 760                                         &i);
 761                         if (error)
 762                                 goto done;
 763                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
 764                         bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
 765                         error = xfs_btree_insert(bma->cur, &i);
 766                         if (error)
 767                                 goto done;
 768                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 769                 }
 770                 if (bma->ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
 771                     bma->ip->i_d.di_nextents > bma->ip->i_df.if_ext_max) {
 772                         error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
 773                                 bma->firstblock, bma->flist, &bma->cur, 1,
 774                                 &tmp_rval, XFS_DATA_FORK);
 775                         rval |= tmp_rval;
 776                         if (error)
 777                                 goto done;
 778                 }
 779                 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
 780                         startblockval(PREV.br_startblock) -
 781                         (bma->cur ? bma->cur->bc_private.b.allocated : 0));
 782                 ep = xfs_iext_get_ext(ifp, bma->idx);
 783                 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
 784                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
 785
 786                 bma->idx++;
 787                 break;
 788
 789         case 0:
 790                 /*
 791                  * Filling in the middle part of a previous delayed allocation.
 792                  * Contiguity is impossible here.
 793                  * This case is avoided almost all the time.
 794                  *
 795                  * We start with a delayed allocation:
 796                  *
 797                  * +ddddddddddddddddddddddddddddddddddddddddddddddddddddddd+
 798                  *  PREV @ idx
 799                  *
 800                  * and we are allocating:
 801                  *                     +rrrrrrrrrrrrrrrrr+
 802                  *                            new
 803                  *
 804                  * and we set it up for insertion as:
 805                  * +ddddddddddddddddddd+rrrrrrrrrrrrrrrrr+ddddddddddddddddd+
 806                  *                            new
 807                  *  PREV @ idx          LEFT              RIGHT
 808                  *                      inserted at idx + 1
 809                  */
 810                 temp = new->br_startoff - PREV.br_startoff;
 811                 temp2 = PREV.br_startoff + PREV.br_blockcount - new_endoff;
 812                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, 0, _THIS_IP_);
 813                 xfs_bmbt_set_blockcount(ep, temp);      /* truncate PREV */
 814                 LEFT = *new;
 815                 RIGHT.br_state = PREV.br_state;
 816                 RIGHT.br_startblock = nullstartblock(
 817                                 (int)xfs_bmap_worst_indlen(bma->ip, temp2));
 818                 RIGHT.br_startoff = new_endoff;
 819                 RIGHT.br_blockcount = temp2;
 820                 /* insert LEFT (r[0]) and RIGHT (r[1]) at the same time */
 821                 xfs_iext_insert(bma->ip, bma->idx + 1, 2, &LEFT, state);
 822                 bma->ip->i_d.di_nextents++;
 823                 if (bma->cur == NULL)
 824                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
 825                 else {
 826                         rval = XFS_ILOG_CORE;
 827                         error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
 828                                         new->br_startblock, new->br_blockcount,
 829                                         &i);
 830                         if (error)
 831                                 goto done;
 832                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
 833                         bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
 834                         error = xfs_btree_insert(bma->cur, &i);
 835                         if (error)
 836                                 goto done;
 837                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 838                 }
 839                 if (bma->ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
 840                     bma->ip->i_d.di_nextents > bma->ip->i_df.if_ext_max) {
 841                         error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
 842                                         bma->firstblock, bma->flist, &bma->cur,
 843                                         1, &tmp_rval, XFS_DATA_FORK);
 844                         rval |= tmp_rval;
 845                         if (error)
 846                                 goto done;
 847                 }
 848                 temp = xfs_bmap_worst_indlen(bma->ip, temp);
 849                 temp2 = xfs_bmap_worst_indlen(bma->ip, temp2);
 850                 diff = (int)(temp + temp2 - startblockval(PREV.br_startblock) -
 851                         (bma->cur ? bma->cur->bc_private.b.allocated : 0));
 852                 if (diff > 0) {
 853                         error = xfs_icsb_modify_counters(bma->ip->i_mount,
 854                                         XFS_SBS_FDBLOCKS,
 855                                         -((int64_t)diff), 0);
 856                         ASSERT(!error);
 857                         if (error)
 858                                 goto done;
 859                 }
 860
 861                 ep = xfs_iext_get_ext(ifp, bma->idx);
 862                 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
 863                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
 864                 trace_xfs_bmap_pre_update(bma->ip, bma->idx + 2, state, _THIS_IP_);
 865                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, bma->idx + 2),
 866                         nullstartblock((int)temp2));
 867                 trace_xfs_bmap_post_update(bma->ip, bma->idx + 2, state, _THIS_IP_);
 868
 869                 bma->idx++;
 870                 da_new = temp + temp2;
 871                 break;
 872
 873         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
 874         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
 875         case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
 876         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
 877         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
 878         case BMAP_LEFT_CONTIG:
 879         case BMAP_RIGHT_CONTIG:
 880                 /*
 881                  * These cases are all impossible.
 882                  */
 883                 ASSERT(0);
 884         }
 885
 886         /* convert to a btree if necessary */
 887         if (XFS_IFORK_FORMAT(bma->ip, XFS_DATA_FORK) == XFS_DINODE_FMT_EXTENTS &&
 888             XFS_IFORK_NEXTENTS(bma->ip, XFS_DATA_FORK) > ifp->if_ext_max) {
 889                 int     tmp_logflags;   /* partial log flag return val */
 890
 891                 ASSERT(bma->cur == NULL);
 892                 error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
 893                                 bma->firstblock, bma->flist, &bma->cur,
 894                                 da_old > 0, &tmp_logflags, XFS_DATA_FORK);
 895                 bma->logflags |= tmp_logflags;
 896                 if (error)
 897                         goto done;
 898         }
 899
 900         /* adjust for changes in reserved delayed indirect blocks */
 901         if (da_old || da_new) {
 902                 temp = da_new;
 903                 if (bma->cur)
 904                         temp += bma->cur->bc_private.b.allocated;
 905                 ASSERT(temp <= da_old);
 906                 if (temp < da_old)
 907                         xfs_icsb_modify_counters(bma->ip->i_mount,
 908                                         XFS_SBS_FDBLOCKS,
 909                                         (int64_t)(da_old - temp), 0);
 910         }
 911
 912         /* clear out the allocated field, done with it now in any case. */
 913         if (bma->cur)
 914                 bma->cur->bc_private.b.allocated = 0;
 915
 916         xfs_bmap_check_leaf_extents(bma->cur, bma->ip, XFS_DATA_FORK);
 917 done:
 918         bma->logflags |= rval;
 919         return error;
 920 #undef  LEFT
 921 #undef  RIGHT
 922 #undef  PREV
 923 }
 924
 925 /*
 926  * Convert an unwritten allocation to a real allocation or vice versa.
 927  */
 928 STATIC int                              /* error */
 929 xfs_bmap_add_extent_unwritten_real(
 930         struct xfs_trans        *tp,
 931         xfs_inode_t             *ip,    /* incore inode pointer */
 932         xfs_extnum_t            *idx,   /* extent number to update/insert */
 933         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
 934         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
 935         xfs_fsblock_t           *first, /* pointer to firstblock variable */
 936         xfs_bmap_free_t         *flist, /* list of extents to be freed */
 937         int                     *logflagsp) /* inode logging flags */
 938 {
 939         xfs_btree_cur_t         *cur;   /* btree cursor */
 940         xfs_bmbt_rec_host_t     *ep;    /* extent entry for idx */
 941         int                     error;  /* error return value */
 942         int                     i;      /* temp state */
 943         xfs_ifork_t             *ifp;   /* inode fork pointer */
 944         xfs_fileoff_t           new_endoff;     /* end offset of new entry */
 945         xfs_exntst_t            newext; /* new extent state */
 946         xfs_exntst_t            oldext; /* old extent state */
 947         xfs_bmbt_irec_t         r[3];   /* neighbor extent entries */
 948                                         /* left is 0, right is 1, prev is 2 */
 949         int                     rval=0; /* return value (logging flags) */
 950         int                     state = 0;/* state bits, accessed thru macros */
 951
 952         *logflagsp = 0;
 953
 954         cur = *curp;
 955         ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
 956
 957         ASSERT(*idx >= 0);
 958         ASSERT(*idx <= ifp->if_bytes / sizeof(struct xfs_bmbt_rec));
 959         ASSERT(!isnullstartblock(new->br_startblock));
 960
 961         XFS_STATS_INC(xs_add_exlist);
 962
 963 #define LEFT            r[0]
 964 #define RIGHT           r[1]
 965 #define PREV            r[2]
 966
 967         /*
 968          * Set up a bunch of variables to make the tests simpler.
 969          */
 970         error = 0;
 971         ep = xfs_iext_get_ext(ifp, *idx);
 972         xfs_bmbt_get_all(ep, &PREV);
 973         newext = new->br_state;
 974         oldext = (newext == XFS_EXT_UNWRITTEN) ?
 975                 XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
 976         ASSERT(PREV.br_state == oldext);
 977         new_endoff = new->br_startoff + new->br_blockcount;
 978         ASSERT(PREV.br_startoff <= new->br_startoff);
 979         ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
 980
 981         /*
 982          * Set flags determining what part of the previous oldext allocation
 983          * extent is being replaced by a newext allocation.
 984          */
 985         if (PREV.br_startoff == new->br_startoff)
 986                 state |= BMAP_LEFT_FILLING;
 987         if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
 988                 state |= BMAP_RIGHT_FILLING;
 989
 990         /*
 991          * Check and set flags if this segment has a left neighbor.
 992          * Don't set contiguous if the combined extent would be too large.
 993          */
 994         if (*idx > 0) {
 995                 state |= BMAP_LEFT_VALID;
 996                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &LEFT);
 997
 998                 if (isnullstartblock(LEFT.br_startblock))
 999                         state |= BMAP_LEFT_DELAY;
1000         }
1001
1002         if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
1003             LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
1004             LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
1005             LEFT.br_state == newext &&
1006             LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
1007                 state |= BMAP_LEFT_CONTIG;
1008
1009         /*
1010          * Check and set flags if this segment has a right neighbor.
1011          * Don't set contiguous if the combined extent would be too large.
1012          * Also check for all-three-contiguous being too large.
1013          */
1014         if (*idx < ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1) {
1015                 state |= BMAP_RIGHT_VALID;
1016                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx + 1), &RIGHT);
1017                 if (isnullstartblock(RIGHT.br_startblock))
1018                         state |= BMAP_RIGHT_DELAY;
1019         }
1020
1021         if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
1022             new_endoff == RIGHT.br_startoff &&
1023             new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
1024             newext == RIGHT.br_state &&
1025             new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
1026             ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1027                        BMAP_RIGHT_FILLING)) !=
1028                       (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1029                        BMAP_RIGHT_FILLING) ||
1030              LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
1031                         <= MAXEXTLEN))
1032                 state |= BMAP_RIGHT_CONTIG;
1033
1034         /*
1035          * Switch out based on the FILLING and CONTIG state bits.
1036          */
1037         switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1038                          BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
1039         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1040              BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1041                 /*
1042                  * Setting all of a previous oldext extent to newext.
1043                  * The left and right neighbors are both contiguous with new.
1044                  */
1045                 --*idx;
1046
1047                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1048                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
1049                         LEFT.br_blockcount + PREV.br_blockcount +
1050                         RIGHT.br_blockcount);
1051                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1052
1053                 xfs_iext_remove(ip, *idx + 1, 2, state);
1054                 ip->i_d.di_nextents -= 2;
1055                 if (cur == NULL)
1056                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1057                 else {
1058                         rval = XFS_ILOG_CORE;
1059                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
1060                                         RIGHT.br_startblock,
1061                                         RIGHT.br_blockcount, &i)))
1062                                 goto done;
1063                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1064                         if ((error = xfs_btree_delete(cur, &i)))
1065                                 goto done;
1066                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1067                         if ((error = xfs_btree_decrement(cur, 0, &i)))
1068                                 goto done;
1069                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1070                         if ((error = xfs_btree_delete(cur, &i)))
1071                                 goto done;
1072                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1073                         if ((error = xfs_btree_decrement(cur, 0, &i)))
1074                                 goto done;
1075                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1076                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
1077                                 LEFT.br_startblock,
1078                                 LEFT.br_blockcount + PREV.br_blockcount +
1079                                 RIGHT.br_blockcount, LEFT.br_state)))
1080                                 goto done;
1081                 }
1082                 break;
1083
1084         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1085                 /*
1086                  * Setting all of a previous oldext extent to newext.
1087                  * The left neighbor is contiguous, the right is not.
1088                  */
1089                 --*idx;
1090
1091                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1092                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
1093                         LEFT.br_blockcount + PREV.br_blockcount);
1094                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1095
1096                 xfs_iext_remove(ip, *idx + 1, 1, state);
1097                 ip->i_d.di_nextents--;
1098                 if (cur == NULL)
1099                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1100                 else {
1101                         rval = XFS_ILOG_CORE;
1102                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1103                                         PREV.br_startblock, PREV.br_blockcount,
1104                                         &i)))
1105                                 goto done;
1106                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1107                         if ((error = xfs_btree_delete(cur, &i)))
1108                                 goto done;
1109                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1110                         if ((error = xfs_btree_decrement(cur, 0, &i)))
1111                                 goto done;
1112                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1113                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
1114                                 LEFT.br_startblock,
1115                                 LEFT.br_blockcount + PREV.br_blockcount,
1116                                 LEFT.br_state)))
1117                                 goto done;
1118                 }
1119                 break;
1120
1121         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1122                 /*
1123                  * Setting all of a previous oldext extent to newext.
1124                  * The right neighbor is contiguous, the left is not.
1125                  */
1126                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1127                 xfs_bmbt_set_blockcount(ep,
1128                         PREV.br_blockcount + RIGHT.br_blockcount);
1129                 xfs_bmbt_set_state(ep, newext);
1130                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1131                 xfs_iext_remove(ip, *idx + 1, 1, state);
1132                 ip->i_d.di_nextents--;
1133                 if (cur == NULL)
1134                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1135                 else {
1136                         rval = XFS_ILOG_CORE;
1137                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
1138                                         RIGHT.br_startblock,
1139                                         RIGHT.br_blockcount, &i)))
1140                                 goto done;
1141                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1142                         if ((error = xfs_btree_delete(cur, &i)))
1143                                 goto done;
1144                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1145                         if ((error = xfs_btree_decrement(cur, 0, &i)))
1146                                 goto done;
1147                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1148                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
1149                                 new->br_startblock,
1150                                 new->br_blockcount + RIGHT.br_blockcount,
1151                                 newext)))
1152                                 goto done;
1153                 }
1154                 break;
1155
1156         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
1157                 /*
1158                  * Setting all of a previous oldext extent to newext.
1159                  * Neither the left nor right neighbors are contiguous with
1160                  * the new one.
1161                  */
1162                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1163                 xfs_bmbt_set_state(ep, newext);
1164                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1165
1166                 if (cur == NULL)
1167                         rval = XFS_ILOG_DEXT;
1168                 else {
1169                         rval = 0;
1170                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1171                                         new->br_startblock, new->br_blockcount,
1172                                         &i)))
1173                                 goto done;
1174                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1175                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
1176                                 new->br_startblock, new->br_blockcount,
1177                                 newext)))
1178                                 goto done;
1179                 }
1180                 break;
1181
1182         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
1183                 /*
1184                  * Setting the first part of a previous oldext extent to newext.
1185                  * The left neighbor is contiguous.
1186                  */
1187                 trace_xfs_bmap_pre_update(ip, *idx - 1, state, _THIS_IP_);
1188                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx - 1),
1189                         LEFT.br_blockcount + new->br_blockcount);
1190                 xfs_bmbt_set_startoff(ep,
1191                         PREV.br_startoff + new->br_blockcount);
1192                 trace_xfs_bmap_post_update(ip, *idx - 1, state, _THIS_IP_);
1193
1194                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1195                 xfs_bmbt_set_startblock(ep,
1196                         new->br_startblock + new->br_blockcount);
1197                 xfs_bmbt_set_blockcount(ep,
1198                         PREV.br_blockcount - new->br_blockcount);
1199                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1200
1201                 --*idx;
1202
1203                 if (cur == NULL)
1204                         rval = XFS_ILOG_DEXT;
1205                 else {
1206                         rval = 0;
1207                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1208                                         PREV.br_startblock, PREV.br_blockcount,
1209                                         &i)))
1210                                 goto done;
1211                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1212                         if ((error = xfs_bmbt_update(cur,
1213                                 PREV.br_startoff + new->br_blockcount,
1214                                 PREV.br_startblock + new->br_blockcount,
1215                                 PREV.br_blockcount - new->br_blockcount,
1216                                 oldext)))
1217                                 goto done;
1218                         if ((error = xfs_btree_decrement(cur, 0, &i)))
1219                                 goto done;
1220                         error = xfs_bmbt_update(cur, LEFT.br_startoff,
1221                                 LEFT.br_startblock,
1222                                 LEFT.br_blockcount + new->br_blockcount,
1223                                 LEFT.br_state);
1224                         if (error)
1225                                 goto done;
1226                 }
1227                 break;
1228
1229         case BMAP_LEFT_FILLING:
1230                 /*
1231                  * Setting the first part of a previous oldext extent to newext.
1232                  * The left neighbor is not contiguous.
1233                  */
1234                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1235                 ASSERT(ep && xfs_bmbt_get_state(ep) == oldext);
1236                 xfs_bmbt_set_startoff(ep, new_endoff);
1237                 xfs_bmbt_set_blockcount(ep,
1238                         PREV.br_blockcount - new->br_blockcount);
1239                 xfs_bmbt_set_startblock(ep,
1240                         new->br_startblock + new->br_blockcount);
1241                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1242
1243                 xfs_iext_insert(ip, *idx, 1, new, state);
1244                 ip->i_d.di_nextents++;
1245                 if (cur == NULL)
1246                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1247                 else {
1248                         rval = XFS_ILOG_CORE;
1249                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1250                                         PREV.br_startblock, PREV.br_blockcount,
1251                                         &i)))
1252                                 goto done;
1253                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1254                         if ((error = xfs_bmbt_update(cur,
1255                                 PREV.br_startoff + new->br_blockcount,
1256                                 PREV.br_startblock + new->br_blockcount,
1257                                 PREV.br_blockcount - new->br_blockcount,
1258                                 oldext)))
1259                                 goto done;
1260                         cur->bc_rec.b = *new;
1261                         if ((error = xfs_btree_insert(cur, &i)))
1262                                 goto done;
1263                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1264                 }
1265                 break;
1266
1267         case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1268                 /*
1269                  * Setting the last part of a previous oldext extent to newext.
1270                  * The right neighbor is contiguous with the new allocation.
1271                  */
1272                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1273                 xfs_bmbt_set_blockcount(ep,
1274                         PREV.br_blockcount - new->br_blockcount);
1275                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1276
1277                 ++*idx;
1278
1279                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1280                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx),
1281                         new->br_startoff, new->br_startblock,
1282                         new->br_blockcount + RIGHT.br_blockcount, newext);
1283                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1284
1285                 if (cur == NULL)
1286                         rval = XFS_ILOG_DEXT;
1287                 else {
1288                         rval = 0;
1289                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1290                                         PREV.br_startblock,
1291                                         PREV.br_blockcount, &i)))
1292                                 goto done;
1293                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1294                         if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
1295                                 PREV.br_startblock,
1296                                 PREV.br_blockcount - new->br_blockcount,
1297                                 oldext)))
1298                                 goto done;
1299                         if ((error = xfs_btree_increment(cur, 0, &i)))
1300                                 goto done;
1301                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
1302                                 new->br_startblock,
1303                                 new->br_blockcount + RIGHT.br_blockcount,
1304                                 newext)))
1305                                 goto done;
1306                 }
1307                 break;
1308
1309         case BMAP_RIGHT_FILLING:
1310                 /*
1311                  * Setting the last part of a previous oldext extent to newext.
1312                  * The right neighbor is not contiguous.
1313                  */
1314                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1315                 xfs_bmbt_set_blockcount(ep,
1316                         PREV.br_blockcount - new->br_blockcount);
1317                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1318
1319                 ++*idx;
1320                 xfs_iext_insert(ip, *idx, 1, new, state);
1321
1322                 ip->i_d.di_nextents++;
1323                 if (cur == NULL)
1324                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1325                 else {
1326                         rval = XFS_ILOG_CORE;
1327                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1328                                         PREV.br_startblock, PREV.br_blockcount,
1329                                         &i)))
1330                                 goto done;
1331                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1332                         if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
1333                                 PREV.br_startblock,
1334                                 PREV.br_blockcount - new->br_blockcount,
1335                                 oldext)))
1336                                 goto done;
1337                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1338                                         new->br_startblock, new->br_blockcount,
1339                                         &i)))
1340                                 goto done;
1341                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1342                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
1343                         if ((error = xfs_btree_insert(cur, &i)))
1344                                 goto done;
1345                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1346                 }
1347                 break;
1348
1349         case 0:
1350                 /*
1351                  * Setting the middle part of a previous oldext extent to
1352                  * newext.  Contiguity is impossible here.
1353                  * One extent becomes three extents.
1354                  */
1355                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1356                 xfs_bmbt_set_blockcount(ep,
1357                         new->br_startoff - PREV.br_startoff);
1358                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1359
1360                 r[0] = *new;
1361                 r[1].br_startoff = new_endoff;
1362                 r[1].br_blockcount =
1363                         PREV.br_startoff + PREV.br_blockcount - new_endoff;
1364                 r[1].br_startblock = new->br_startblock + new->br_blockcount;
1365                 r[1].br_state = oldext;
1366
1367                 ++*idx;
1368                 xfs_iext_insert(ip, *idx, 2, &r[0], state);
1369
1370                 ip->i_d.di_nextents += 2;
1371                 if (cur == NULL)
1372                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1373                 else {
1374                         rval = XFS_ILOG_CORE;
1375                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1376                                         PREV.br_startblock, PREV.br_blockcount,
1377                                         &i)))
1378                                 goto done;
1379                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1380                         /* new right extent - oldext */
1381                         if ((error = xfs_bmbt_update(cur, r[1].br_startoff,
1382                                 r[1].br_startblock, r[1].br_blockcount,
1383                                 r[1].br_state)))
1384                                 goto done;
1385                         /* new left extent - oldext */
1386                         cur->bc_rec.b = PREV;
1387                         cur->bc_rec.b.br_blockcount =
1388                                 new->br_startoff - PREV.br_startoff;
1389                         if ((error = xfs_btree_insert(cur, &i)))
1390                                 goto done;
1391                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1392                         /*
1393                          * Reset the cursor to the position of the new extent
1394                          * we are about to insert as we can't trust it after
1395                          * the previous insert.
1396                          */
1397                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1398                                         new->br_startblock, new->br_blockcount,
1399                                         &i)))
1400                                 goto done;
1401                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1402                         /* new middle extent - newext */
1403                         cur->bc_rec.b.br_state = new->br_state;
1404                         if ((error = xfs_btree_insert(cur, &i)))
1405                                 goto done;
1406                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1407                 }
1408                 break;
1409
1410         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1411         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1412         case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
1413         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1414         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1415         case BMAP_LEFT_CONTIG:
1416         case BMAP_RIGHT_CONTIG:
1417                 /*
1418                  * These cases are all impossible.
1419                  */
1420                 ASSERT(0);
1421         }
1422
1423         /* convert to a btree if necessary */
1424         if (XFS_IFORK_FORMAT(ip, XFS_DATA_FORK) == XFS_DINODE_FMT_EXTENTS &&
1425             XFS_IFORK_NEXTENTS(ip, XFS_DATA_FORK) > ifp->if_ext_max) {
1426                 int     tmp_logflags;   /* partial log flag return val */
1427
1428                 ASSERT(cur == NULL);
1429                 error = xfs_bmap_extents_to_btree(tp, ip, first, flist, &cur,
1430                                 0, &tmp_logflags, XFS_DATA_FORK);
1431                 *logflagsp |= tmp_logflags;
1432                 if (error)
1433                         goto done;
1434         }
1435
1436         /* clear out the allocated field, done with it now in any case. */
1437         if (cur) {
1438                 cur->bc_private.b.allocated = 0;
1439                 *curp = cur;
1440         }
1441
1442         xfs_bmap_check_leaf_extents(*curp, ip, XFS_DATA_FORK);
1443 done:
1444         *logflagsp |= rval;
1445         return error;
1446 #undef  LEFT
1447 #undef  RIGHT
1448 #undef  PREV
1449 }
1450
1451 /*
1452  * Convert a hole to a delayed allocation.
1453  */
1454 STATIC void
1455 xfs_bmap_add_extent_hole_delay(
1456         xfs_inode_t             *ip,    /* incore inode pointer */
1457         xfs_extnum_t            *idx,   /* extent number to update/insert */
1458         xfs_bmbt_irec_t         *new)   /* new data to add to file extents */
1459 {
1460         xfs_ifork_t             *ifp;   /* inode fork pointer */
1461         xfs_bmbt_irec_t         left;   /* left neighbor extent entry */
1462         xfs_filblks_t           newlen=0;       /* new indirect size */
1463         xfs_filblks_t           oldlen=0;       /* old indirect size */
1464         xfs_bmbt_irec_t         right;  /* right neighbor extent entry */
1465         int                     state;  /* state bits, accessed thru macros */
1466         xfs_filblks_t           temp=0; /* temp for indirect calculations */
1467
1468         ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
1469         state = 0;
1470         ASSERT(isnullstartblock(new->br_startblock));
1471
1472         /*
1473          * Check and set flags if this segment has a left neighbor
1474          */
1475         if (*idx > 0) {
1476                 state |= BMAP_LEFT_VALID;
1477                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &left);
1478
1479                 if (isnullstartblock(left.br_startblock))
1480                         state |= BMAP_LEFT_DELAY;
1481         }
1482
1483         /*
1484          * Check and set flags if the current (right) segment exists.
1485          * If it doesn't exist, we're converting the hole at end-of-file.
1486          */
1487         if (*idx < ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t)) {
1488                 state |= BMAP_RIGHT_VALID;
1489                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx), &right);
1490
1491                 if (isnullstartblock(right.br_startblock))
1492                         state |= BMAP_RIGHT_DELAY;
1493         }
1494
1495         /*
1496          * Set contiguity flags on the left and right neighbors.
1497          * Don't let extents get too large, even if the pieces are contiguous.
1498          */
1499         if ((state & BMAP_LEFT_VALID) && (state & BMAP_LEFT_DELAY) &&
1500             left.br_startoff + left.br_blockcount == new->br_startoff &&
1501             left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
1502                 state |= BMAP_LEFT_CONTIG;
1503
1504         if ((state & BMAP_RIGHT_VALID) && (state & BMAP_RIGHT_DELAY) &&
1505             new->br_startoff + new->br_blockcount == right.br_startoff &&
1506             new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
1507             (!(state & BMAP_LEFT_CONTIG) ||
1508              (left.br_blockcount + new->br_blockcount +
1509               right.br_blockcount <= MAXEXTLEN)))
1510                 state |= BMAP_RIGHT_CONTIG;
1511
1512         /*
1513          * Switch out based on the contiguity flags.
1514          */
1515         switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
1516         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1517                 /*
1518                  * New allocation is contiguous with delayed allocations
1519                  * on the left and on the right.
1520                  * Merge all three into a single extent record.
1521                  */
1522                 --*idx;
1523                 temp = left.br_blockcount + new->br_blockcount +
1524                         right.br_blockcount;
1525
1526                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1527                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx), temp);
1528                 oldlen = startblockval(left.br_startblock) +
1529                         startblockval(new->br_startblock) +
1530                         startblockval(right.br_startblock);
1531                 newlen = xfs_bmap_worst_indlen(ip, temp);
1532                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, *idx),
1533                         nullstartblock((int)newlen));
1534                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1535
1536                 xfs_iext_remove(ip, *idx + 1, 1, state);
1537                 break;
1538
1539         case BMAP_LEFT_CONTIG:
1540                 /*
1541                  * New allocation is contiguous with a delayed allocation
1542                  * on the left.
1543                  * Merge the new allocation with the left neighbor.
1544                  */
1545                 --*idx;
1546                 temp = left.br_blockcount + new->br_blockcount;
1547
1548                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1549                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx), temp);
1550                 oldlen = startblockval(left.br_startblock) +
1551                         startblockval(new->br_startblock);
1552                 newlen = xfs_bmap_worst_indlen(ip, temp);
1553                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, *idx),
1554                         nullstartblock((int)newlen));
1555                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1556                 break;
1557
1558         case BMAP_RIGHT_CONTIG:
1559                 /*
1560                  * New allocation is contiguous with a delayed allocation
1561                  * on the right.
1562                  * Merge the new allocation with the right neighbor.
1563                  */
1564                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1565                 temp = new->br_blockcount + right.br_blockcount;
1566                 oldlen = startblockval(new->br_startblock) +
1567                         startblockval(right.br_startblock);
1568                 newlen = xfs_bmap_worst_indlen(ip, temp);
1569                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx),
1570                         new->br_startoff,
1571                         nullstartblock((int)newlen), temp, right.br_state);
1572                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1573                 break;
1574
1575         case 0:
1576                 /*
1577                  * New allocation is not contiguous with another
1578                  * delayed allocation.
1579                  * Insert a new entry.
1580                  */
1581                 oldlen = newlen = 0;
1582                 xfs_iext_insert(ip, *idx, 1, new, state);
1583                 break;
1584         }
1585         if (oldlen != newlen) {
1586                 ASSERT(oldlen > newlen);
1587                 xfs_icsb_modify_counters(ip->i_mount, XFS_SBS_FDBLOCKS,
1588                         (int64_t)(oldlen - newlen), 0);
1589                 /*
1590                  * Nothing to do for disk quota accounting here.
1591                  */
1592         }
1593 }
1594
1595 /*
1596  * Convert a hole to a real allocation.
1597  */
1598 STATIC int                              /* error */
1599 xfs_bmap_add_extent_hole_real(
1600         struct xfs_bmalloca     *bma,
1601         int                     whichfork)
1602 {
1603         struct xfs_bmbt_irec    *new = &bma->got;
1604         int                     error;  /* error return value */
1605         int                     i;      /* temp state */
1606         xfs_ifork_t             *ifp;   /* inode fork pointer */
1607         xfs_bmbt_irec_t         left;   /* left neighbor extent entry */
1608         xfs_bmbt_irec_t         right;  /* right neighbor extent entry */
1609         int                     rval=0; /* return value (logging flags) */
1610         int                     state;  /* state bits, accessed thru macros */
1611
1612         ifp = XFS_IFORK_PTR(bma->ip, whichfork);
1613
1614         ASSERT(bma->idx >= 0);
1615         ASSERT(bma->idx <= ifp->if_bytes / sizeof(struct xfs_bmbt_rec));
1616         ASSERT(!isnullstartblock(new->br_startblock));
1617         ASSERT(!bma->cur ||
1618                !(bma->cur->bc_private.b.flags & XFS_BTCUR_BPRV_WASDEL));
1619
1620         XFS_STATS_INC(xs_add_exlist);
1621
1622         state = 0;
1623         if (whichfork == XFS_ATTR_FORK)
1624                 state |= BMAP_ATTRFORK;
1625
1626         /*
1627          * Check and set flags if this segment has a left neighbor.
1628          */
1629         if (bma->idx > 0) {
1630                 state |= BMAP_LEFT_VALID;
1631                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx - 1), &left);
1632                 if (isnullstartblock(left.br_startblock))
1633                         state |= BMAP_LEFT_DELAY;
1634         }
1635
1636         /*
1637          * Check and set flags if this segment has a current value.
1638          * Not true if we're inserting into the "hole" at eof.
1639          */
1640         if (bma->idx < ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)) {
1641                 state |= BMAP_RIGHT_VALID;
1642                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx), &right);
1643                 if (isnullstartblock(right.br_startblock))
1644                         state |= BMAP_RIGHT_DELAY;
1645         }
1646
1647         /*
1648          * We're inserting a real allocation between "left" and "right".
1649          * Set the contiguity flags.  Don't let extents get too large.
1650          */
1651         if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
1652             left.br_startoff + left.br_blockcount == new->br_startoff &&
1653             left.br_startblock + left.br_blockcount == new->br_startblock &&
1654             left.br_state == new->br_state &&
1655             left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
1656                 state |= BMAP_LEFT_CONTIG;
1657
1658         if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
1659             new->br_startoff + new->br_blockcount == right.br_startoff &&
1660             new->br_startblock + new->br_blockcount == right.br_startblock &&
1661             new->br_state == right.br_state &&
1662             new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
1663             (!(state & BMAP_LEFT_CONTIG) ||
1664              left.br_blockcount + new->br_blockcount +
1665              right.br_blockcount <= MAXEXTLEN))
1666                 state |= BMAP_RIGHT_CONTIG;
1667
1668         error = 0;
1669         /*
1670          * Select which case we're in here, and implement it.
1671          */
1672         switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
1673         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1674                 /*
1675                  * New allocation is contiguous with real allocations on the
1676                  * left and on the right.
1677                  * Merge all three into a single extent record.
1678                  */
1679                 --bma->idx;
1680                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1681                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
1682                         left.br_blockcount + new->br_blockcount +
1683                         right.br_blockcount);
1684                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1685
1686                 xfs_iext_remove(bma->ip, bma->idx + 1, 1, state);
1687
1688                 XFS_IFORK_NEXT_SET(bma->ip, whichfork,
1689                         XFS_IFORK_NEXTENTS(bma->ip, whichfork) - 1);
1690                 if (bma->cur == NULL) {
1691                         rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
1692                 } else {
1693                         rval = XFS_ILOG_CORE;
1694                         error = xfs_bmbt_lookup_eq(bma->cur, right.br_startoff,
1695                                         right.br_startblock, right.br_blockcount,
1696                                         &i);
1697                         if (error)
1698                                 goto done;
1699                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1700                         error = xfs_btree_delete(bma->cur, &i);
1701                         if (error)
1702                                 goto done;
1703                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1704                         error = xfs_btree_decrement(bma->cur, 0, &i);
1705                         if (error)
1706                                 goto done;
1707                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1708                         error = xfs_bmbt_update(bma->cur, left.br_startoff,
1709                                         left.br_startblock,
1710                                         left.br_blockcount +
1711                                                 new->br_blockcount +
1712                                                 right.br_blockcount,
1713                                         left.br_state);
1714                         if (error)
1715                                 goto done;
1716                 }
1717                 break;
1718
1719         case BMAP_LEFT_CONTIG:
1720                 /*
1721                  * New allocation is contiguous with a real allocation
1722                  * on the left.
1723                  * Merge the new allocation with the left neighbor.
1724                  */
1725                 --bma->idx;
1726                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1727                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
1728                         left.br_blockcount + new->br_blockcount);
1729                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1730
1731                 if (bma->cur == NULL) {
1732                         rval = xfs_ilog_fext(whichfork);
1733                 } else {
1734                         rval = 0;
1735                         error = xfs_bmbt_lookup_eq(bma->cur, left.br_startoff,
1736                                         left.br_startblock, left.br_blockcount,
1737                                         &i);
1738                         if (error)
1739                                 goto done;
1740                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1741                         error = xfs_bmbt_update(bma->cur, left.br_startoff,
1742                                         left.br_startblock,
1743                                         left.br_blockcount +
1744                                                 new->br_blockcount,
1745                                         left.br_state);
1746                         if (error)
1747                                 goto done;
1748                 }
1749                 break;
1750
1751         case BMAP_RIGHT_CONTIG:
1752                 /*
1753                  * New allocation is contiguous with a real allocation
1754                  * on the right.
1755                  * Merge the new allocation with the right neighbor.
1756                  */
1757                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1758                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, bma->idx),
1759                         new->br_startoff, new->br_startblock,
1760                         new->br_blockcount + right.br_blockcount,
1761                         right.br_state);
1762                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1763
1764                 if (bma->cur == NULL) {
1765                         rval = xfs_ilog_fext(whichfork);
1766                 } else {
1767                         rval = 0;
1768                         error = xfs_bmbt_lookup_eq(bma->cur,
1769                                         right.br_startoff,
1770                                         right.br_startblock,
1771                                         right.br_blockcount, &i);
1772                         if (error)
1773                                 goto done;
1774                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1775                         error = xfs_bmbt_update(bma->cur, new->br_startoff,
1776                                         new->br_startblock,
1777                                         new->br_blockcount +
1778                                                 right.br_blockcount,
1779                                         right.br_state);
1780                         if (error)
1781                                 goto done;
1782                 }
1783                 break;
1784
1785         case 0:
1786                 /*
1787                  * New allocation is not contiguous with another
1788                  * real allocation.
1789                  * Insert a new entry.
1790                  */
1791                 xfs_iext_insert(bma->ip, bma->idx, 1, new, state);
1792                 XFS_IFORK_NEXT_SET(bma->ip, whichfork,
1793                         XFS_IFORK_NEXTENTS(bma->ip, whichfork) + 1);
1794                 if (bma->cur == NULL) {
1795                         rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
1796                 } else {
1797                         rval = XFS_ILOG_CORE;
1798                         error = xfs_bmbt_lookup_eq(bma->cur,
1799                                         new->br_startoff,
1800                                         new->br_startblock,
1801                                         new->br_blockcount, &i);
1802                         if (error)
1803                                 goto done;
1804                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1805                         bma->cur->bc_rec.b.br_state = new->br_state;
1806                         error = xfs_btree_insert(bma->cur, &i);
1807                         if (error)
1808                                 goto done;
1809                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1810                 }
1811                 break;
1812         }
1813
1814         /* convert to a btree if necessary */
1815         if (XFS_IFORK_FORMAT(bma->ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
1816             XFS_IFORK_NEXTENTS(bma->ip, whichfork) > ifp->if_ext_max) {
1817                 int     tmp_logflags;   /* partial log flag return val */
1818
1819                 ASSERT(bma->cur == NULL);
1820                 error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
1821                                 bma->firstblock, bma->flist, &bma->cur,
1822                                 0, &tmp_logflags, whichfork);
1823                 bma->logflags |= tmp_logflags;
1824                 if (error)
1825                         goto done;
1826         }
1827
1828         /* clear out the allocated field, done with it now in any case. */
1829         if (bma->cur)
1830                 bma->cur->bc_private.b.allocated = 0;
1831
1832         xfs_bmap_check_leaf_extents(bma->cur, bma->ip, whichfork);
1833 done:
1834         bma->logflags |= rval;
1835         return error;
1836 }
1837
1838 /*
1839  * Adjust the size of the new extent based on di_extsize and rt extsize.
1840  */
1841 STATIC int
1842 xfs_bmap_extsize_align(
1843         xfs_mount_t     *mp,
1844         xfs_bmbt_irec_t *gotp,          /* next extent pointer */
1845         xfs_bmbt_irec_t *prevp,         /* previous extent pointer */
1846         xfs_extlen_t    extsz,          /* align to this extent size */
1847         int             rt,             /* is this a realtime inode? */
1848         int             eof,            /* is extent at end-of-file? */
1849         int             delay,          /* creating delalloc extent? */
1850         int             convert,        /* overwriting unwritten extent? */
1851         xfs_fileoff_t   *offp,          /* in/out: aligned offset */
1852         xfs_extlen_t    *lenp)          /* in/out: aligned length */
1853 {
1854         xfs_fileoff_t   orig_off;       /* original offset */
1855         xfs_extlen_t    orig_alen;      /* original length */
1856         xfs_fileoff_t   orig_end;       /* original off+len */
1857         xfs_fileoff_t   nexto;          /* next file offset */
1858         xfs_fileoff_t   prevo;          /* previous file offset */
1859         xfs_fileoff_t   align_off;      /* temp for offset */
1860         xfs_extlen_t    align_alen;     /* temp for length */
1861         xfs_extlen_t    temp;           /* temp for calculations */
1862
1863         if (convert)
1864                 return 0;
1865
1866         orig_off = align_off = *offp;
1867         orig_alen = align_alen = *lenp;
1868         orig_end = orig_off + orig_alen;
1869
1870         /*
1871          * If this request overlaps an existing extent, then don't
1872          * attempt to perform any additional alignment.
1873          */
1874         if (!delay && !eof &&
1875             (orig_off >= gotp->br_startoff) &&
1876             (orig_end <= gotp->br_startoff + gotp->br_blockcount)) {
1877                 return 0;
1878         }
1879
1880         /*
1881          * If the file offset is unaligned vs. the extent size
1882          * we need to align it.  This will be possible unless
1883          * the file was previously written with a kernel that didn't
1884          * perform this alignment, or if a truncate shot us in the
1885          * foot.
1886          */
1887         temp = do_mod(orig_off, extsz);
1888         if (temp) {
1889                 align_alen += temp;
1890                 align_off -= temp;
1891         }
1892         /*
1893          * Same adjustment for the end of the requested area.
1894          */
1895         if ((temp = (align_alen % extsz))) {
1896                 align_alen += extsz - temp;
1897         }
1898         /*
1899          * If the previous block overlaps with this proposed allocation
1900          * then move the start forward without adjusting the length.
1901          */
1902         if (prevp->br_startoff != NULLFILEOFF) {
1903                 if (prevp->br_startblock == HOLESTARTBLOCK)
1904                         prevo = prevp->br_startoff;
1905                 else
1906                         prevo = prevp->br_startoff + prevp->br_blockcount;
1907         } else
1908                 prevo = 0;
1909         if (align_off != orig_off && align_off < prevo)
1910                 align_off = prevo;
1911         /*
1912          * If the next block overlaps with this proposed allocation
1913          * then move the start back without adjusting the length,
1914          * but not before offset 0.
1915          * This may of course make the start overlap previous block,
1916          * and if we hit the offset 0 limit then the next block
1917          * can still overlap too.
1918          */
1919         if (!eof && gotp->br_startoff != NULLFILEOFF) {
1920                 if ((delay && gotp->br_startblock == HOLESTARTBLOCK) ||
1921                     (!delay && gotp->br_startblock == DELAYSTARTBLOCK))
1922                         nexto = gotp->br_startoff + gotp->br_blockcount;
1923                 else
1924                         nexto = gotp->br_startoff;
1925         } else
1926                 nexto = NULLFILEOFF;
1927         if (!eof &&
1928             align_off + align_alen != orig_end &&
1929             align_off + align_alen > nexto)
1930                 align_off = nexto > align_alen ? nexto - align_alen : 0;
1931         /*
1932          * If we're now overlapping the next or previous extent that
1933          * means we can't fit an extsz piece in this hole.  Just move
1934          * the start forward to the first valid spot and set
1935          * the length so we hit the end.
1936          */
1937         if (align_off != orig_off && align_off < prevo)
1938                 align_off = prevo;
1939         if (align_off + align_alen != orig_end &&
1940             align_off + align_alen > nexto &&
1941             nexto != NULLFILEOFF) {
1942                 ASSERT(nexto > prevo);
1943                 align_alen = nexto - align_off;
1944         }
1945
1946         /*
1947          * If realtime, and the result isn't a multiple of the realtime
1948          * extent size we need to remove blocks until it is.
1949          */
1950         if (rt && (temp = (align_alen % mp->m_sb.sb_rextsize))) {
1951                 /*
1952                  * We're not covering the original request, or
1953                  * we won't be able to once we fix the length.
1954                  */
1955                 if (orig_off < align_off ||
1956                     orig_end > align_off + align_alen ||
1957                     align_alen - temp < orig_alen)
1958                         return XFS_ERROR(EINVAL);
1959                 /*
1960                  * Try to fix it by moving the start up.
1961                  */
1962                 if (align_off + temp <= orig_off) {
1963                         align_alen -= temp;
1964                         align_off += temp;
1965                 }
1966                 /*
1967                  * Try to fix it by moving the end in.
1968                  */
1969                 else if (align_off + align_alen - temp >= orig_end)
1970                         align_alen -= temp;
1971                 /*
1972                  * Set the start to the minimum then trim the length.
1973                  */
1974                 else {
1975                         align_alen -= orig_off - align_off;
1976                         align_off = orig_off;
1977                         align_alen -= align_alen % mp->m_sb.sb_rextsize;
1978                 }
1979                 /*
1980                  * Result doesn't cover the request, fail it.
1981                  */
1982                 if (orig_off < align_off || orig_end > align_off + align_alen)
1983                         return XFS_ERROR(EINVAL);
1984         } else {
1985                 ASSERT(orig_off >= align_off);
1986                 ASSERT(orig_end <= align_off + align_alen);
1987         }
1988
1989 #ifdef DEBUG
1990         if (!eof && gotp->br_startoff != NULLFILEOFF)
1991                 ASSERT(align_off + align_alen <= gotp->br_startoff);
1992         if (prevp->br_startoff != NULLFILEOFF)
1993                 ASSERT(align_off >= prevp->br_startoff + prevp->br_blockcount);
1994 #endif
1995
1996         *lenp = align_alen;
1997         *offp = align_off;
1998         return 0;
1999 }
2000
2001 #define XFS_ALLOC_GAP_UNITS     4
2002
2003 STATIC void
2004 xfs_bmap_adjacent(
2005         xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
2006 {
2007         xfs_fsblock_t   adjust;         /* adjustment to block numbers */
2008         xfs_agnumber_t  fb_agno;        /* ag number of ap->firstblock */
2009         xfs_mount_t     *mp;            /* mount point structure */
2010         int             nullfb;         /* true if ap->firstblock isn't set */
2011         int             rt;             /* true if inode is realtime */
2012
2013 #define ISVALID(x,y)    \
2014         (rt ? \
2015                 (x) < mp->m_sb.sb_rblocks : \
2016                 XFS_FSB_TO_AGNO(mp, x) == XFS_FSB_TO_AGNO(mp, y) && \
2017                 XFS_FSB_TO_AGNO(mp, x) < mp->m_sb.sb_agcount && \
2018                 XFS_FSB_TO_AGBNO(mp, x) < mp->m_sb.sb_agblocks)
2019
2020         mp = ap->ip->i_mount;
2021         nullfb = *ap->firstblock == NULLFSBLOCK;
2022         rt = XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata;
2023         fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, *ap->firstblock);
2024         /*
2025          * If allocating at eof, and there's a previous real block,
2026          * try to use its last block as our starting point.
2027          */
2028         if (ap->eof && ap->prev.br_startoff != NULLFILEOFF &&
2029             !isnullstartblock(ap->prev.br_startblock) &&
2030             ISVALID(ap->prev.br_startblock + ap->prev.br_blockcount,
2031                     ap->prev.br_startblock)) {
2032                 ap->blkno = ap->prev.br_startblock + ap->prev.br_blockcount;
2033                 /*
2034                  * Adjust for the gap between prevp and us.
2035                  */
2036                 adjust = ap->offset -
2037                         (ap->prev.br_startoff + ap->prev.br_blockcount);
2038                 if (adjust &&
2039                     ISVALID(ap->blkno + adjust, ap->prev.br_startblock))
2040                         ap->blkno += adjust;
2041         }
2042         /*
2043          * If not at eof, then compare the two neighbor blocks.
2044          * Figure out whether either one gives us a good starting point,
2045          * and pick the better one.
2046          */
2047         else if (!ap->eof) {
2048                 xfs_fsblock_t   gotbno;         /* right side block number */
2049                 xfs_fsblock_t   gotdiff=0;      /* right side difference */
2050                 xfs_fsblock_t   prevbno;        /* left side block number */
2051                 xfs_fsblock_t   prevdiff=0;     /* left side difference */
2052
2053                 /*
2054                  * If there's a previous (left) block, select a requested
2055                  * start block based on it.
2056                  */
2057                 if (ap->prev.br_startoff != NULLFILEOFF &&
2058                     !isnullstartblock(ap->prev.br_startblock) &&
2059                     (prevbno = ap->prev.br_startblock +
2060                                ap->prev.br_blockcount) &&
2061                     ISVALID(prevbno, ap->prev.br_startblock)) {
2062                         /*
2063                          * Calculate gap to end of previous block.
2064                          */
2065                         adjust = prevdiff = ap->offset -
2066                                 (ap->prev.br_startoff +
2067                                  ap->prev.br_blockcount);
2068                         /*
2069                          * Figure the startblock based on the previous block's
2070                          * end and the gap size.
2071                          * Heuristic!
2072                          * If the gap is large relative to the piece we're
2073                          * allocating, or using it gives us an invalid block
2074                          * number, then just use the end of the previous block.
2075                          */
2076                         if (prevdiff <= XFS_ALLOC_GAP_UNITS * ap->length &&
2077                             ISVALID(prevbno + prevdiff,
2078                                     ap->prev.br_startblock))
2079                                 prevbno += adjust;
2080                         else
2081                                 prevdiff += adjust;
2082                         /*
2083                          * If the firstblock forbids it, can't use it,
2084                          * must use default.
2085                          */
2086                         if (!rt && !nullfb &&
2087                             XFS_FSB_TO_AGNO(mp, prevbno) != fb_agno)
2088                                 prevbno = NULLFSBLOCK;
2089                 }
2090                 /*
2091                  * No previous block or can't follow it, just default.
2092                  */
2093                 else
2094                         prevbno = NULLFSBLOCK;
2095                 /*
2096                  * If there's a following (right) block, select a requested
2097                  * start block based on it.
2098                  */
2099                 if (!isnullstartblock(ap->got.br_startblock)) {
2100                         /*
2101                          * Calculate gap to start of next block.
2102                          */
2103                         adjust = gotdiff = ap->got.br_startoff - ap->offset;
2104                         /*
2105                          * Figure the startblock based on the next block's
2106                          * start and the gap size.
2107                          */
2108                         gotbno = ap->got.br_startblock;
2109                         /*
2110                          * Heuristic!
2111                          * If the gap is large relative to the piece we're
2112                          * allocating, or using it gives us an invalid block
2113                          * number, then just use the start of the next block
2114                          * offset by our length.
2115                          */
2116                         if (gotdiff <= XFS_ALLOC_GAP_UNITS * ap->length &&
2117                             ISVALID(gotbno - gotdiff, gotbno))
2118                                 gotbno -= adjust;
2119                         else if (ISVALID(gotbno - ap->length, gotbno)) {
2120                                 gotbno -= ap->length;
2121                                 gotdiff += adjust - ap->length;
2122                         } else
2123                                 gotdiff += adjust;
2124                         /*
2125                          * If the firstblock forbids it, can't use it,
2126                          * must use default.
2127                          */
2128                         if (!rt && !nullfb &&
2129                             XFS_FSB_TO_AGNO(mp, gotbno) != fb_agno)
2130                                 gotbno = NULLFSBLOCK;
2131                 }
2132                 /*
2133                  * No next block, just default.
2134                  */
2135                 else
2136                         gotbno = NULLFSBLOCK;
2137                 /*
2138                  * If both valid, pick the better one, else the only good
2139                  * one, else ap->blkno is already set (to 0 or the inode block).
2140                  */
2141                 if (prevbno != NULLFSBLOCK && gotbno != NULLFSBLOCK)
2142                         ap->blkno = prevdiff <= gotdiff ? prevbno : gotbno;
2143                 else if (prevbno != NULLFSBLOCK)
2144                         ap->blkno = prevbno;
2145                 else if (gotbno != NULLFSBLOCK)
2146                         ap->blkno = gotbno;
2147         }
2148 #undef ISVALID
2149 }
2150
2151 STATIC int
2152 xfs_bmap_rtalloc(
2153         xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
2154 {
2155         xfs_alloctype_t atype = 0;      /* type for allocation routines */
2156         int             error;          /* error return value */
2157         xfs_mount_t     *mp;            /* mount point structure */
2158         xfs_extlen_t    prod = 0;       /* product factor for allocators */
2159         xfs_extlen_t    ralen = 0;      /* realtime allocation length */
2160         xfs_extlen_t    align;          /* minimum allocation alignment */
2161         xfs_rtblock_t   rtb;
2162
2163         mp = ap->ip->i_mount;
2164         align = xfs_get_extsz_hint(ap->ip);
2165         prod = align / mp->m_sb.sb_rextsize;
2166         error = xfs_bmap_extsize_align(mp, &ap->got, &ap->prev,
2167                                         align, 1, ap->eof, 0,
2168                                         ap->conv, &ap->offset, &ap->length);
2169         if (error)
2170                 return error;
2171         ASSERT(ap->length);
2172         ASSERT(ap->length % mp->m_sb.sb_rextsize == 0);
2173
2174         /*
2175          * If the offset & length are not perfectly aligned
2176          * then kill prod, it will just get us in trouble.
2177          */
2178         if (do_mod(ap->offset, align) || ap->length % align)
2179                 prod = 1;
2180         /*
2181          * Set ralen to be the actual requested length in rtextents.
2182          */
2183         ralen = ap->length / mp->m_sb.sb_rextsize;
2184         /*
2185          * If the old value was close enough to MAXEXTLEN that
2186          * we rounded up to it, cut it back so it's valid again.
2187          * Note that if it's a really large request (bigger than
2188          * MAXEXTLEN), we don't hear about that number, and can't
2189          * adjust the starting point to match it.
2190          */
2191         if (ralen * mp->m_sb.sb_rextsize >= MAXEXTLEN)
2192                 ralen = MAXEXTLEN / mp->m_sb.sb_rextsize;
2193
2194         /*
2195          * Lock out other modifications to the RT bitmap inode.
2196          */
2197         xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL);
2198         xfs_trans_ijoin(ap->tp, mp->m_rbmip, XFS_ILOCK_EXCL);
2199
2200         /*
2201          * If it's an allocation to an empty file at offset 0,
2202          * pick an extent that will space things out in the rt area.
2203          */
2204         if (ap->eof && ap->offset == 0) {
2205                 xfs_rtblock_t uninitialized_var(rtx); /* realtime extent no */
2206
2207                 error = xfs_rtpick_extent(mp, ap->tp, ralen, &rtx);
2208                 if (error)
2209                         return error;
2210                 ap->blkno = rtx * mp->m_sb.sb_rextsize;
2211         } else {
2212                 ap->blkno = 0;
2213         }
2214
2215         xfs_bmap_adjacent(ap);
2216
2217         /*
2218          * Realtime allocation, done through xfs_rtallocate_extent.
2219          */
2220         atype = ap->blkno == 0 ?  XFS_ALLOCTYPE_ANY_AG : XFS_ALLOCTYPE_NEAR_BNO;
2221         do_div(ap->blkno, mp->m_sb.sb_rextsize);
2222         rtb = ap->blkno;
2223         ap->length = ralen;
2224         if ((error = xfs_rtallocate_extent(ap->tp, ap->blkno, 1, ap->length,
2225                                 &ralen, atype, ap->wasdel, prod, &rtb)))
2226                 return error;
2227         if (rtb == NULLFSBLOCK && prod > 1 &&
2228             (error = xfs_rtallocate_extent(ap->tp, ap->blkno, 1,
2229                                            ap->length, &ralen, atype,
2230                                            ap->wasdel, 1, &rtb)))
2231                 return error;
2232         ap->blkno = rtb;
2233         if (ap->blkno != NULLFSBLOCK) {
2234                 ap->blkno *= mp->m_sb.sb_rextsize;
2235                 ralen *= mp->m_sb.sb_rextsize;
2236                 ap->length = ralen;
2237                 ap->ip->i_d.di_nblocks += ralen;
2238                 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
2239                 if (ap->wasdel)
2240                         ap->ip->i_delayed_blks -= ralen;
2241                 /*
2242                  * Adjust the disk quota also. This was reserved
2243                  * earlier.
2244                  */
2245                 xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
2246                         ap->wasdel ? XFS_TRANS_DQ_DELRTBCOUNT :
2247                                         XFS_TRANS_DQ_RTBCOUNT, (long) ralen);
2248         } else {
2249                 ap->length = 0;
2250         }
2251         return 0;
2252 }
2253
2254 STATIC int
2255 xfs_bmap_btalloc_nullfb(
2256         struct xfs_bmalloca     *ap,
2257         struct xfs_alloc_arg    *args,
2258         xfs_extlen_t            *blen)
2259 {
2260         struct xfs_mount        *mp = ap->ip->i_mount;
2261         struct xfs_perag        *pag;
2262         xfs_agnumber_t          ag, startag;
2263         int                     notinit = 0;
2264         int                     error;
2265
2266         if (ap->userdata && xfs_inode_is_filestream(ap->ip))
2267                 args->type = XFS_ALLOCTYPE_NEAR_BNO;
2268         else
2269                 args->type = XFS_ALLOCTYPE_START_BNO;
2270         args->total = ap->total;
2271
2272         /*
2273          * Search for an allocation group with a single extent large enough
2274          * for the request.  If one isn't found, then adjust the minimum
2275          * allocation size to the largest space found.
2276          */
2277         startag = ag = XFS_FSB_TO_AGNO(mp, args->fsbno);
2278         if (startag == NULLAGNUMBER)
2279                 startag = ag = 0;
2280
2281         pag = xfs_perag_get(mp, ag);
2282         while (*blen < args->maxlen) {
2283                 if (!pag->pagf_init) {
2284                         error = xfs_alloc_pagf_init(mp, args->tp, ag,
2285                                                     XFS_ALLOC_FLAG_TRYLOCK);
2286                         if (error) {
2287                                 xfs_perag_put(pag);
2288                                 return error;
2289                         }
2290                 }
2291
2292                 /*
2293                  * See xfs_alloc_fix_freelist...
2294                  */
2295                 if (pag->pagf_init) {
2296                         xfs_extlen_t    longest;
2297                         longest = xfs_alloc_longest_free_extent(mp, pag);
2298                         if (*blen < longest)
2299                                 *blen = longest;
2300                 } else
2301                         notinit = 1;
2302
2303                 if (xfs_inode_is_filestream(ap->ip)) {
2304                         if (*blen >= args->maxlen)
2305                                 break;
2306
2307                         if (ap->userdata) {
2308                                 /*
2309                                  * If startag is an invalid AG, we've
2310                                  * come here once before and
2311                                  * xfs_filestream_new_ag picked the
2312                                  * best currently available.
2313                                  *
2314                                  * Don't continue looping, since we
2315                                  * could loop forever.
2316                                  */
2317                                 if (startag == NULLAGNUMBER)
2318                                         break;
2319
2320                                 error = xfs_filestream_new_ag(ap, &ag);
2321                                 xfs_perag_put(pag);
2322                                 if (error)
2323                                         return error;
2324
2325                                 /* loop again to set 'blen'*/
2326                                 startag = NULLAGNUMBER;
2327                                 pag = xfs_perag_get(mp, ag);
2328                                 continue;
2329                         }
2330                 }
2331                 if (++ag == mp->m_sb.sb_agcount)
2332                         ag = 0;
2333                 if (ag == startag)
2334                         break;
2335                 xfs_perag_put(pag);
2336                 pag = xfs_perag_get(mp, ag);
2337         }
2338         xfs_perag_put(pag);
2339
2340         /*
2341          * Since the above loop did a BUF_TRYLOCK, it is
2342          * possible that there is space for this request.
2343          */
2344         if (notinit || *blen < ap->minlen)
2345                 args->minlen = ap->minlen;
2346         /*
2347          * If the best seen length is less than the request
2348          * length, use the best as the minimum.
2349          */
2350         else if (*blen < args->maxlen)
2351                 args->minlen = *blen;
2352         /*
2353          * Otherwise we've seen an extent as big as maxlen,
2354          * use that as the minimum.
2355          */
2356         else
2357                 args->minlen = args->maxlen;
2358
2359         /*
2360          * set the failure fallback case to look in the selected
2361          * AG as the stream may have moved.
2362          */
2363         if (xfs_inode_is_filestream(ap->ip))
2364                 ap->blkno = args->fsbno = XFS_AGB_TO_FSB(mp, ag, 0);
2365
2366         return 0;
2367 }
2368
2369 STATIC int
2370 xfs_bmap_btalloc(
2371         xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
2372 {
2373         xfs_mount_t     *mp;            /* mount point structure */
2374         xfs_alloctype_t atype = 0;      /* type for allocation routines */
2375         xfs_extlen_t    align;          /* minimum allocation alignment */
2376         xfs_agnumber_t  fb_agno;        /* ag number of ap->firstblock */
2377         xfs_agnumber_t  ag;
2378         xfs_alloc_arg_t args;
2379         xfs_extlen_t    blen;
2380         xfs_extlen_t    nextminlen = 0;
2381         int             nullfb;         /* true if ap->firstblock isn't set */
2382         int             isaligned;
2383         int             tryagain;
2384         int             error;
2385
2386         ASSERT(ap->length);
2387
2388         mp = ap->ip->i_mount;
2389         align = ap->userdata ? xfs_get_extsz_hint(ap->ip) : 0;
2390         if (unlikely(align)) {
2391                 error = xfs_bmap_extsize_align(mp, &ap->got, &ap->prev,
2392                                                 align, 0, ap->eof, 0, ap->conv,
2393                                                 &ap->offset, &ap->length);
2394                 ASSERT(!error);
2395                 ASSERT(ap->length);
2396         }
2397         nullfb = *ap->firstblock == NULLFSBLOCK;
2398         fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, *ap->firstblock);
2399         if (nullfb) {
2400                 if (ap->userdata && xfs_inode_is_filestream(ap->ip)) {
2401                         ag = xfs_filestream_lookup_ag(ap->ip);
2402                         ag = (ag != NULLAGNUMBER) ? ag : 0;
2403                         ap->blkno = XFS_AGB_TO_FSB(mp, ag, 0);
2404                 } else {
2405                         ap->blkno = XFS_INO_TO_FSB(mp, ap->ip->i_ino);
2406                 }
2407         } else
2408                 ap->blkno = *ap->firstblock;
2409
2410         xfs_bmap_adjacent(ap);
2411
2412         /*
2413          * If allowed, use ap->blkno; otherwise must use firstblock since
2414          * it's in the right allocation group.
2415          */
2416         if (nullfb || XFS_FSB_TO_AGNO(mp, ap->blkno) == fb_agno)
2417                 ;
2418         else
2419                 ap->blkno = *ap->firstblock;
2420         /*
2421          * Normal allocation, done through xfs_alloc_vextent.
2422          */
2423         tryagain = isaligned = 0;
2424         args.tp = ap->tp;
2425         args.mp = mp;
2426         args.fsbno = ap->blkno;
2427
2428         /* Trim the allocation back to the maximum an AG can fit. */
2429         args.maxlen = MIN(ap->length, XFS_ALLOC_AG_MAX_USABLE(mp));
2430         args.firstblock = *ap->firstblock;
2431         blen = 0;
2432         if (nullfb) {
2433                 error = xfs_bmap_btalloc_nullfb(ap, &args, &blen);
2434                 if (error)
2435                         return error;
2436         } else if (ap->flist->xbf_low) {
2437                 if (xfs_inode_is_filestream(ap->ip))
2438                         args.type = XFS_ALLOCTYPE_FIRST_AG;
2439                 else
2440                         args.type = XFS_ALLOCTYPE_START_BNO;
2441                 args.total = args.minlen = ap->minlen;
2442         } else {
2443                 args.type = XFS_ALLOCTYPE_NEAR_BNO;
2444                 args.total = ap->total;
2445                 args.minlen = ap->minlen;
2446         }
2447         /* apply extent size hints if obtained earlier */
2448         if (unlikely(align)) {
2449                 args.prod = align;
2450                 if ((args.mod = (xfs_extlen_t)do_mod(ap->offset, args.prod)))
2451                         args.mod = (xfs_extlen_t)(args.prod - args.mod);
2452         } else if (mp->m_sb.sb_blocksize >= PAGE_CACHE_SIZE) {
2453                 args.prod = 1;
2454                 args.mod = 0;
2455         } else {
2456                 args.prod = PAGE_CACHE_SIZE >> mp->m_sb.sb_blocklog;
2457                 if ((args.mod = (xfs_extlen_t)(do_mod(ap->offset, args.prod))))
2458                         args.mod = (xfs_extlen_t)(args.prod - args.mod);
2459         }
2460         /*
2461          * If we are not low on available data blocks, and the
2462          * underlying logical volume manager is a stripe, and
2463          * the file offset is zero then try to allocate data
2464          * blocks on stripe unit boundary.
2465          * NOTE: ap->aeof is only set if the allocation length
2466          * is >= the stripe unit and the allocation offset is
2467          * at the end of file.
2468          */
2469         if (!ap->flist->xbf_low && ap->aeof) {
2470                 if (!ap->offset) {
2471                         args.alignment = mp->m_dalign;
2472                         atype = args.type;
2473                         isaligned = 1;
2474                         /*
2475                          * Adjust for alignment
2476                          */
2477                         if (blen > args.alignment && blen <= args.maxlen)
2478                                 args.minlen = blen - args.alignment;
2479                         args.minalignslop = 0;
2480                 } else {
2481                         /*
2482                          * First try an exact bno allocation.
2483                          * If it fails then do a near or start bno
2484                          * allocation with alignment turned on.
2485                          */
2486                         atype = args.type;
2487                         tryagain = 1;
2488                         args.type = XFS_ALLOCTYPE_THIS_BNO;
2489                         args.alignment = 1;
2490                         /*
2491                          * Compute the minlen+alignment for the
2492                          * next case.  Set slop so that the value
2493                          * of minlen+alignment+slop doesn't go up
2494                          * between the calls.
2495                          */
2496                         if (blen > mp->m_dalign && blen <= args.maxlen)
2497                                 nextminlen = blen - mp->m_dalign;
2498                         else
2499                                 nextminlen = args.minlen;
2500                         if (nextminlen + mp->m_dalign > args.minlen + 1)
2501                                 args.minalignslop =
2502                                         nextminlen + mp->m_dalign -
2503                                         args.minlen - 1;
2504                         else
2505                                 args.minalignslop = 0;
2506                 }
2507         } else {
2508                 args.alignment = 1;
2509                 args.minalignslop = 0;
2510         }
2511         args.minleft = ap->minleft;
2512         args.wasdel = ap->wasdel;
2513         args.isfl = 0;
2514         args.userdata = ap->userdata;
2515         if ((error = xfs_alloc_vextent(&args)))
2516                 return error;
2517         if (tryagain && args.fsbno == NULLFSBLOCK) {
2518                 /*
2519                  * Exact allocation failed. Now try with alignment
2520                  * turned on.
2521                  */
2522                 args.type = atype;
2523                 args.fsbno = ap->blkno;
2524                 args.alignment = mp->m_dalign;
2525                 args.minlen = nextminlen;
2526                 args.minalignslop = 0;
2527                 isaligned = 1;
2528                 if ((error = xfs_alloc_vextent(&args)))
2529                         return error;
2530         }
2531         if (isaligned && args.fsbno == NULLFSBLOCK) {
2532                 /*
2533                  * allocation failed, so turn off alignment and
2534                  * try again.
2535                  */
2536                 args.type = atype;
2537                 args.fsbno = ap->blkno;
2538                 args.alignment = 0;
2539                 if ((error = xfs_alloc_vextent(&args)))
2540                         return error;
2541         }
2542         if (args.fsbno == NULLFSBLOCK && nullfb &&
2543             args.minlen > ap->minlen) {
2544                 args.minlen = ap->minlen;
2545                 args.type = XFS_ALLOCTYPE_START_BNO;
2546                 args.fsbno = ap->blkno;
2547                 if ((error = xfs_alloc_vextent(&args)))
2548                         return error;
2549         }
2550         if (args.fsbno == NULLFSBLOCK && nullfb) {
2551                 args.fsbno = 0;
2552                 args.type = XFS_ALLOCTYPE_FIRST_AG;
2553                 args.total = ap->minlen;
2554                 args.minleft = 0;
2555                 if ((error = xfs_alloc_vextent(&args)))
2556                         return error;
2557                 ap->flist->xbf_low = 1;
2558         }
2559         if (args.fsbno != NULLFSBLOCK) {
2560                 /*
2561                  * check the allocation happened at the same or higher AG than
2562                  * the first block that was allocated.
2563                  */
2564                 ASSERT(*ap->firstblock == NULLFSBLOCK ||
2565                        XFS_FSB_TO_AGNO(mp, *ap->firstblock) ==
2566                        XFS_FSB_TO_AGNO(mp, args.fsbno) ||
2567                        (ap->flist->xbf_low &&
2568                         XFS_FSB_TO_AGNO(mp, *ap->firstblock) <
2569                         XFS_FSB_TO_AGNO(mp, args.fsbno)));
2570
2571                 ap->blkno = args.fsbno;
2572                 if (*ap->firstblock == NULLFSBLOCK)
2573                         *ap->firstblock = args.fsbno;
2574                 ASSERT(nullfb || fb_agno == args.agno ||
2575                        (ap->flist->xbf_low && fb_agno < args.agno));
2576                 ap->length = args.len;
2577                 ap->ip->i_d.di_nblocks += args.len;
2578                 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
2579                 if (ap->wasdel)
2580                         ap->ip->i_delayed_blks -= args.len;
2581                 /*
2582                  * Adjust the disk quota also. This was reserved
2583                  * earlier.
2584                  */
2585                 xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
2586                         ap->wasdel ? XFS_TRANS_DQ_DELBCOUNT :
2587                                         XFS_TRANS_DQ_BCOUNT,
2588                         (long) args.len);
2589         } else {
2590                 ap->blkno = NULLFSBLOCK;
2591                 ap->length = 0;
2592         }
2593         return 0;
2594 }
2595
2596 /*
2597  * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
2598  * It figures out where to ask the underlying allocator to put the new extent.
2599  */
2600 STATIC int
2601 xfs_bmap_alloc(
2602         xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
2603 {
2604         if (XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata)
2605                 return xfs_bmap_rtalloc(ap);
2606         return xfs_bmap_btalloc(ap);
2607 }
2608
2609 /*
2610  * Transform a btree format file with only one leaf node, where the
2611  * extents list will fit in the inode, into an extents format file.
2612  * Since the file extents are already in-core, all we have to do is
2613  * give up the space for the btree root and pitch the leaf block.
2614  */
2615 STATIC int                              /* error */
2616 xfs_bmap_btree_to_extents(
2617         xfs_trans_t             *tp,    /* transaction pointer */
2618         xfs_inode_t             *ip,    /* incore inode pointer */
2619         xfs_btree_cur_t         *cur,   /* btree cursor */
2620         int                     *logflagsp, /* inode logging flags */
2621         int                     whichfork)  /* data or attr fork */
2622 {
2623         /* REFERENCED */
2624         struct xfs_btree_block  *cblock;/* child btree block */
2625         xfs_fsblock_t           cbno;   /* child block number */
2626         xfs_buf_t               *cbp;   /* child block's buffer */
2627         int                     error;  /* error return value */
2628         xfs_ifork_t             *ifp;   /* inode fork data */
2629         xfs_mount_t             *mp;    /* mount point structure */
2630         __be64                  *pp;    /* ptr to block address */
2631         struct xfs_btree_block  *rblock;/* root btree block */
2632
2633         mp = ip->i_mount;
2634         ifp = XFS_IFORK_PTR(ip, whichfork);
2635         ASSERT(ifp->if_flags & XFS_IFEXTENTS);
2636         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
2637         rblock = ifp->if_broot;
2638         ASSERT(be16_to_cpu(rblock->bb_level) == 1);
2639         ASSERT(be16_to_cpu(rblock->bb_numrecs) == 1);
2640         ASSERT(xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0) == 1);
2641         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, rblock, 1, ifp->if_broot_bytes);
2642         cbno = be64_to_cpu(*pp);
2643         *logflagsp = 0;
2644 #ifdef DEBUG
2645         if ((error = xfs_btree_check_lptr(cur, cbno, 1)))
2646                 return error;
2647 #endif
2648         if ((error = xfs_btree_read_bufl(mp, tp, cbno, 0, &cbp,
2649                         XFS_BMAP_BTREE_REF)))
2650                 return error;
2651         cblock = XFS_BUF_TO_BLOCK(cbp);
2652         if ((error = xfs_btree_check_block(cur, cblock, 0, cbp)))
2653                 return error;
2654         xfs_bmap_add_free(cbno, 1, cur->bc_private.b.flist, mp);
2655         ip->i_d.di_nblocks--;
2656         xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, -1L);
2657         xfs_trans_binval(tp, cbp);
2658         if (cur->bc_bufs[0] == cbp)
2659                 cur->bc_bufs[0] = NULL;
2660         xfs_iroot_realloc(ip, -1, whichfork);
2661         ASSERT(ifp->if_broot == NULL);
2662         ASSERT((ifp->if_flags & XFS_IFBROOT) == 0);
2663         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
2664         *logflagsp = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
2665         return 0;
2666 }
2667
2668 /*
2669  * Called by xfs_bmapi to update file extent records and the btree
2670  * after removing space (or undoing a delayed allocation).
2671  */
2672 STATIC int                              /* error */
2673 xfs_bmap_del_extent(
2674         xfs_inode_t             *ip,    /* incore inode pointer */
2675         xfs_trans_t             *tp,    /* current transaction pointer */
2676         xfs_extnum_t            *idx,   /* extent number to update/delete */
2677         xfs_bmap_free_t         *flist, /* list of extents to be freed */
2678         xfs_btree_cur_t         *cur,   /* if null, not a btree */
2679         xfs_bmbt_irec_t         *del,   /* data to remove from extents */
2680         int                     *logflagsp, /* inode logging flags */
2681         int                     whichfork) /* data or attr fork */
2682 {
2683         xfs_filblks_t           da_new; /* new delay-alloc indirect blocks */
2684         xfs_filblks_t           da_old; /* old delay-alloc indirect blocks */
2685         xfs_fsblock_t           del_endblock=0; /* first block past del */
2686         xfs_fileoff_t           del_endoff;     /* first offset past del */
2687         int                     delay;  /* current block is delayed allocated */
2688         int                     do_fx;  /* free extent at end of routine */
2689         xfs_bmbt_rec_host_t     *ep;    /* current extent entry pointer */
2690         int                     error;  /* error return value */
2691         int                     flags;  /* inode logging flags */
2692         xfs_bmbt_irec_t         got;    /* current extent entry */
2693         xfs_fileoff_t           got_endoff;     /* first offset past got */
2694         int                     i;      /* temp state */
2695         xfs_ifork_t             *ifp;   /* inode fork pointer */
2696         xfs_mount_t             *mp;    /* mount structure */
2697         xfs_filblks_t           nblks;  /* quota/sb block count */
2698         xfs_bmbt_irec_t         new;    /* new record to be inserted */
2699         /* REFERENCED */
2700         uint                    qfield; /* quota field to update */
2701         xfs_filblks_t           temp;   /* for indirect length calculations */
2702         xfs_filblks_t           temp2;  /* for indirect length calculations */
2703         int                     state = 0;
2704
2705         XFS_STATS_INC(xs_del_exlist);
2706
2707         if (whichfork == XFS_ATTR_FORK)
2708                 state |= BMAP_ATTRFORK;
2709
2710         mp = ip->i_mount;
2711         ifp = XFS_IFORK_PTR(ip, whichfork);
2712         ASSERT((*idx >= 0) && (*idx < ifp->if_bytes /
2713                 (uint)sizeof(xfs_bmbt_rec_t)));
2714         ASSERT(del->br_blockcount > 0);
2715         ep = xfs_iext_get_ext(ifp, *idx);
2716         xfs_bmbt_get_all(ep, &got);
2717         ASSERT(got.br_startoff <= del->br_startoff);
2718         del_endoff = del->br_startoff + del->br_blockcount;
2719         got_endoff = got.br_startoff + got.br_blockcount;
2720         ASSERT(got_endoff >= del_endoff);
2721         delay = isnullstartblock(got.br_startblock);
2722         ASSERT(isnullstartblock(del->br_startblock) == delay);
2723         flags = 0;
2724         qfield = 0;
2725         error = 0;
2726         /*
2727          * If deleting a real allocation, must free up the disk space.
2728          */
2729         if (!delay) {
2730                 flags = XFS_ILOG_CORE;
2731                 /*
2732                  * Realtime allocation.  Free it and record di_nblocks update.
2733                  */
2734                 if (whichfork == XFS_DATA_FORK && XFS_IS_REALTIME_INODE(ip)) {
2735                         xfs_fsblock_t   bno;
2736                         xfs_filblks_t   len;
2737
2738                         ASSERT(do_mod(del->br_blockcount,
2739                                       mp->m_sb.sb_rextsize) == 0);
2740                         ASSERT(do_mod(del->br_startblock,
2741                                       mp->m_sb.sb_rextsize) == 0);
2742                         bno = del->br_startblock;
2743                         len = del->br_blockcount;
2744                         do_div(bno, mp->m_sb.sb_rextsize);
2745                         do_div(len, mp->m_sb.sb_rextsize);
2746                         error = xfs_rtfree_extent(tp, bno, (xfs_extlen_t)len);
2747                         if (error)
2748                                 goto done;
2749                         do_fx = 0;
2750                         nblks = len * mp->m_sb.sb_rextsize;
2751                         qfield = XFS_TRANS_DQ_RTBCOUNT;
2752                 }
2753                 /*
2754                  * Ordinary allocation.
2755                  */
2756                 else {
2757                         do_fx = 1;
2758                         nblks = del->br_blockcount;
2759                         qfield = XFS_TRANS_DQ_BCOUNT;
2760                 }
2761                 /*
2762                  * Set up del_endblock and cur for later.
2763                  */
2764                 del_endblock = del->br_startblock + del->br_blockcount;
2765                 if (cur) {
2766                         if ((error = xfs_bmbt_lookup_eq(cur, got.br_startoff,
2767                                         got.br_startblock, got.br_blockcount,
2768                                         &i)))
2769                                 goto done;
2770                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2771                 }
2772                 da_old = da_new = 0;
2773         } else {
2774                 da_old = startblockval(got.br_startblock);
2775                 da_new = 0;
2776                 nblks = 0;
2777                 do_fx = 0;
2778         }
2779         /*
2780          * Set flag value to use in switch statement.
2781          * Left-contig is 2, right-contig is 1.
2782          */
2783         switch (((got.br_startoff == del->br_startoff) << 1) |
2784                 (got_endoff == del_endoff)) {
2785         case 3:
2786                 /*
2787                  * Matches the whole extent.  Delete the entry.
2788                  */
2789                 xfs_iext_remove(ip, *idx, 1,
2790                                 whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0);
2791                 --*idx;
2792                 if (delay)
2793                         break;
2794
2795                 XFS_IFORK_NEXT_SET(ip, whichfork,
2796                         XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
2797                 flags |= XFS_ILOG_CORE;
2798                 if (!cur) {
2799                         flags |= xfs_ilog_fext(whichfork);
2800                         break;
2801                 }
2802                 if ((error = xfs_btree_delete(cur, &i)))
2803                         goto done;
2804                 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2805                 break;
2806
2807         case 2:
2808                 /*
2809                  * Deleting the first part of the extent.
2810                  */
2811                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2812                 xfs_bmbt_set_startoff(ep, del_endoff);
2813                 temp = got.br_blockcount - del->br_blockcount;
2814                 xfs_bmbt_set_blockcount(ep, temp);
2815                 if (delay) {
2816                         temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2817                                 da_old);
2818                         xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
2819                         trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2820                         da_new = temp;
2821                         break;
2822                 }
2823                 xfs_bmbt_set_startblock(ep, del_endblock);
2824                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2825                 if (!cur) {
2826                         flags |= xfs_ilog_fext(whichfork);
2827                         break;
2828                 }
2829                 if ((error = xfs_bmbt_update(cur, del_endoff, del_endblock,
2830                                 got.br_blockcount - del->br_blockcount,
2831                                 got.br_state)))
2832                         goto done;
2833                 break;
2834
2835         case 1:
2836                 /*
2837                  * Deleting the last part of the extent.
2838                  */
2839                 temp = got.br_blockcount - del->br_blockcount;
2840                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2841                 xfs_bmbt_set_blockcount(ep, temp);
2842                 if (delay) {
2843                         temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2844                                 da_old);
2845                         xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
2846                         trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2847                         da_new = temp;
2848                         break;
2849                 }
2850                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2851                 if (!cur) {
2852                         flags |= xfs_ilog_fext(whichfork);
2853                         break;
2854                 }
2855                 if ((error = xfs_bmbt_update(cur, got.br_startoff,
2856                                 got.br_startblock,
2857                                 got.br_blockcount - del->br_blockcount,
2858                                 got.br_state)))
2859                         goto done;
2860                 break;
2861
2862         case 0:
2863                 /*
2864                  * Deleting the middle of the extent.
2865                  */
2866                 temp = del->br_startoff - got.br_startoff;
2867                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2868                 xfs_bmbt_set_blockcount(ep, temp);
2869                 new.br_startoff = del_endoff;
2870                 temp2 = got_endoff - del_endoff;
2871                 new.br_blockcount = temp2;
2872                 new.br_state = got.br_state;
2873                 if (!delay) {
2874                         new.br_startblock = del_endblock;
2875                         flags |= XFS_ILOG_CORE;
2876                         if (cur) {
2877                                 if ((error = xfs_bmbt_update(cur,
2878                                                 got.br_startoff,
2879                                                 got.br_startblock, temp,
2880                                                 got.br_state)))
2881                                         goto done;
2882                                 if ((error = xfs_btree_increment(cur, 0, &i)))
2883                                         goto done;
2884                                 cur->bc_rec.b = new;
2885                                 error = xfs_btree_insert(cur, &i);
2886                                 if (error && error != ENOSPC)
2887                                         goto done;
2888                                 /*
2889                                  * If get no-space back from btree insert,
2890                                  * it tried a split, and we have a zero
2891                                  * block reservation.
2892                                  * Fix up our state and return the error.
2893                                  */
2894                                 if (error == ENOSPC) {
2895                                         /*
2896                                          * Reset the cursor, don't trust
2897                                          * it after any insert operation.
2898                                          */
2899                                         if ((error = xfs_bmbt_lookup_eq(cur,
2900                                                         got.br_startoff,
2901                                                         got.br_startblock,
2902                                                         temp, &i)))
2903                                                 goto done;
2904                                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2905                                         /*
2906                                          * Update the btree record back
2907                                          * to the original value.
2908                                          */
2909                                         if ((error = xfs_bmbt_update(cur,
2910                                                         got.br_startoff,
2911                                                         got.br_startblock,
2912                                                         got.br_blockcount,
2913                                                         got.br_state)))
2914                                                 goto done;
2915                                         /*
2916                                          * Reset the extent record back
2917                                          * to the original value.
2918                                          */
2919                                         xfs_bmbt_set_blockcount(ep,
2920                                                 got.br_blockcount);
2921                                         flags = 0;
2922                                         error = XFS_ERROR(ENOSPC);
2923                                         goto done;
2924                                 }
2925                                 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2926                         } else
2927                                 flags |= xfs_ilog_fext(whichfork);
2928                         XFS_IFORK_NEXT_SET(ip, whichfork,
2929                                 XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
2930                 } else {
2931                         ASSERT(whichfork == XFS_DATA_FORK);
2932                         temp = xfs_bmap_worst_indlen(ip, temp);
2933                         xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
2934                         temp2 = xfs_bmap_worst_indlen(ip, temp2);
2935                         new.br_startblock = nullstartblock((int)temp2);
2936                         da_new = temp + temp2;
2937                         while (da_new > da_old) {
2938                                 if (temp) {
2939                                         temp--;
2940                                         da_new--;
2941                                         xfs_bmbt_set_startblock(ep,
2942                                                 nullstartblock((int)temp));
2943                                 }
2944                                 if (da_new == da_old)
2945                                         break;
2946                                 if (temp2) {
2947                                         temp2--;
2948                                         da_new--;
2949                                         new.br_startblock =
2950                                                 nullstartblock((int)temp2);
2951                                 }
2952                         }
2953                 }
2954                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2955                 xfs_iext_insert(ip, *idx + 1, 1, &new, state);
2956                 ++*idx;
2957                 break;
2958         }
2959         /*
2960          * If we need to, add to list of extents to delete.
2961          */
2962         if (do_fx)
2963                 xfs_bmap_add_free(del->br_startblock, del->br_blockcount, flist,
2964                         mp);
2965         /*
2966          * Adjust inode # blocks in the file.
2967          */
2968         if (nblks)
2969                 ip->i_d.di_nblocks -= nblks;
2970         /*
2971          * Adjust quota data.
2972          */
2973         if (qfield)
2974                 xfs_trans_mod_dquot_byino(tp, ip, qfield, (long)-nblks);
2975
2976         /*
2977          * Account for change in delayed indirect blocks.
2978          * Nothing to do for disk quota accounting here.
2979          */
2980         ASSERT(da_old >= da_new);
2981         if (da_old > da_new) {
2982                 xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
2983                         (int64_t)(da_old - da_new), 0);
2984         }
2985 done:
2986         *logflagsp = flags;
2987         return error;
2988 }
2989
2990 /*
2991  * Remove the entry "free" from the free item list.  Prev points to the
2992  * previous entry, unless "free" is the head of the list.
2993  */
2994 STATIC void
2995 xfs_bmap_del_free(
2996         xfs_bmap_free_t         *flist, /* free item list header */
2997         xfs_bmap_free_item_t    *prev,  /* previous item on list, if any */
2998         xfs_bmap_free_item_t    *free)  /* list item to be freed */
2999 {
3000         if (prev)
3001                 prev->xbfi_next = free->xbfi_next;
3002         else
3003                 flist->xbf_first = free->xbfi_next;
3004         flist->xbf_count--;
3005         kmem_zone_free(xfs_bmap_free_item_zone, free);
3006 }
3007
3008 /*
3009  * Convert an extents-format file into a btree-format file.
3010  * The new file will have a root block (in the inode) and a single child block.
3011  */
3012 STATIC int                                      /* error */
3013 xfs_bmap_extents_to_btree(
3014         xfs_trans_t             *tp,            /* transaction pointer */
3015         xfs_inode_t             *ip,            /* incore inode pointer */
3016         xfs_fsblock_t           *firstblock,    /* first-block-allocated */
3017         xfs_bmap_free_t         *flist,         /* blocks freed in xaction */
3018         xfs_btree_cur_t         **curp,         /* cursor returned to caller */
3019         int                     wasdel,         /* converting a delayed alloc */
3020         int                     *logflagsp,     /* inode logging flags */
3021         int                     whichfork)      /* data or attr fork */
3022 {
3023         struct xfs_btree_block  *ablock;        /* allocated (child) bt block */
3024         xfs_buf_t               *abp;           /* buffer for ablock */
3025         xfs_alloc_arg_t         args;           /* allocation arguments */
3026         xfs_bmbt_rec_t          *arp;           /* child record pointer */
3027         struct xfs_btree_block  *block;         /* btree root block */
3028         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
3029         xfs_bmbt_rec_host_t     *ep;            /* extent record pointer */
3030         int                     error;          /* error return value */
3031         xfs_extnum_t            i, cnt;         /* extent record index */
3032         xfs_ifork_t             *ifp;           /* inode fork pointer */
3033         xfs_bmbt_key_t          *kp;            /* root block key pointer */
3034         xfs_mount_t             *mp;            /* mount structure */
3035         xfs_extnum_t            nextents;       /* number of file extents */
3036         xfs_bmbt_ptr_t          *pp;            /* root block address pointer */
3037
3038         ifp = XFS_IFORK_PTR(ip, whichfork);
3039         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS);
3040         ASSERT(ifp->if_ext_max ==
3041                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
3042         /*
3043          * Make space in the inode incore.
3044          */
3045         xfs_iroot_realloc(ip, 1, whichfork);
3046         ifp->if_flags |= XFS_IFBROOT;
3047
3048         /*
3049          * Fill in the root.
3050          */
3051         block = ifp->if_broot;
3052         block->bb_magic = cpu_to_be32(XFS_BMAP_MAGIC);
3053         block->bb_level = cpu_to_be16(1);
3054         block->bb_numrecs = cpu_to_be16(1);
3055         block->bb_u.l.bb_leftsib = cpu_to_be64(NULLDFSBNO);
3056         block->bb_u.l.bb_rightsib = cpu_to_be64(NULLDFSBNO);
3057
3058         /*
3059          * Need a cursor.  Can't allocate until bb_level is filled in.
3060          */
3061         mp = ip->i_mount;
3062         cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
3063         cur->bc_private.b.firstblock = *firstblock;
3064         cur->bc_private.b.flist = flist;
3065         cur->bc_private.b.flags = wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
3066         /*
3067          * Convert to a btree with two levels, one record in root.
3068          */
3069         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_BTREE);
3070         args.tp = tp;
3071         args.mp = mp;
3072         args.firstblock = *firstblock;
3073         if (*firstblock == NULLFSBLOCK) {
3074                 args.type = XFS_ALLOCTYPE_START_BNO;
3075                 args.fsbno = XFS_INO_TO_FSB(mp, ip->i_ino);
3076         } else if (flist->xbf_low) {
3077                 args.type = XFS_ALLOCTYPE_START_BNO;
3078                 args.fsbno = *firstblock;
3079         } else {
3080                 args.type = XFS_ALLOCTYPE_NEAR_BNO;
3081                 args.fsbno = *firstblock;
3082         }
3083         args.minlen = args.maxlen = args.prod = 1;
3084         args.total = args.minleft = args.alignment = args.mod = args.isfl =
3085                 args.minalignslop = 0;
3086         args.wasdel = wasdel;
3087         *logflagsp = 0;
3088         if ((error = xfs_alloc_vextent(&args))) {
3089                 xfs_iroot_realloc(ip, -1, whichfork);
3090                 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
3091                 return error;
3092         }
3093         /*
3094          * Allocation can't fail, the space was reserved.
3095          */
3096         ASSERT(args.fsbno != NULLFSBLOCK);
3097         ASSERT(*firstblock == NULLFSBLOCK ||
3098                args.agno == XFS_FSB_TO_AGNO(mp, *firstblock) ||
3099                (flist->xbf_low &&
3100                 args.agno > XFS_FSB_TO_AGNO(mp, *firstblock)));
3101         *firstblock = cur->bc_private.b.firstblock = args.fsbno;
3102         cur->bc_private.b.allocated++;
3103         ip->i_d.di_nblocks++;
3104         xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, 1L);
3105         abp = xfs_btree_get_bufl(mp, tp, args.fsbno, 0);
3106         /*
3107          * Fill in the child block.
3108          */
3109         ablock = XFS_BUF_TO_BLOCK(abp);
3110         ablock->bb_magic = cpu_to_be32(XFS_BMAP_MAGIC);
3111         ablock->bb_level = 0;
3112         ablock->bb_u.l.bb_leftsib = cpu_to_be64(NULLDFSBNO);
3113         ablock->bb_u.l.bb_rightsib = cpu_to_be64(NULLDFSBNO);
3114         arp = XFS_BMBT_REC_ADDR(mp, ablock, 1);
3115         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3116         for (cnt = i = 0; i < nextents; i++) {
3117                 ep = xfs_iext_get_ext(ifp, i);
3118                 if (!isnullstartblock(xfs_bmbt_get_startblock(ep))) {
3119                         arp->l0 = cpu_to_be64(ep->l0);
3120                         arp->l1 = cpu_to_be64(ep->l1);
3121                         arp++; cnt++;
3122                 }
3123         }
3124         ASSERT(cnt == XFS_IFORK_NEXTENTS(ip, whichfork));
3125         xfs_btree_set_numrecs(ablock, cnt);
3126
3127         /*
3128          * Fill in the root key and pointer.
3129          */
3130         kp = XFS_BMBT_KEY_ADDR(mp, block, 1);
3131         arp = XFS_BMBT_REC_ADDR(mp, ablock, 1);
3132         kp->br_startoff = cpu_to_be64(xfs_bmbt_disk_get_startoff(arp));
3133         pp = XFS_BMBT_PTR_ADDR(mp, block, 1, xfs_bmbt_get_maxrecs(cur,
3134                                                 be16_to_cpu(block->bb_level)));
3135         *pp = cpu_to_be64(args.fsbno);
3136
3137         /*
3138          * Do all this logging at the end so that
3139          * the root is at the right level.
3140          */
3141         xfs_btree_log_block(cur, abp, XFS_BB_ALL_BITS);
3142         xfs_btree_log_recs(cur, abp, 1, be16_to_cpu(ablock->bb_numrecs));
3143         ASSERT(*curp == NULL);
3144         *curp = cur;
3145         *logflagsp = XFS_ILOG_CORE | xfs_ilog_fbroot(whichfork);
3146         return 0;
3147 }
3148
3149 /*
3150  * Calculate the default attribute fork offset for newly created inodes.
3151  */
3152 uint
3153 xfs_default_attroffset(
3154         struct xfs_inode        *ip)
3155 {
3156         struct xfs_mount        *mp = ip->i_mount;
3157         uint                    offset;
3158
3159         if (mp->m_sb.sb_inodesize == 256) {
3160                 offset = XFS_LITINO(mp) -
3161                                 XFS_BMDR_SPACE_CALC(MINABTPTRS);
3162         } else {
3163                 offset = XFS_BMDR_SPACE_CALC(6 * MINABTPTRS);
3164         }
3165
3166         ASSERT(offset < XFS_LITINO(mp));
3167         return offset;
3168 }
3169
3170 /*
3171  * Helper routine to reset inode di_forkoff field when switching
3172  * attribute fork from local to extent format - we reset it where
3173  * possible to make space available for inline data fork extents.
3174  */
3175 STATIC void
3176 xfs_bmap_forkoff_reset(
3177         xfs_mount_t     *mp,
3178         xfs_inode_t     *ip,
3179         int             whichfork)
3180 {
3181         if (whichfork == XFS_ATTR_FORK &&
3182             ip->i_d.di_format != XFS_DINODE_FMT_DEV &&
3183             ip->i_d.di_format != XFS_DINODE_FMT_UUID &&
3184             ip->i_d.di_format != XFS_DINODE_FMT_BTREE) {
3185                 uint    dfl_forkoff = xfs_default_attroffset(ip) >> 3;
3186
3187                 if (dfl_forkoff > ip->i_d.di_forkoff) {
3188                         ip->i_d.di_forkoff = dfl_forkoff;
3189                         ip->i_df.if_ext_max =
3190                                 XFS_IFORK_DSIZE(ip) / sizeof(xfs_bmbt_rec_t);
3191                         ip->i_afp->if_ext_max =
3192                                 XFS_IFORK_ASIZE(ip) / sizeof(xfs_bmbt_rec_t);
3193                 }
3194         }
3195 }
3196
3197 /*
3198  * Convert a local file to an extents file.
3199  * This code is out of bounds for data forks of regular files,
3200  * since the file data needs to get logged so things will stay consistent.
3201  * (The bmap-level manipulations are ok, though).
3202  */
3203 STATIC int                              /* error */
3204 xfs_bmap_local_to_extents(
3205         xfs_trans_t     *tp,            /* transaction pointer */
3206         xfs_inode_t     *ip,            /* incore inode pointer */
3207         xfs_fsblock_t   *firstblock,    /* first block allocated in xaction */
3208         xfs_extlen_t    total,          /* total blocks needed by transaction */
3209         int             *logflagsp,     /* inode logging flags */
3210         int             whichfork)      /* data or attr fork */
3211 {
3212         int             error;          /* error return value */
3213         int             flags;          /* logging flags returned */
3214         xfs_ifork_t     *ifp;           /* inode fork pointer */
3215
3216         /*
3217          * We don't want to deal with the case of keeping inode data inline yet.
3218          * So sending the data fork of a regular inode is invalid.
3219          */
3220         ASSERT(!(S_ISREG(ip->i_d.di_mode) && whichfork == XFS_DATA_FORK));
3221         ifp = XFS_IFORK_PTR(ip, whichfork);
3222         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
3223         flags = 0;
3224         error = 0;
3225         if (ifp->if_bytes) {
3226                 xfs_alloc_arg_t args;   /* allocation arguments */
3227                 xfs_buf_t       *bp;    /* buffer for extent block */
3228                 xfs_bmbt_rec_host_t *ep;/* extent record pointer */
3229
3230                 args.tp = tp;
3231                 args.mp = ip->i_mount;
3232                 args.firstblock = *firstblock;
3233                 ASSERT((ifp->if_flags &
3234                         (XFS_IFINLINE|XFS_IFEXTENTS|XFS_IFEXTIREC)) == XFS_IFINLINE);
3235                 /*
3236                  * Allocate a block.  We know we need only one, since the
3237                  * file currently fits in an inode.
3238                  */
3239                 if (*firstblock == NULLFSBLOCK) {
3240                         args.fsbno = XFS_INO_TO_FSB(args.mp, ip->i_ino);
3241                         args.type = XFS_ALLOCTYPE_START_BNO;
3242                 } else {
3243                         args.fsbno = *firstblock;
3244                         args.type = XFS_ALLOCTYPE_NEAR_BNO;
3245                 }
3246                 args.total = total;
3247                 args.mod = args.minleft = args.alignment = args.wasdel =
3248                         args.isfl = args.minalignslop = 0;
3249                 args.minlen = args.maxlen = args.prod = 1;
3250                 if ((error = xfs_alloc_vextent(&args)))
3251                         goto done;
3252                 /*
3253                  * Can't fail, the space was reserved.
3254                  */
3255                 ASSERT(args.fsbno != NULLFSBLOCK);
3256                 ASSERT(args.len == 1);
3257                 *firstblock = args.fsbno;
3258                 bp = xfs_btree_get_bufl(args.mp, tp, args.fsbno, 0);
3259                 memcpy(bp->b_addr, ifp->if_u1.if_data, ifp->if_bytes);
3260                 xfs_trans_log_buf(tp, bp, 0, ifp->if_bytes - 1);
3261                 xfs_bmap_forkoff_reset(args.mp, ip, whichfork);
3262                 xfs_idata_realloc(ip, -ifp->if_bytes, whichfork);
3263                 xfs_iext_add(ifp, 0, 1);
3264                 ep = xfs_iext_get_ext(ifp, 0);
3265                 xfs_bmbt_set_allf(ep, 0, args.fsbno, 1, XFS_EXT_NORM);
3266                 trace_xfs_bmap_post_update(ip, 0,
3267                                 whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0,
3268                                 _THIS_IP_);
3269                 XFS_IFORK_NEXT_SET(ip, whichfork, 1);
3270                 ip->i_d.di_nblocks = 1;
3271                 xfs_trans_mod_dquot_byino(tp, ip,
3272                         XFS_TRANS_DQ_BCOUNT, 1L);
3273                 flags |= xfs_ilog_fext(whichfork);
3274         } else {
3275                 ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) == 0);
3276                 xfs_bmap_forkoff_reset(ip->i_mount, ip, whichfork);
3277         }
3278         ifp->if_flags &= ~XFS_IFINLINE;
3279         ifp->if_flags |= XFS_IFEXTENTS;
3280         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
3281         flags |= XFS_ILOG_CORE;
3282 done:
3283         *logflagsp = flags;
3284         return error;
3285 }
3286
3287 /*
3288  * Search the extent records for the entry containing block bno.
3289  * If bno lies in a hole, point to the next entry.  If bno lies
3290  * past eof, *eofp will be set, and *prevp will contain the last
3291  * entry (null if none).  Else, *lastxp will be set to the index
3292  * of the found entry; *gotp will contain the entry.
3293  */
3294 STATIC xfs_bmbt_rec_host_t *            /* pointer to found extent entry */
3295 xfs_bmap_search_multi_extents(
3296         xfs_ifork_t     *ifp,           /* inode fork pointer */
3297         xfs_fileoff_t   bno,            /* block number searched for */
3298         int             *eofp,          /* out: end of file found */
3299         xfs_extnum_t    *lastxp,        /* out: last extent index */
3300         xfs_bmbt_irec_t *gotp,          /* out: extent entry found */
3301         xfs_bmbt_irec_t *prevp)         /* out: previous extent entry found */
3302 {
3303         xfs_bmbt_rec_host_t *ep;                /* extent record pointer */
3304         xfs_extnum_t    lastx;          /* last extent index */
3305
3306         /*
3307          * Initialize the extent entry structure to catch access to
3308          * uninitialized br_startblock field.
3309          */
3310         gotp->br_startoff = 0xffa5a5a5a5a5a5a5LL;
3311         gotp->br_blockcount = 0xa55a5a5a5a5a5a5aLL;
3312         gotp->br_state = XFS_EXT_INVALID;
3313 #if XFS_BIG_BLKNOS
3314         gotp->br_startblock = 0xffffa5a5a5a5a5a5LL;
3315 #else
3316         gotp->br_startblock = 0xffffa5a5;
3317 #endif
3318         prevp->br_startoff = NULLFILEOFF;
3319
3320         ep = xfs_iext_bno_to_ext(ifp, bno, &lastx);
3321         if (lastx > 0) {
3322                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx - 1), prevp);
3323         }
3324         if (lastx < (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t))) {
3325                 xfs_bmbt_get_all(ep, gotp);
3326                 *eofp = 0;
3327         } else {
3328                 if (lastx > 0) {
3329                         *gotp = *prevp;
3330                 }
3331                 *eofp = 1;
3332                 ep = NULL;
3333         }
3334         *lastxp = lastx;
3335         return ep;
3336 }
3337
3338 /*
3339  * Search the extents list for the inode, for the extent containing bno.
3340  * If bno lies in a hole, point to the next entry.  If bno lies past eof,
3341  * *eofp will be set, and *prevp will contain the last entry (null if none).
3342  * Else, *lastxp will be set to the index of the found
3343  * entry; *gotp will contain the entry.
3344  */
3345 STATIC xfs_bmbt_rec_host_t *                 /* pointer to found extent entry */
3346 xfs_bmap_search_extents(
3347         xfs_inode_t     *ip,            /* incore inode pointer */
3348         xfs_fileoff_t   bno,            /* block number searched for */
3349         int             fork,           /* data or attr fork */
3350         int             *eofp,          /* out: end of file found */
3351         xfs_extnum_t    *lastxp,        /* out: last extent index */
3352         xfs_bmbt_irec_t *gotp,          /* out: extent entry found */
3353         xfs_bmbt_irec_t *prevp)         /* out: previous extent entry found */
3354 {
3355         xfs_ifork_t     *ifp;           /* inode fork pointer */
3356         xfs_bmbt_rec_host_t  *ep;            /* extent record pointer */
3357
3358         XFS_STATS_INC(xs_look_exlist);
3359         ifp = XFS_IFORK_PTR(ip, fork);
3360
3361         ep = xfs_bmap_search_multi_extents(ifp, bno, eofp, lastxp, gotp, prevp);
3362
3363         if (unlikely(!(gotp->br_startblock) && (*lastxp != NULLEXTNUM) &&
3364                      !(XFS_IS_REALTIME_INODE(ip) && fork == XFS_DATA_FORK))) {
3365                 xfs_alert_tag(ip->i_mount, XFS_PTAG_FSBLOCK_ZERO,
3366                                 "Access to block zero in inode %llu "
3367                                 "start_block: %llx start_off: %llx "
3368                                 "blkcnt: %llx extent-state: %x lastx: %x\n",
3369                         (unsigned long long)ip->i_ino,
3370                         (unsigned long long)gotp->br_startblock,
3371                         (unsigned long long)gotp->br_startoff,
3372                         (unsigned long long)gotp->br_blockcount,
3373                         gotp->br_state, *lastxp);
3374                 *lastxp = NULLEXTNUM;
3375                 *eofp = 1;
3376                 return NULL;
3377         }
3378         return ep;
3379 }
3380
3381 /*
3382  * Compute the worst-case number of indirect blocks that will be used
3383  * for ip's delayed extent of length "len".
3384  */
3385 STATIC xfs_filblks_t
3386 xfs_bmap_worst_indlen(
3387         xfs_inode_t     *ip,            /* incore inode pointer */
3388         xfs_filblks_t   len)            /* delayed extent length */
3389 {
3390         int             level;          /* btree level number */
3391         int             maxrecs;        /* maximum record count at this level */
3392         xfs_mount_t     *mp;            /* mount structure */
3393         xfs_filblks_t   rval;           /* return value */
3394
3395         mp = ip->i_mount;
3396         maxrecs = mp->m_bmap_dmxr[0];
3397         for (level = 0, rval = 0;
3398              level < XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK);
3399              level++) {
3400                 len += maxrecs - 1;
3401                 do_div(len, maxrecs);
3402                 rval += len;
3403                 if (len == 1)
3404                         return rval + XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK) -
3405                                 level - 1;
3406                 if (level == 0)
3407                         maxrecs = mp->m_bmap_dmxr[1];
3408         }
3409         return rval;
3410 }
3411
3412 /*
3413  * Convert inode from non-attributed to attributed.
3414  * Must not be in a transaction, ip must not be locked.
3415  */
3416 int                                             /* error code */
3417 xfs_bmap_add_attrfork(
3418         xfs_inode_t             *ip,            /* incore inode pointer */
3419         int                     size,           /* space new attribute needs */
3420         int                     rsvd)           /* xact may use reserved blks */
3421 {
3422         xfs_fsblock_t           firstblock;     /* 1st block/ag allocated */
3423         xfs_bmap_free_t         flist;          /* freed extent records */
3424         xfs_mount_t             *mp;            /* mount structure */
3425         xfs_trans_t             *tp;            /* transaction pointer */
3426         int                     blks;           /* space reservation */
3427         int                     version = 1;    /* superblock attr version */
3428         int                     committed;      /* xaction was committed */
3429         int                     logflags;       /* logging flags */
3430         int                     error;          /* error return value */
3431
3432         ASSERT(XFS_IFORK_Q(ip) == 0);
3433         ASSERT(ip->i_df.if_ext_max ==
3434                XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t));
3435
3436         mp = ip->i_mount;
3437         ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
3438         tp = xfs_trans_alloc(mp, XFS_TRANS_ADDAFORK);
3439         blks = XFS_ADDAFORK_SPACE_RES(mp);
3440         if (rsvd)
3441                 tp->t_flags |= XFS_TRANS_RESERVE;
3442         if ((error = xfs_trans_reserve(tp, blks, XFS_ADDAFORK_LOG_RES(mp), 0,
3443                         XFS_TRANS_PERM_LOG_RES, XFS_ADDAFORK_LOG_COUNT)))
3444                 goto error0;
3445         xfs_ilock(ip, XFS_ILOCK_EXCL);
3446         error = xfs_trans_reserve_quota_nblks(tp, ip, blks, 0, rsvd ?
3447                         XFS_QMOPT_RES_REGBLKS | XFS_QMOPT_FORCE_RES :
3448                         XFS_QMOPT_RES_REGBLKS);
3449         if (error) {
3450                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3451                 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES);
3452                 return error;
3453         }
3454         if (XFS_IFORK_Q(ip))
3455                 goto error1;
3456         if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS) {
3457                 /*
3458                  * For inodes coming from pre-6.2 filesystems.
3459                  */
3460                 ASSERT(ip->i_d.di_aformat == 0);
3461                 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
3462         }
3463         ASSERT(ip->i_d.di_anextents == 0);
3464
3465         xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
3466         xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
3467
3468         switch (ip->i_d.di_format) {
3469         case XFS_DINODE_FMT_DEV:
3470                 ip->i_d.di_forkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
3471                 break;
3472         case XFS_DINODE_FMT_UUID:
3473                 ip->i_d.di_forkoff = roundup(sizeof(uuid_t), 8) >> 3;
3474                 break;
3475         case XFS_DINODE_FMT_LOCAL:
3476         case XFS_DINODE_FMT_EXTENTS:
3477         case XFS_DINODE_FMT_BTREE:
3478                 ip->i_d.di_forkoff = xfs_attr_shortform_bytesfit(ip, size);
3479                 if (!ip->i_d.di_forkoff)
3480                         ip->i_d.di_forkoff = xfs_default_attroffset(ip) >> 3;
3481                 else if (mp->m_flags & XFS_MOUNT_ATTR2)
3482                         version = 2;
3483                 break;
3484         default:
3485                 ASSERT(0);
3486                 error = XFS_ERROR(EINVAL);
3487                 goto error1;
3488         }
3489         ip->i_df.if_ext_max =
3490                 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
3491         ASSERT(ip->i_afp == NULL);
3492         ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP);
3493         ip->i_afp->if_ext_max =
3494                 XFS_IFORK_ASIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
3495         ip->i_afp->if_flags = XFS_IFEXTENTS;
3496         logflags = 0;
3497         xfs_bmap_init(&flist, &firstblock);
3498         switch (ip->i_d.di_format) {
3499         case XFS_DINODE_FMT_LOCAL:
3500                 error = xfs_bmap_add_attrfork_local(tp, ip, &firstblock, &flist,
3501                         &logflags);
3502                 break;
3503         case XFS_DINODE_FMT_EXTENTS:
3504                 error = xfs_bmap_add_attrfork_extents(tp, ip, &firstblock,
3505                         &flist, &logflags);
3506                 break;
3507         case XFS_DINODE_FMT_BTREE:
3508                 error = xfs_bmap_add_attrfork_btree(tp, ip, &firstblock, &flist,
3509                         &logflags);
3510                 break;
3511         default:
3512                 error = 0;
3513                 break;
3514         }
3515         if (logflags)
3516                 xfs_trans_log_inode(tp, ip, logflags);
3517         if (error)
3518                 goto error2;
3519         if (!xfs_sb_version_hasattr(&mp->m_sb) ||
3520            (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2)) {
3521                 __int64_t sbfields = 0;
3522
3523                 spin_lock(&mp->m_sb_lock);
3524                 if (!xfs_sb_version_hasattr(&mp->m_sb)) {
3525                         xfs_sb_version_addattr(&mp->m_sb);
3526                         sbfields |= XFS_SB_VERSIONNUM;
3527                 }
3528                 if (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2) {
3529                         xfs_sb_version_addattr2(&mp->m_sb);
3530                         sbfields |= (XFS_SB_VERSIONNUM | XFS_SB_FEATURES2);
3531                 }
3532                 if (sbfields) {
3533                         spin_unlock(&mp->m_sb_lock);
3534                         xfs_mod_sb(tp, sbfields);
3535                 } else
3536                         spin_unlock(&mp->m_sb_lock);
3537         }
3538         if ((error = xfs_bmap_finish(&tp, &flist, &committed)))
3539                 goto error2;
3540         error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
3541         ASSERT(ip->i_df.if_ext_max ==
3542                XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t));
3543         return error;
3544 error2:
3545         xfs_bmap_cancel(&flist);
3546 error1:
3547         xfs_iunlock(ip, XFS_ILOCK_EXCL);
3548 error0:
3549         xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
3550         ASSERT(ip->i_df.if_ext_max ==
3551                XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t));
3552         return error;
3553 }
3554
3555 /*
3556  * Add the extent to the list of extents to be free at transaction end.
3557  * The list is maintained sorted (by block number).
3558  */
3559 /* ARGSUSED */
3560 void
3561 xfs_bmap_add_free(
3562         xfs_fsblock_t           bno,            /* fs block number of extent */
3563         xfs_filblks_t           len,            /* length of extent */
3564         xfs_bmap_free_t         *flist,         /* list of extents */
3565         xfs_mount_t             *mp)            /* mount point structure */
3566 {
3567         xfs_bmap_free_item_t    *cur;           /* current (next) element */
3568         xfs_bmap_free_item_t    *new;           /* new element */
3569         xfs_bmap_free_item_t    *prev;          /* previous element */
3570 #ifdef DEBUG
3571         xfs_agnumber_t          agno;
3572         xfs_agblock_t           agbno;
3573
3574         ASSERT(bno != NULLFSBLOCK);
3575         ASSERT(len > 0);
3576         ASSERT(len <= MAXEXTLEN);
3577         ASSERT(!isnullstartblock(bno));
3578         agno = XFS_FSB_TO_AGNO(mp, bno);
3579         agbno = XFS_FSB_TO_AGBNO(mp, bno);
3580         ASSERT(agno < mp->m_sb.sb_agcount);
3581         ASSERT(agbno < mp->m_sb.sb_agblocks);
3582         ASSERT(len < mp->m_sb.sb_agblocks);
3583         ASSERT(agbno + len <= mp->m_sb.sb_agblocks);
3584 #endif
3585         ASSERT(xfs_bmap_free_item_zone != NULL);
3586         new = kmem_zone_alloc(xfs_bmap_free_item_zone, KM_SLEEP);
3587         new->xbfi_startblock = bno;
3588         new->xbfi_blockcount = (xfs_extlen_t)len;
3589         for (prev = NULL, cur = flist->xbf_first;
3590              cur != NULL;
3591              prev = cur, cur = cur->xbfi_next) {
3592                 if (cur->xbfi_startblock >= bno)
3593                         break;
3594         }
3595         if (prev)
3596                 prev->xbfi_next = new;
3597         else
3598                 flist->xbf_first = new;
3599         new->xbfi_next = cur;
3600         flist->xbf_count++;
3601 }
3602
3603 /*
3604  * Compute and fill in the value of the maximum depth of a bmap btree
3605  * in this filesystem.  Done once, during mount.
3606  */
3607 void
3608 xfs_bmap_compute_maxlevels(
3609         xfs_mount_t     *mp,            /* file system mount structure */
3610         int             whichfork)      /* data or attr fork */
3611 {
3612         int             level;          /* btree level */
3613         uint            maxblocks;      /* max blocks at this level */
3614         uint            maxleafents;    /* max leaf entries possible */
3615         int             maxrootrecs;    /* max records in root block */
3616         int             minleafrecs;    /* min records in leaf block */
3617         int             minnoderecs;    /* min records in node block */
3618         int             sz;             /* root block size */
3619
3620         /*
3621          * The maximum number of extents in a file, hence the maximum
3622          * number of leaf entries, is controlled by the type of di_nextents
3623          * (a signed 32-bit number, xfs_extnum_t), or by di_anextents
3624          * (a signed 16-bit number, xfs_aextnum_t).
3625          *
3626          * Note that we can no longer assume that if we are in ATTR1 that
3627          * the fork offset of all the inodes will be
3628          * (xfs_default_attroffset(ip) >> 3) because we could have mounted
3629          * with ATTR2 and then mounted back with ATTR1, keeping the
3630          * di_forkoff's fixed but probably at various positions. Therefore,
3631          * for both ATTR1 and ATTR2 we have to assume the worst case scenario
3632          * of a minimum size available.
3633          */
3634         if (whichfork == XFS_DATA_FORK) {
3635                 maxleafents = MAXEXTNUM;
3636                 sz = XFS_BMDR_SPACE_CALC(MINDBTPTRS);
3637         } else {
3638                 maxleafents = MAXAEXTNUM;
3639                 sz = XFS_BMDR_SPACE_CALC(MINABTPTRS);
3640         }
3641         maxrootrecs = xfs_bmdr_maxrecs(mp, sz, 0);
3642         minleafrecs = mp->m_bmap_dmnr[0];
3643         minnoderecs = mp->m_bmap_dmnr[1];
3644         maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs;
3645         for (level = 1; maxblocks > 1; level++) {
3646                 if (maxblocks <= maxrootrecs)
3647                         maxblocks = 1;
3648                 else
3649                         maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs;
3650         }
3651         mp->m_bm_maxlevels[whichfork] = level;
3652 }
3653
3654 /*
3655  * Routine to be called at transaction's end by xfs_bmapi, xfs_bunmapi
3656  * caller.  Frees all the extents that need freeing, which must be done
3657  * last due to locking considerations.  We never free any extents in
3658  * the first transaction.
3659  *
3660  * Return 1 if the given transaction was committed and a new one
3661  * started, and 0 otherwise in the committed parameter.
3662  */
3663 int                                             /* error */
3664 xfs_bmap_finish(
3665         xfs_trans_t             **tp,           /* transaction pointer addr */
3666         xfs_bmap_free_t         *flist,         /* i/o: list extents to free */
3667         int                     *committed)     /* xact committed or not */
3668 {
3669         xfs_efd_log_item_t      *efd;           /* extent free data */
3670         xfs_efi_log_item_t      *efi;           /* extent free intention */
3671         int                     error;          /* error return value */
3672         xfs_bmap_free_item_t    *free;          /* free extent item */
3673         unsigned int            logres;         /* new log reservation */
3674         unsigned int            logcount;       /* new log count */
3675         xfs_mount_t             *mp;            /* filesystem mount structure */
3676         xfs_bmap_free_item_t    *next;          /* next item on free list */
3677         xfs_trans_t             *ntp;           /* new transaction pointer */
3678
3679         ASSERT((*tp)->t_flags & XFS_TRANS_PERM_LOG_RES);
3680         if (flist->xbf_count == 0) {
3681                 *committed = 0;
3682                 return 0;
3683         }
3684         ntp = *tp;
3685         efi = xfs_trans_get_efi(ntp, flist->xbf_count);
3686         for (free = flist->xbf_first; free; free = free->xbfi_next)
3687                 xfs_trans_log_efi_extent(ntp, efi, free->xbfi_startblock,
3688                         free->xbfi_blockcount);
3689         logres = ntp->t_log_res;
3690         logcount = ntp->t_log_count;
3691         ntp = xfs_trans_dup(*tp);
3692         error = xfs_trans_commit(*tp, 0);
3693         *tp = ntp;
3694         *committed = 1;
3695         /*
3696          * We have a new transaction, so we should return committed=1,
3697          * even though we're returning an error.
3698          */
3699         if (error)
3700                 return error;
3701
3702         /*
3703          * transaction commit worked ok so we can drop the extra ticket
3704          * reference that we gained in xfs_trans_dup()
3705          */
3706         xfs_log_ticket_put(ntp->t_ticket);
3707
3708         if ((error = xfs_trans_reserve(ntp, 0, logres, 0, XFS_TRANS_PERM_LOG_RES,
3709                         logcount)))
3710                 return error;
3711         efd = xfs_trans_get_efd(ntp, efi, flist->xbf_count);
3712         for (free = flist->xbf_first; free != NULL; free = next) {
3713                 next = free->xbfi_next;
3714                 if ((error = xfs_free_extent(ntp, free->xbfi_startblock,
3715                                 free->xbfi_blockcount))) {
3716                         /*
3717                          * The bmap free list will be cleaned up at a
3718                          * higher level.  The EFI will be canceled when
3719                          * this transaction is aborted.
3720                          * Need to force shutdown here to make sure it
3721                          * happens, since this transaction may not be
3722                          * dirty yet.
3723                          */
3724                         mp = ntp->t_mountp;
3725                         if (!XFS_FORCED_SHUTDOWN(mp))
3726                                 xfs_force_shutdown(mp,
3727                                                    (error == EFSCORRUPTED) ?
3728                                                    SHUTDOWN_CORRUPT_INCORE :
3729                                                    SHUTDOWN_META_IO_ERROR);
3730                         return error;
3731                 }
3732                 xfs_trans_log_efd_extent(ntp, efd, free->xbfi_startblock,
3733                         free->xbfi_blockcount);
3734                 xfs_bmap_del_free(flist, NULL, free);
3735         }
3736         return 0;
3737 }
3738
3739 /*
3740  * Free up any items left in the list.
3741  */
3742 void
3743 xfs_bmap_cancel(
3744         xfs_bmap_free_t         *flist) /* list of bmap_free_items */
3745 {
3746         xfs_bmap_free_item_t    *free;  /* free list item */
3747         xfs_bmap_free_item_t    *next;
3748
3749         if (flist->xbf_count == 0)
3750                 return;
3751         ASSERT(flist->xbf_first != NULL);
3752         for (free = flist->xbf_first; free; free = next) {
3753                 next = free->xbfi_next;
3754                 xfs_bmap_del_free(flist, NULL, free);
3755         }
3756         ASSERT(flist->xbf_count == 0);
3757 }
3758
3759 /*
3760  * Returns the file-relative block number of the first unused block(s)
3761  * in the file with at least "len" logically contiguous blocks free.
3762  * This is the lowest-address hole if the file has holes, else the first block
3763  * past the end of file.
3764  * Return 0 if the file is currently local (in-inode).
3765  */
3766 int                                             /* error */
3767 xfs_bmap_first_unused(
3768         xfs_trans_t     *tp,                    /* transaction pointer */
3769         xfs_inode_t     *ip,                    /* incore inode */
3770         xfs_extlen_t    len,                    /* size of hole to find */
3771         xfs_fileoff_t   *first_unused,          /* unused block */
3772         int             whichfork)              /* data or attr fork */
3773 {
3774         int             error;                  /* error return value */
3775         int             idx;                    /* extent record index */
3776         xfs_ifork_t     *ifp;                   /* inode fork pointer */
3777         xfs_fileoff_t   lastaddr;               /* last block number seen */
3778         xfs_fileoff_t   lowest;                 /* lowest useful block */
3779         xfs_fileoff_t   max;                    /* starting useful block */
3780         xfs_fileoff_t   off;                    /* offset for this block */
3781         xfs_extnum_t    nextents;               /* number of extent entries */
3782
3783         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE ||
3784                XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ||
3785                XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
3786         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
3787                 *first_unused = 0;
3788                 return 0;
3789         }
3790         ifp = XFS_IFORK_PTR(ip, whichfork);
3791         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
3792             (error = xfs_iread_extents(tp, ip, whichfork)))
3793                 return error;
3794         lowest = *first_unused;
3795         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3796         for (idx = 0, lastaddr = 0, max = lowest; idx < nextents; idx++) {
3797                 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, idx);
3798                 off = xfs_bmbt_get_startoff(ep);
3799                 /*
3800                  * See if the hole before this extent will work.
3801                  */
3802                 if (off >= lowest + len && off - max >= len) {
3803                         *first_unused = max;
3804                         return 0;
3805                 }
3806                 lastaddr = off + xfs_bmbt_get_blockcount(ep);
3807                 max = XFS_FILEOFF_MAX(lastaddr, lowest);
3808         }
3809         *first_unused = max;
3810         return 0;
3811 }
3812
3813 /*
3814  * Returns the file-relative block number of the last block + 1 before
3815  * last_block (input value) in the file.
3816  * This is not based on i_size, it is based on the extent records.
3817  * Returns 0 for local files, as they do not have extent records.
3818  */
3819 int                                             /* error */
3820 xfs_bmap_last_before(
3821         xfs_trans_t     *tp,                    /* transaction pointer */
3822         xfs_inode_t     *ip,                    /* incore inode */
3823         xfs_fileoff_t   *last_block,            /* last block */
3824         int             whichfork)              /* data or attr fork */
3825 {
3826         xfs_fileoff_t   bno;                    /* input file offset */
3827         int             eof;                    /* hit end of file */
3828         xfs_bmbt_rec_host_t *ep;                /* pointer to last extent */
3829         int             error;                  /* error return value */
3830         xfs_bmbt_irec_t got;                    /* current extent value */
3831         xfs_ifork_t     *ifp;                   /* inode fork pointer */
3832         xfs_extnum_t    lastx;                  /* last extent used */
3833         xfs_bmbt_irec_t prev;                   /* previous extent value */
3834
3835         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
3836             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
3837             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL)
3838                return XFS_ERROR(EIO);
3839         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
3840                 *last_block = 0;
3841                 return 0;
3842         }
3843         ifp = XFS_IFORK_PTR(ip, whichfork);
3844         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
3845             (error = xfs_iread_extents(tp, ip, whichfork)))
3846                 return error;
3847         bno = *last_block - 1;
3848         ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
3849                 &prev);
3850         if (eof || xfs_bmbt_get_startoff(ep) > bno) {
3851                 if (prev.br_startoff == NULLFILEOFF)
3852                         *last_block = 0;
3853                 else
3854                         *last_block = prev.br_startoff + prev.br_blockcount;
3855         }
3856         /*
3857          * Otherwise *last_block is already the right answer.
3858          */
3859         return 0;
3860 }
3861
3862 STATIC int
3863 xfs_bmap_last_extent(
3864         struct xfs_trans        *tp,
3865         struct xfs_inode        *ip,
3866         int                     whichfork,
3867         struct xfs_bmbt_irec    *rec,
3868         int                     *is_empty)
3869 {
3870         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, whichfork);
3871         int                     error;
3872         int                     nextents;
3873
3874         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
3875                 error = xfs_iread_extents(tp, ip, whichfork);
3876                 if (error)
3877                         return error;
3878         }
3879
3880         nextents = ifp->if_bytes / sizeof(xfs_bmbt_rec_t);
3881         if (nextents == 0) {
3882                 *is_empty = 1;
3883                 return 0;
3884         }
3885
3886         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, nextents - 1), rec);
3887         *is_empty = 0;
3888         return 0;
3889 }
3890
3891 /*
3892  * Check the last inode extent to determine whether this allocation will result
3893  * in blocks being allocated at the end of the file. When we allocate new data
3894  * blocks at the end of the file which do not start at the previous data block,
3895  * we will try to align the new blocks at stripe unit boundaries.
3896  *
3897  * Returns 0 in bma->aeof if the file (fork) is empty as any new write will be
3898  * at, or past the EOF.
3899  */
3900 STATIC int
3901 xfs_bmap_isaeof(
3902         struct xfs_bmalloca     *bma,
3903         int                     whichfork)
3904 {
3905         struct xfs_bmbt_irec    rec;
3906         int                     is_empty;
3907         int                     error;
3908
3909         bma->aeof = 0;
3910         error = xfs_bmap_last_extent(NULL, bma->ip, whichfork, &rec,
3911                                      &is_empty);
3912         if (error || is_empty)
3913                 return error;
3914
3915         /*
3916          * Check if we are allocation or past the last extent, or at least into
3917          * the last delayed allocated extent.
3918          */
3919         bma->aeof = bma->offset >= rec.br_startoff + rec.br_blockcount ||
3920                 (bma->offset >= rec.br_startoff &&
3921                  isnullstartblock(rec.br_startblock));
3922         return 0;
3923 }
3924
3925 /*
3926  * Check if the endoff is outside the last extent. If so the caller will grow
3927  * the allocation to a stripe unit boundary.  All offsets are considered outside
3928  * the end of file for an empty fork, so 1 is returned in *eof in that case.
3929  */
3930 int
3931 xfs_bmap_eof(
3932         struct xfs_inode        *ip,
3933         xfs_fileoff_t           endoff,
3934         int                     whichfork,
3935         int                     *eof)
3936 {
3937         struct xfs_bmbt_irec    rec;
3938         int                     error;
3939
3940         error = xfs_bmap_last_extent(NULL, ip, whichfork, &rec, eof);
3941         if (error || *eof)
3942                 return error;
3943
3944         *eof = endoff >= rec.br_startoff + rec.br_blockcount;
3945         return 0;
3946 }
3947
3948 /*
3949  * Returns the file-relative block number of the first block past eof in
3950  * the file.  This is not based on i_size, it is based on the extent records.
3951  * Returns 0 for local files, as they do not have extent records.
3952  */
3953 int
3954 xfs_bmap_last_offset(
3955         struct xfs_trans        *tp,
3956         struct xfs_inode        *ip,
3957         xfs_fileoff_t           *last_block,
3958         int                     whichfork)
3959 {
3960         struct xfs_bmbt_irec    rec;
3961         int                     is_empty;
3962         int                     error;
3963
3964         *last_block = 0;
3965
3966         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL)
3967                 return 0;
3968
3969         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
3970             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
3971                return XFS_ERROR(EIO);
3972
3973         error = xfs_bmap_last_extent(NULL, ip, whichfork, &rec, &is_empty);
3974         if (error || is_empty)
3975                 return error;
3976
3977         *last_block = rec.br_startoff + rec.br_blockcount;
3978         return 0;
3979 }
3980
3981 /*
3982  * Returns whether the selected fork of the inode has exactly one
3983  * block or not.  For the data fork we check this matches di_size,
3984  * implying the file's range is 0..bsize-1.
3985  */
3986 int                                     /* 1=>1 block, 0=>otherwise */
3987 xfs_bmap_one_block(
3988         xfs_inode_t     *ip,            /* incore inode */
3989         int             whichfork)      /* data or attr fork */
3990 {
3991         xfs_bmbt_rec_host_t *ep;        /* ptr to fork's extent */
3992         xfs_ifork_t     *ifp;           /* inode fork pointer */
3993         int             rval;           /* return value */
3994         xfs_bmbt_irec_t s;              /* internal version of extent */
3995
3996 #ifndef DEBUG
3997         if (whichfork == XFS_DATA_FORK) {
3998                 return S_ISREG(ip->i_d.di_mode) ?
3999                         (ip->i_size == ip->i_mount->m_sb.sb_blocksize) :
4000                         (ip->i_d.di_size == ip->i_mount->m_sb.sb_blocksize);
4001         }
4002 #endif  /* !DEBUG */
4003         if (XFS_IFORK_NEXTENTS(ip, whichfork) != 1)
4004                 return 0;
4005         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
4006                 return 0;
4007         ifp = XFS_IFORK_PTR(ip, whichfork);
4008         ASSERT(ifp->if_flags & XFS_IFEXTENTS);
4009         ep = xfs_iext_get_ext(ifp, 0);
4010         xfs_bmbt_get_all(ep, &s);
4011         rval = s.br_startoff == 0 && s.br_blockcount == 1;
4012         if (rval && whichfork == XFS_DATA_FORK)
4013                 ASSERT(ip->i_size == ip->i_mount->m_sb.sb_blocksize);
4014         return rval;
4015 }
4016
4017 STATIC int
4018 xfs_bmap_sanity_check(
4019         struct xfs_mount        *mp,
4020         struct xfs_buf          *bp,
4021         int                     level)
4022 {
4023         struct xfs_btree_block  *block = XFS_BUF_TO_BLOCK(bp);
4024
4025         if (block->bb_magic != cpu_to_be32(XFS_BMAP_MAGIC) ||
4026             be16_to_cpu(block->bb_level) != level ||
4027             be16_to_cpu(block->bb_numrecs) == 0 ||
4028             be16_to_cpu(block->bb_numrecs) > mp->m_bmap_dmxr[level != 0])
4029                 return 0;
4030         return 1;
4031 }
4032
4033 /*
4034  * Read in the extents to if_extents.
4035  * All inode fields are set up by caller, we just traverse the btree
4036  * and copy the records in. If the file system cannot contain unwritten
4037  * extents, the records are checked for no "state" flags.
4038  */
4039 int                                     /* error */
4040 xfs_bmap_read_extents(
4041         xfs_trans_t             *tp,    /* transaction pointer */
4042         xfs_inode_t             *ip,    /* incore inode */
4043         int                     whichfork) /* data or attr fork */
4044 {
4045         struct xfs_btree_block  *block; /* current btree block */
4046         xfs_fsblock_t           bno;    /* block # of "block" */
4047         xfs_buf_t               *bp;    /* buffer for "block" */
4048         int                     error;  /* error return value */
4049         xfs_exntfmt_t           exntf;  /* XFS_EXTFMT_NOSTATE, if checking */
4050         xfs_extnum_t            i, j;   /* index into the extents list */
4051         xfs_ifork_t             *ifp;   /* fork structure */
4052         int                     level;  /* btree level, for checking */
4053         xfs_mount_t             *mp;    /* file system mount structure */
4054         __be64                  *pp;    /* pointer to block address */
4055         /* REFERENCED */
4056         xfs_extnum_t            room;   /* number of entries there's room for */
4057
4058         bno = NULLFSBLOCK;
4059         mp = ip->i_mount;
4060         ifp = XFS_IFORK_PTR(ip, whichfork);
4061         exntf = (whichfork != XFS_DATA_FORK) ? XFS_EXTFMT_NOSTATE :
4062                                         XFS_EXTFMT_INODE(ip);
4063         block = ifp->if_broot;
4064         /*
4065          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
4066          */
4067         level = be16_to_cpu(block->bb_level);
4068         ASSERT(level > 0);
4069         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
4070         bno = be64_to_cpu(*pp);
4071         ASSERT(bno != NULLDFSBNO);
4072         ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
4073         ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
4074         /*
4075          * Go down the tree until leaf level is reached, following the first
4076          * pointer (leftmost) at each level.
4077          */
4078         while (level-- > 0) {
4079                 if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
4080                                 XFS_BMAP_BTREE_REF)))
4081                         return error;
4082                 block = XFS_BUF_TO_BLOCK(bp);
4083                 XFS_WANT_CORRUPTED_GOTO(
4084                         xfs_bmap_sanity_check(mp, bp, level),
4085                         error0);
4086                 if (level == 0)
4087                         break;
4088                 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
4089                 bno = be64_to_cpu(*pp);
4090                 XFS_WANT_CORRUPTED_GOTO(XFS_FSB_SANITY_CHECK(mp, bno), error0);
4091                 xfs_trans_brelse(tp, bp);
4092         }
4093         /*
4094          * Here with bp and block set to the leftmost leaf node in the tree.
4095          */
4096         room = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4097         i = 0;
4098         /*
4099          * Loop over all leaf nodes.  Copy information to the extent records.
4100          */
4101         for (;;) {
4102                 xfs_bmbt_rec_t  *frp;
4103                 xfs_fsblock_t   nextbno;
4104                 xfs_extnum_t    num_recs;
4105                 xfs_extnum_t    start;
4106
4107                 num_recs = xfs_btree_get_numrecs(block);
4108                 if (unlikely(i + num_recs > room)) {
4109                         ASSERT(i + num_recs <= room);
4110                         xfs_warn(ip->i_mount,
4111                                 "corrupt dinode %Lu, (btree extents).",
4112                                 (unsigned long long) ip->i_ino);
4113                         XFS_CORRUPTION_ERROR("xfs_bmap_read_extents(1)",
4114                                 XFS_ERRLEVEL_LOW, ip->i_mount, block);
4115                         goto error0;
4116                 }
4117                 XFS_WANT_CORRUPTED_GOTO(
4118                         xfs_bmap_sanity_check(mp, bp, 0),
4119                         error0);
4120                 /*
4121                  * Read-ahead the next leaf block, if any.
4122                  */
4123                 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
4124                 if (nextbno != NULLFSBLOCK)
4125                         xfs_btree_reada_bufl(mp, nextbno, 1);
4126                 /*
4127                  * Copy records into the extent records.
4128                  */
4129                 frp = XFS_BMBT_REC_ADDR(mp, block, 1);
4130                 start = i;
4131                 for (j = 0; j < num_recs; j++, i++, frp++) {
4132                         xfs_bmbt_rec_host_t *trp = xfs_iext_get_ext(ifp, i);
4133                         trp->l0 = be64_to_cpu(frp->l0);
4134                         trp->l1 = be64_to_cpu(frp->l1);
4135                 }
4136                 if (exntf == XFS_EXTFMT_NOSTATE) {
4137                         /*
4138                          * Check all attribute bmap btree records and
4139                          * any "older" data bmap btree records for a
4140                          * set bit in the "extent flag" position.
4141                          */
4142                         if (unlikely(xfs_check_nostate_extents(ifp,
4143                                         start, num_recs))) {
4144                                 XFS_ERROR_REPORT("xfs_bmap_read_extents(2)",
4145                                                  XFS_ERRLEVEL_LOW,
4146                                                  ip->i_mount);
4147                                 goto error0;
4148                         }
4149                 }
4150                 xfs_trans_brelse(tp, bp);
4151                 bno = nextbno;
4152                 /*
4153                  * If we've reached the end, stop.
4154                  */
4155                 if (bno == NULLFSBLOCK)
4156                         break;
4157                 if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
4158                                 XFS_BMAP_BTREE_REF)))
4159                         return error;
4160                 block = XFS_BUF_TO_BLOCK(bp);
4161         }
4162         ASSERT(i == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)));
4163         ASSERT(i == XFS_IFORK_NEXTENTS(ip, whichfork));
4164         XFS_BMAP_TRACE_EXLIST(ip, i, whichfork);
4165         return 0;
4166 error0:
4167         xfs_trans_brelse(tp, bp);
4168         return XFS_ERROR(EFSCORRUPTED);
4169 }
4170
4171 #ifdef DEBUG
4172 /*
4173  * Add bmap trace insert entries for all the contents of the extent records.
4174  */
4175 void
4176 xfs_bmap_trace_exlist(
4177         xfs_inode_t     *ip,            /* incore inode pointer */
4178         xfs_extnum_t    cnt,            /* count of entries in the list */
4179         int             whichfork,      /* data or attr fork */
4180         unsigned long   caller_ip)
4181 {
4182         xfs_extnum_t    idx;            /* extent record index */
4183         xfs_ifork_t     *ifp;           /* inode fork pointer */
4184         int             state = 0;
4185
4186         if (whichfork == XFS_ATTR_FORK)
4187                 state |= BMAP_ATTRFORK;
4188
4189         ifp = XFS_IFORK_PTR(ip, whichfork);
4190         ASSERT(cnt == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)));
4191         for (idx = 0; idx < cnt; idx++)
4192                 trace_xfs_extlist(ip, idx, whichfork, caller_ip);
4193 }
4194
4195 /*
4196  * Validate that the bmbt_irecs being returned from bmapi are valid
4197  * given the callers original parameters.  Specifically check the
4198  * ranges of the returned irecs to ensure that they only extent beyond
4199  * the given parameters if the XFS_BMAPI_ENTIRE flag was set.
4200  */
4201 STATIC void
4202 xfs_bmap_validate_ret(
4203         xfs_fileoff_t           bno,
4204         xfs_filblks_t           len,
4205         int                     flags,
4206         xfs_bmbt_irec_t         *mval,
4207         int                     nmap,
4208         int                     ret_nmap)
4209 {
4210         int                     i;              /* index to map values */
4211
4212         ASSERT(ret_nmap <= nmap);
4213
4214         for (i = 0; i < ret_nmap; i++) {
4215                 ASSERT(mval[i].br_blockcount > 0);
4216                 if (!(flags & XFS_BMAPI_ENTIRE)) {
4217                         ASSERT(mval[i].br_startoff >= bno);
4218                         ASSERT(mval[i].br_blockcount <= len);
4219                         ASSERT(mval[i].br_startoff + mval[i].br_blockcount <=
4220                                bno + len);
4221                 } else {
4222                         ASSERT(mval[i].br_startoff < bno + len);
4223                         ASSERT(mval[i].br_startoff + mval[i].br_blockcount >
4224                                bno);
4225                 }
4226                 ASSERT(i == 0 ||
4227                        mval[i - 1].br_startoff + mval[i - 1].br_blockcount ==
4228                        mval[i].br_startoff);
4229                 ASSERT(mval[i].br_startblock != DELAYSTARTBLOCK &&
4230                        mval[i].br_startblock != HOLESTARTBLOCK);
4231                 ASSERT(mval[i].br_state == XFS_EXT_NORM ||
4232                        mval[i].br_state == XFS_EXT_UNWRITTEN);
4233         }
4234 }
4235 #endif /* DEBUG */
4236
4237
4238 /*
4239  * Trim the returned map to the required bounds
4240  */
4241 STATIC void
4242 xfs_bmapi_trim_map(
4243         struct xfs_bmbt_irec    *mval,
4244         struct xfs_bmbt_irec    *got,
4245         xfs_fileoff_t           *bno,
4246         xfs_filblks_t           len,
4247         xfs_fileoff_t           obno,
4248         xfs_fileoff_t           end,
4249         int                     n,
4250         int                     flags)
4251 {
4252         if ((flags & XFS_BMAPI_ENTIRE) ||
4253             got->br_startoff + got->br_blockcount <= obno) {
4254                 *mval = *got;
4255                 if (isnullstartblock(got->br_startblock))
4256                         mval->br_startblock = DELAYSTARTBLOCK;
4257                 return;
4258         }
4259
4260         if (obno > *bno)
4261                 *bno = obno;
4262         ASSERT((*bno >= obno) || (n == 0));
4263         ASSERT(*bno < end);
4264         mval->br_startoff = *bno;
4265         if (isnullstartblock(got->br_startblock))
4266                 mval->br_startblock = DELAYSTARTBLOCK;
4267         else
4268                 mval->br_startblock = got->br_startblock +
4269                                         (*bno - got->br_startoff);
4270         /*
4271          * Return the minimum of what we got and what we asked for for
4272          * the length.  We can use the len variable here because it is
4273          * modified below and we could have been there before coming
4274          * here if the first part of the allocation didn't overlap what
4275          * was asked for.
4276          */
4277         mval->br_blockcount = XFS_FILBLKS_MIN(end - *bno,
4278                         got->br_blockcount - (*bno - got->br_startoff));
4279         mval->br_state = got->br_state;
4280         ASSERT(mval->br_blockcount <= len);
4281         return;
4282 }
4283
4284 /*
4285  * Update and validate the extent map to return
4286  */
4287 STATIC void
4288 xfs_bmapi_update_map(
4289         struct xfs_bmbt_irec    **map,
4290         xfs_fileoff_t           *bno,
4291         xfs_filblks_t           *len,
4292         xfs_fileoff_t           obno,
4293         xfs_fileoff_t           end,
4294         int                     *n,
4295         int                     flags)
4296 {
4297         xfs_bmbt_irec_t *mval = *map;
4298
4299         ASSERT((flags & XFS_BMAPI_ENTIRE) ||
4300                ((mval->br_startoff + mval->br_blockcount) <= end));
4301         ASSERT((flags & XFS_BMAPI_ENTIRE) || (mval->br_blockcount <= *len) ||
4302                (mval->br_startoff < obno));
4303
4304         *bno = mval->br_startoff + mval->br_blockcount;
4305         *len = end - *bno;
4306         if (*n > 0 && mval->br_startoff == mval[-1].br_startoff) {
4307                 /* update previous map with new information */
4308                 ASSERT(mval->br_startblock == mval[-1].br_startblock);
4309                 ASSERT(mval->br_blockcount > mval[-1].br_blockcount);
4310                 ASSERT(mval->br_state == mval[-1].br_state);
4311                 mval[-1].br_blockcount = mval->br_blockcount;
4312                 mval[-1].br_state = mval->br_state;
4313         } else if (*n > 0 && mval->br_startblock != DELAYSTARTBLOCK &&
4314                    mval[-1].br_startblock != DELAYSTARTBLOCK &&
4315                    mval[-1].br_startblock != HOLESTARTBLOCK &&
4316                    mval->br_startblock == mval[-1].br_startblock +
4317                                           mval[-1].br_blockcount &&
4318                    ((flags & XFS_BMAPI_IGSTATE) ||
4319                         mval[-1].br_state == mval->br_state)) {
4320                 ASSERT(mval->br_startoff ==
4321                        mval[-1].br_startoff + mval[-1].br_blockcount);
4322                 mval[-1].br_blockcount += mval->br_blockcount;
4323         } else if (*n > 0 &&
4324                    mval->br_startblock == DELAYSTARTBLOCK &&
4325                    mval[-1].br_startblock == DELAYSTARTBLOCK &&
4326                    mval->br_startoff ==
4327                    mval[-1].br_startoff + mval[-1].br_blockcount) {
4328                 mval[-1].br_blockcount += mval->br_blockcount;
4329                 mval[-1].br_state = mval->br_state;
4330         } else if (!((*n == 0) &&
4331                      ((mval->br_startoff + mval->br_blockcount) <=
4332                       obno))) {
4333                 mval++;
4334                 (*n)++;
4335         }
4336         *map = mval;
4337 }
4338
4339 /*
4340  * Map file blocks to filesystem blocks without allocation.
4341  */
4342 int
4343 xfs_bmapi_read(
4344         struct xfs_inode        *ip,
4345         xfs_fileoff_t           bno,
4346         xfs_filblks_t           len,
4347         struct xfs_bmbt_irec    *mval,
4348         int                     *nmap,
4349         int                     flags)
4350 {
4351         struct xfs_mount        *mp = ip->i_mount;
4352         struct xfs_ifork        *ifp;
4353         struct xfs_bmbt_irec    got;
4354         struct xfs_bmbt_irec    prev;
4355         xfs_fileoff_t           obno;
4356         xfs_fileoff_t           end;
4357         xfs_extnum_t            lastx;
4358         int                     error;
4359         int                     eof;
4360         int                     n = 0;
4361         int                     whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4362                                                 XFS_ATTR_FORK : XFS_DATA_FORK;
4363
4364         ASSERT(*nmap >= 1);
4365         ASSERT(!(flags & ~(XFS_BMAPI_ATTRFORK|XFS_BMAPI_ENTIRE|
4366                            XFS_BMAPI_IGSTATE)));
4367
4368         if (unlikely(XFS_TEST_ERROR(
4369             (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4370              XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
4371              mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4372                 XFS_ERROR_REPORT("xfs_bmapi_read", XFS_ERRLEVEL_LOW, mp);
4373                 return XFS_ERROR(EFSCORRUPTED);
4374         }
4375
4376         if (XFS_FORCED_SHUTDOWN(mp))
4377                 return XFS_ERROR(EIO);
4378
4379         XFS_STATS_INC(xs_blk_mapr);
4380
4381         ifp = XFS_IFORK_PTR(ip, whichfork);
4382         ASSERT(ifp->if_ext_max ==
4383                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
4384
4385         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4386                 error = xfs_iread_extents(NULL, ip, whichfork);
4387                 if (error)
4388                         return error;
4389         }
4390
4391         xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got, &prev);
4392         end = bno + len;
4393         obno = bno;
4394
4395         while (bno < end && n < *nmap) {
4396                 /* Reading past eof, act as though there's a hole up to end. */
4397                 if (eof)
4398                         got.br_startoff = end;
4399                 if (got.br_startoff > bno) {
4400                         /* Reading in a hole.  */
4401                         mval->br_startoff = bno;
4402                         mval->br_startblock = HOLESTARTBLOCK;
4403                         mval->br_blockcount =
4404                                 XFS_FILBLKS_MIN(len, got.br_startoff - bno);
4405                         mval->br_state = XFS_EXT_NORM;
4406                         bno += mval->br_blockcount;
4407                         len -= mval->br_blockcount;
4408                         mval++;
4409                         n++;
4410                         continue;
4411                 }
4412
4413                 /* set up the extent map to return. */
4414                 xfs_bmapi_trim_map(mval, &got, &bno, len, obno, end, n, flags);
4415                 xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4416
4417                 /* If we're done, stop now. */
4418                 if (bno >= end || n >= *nmap)
4419                         break;
4420
4421                 /* Else go on to the next record. */
4422                 if (++lastx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t))
4423                         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx), &got);
4424                 else
4425                         eof = 1;
4426         }
4427         *nmap = n;
4428         return 0;
4429 }
4430
4431 STATIC int
4432 xfs_bmapi_reserve_delalloc(
4433         struct xfs_inode        *ip,
4434         xfs_fileoff_t           aoff,
4435         xfs_filblks_t           len,
4436         struct xfs_bmbt_irec    *got,
4437         struct xfs_bmbt_irec    *prev,
4438         xfs_extnum_t            *lastx,
4439         int                     eof)
4440 {
4441         struct xfs_mount        *mp = ip->i_mount;
4442         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
4443         xfs_extlen_t            alen;
4444         xfs_extlen_t            indlen;
4445         char                    rt = XFS_IS_REALTIME_INODE(ip);
4446         xfs_extlen_t            extsz;
4447         int                     error;
4448
4449         alen = XFS_FILBLKS_MIN(len, MAXEXTLEN);
4450         if (!eof)
4451                 alen = XFS_FILBLKS_MIN(alen, got->br_startoff - aoff);
4452
4453         /* Figure out the extent size, adjust alen */
4454         extsz = xfs_get_extsz_hint(ip);
4455         if (extsz) {
4456                 /*
4457                  * Make sure we don't exceed a single extent length when we
4458                  * align the extent by reducing length we are going to
4459                  * allocate by the maximum amount extent size aligment may
4460                  * require.
4461                  */
4462                 alen = XFS_FILBLKS_MIN(len, MAXEXTLEN - (2 * extsz - 1));
4463                 error = xfs_bmap_extsize_align(mp, got, prev, extsz, rt, eof,
4464                                                1, 0, &aoff, &alen);
4465                 ASSERT(!error);
4466         }
4467
4468         if (rt)
4469                 extsz = alen / mp->m_sb.sb_rextsize;
4470
4471         /*
4472          * Make a transaction-less quota reservation for delayed allocation
4473          * blocks.  This number gets adjusted later.  We return if we haven't
4474          * allocated blocks already inside this loop.
4475          */
4476         error = xfs_trans_reserve_quota_nblks(NULL, ip, (long)alen, 0,
4477                         rt ? XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4478         if (error)
4479                 return error;
4480
4481         /*
4482          * Split changing sb for alen and indlen since they could be coming
4483          * from different places.
4484          */
4485         indlen = (xfs_extlen_t)xfs_bmap_worst_indlen(ip, alen);
4486         ASSERT(indlen > 0);
4487
4488         if (rt) {
4489                 error = xfs_mod_incore_sb(mp, XFS_SBS_FREXTENTS,
4490                                           -((int64_t)extsz), 0);
4491         } else {
4492                 error = xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
4493                                                  -((int64_t)alen), 0);
4494         }
4495
4496         if (error)
4497                 goto out_unreserve_quota;
4498
4499         error = xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
4500                                          -((int64_t)indlen), 0);
4501         if (error)
4502                 goto out_unreserve_blocks;
4503
4504
4505         ip->i_delayed_blks += alen;
4506
4507         got->br_startoff = aoff;
4508         got->br_startblock = nullstartblock(indlen);
4509         got->br_blockcount = alen;
4510         got->br_state = XFS_EXT_NORM;
4511         xfs_bmap_add_extent_hole_delay(ip, lastx, got);
4512
4513         /*
4514          * Update our extent pointer, given that xfs_bmap_add_extent_hole_delay
4515          * might have merged it into one of the neighbouring ones.
4516          */
4517         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *lastx), got);
4518
4519         ASSERT(got->br_startoff <= aoff);
4520         ASSERT(got->br_startoff + got->br_blockcount >= aoff + alen);
4521         ASSERT(isnullstartblock(got->br_startblock));
4522         ASSERT(got->br_state == XFS_EXT_NORM);
4523         return 0;
4524
4525 out_unreserve_blocks:
4526         if (rt)
4527                 xfs_mod_incore_sb(mp, XFS_SBS_FREXTENTS, extsz, 0);
4528         else
4529                 xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS, alen, 0);
4530 out_unreserve_quota:
4531         if (XFS_IS_QUOTA_ON(mp))
4532                 xfs_trans_unreserve_quota_nblks(NULL, ip, alen, 0, rt ?
4533                                 XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4534         return error;
4535 }
4536
4537 /*
4538  * Map file blocks to filesystem blocks, adding delayed allocations as needed.
4539  */
4540 int
4541 xfs_bmapi_delay(
4542         struct xfs_inode        *ip,    /* incore inode */
4543         xfs_fileoff_t           bno,    /* starting file offs. mapped */
4544         xfs_filblks_t           len,    /* length to map in file */
4545         struct xfs_bmbt_irec    *mval,  /* output: map values */
4546         int                     *nmap,  /* i/o: mval size/count */
4547         int                     flags)  /* XFS_BMAPI_... */
4548 {
4549         struct xfs_mount        *mp = ip->i_mount;
4550         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
4551         struct xfs_bmbt_irec    got;    /* current file extent record */
4552         struct xfs_bmbt_irec    prev;   /* previous file extent record */
4553         xfs_fileoff_t           obno;   /* old block number (offset) */
4554         xfs_fileoff_t           end;    /* end of mapped file region */
4555         xfs_extnum_t            lastx;  /* last useful extent number */
4556         int                     eof;    /* we've hit the end of extents */
4557         int                     n = 0;  /* current extent index */
4558         int                     error = 0;
4559
4560         ASSERT(*nmap >= 1);
4561         ASSERT(*nmap <= XFS_BMAP_MAX_NMAP);
4562         ASSERT(!(flags & ~XFS_BMAPI_ENTIRE));
4563
4564         if (unlikely(XFS_TEST_ERROR(
4565             (XFS_IFORK_FORMAT(ip, XFS_DATA_FORK) != XFS_DINODE_FMT_EXTENTS &&
4566              XFS_IFORK_FORMAT(ip, XFS_DATA_FORK) != XFS_DINODE_FMT_BTREE),
4567              mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4568                 XFS_ERROR_REPORT("xfs_bmapi_delay", XFS_ERRLEVEL_LOW, mp);
4569                 return XFS_ERROR(EFSCORRUPTED);
4570         }
4571
4572         if (XFS_FORCED_SHUTDOWN(mp))
4573                 return XFS_ERROR(EIO);
4574
4575         XFS_STATS_INC(xs_blk_mapw);
4576
4577         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4578                 error = xfs_iread_extents(NULL, ip, XFS_DATA_FORK);
4579                 if (error)
4580                         return error;
4581         }
4582
4583         xfs_bmap_search_extents(ip, bno, XFS_DATA_FORK, &eof, &lastx, &got, &prev);
4584         end = bno + len;
4585         obno = bno;
4586
4587         while (bno < end && n < *nmap) {
4588                 if (eof || got.br_startoff > bno) {
4589                         error = xfs_bmapi_reserve_delalloc(ip, bno, len, &got,
4590                                                            &prev, &lastx, eof);
4591                         if (error) {
4592                                 if (n == 0) {
4593                                         *nmap = 0;
4594                                         return error;
4595                                 }
4596                                 break;
4597                         }
4598                 }
4599
4600                 /* set up the extent map to return. */
4601                 xfs_bmapi_trim_map(mval, &got, &bno, len, obno, end, n, flags);
4602                 xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4603
4604                 /* If we're done, stop now. */
4605                 if (bno >= end || n >= *nmap)
4606                         break;
4607
4608                 /* Else go on to the next record. */
4609                 prev = got;
4610                 if (++lastx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t))
4611                         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx), &got);
4612                 else
4613                         eof = 1;
4614         }
4615
4616         *nmap = n;
4617         return 0;
4618 }
4619
4620
4621 STATIC int
4622 xfs_bmapi_allocate(
4623         struct xfs_bmalloca     *bma,
4624         int                     flags)
4625 {
4626         struct xfs_mount        *mp = bma->ip->i_mount;
4627         int                     whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4628                                                 XFS_ATTR_FORK : XFS_DATA_FORK;
4629         struct xfs_ifork        *ifp = XFS_IFORK_PTR(bma->ip, whichfork);
4630         int                     tmp_logflags = 0;
4631         int                     error;
4632         int                     rt;
4633
4634         ASSERT(bma->length > 0);
4635
4636         rt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(bma->ip);
4637
4638         /*
4639          * For the wasdelay case, we could also just allocate the stuff asked
4640          * for in this bmap call but that wouldn't be as good.
4641          */
4642         if (bma->wasdel) {
4643                 bma->length = (xfs_extlen_t)bma->got.br_blockcount;
4644                 bma->offset = bma->got.br_startoff;
4645                 if (bma->idx != NULLEXTNUM && bma->idx) {
4646                         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx - 1),
4647                                          &bma->prev);
4648                 }
4649         } else {
4650                 bma->length = XFS_FILBLKS_MIN(bma->length, MAXEXTLEN);
4651                 if (!bma->eof)
4652                         bma->length = XFS_FILBLKS_MIN(bma->length,
4653                                         bma->got.br_startoff - bma->offset);
4654         }
4655
4656         /*
4657          * Indicate if this is the first user data in the file, or just any
4658          * user data.
4659          */
4660         if (!(flags & XFS_BMAPI_METADATA)) {
4661                 bma->userdata = (bma->offset == 0) ?
4662                         XFS_ALLOC_INITIAL_USER_DATA : XFS_ALLOC_USERDATA;
4663         }
4664
4665         bma->minlen = (flags & XFS_BMAPI_CONTIG) ? bma->length : 1;
4666
4667         /*
4668          * Only want to do the alignment at the eof if it is userdata and
4669          * allocation length is larger than a stripe unit.
4670          */
4671         if (mp->m_dalign && bma->length >= mp->m_dalign &&
4672             !(flags & XFS_BMAPI_METADATA) && whichfork == XFS_DATA_FORK) {
4673                 error = xfs_bmap_isaeof(bma, whichfork);
4674                 if (error)
4675                         return error;
4676         }
4677
4678         error = xfs_bmap_alloc(bma);
4679         if (error)
4680                 return error;
4681
4682         if (bma->flist->xbf_low)
4683                 bma->minleft = 0;
4684         if (bma->cur)
4685                 bma->cur->bc_private.b.firstblock = *bma->firstblock;
4686         if (bma->blkno == NULLFSBLOCK)
4687                 return 0;
4688         if ((ifp->if_flags & XFS_IFBROOT) && !bma->cur) {
4689                 bma->cur = xfs_bmbt_init_cursor(mp, bma->tp, bma->ip, whichfork);
4690                 bma->cur->bc_private.b.firstblock = *bma->firstblock;
4691                 bma->cur->bc_private.b.flist = bma->flist;
4692         }
4693         /*
4694          * Bump the number of extents we've allocated
4695          * in this call.
4696          */
4697         bma->nallocs++;
4698
4699         if (bma->cur)
4700                 bma->cur->bc_private.b.flags =
4701                         bma->wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
4702
4703         bma->got.br_startoff = bma->offset;
4704         bma->got.br_startblock = bma->blkno;
4705         bma->got.br_blockcount = bma->length;
4706         bma->got.br_state = XFS_EXT_NORM;
4707
4708         /*
4709          * A wasdelay extent has been initialized, so shouldn't be flagged
4710          * as unwritten.
4711          */
4712         if (!bma->wasdel && (flags & XFS_BMAPI_PREALLOC) &&
4713             xfs_sb_version_hasextflgbit(&mp->m_sb))
4714                 bma->got.br_state = XFS_EXT_UNWRITTEN;
4715
4716         if (bma->wasdel)
4717                 error = xfs_bmap_add_extent_delay_real(bma);
4718         else
4719                 error = xfs_bmap_add_extent_hole_real(bma, whichfork);
4720
4721         bma->logflags |= tmp_logflags;
4722         if (error)
4723                 return error;
4724
4725         /*
4726          * Update our extent pointer, given that xfs_bmap_add_extent_delay_real
4727          * or xfs_bmap_add_extent_hole_real might have merged it into one of
4728          * the neighbouring ones.
4729          */
4730         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx), &bma->got);
4731
4732         ASSERT(bma->got.br_startoff <= bma->offset);
4733         ASSERT(bma->got.br_startoff + bma->got.br_blockcount >=
4734                bma->offset + bma->length);
4735         ASSERT(bma->got.br_state == XFS_EXT_NORM ||
4736                bma->got.br_state == XFS_EXT_UNWRITTEN);
4737         return 0;
4738 }
4739
4740 STATIC int
4741 xfs_bmapi_convert_unwritten(
4742         struct xfs_bmalloca     *bma,
4743         struct xfs_bmbt_irec    *mval,
4744         xfs_filblks_t           len,
4745         int                     flags)
4746 {
4747         int                     whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4748                                                 XFS_ATTR_FORK : XFS_DATA_FORK;
4749         struct xfs_ifork        *ifp = XFS_IFORK_PTR(bma->ip, whichfork);
4750         int                     tmp_logflags = 0;
4751         int                     error;
4752
4753         /* check if we need to do unwritten->real conversion */
4754         if (mval->br_state == XFS_EXT_UNWRITTEN &&
4755             (flags & XFS_BMAPI_PREALLOC))
4756                 return 0;
4757
4758         /* check if we need to do real->unwritten conversion */
4759         if (mval->br_state == XFS_EXT_NORM &&
4760             (flags & (XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT)) !=
4761                         (XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT))
4762                 return 0;
4763
4764         /*
4765          * Modify (by adding) the state flag, if writing.
4766          */
4767         ASSERT(mval->br_blockcount <= len);
4768         if ((ifp->if_flags & XFS_IFBROOT) && !bma->cur) {
4769                 bma->cur = xfs_bmbt_init_cursor(bma->ip->i_mount, bma->tp,
4770                                         bma->ip, whichfork);
4771                 bma->cur->bc_private.b.firstblock = *bma->firstblock;
4772                 bma->cur->bc_private.b.flist = bma->flist;
4773         }
4774         mval->br_state = (mval->br_state == XFS_EXT_UNWRITTEN)
4775                                 ? XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
4776
4777         error = xfs_bmap_add_extent_unwritten_real(bma->tp, bma->ip, &bma->idx,
4778                         &bma->cur, mval, bma->firstblock, bma->flist,
4779                         &tmp_logflags);
4780         bma->logflags |= tmp_logflags;
4781         if (error)
4782                 return error;
4783
4784         /*
4785          * Update our extent pointer, given that
4786          * xfs_bmap_add_extent_unwritten_real might have merged it into one
4787          * of the neighbouring ones.
4788          */
4789         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx), &bma->got);
4790
4791         /*
4792          * We may have combined previously unwritten space with written space,
4793          * so generate another request.
4794          */
4795         if (mval->br_blockcount < len)
4796                 return EAGAIN;
4797         return 0;
4798 }
4799
4800 /*
4801  * Map file blocks to filesystem blocks, and allocate blocks or convert the
4802  * extent state if necessary.  Details behaviour is controlled by the flags
4803  * parameter.  Only allocates blocks from a single allocation group, to avoid
4804  * locking problems.
4805  *
4806  * The returned value in "firstblock" from the first call in a transaction
4807  * must be remembered and presented to subsequent calls in "firstblock".
4808  * An upper bound for the number of blocks to be allocated is supplied to
4809  * the first call in "total"; if no allocation group has that many free
4810  * blocks then the call will fail (return NULLFSBLOCK in "firstblock").
4811  */
4812 int
4813 xfs_bmapi_write(
4814         struct xfs_trans        *tp,            /* transaction pointer */
4815         struct xfs_inode        *ip,            /* incore inode */
4816         xfs_fileoff_t           bno,            /* starting file offs. mapped */
4817         xfs_filblks_t           len,            /* length to map in file */
4818         int                     flags,          /* XFS_BMAPI_... */
4819         xfs_fsblock_t           *firstblock,    /* first allocated block
4820                                                    controls a.g. for allocs */
4821         xfs_extlen_t            total,          /* total blocks needed */
4822         struct xfs_bmbt_irec    *mval,          /* output: map values */
4823         int                     *nmap,          /* i/o: mval size/count */
4824         struct xfs_bmap_free    *flist)         /* i/o: list extents to free */
4825 {
4826         struct xfs_mount        *mp = ip->i_mount;
4827         struct xfs_ifork        *ifp;
4828         struct xfs_bmalloca     bma = { 0 };    /* args for xfs_bmap_alloc */
4829         xfs_fileoff_t           end;            /* end of mapped file region */
4830         int                     eof;            /* after the end of extents */
4831         int                     error;          /* error return */
4832         int                     n;              /* current extent index */
4833         xfs_fileoff_t           obno;           /* old block number (offset) */
4834         int                     whichfork;      /* data or attr fork */
4835         char                    inhole;         /* current location is hole in file */
4836         char                    wasdelay;       /* old extent was delayed */
4837
4838 #ifdef DEBUG
4839         xfs_fileoff_t           orig_bno;       /* original block number value */
4840         int                     orig_flags;     /* original flags arg value */
4841         xfs_filblks_t           orig_len;       /* original value of len arg */
4842         struct xfs_bmbt_irec    *orig_mval;     /* original value of mval */
4843         int                     orig_nmap;      /* original value of *nmap */
4844
4845         orig_bno = bno;
4846         orig_len = len;
4847         orig_flags = flags;
4848         orig_mval = mval;
4849         orig_nmap = *nmap;
4850 #endif
4851
4852         ASSERT(*nmap >= 1);
4853         ASSERT(*nmap <= XFS_BMAP_MAX_NMAP);
4854         ASSERT(!(flags & XFS_BMAPI_IGSTATE));
4855         ASSERT(tp != NULL);
4856         ASSERT(len > 0);
4857
4858         whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4859                 XFS_ATTR_FORK : XFS_DATA_FORK;
4860
4861         if (unlikely(XFS_TEST_ERROR(
4862             (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4863              XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
4864              XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL),
4865              mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4866                 XFS_ERROR_REPORT("xfs_bmapi_write", XFS_ERRLEVEL_LOW, mp);
4867                 return XFS_ERROR(EFSCORRUPTED);
4868         }
4869
4870         if (XFS_FORCED_SHUTDOWN(mp))
4871                 return XFS_ERROR(EIO);
4872
4873         ifp = XFS_IFORK_PTR(ip, whichfork);
4874         ASSERT(ifp->if_ext_max ==
4875                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
4876
4877         XFS_STATS_INC(xs_blk_mapw);
4878
4879         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
4880                 error = xfs_bmap_local_to_extents(tp, ip, firstblock, total,
4881                                                   &bma.logflags, whichfork);
4882                 if (error)
4883                         goto error0;
4884         }
4885
4886         if (*firstblock == NULLFSBLOCK) {
4887                 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE)
4888                         bma.minleft = be16_to_cpu(ifp->if_broot->bb_level) + 1;
4889                 else
4890                         bma.minleft = 1;
4891         } else {
4892                 bma.minleft = 0;
4893         }
4894
4895         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4896                 error = xfs_iread_extents(tp, ip, whichfork);
4897                 if (error)
4898                         goto error0;
4899         }
4900
4901         xfs_bmap_search_extents(ip, bno, whichfork, &eof, &bma.idx, &bma.got,
4902                                 &bma.prev);
4903         n = 0;
4904         end = bno + len;
4905         obno = bno;
4906
4907         bma.tp = tp;
4908         bma.ip = ip;
4909         bma.total = total;
4910         bma.userdata = 0;
4911         bma.flist = flist;
4912         bma.firstblock = firstblock;
4913
4914         while (bno < end && n < *nmap) {
4915                 inhole = eof || bma.got.br_startoff > bno;
4916                 wasdelay = !inhole && isnullstartblock(bma.got.br_startblock);
4917
4918                 /*
4919                  * First, deal with the hole before the allocated space
4920                  * that we found, if any.
4921                  */
4922                 if (inhole || wasdelay) {
4923                         bma.eof = eof;
4924                         bma.conv = !!(flags & XFS_BMAPI_CONVERT);
4925                         bma.wasdel = wasdelay;
4926                         bma.offset = bno;
4927
4928                         /*
4929                          * There's a 32/64 bit type mismatch between the
4930                          * allocation length request (which can be 64 bits in
4931                          * length) and the bma length request, which is
4932                          * xfs_extlen_t and therefore 32 bits. Hence we have to
4933                          * check for 32-bit overflows and handle them here.
4934                          */
4935                         if (len > (xfs_filblks_t)MAXEXTLEN)
4936                                 bma.length = MAXEXTLEN;
4937                         else
4938                                 bma.length = len;
4939
4940                         ASSERT(len > 0);
4941                         ASSERT(bma.length > 0);
4942                         error = xfs_bmapi_allocate(&bma, flags);
4943                         if (error)
4944                                 goto error0;
4945                         if (bma.blkno == NULLFSBLOCK)
4946                                 break;
4947                 }
4948
4949                 /* Deal with the allocated space we found.  */
4950                 xfs_bmapi_trim_map(mval, &bma.got, &bno, len, obno,
4951                                                         end, n, flags);
4952
4953                 /* Execute unwritten extent conversion if necessary */
4954                 error = xfs_bmapi_convert_unwritten(&bma, mval, len, flags);
4955                 if (error == EAGAIN)
4956                         continue;
4957                 if (error)
4958                         goto error0;
4959
4960                 /* update the extent map to return */
4961                 xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4962
4963                 /*
4964                  * If we're done, stop now.  Stop when we've allocated
4965                  * XFS_BMAP_MAX_NMAP extents no matter what.  Otherwise
4966                  * the transaction may get too big.
4967                  */
4968                 if (bno >= end || n >= *nmap || bma.nallocs >= *nmap)
4969                         break;
4970
4971                 /* Else go on to the next record. */
4972                 bma.prev = bma.got;
4973                 if (++bma.idx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t)) {
4974                         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma.idx),
4975                                          &bma.got);
4976                 } else
4977                         eof = 1;
4978         }
4979         *nmap = n;
4980
4981         /*
4982          * Transform from btree to extents, give it cur.
4983          */
4984         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE &&
4985             XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max) {
4986                 int             tmp_logflags = 0;
4987
4988                 ASSERT(bma.cur);
4989                 error = xfs_bmap_btree_to_extents(tp, ip, bma.cur,
4990                         &tmp_logflags, whichfork);
4991                 bma.logflags |= tmp_logflags;
4992                 if (error)
4993                         goto error0;
4994         }
4995         ASSERT(ifp->if_ext_max ==
4996                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
4997         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE ||
4998                XFS_IFORK_NEXTENTS(ip, whichfork) > ifp->if_ext_max);
4999         error = 0;
5000 error0:
5001         /*
5002          * Log everything.  Do this after conversion, there's no point in
5003          * logging the extent records if we've converted to btree format.
5004          */
5005         if ((bma.logflags & xfs_ilog_fext(whichfork)) &&
5006             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
5007                 bma.logflags &= ~xfs_ilog_fext(whichfork);
5008         else if ((bma.logflags & xfs_ilog_fbroot(whichfork)) &&
5009                  XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
5010                 bma.logflags &= ~xfs_ilog_fbroot(whichfork);
5011         /*
5012          * Log whatever the flags say, even if error.  Otherwise we might miss
5013          * detecting a case where the data is changed, there's an error,
5014          * and it's not logged so we don't shutdown when we should.
5015          */
5016         if (bma.logflags)
5017                 xfs_trans_log_inode(tp, ip, bma.logflags);
5018
5019         if (bma.cur) {
5020                 if (!error) {
5021                         ASSERT(*firstblock == NULLFSBLOCK ||
5022                                XFS_FSB_TO_AGNO(mp, *firstblock) ==
5023                                XFS_FSB_TO_AGNO(mp,
5024                                        bma.cur->bc_private.b.firstblock) ||
5025                                (flist->xbf_low &&
5026                                 XFS_FSB_TO_AGNO(mp, *firstblock) <
5027                                 XFS_FSB_TO_AGNO(mp,
5028                                         bma.cur->bc_private.b.firstblock)));
5029                         *firstblock = bma.cur->bc_private.b.firstblock;
5030                 }
5031                 xfs_btree_del_cursor(bma.cur,
5032                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5033         }
5034         if (!error)
5035                 xfs_bmap_validate_ret(orig_bno, orig_len, orig_flags, orig_mval,
5036                         orig_nmap, *nmap);
5037         return error;
5038 }
5039
5040 /*
5041  * Unmap (remove) blocks from a file.
5042  * If nexts is nonzero then the number of extents to remove is limited to
5043  * that value.  If not all extents in the block range can be removed then
5044  * *done is set.
5045  */
5046 int                                             /* error */
5047 xfs_bunmapi(
5048         xfs_trans_t             *tp,            /* transaction pointer */
5049         struct xfs_inode        *ip,            /* incore inode */
5050         xfs_fileoff_t           bno,            /* starting offset to unmap */
5051         xfs_filblks_t           len,            /* length to unmap in file */
5052         int                     flags,          /* misc flags */
5053         xfs_extnum_t            nexts,          /* number of extents max */
5054         xfs_fsblock_t           *firstblock,    /* first allocated block
5055                                                    controls a.g. for allocs */
5056         xfs_bmap_free_t         *flist,         /* i/o: list extents to free */
5057         int                     *done)          /* set if not done yet */
5058 {
5059         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
5060         xfs_bmbt_irec_t         del;            /* extent being deleted */
5061         int                     eof;            /* is deleting at eof */
5062         xfs_bmbt_rec_host_t     *ep;            /* extent record pointer */
5063         int                     error;          /* error return value */
5064         xfs_extnum_t            extno;          /* extent number in list */
5065         xfs_bmbt_irec_t         got;            /* current extent record */
5066         xfs_ifork_t             *ifp;           /* inode fork pointer */
5067         int                     isrt;           /* freeing in rt area */
5068         xfs_extnum_t            lastx;          /* last extent index used */
5069         int                     logflags;       /* transaction logging flags */
5070         xfs_extlen_t            mod;            /* rt extent offset */
5071         xfs_mount_t             *mp;            /* mount structure */
5072         xfs_extnum_t            nextents;       /* number of file extents */
5073         xfs_bmbt_irec_t         prev;           /* previous extent record */
5074         xfs_fileoff_t           start;          /* first file offset deleted */
5075         int                     tmp_logflags;   /* partial logging flags */
5076         int                     wasdel;         /* was a delayed alloc extent */
5077         int                     whichfork;      /* data or attribute fork */
5078         xfs_fsblock_t           sum;
5079
5080         trace_xfs_bunmap(ip, bno, len, flags, _RET_IP_);
5081
5082         whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
5083                 XFS_ATTR_FORK : XFS_DATA_FORK;
5084         ifp = XFS_IFORK_PTR(ip, whichfork);
5085         if (unlikely(
5086             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
5087             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
5088                 XFS_ERROR_REPORT("xfs_bunmapi", XFS_ERRLEVEL_LOW,
5089                                  ip->i_mount);
5090                 return XFS_ERROR(EFSCORRUPTED);
5091         }
5092         mp = ip->i_mount;
5093         if (XFS_FORCED_SHUTDOWN(mp))
5094                 return XFS_ERROR(EIO);
5095
5096         ASSERT(len > 0);
5097         ASSERT(nexts >= 0);
5098         ASSERT(ifp->if_ext_max ==
5099                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5100         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5101             (error = xfs_iread_extents(tp, ip, whichfork)))
5102                 return error;
5103         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5104         if (nextents == 0) {
5105                 *done = 1;
5106                 return 0;
5107         }
5108         XFS_STATS_INC(xs_blk_unmap);
5109         isrt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
5110         start = bno;
5111         bno = start + len - 1;
5112         ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
5113                 &prev);
5114
5115         /*
5116          * Check to see if the given block number is past the end of the
5117          * file, back up to the last block if so...
5118          */
5119         if (eof) {
5120                 ep = xfs_iext_get_ext(ifp, --lastx);
5121                 xfs_bmbt_get_all(ep, &got);
5122                 bno = got.br_startoff + got.br_blockcount - 1;
5123         }
5124         logflags = 0;
5125         if (ifp->if_flags & XFS_IFBROOT) {
5126                 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
5127                 cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5128                 cur->bc_private.b.firstblock = *firstblock;
5129                 cur->bc_private.b.flist = flist;
5130                 cur->bc_private.b.flags = 0;
5131         } else
5132                 cur = NULL;
5133         extno = 0;
5134         while (bno != (xfs_fileoff_t)-1 && bno >= start && lastx >= 0 &&
5135                (nexts == 0 || extno < nexts)) {
5136                 /*
5137                  * Is the found extent after a hole in which bno lives?
5138                  * Just back up to the previous extent, if so.
5139                  */
5140                 if (got.br_startoff > bno) {
5141                         if (--lastx < 0)
5142                                 break;
5143                         ep = xfs_iext_get_ext(ifp, lastx);
5144                         xfs_bmbt_get_all(ep, &got);
5145                 }
5146                 /*
5147                  * Is the last block of this extent before the range
5148                  * we're supposed to delete?  If so, we're done.
5149                  */
5150                 bno = XFS_FILEOFF_MIN(bno,
5151                         got.br_startoff + got.br_blockcount - 1);
5152                 if (bno < start)
5153                         break;
5154                 /*
5155                  * Then deal with the (possibly delayed) allocated space
5156                  * we found.
5157                  */
5158                 ASSERT(ep != NULL);
5159                 del = got;
5160                 wasdel = isnullstartblock(del.br_startblock);
5161                 if (got.br_startoff < start) {
5162                         del.br_startoff = start;
5163                         del.br_blockcount -= start - got.br_startoff;
5164                         if (!wasdel)
5165                                 del.br_startblock += start - got.br_startoff;
5166                 }
5167                 if (del.br_startoff + del.br_blockcount > bno + 1)
5168                         del.br_blockcount = bno + 1 - del.br_startoff;
5169                 sum = del.br_startblock + del.br_blockcount;
5170                 if (isrt &&
5171                     (mod = do_mod(sum, mp->m_sb.sb_rextsize))) {
5172                         /*
5173                          * Realtime extent not lined up at the end.
5174                          * The extent could have been split into written
5175                          * and unwritten pieces, or we could just be
5176                          * unmapping part of it.  But we can't really
5177                          * get rid of part of a realtime extent.
5178                          */
5179                         if (del.br_state == XFS_EXT_UNWRITTEN ||
5180                             !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5181                                 /*
5182                                  * This piece is unwritten, or we're not
5183                                  * using unwritten extents.  Skip over it.
5184                                  */
5185                                 ASSERT(bno >= mod);
5186                                 bno -= mod > del.br_blockcount ?
5187                                         del.br_blockcount : mod;
5188                                 if (bno < got.br_startoff) {
5189                                         if (--lastx >= 0)
5190                                                 xfs_bmbt_get_all(xfs_iext_get_ext(
5191                                                         ifp, lastx), &got);
5192                                 }
5193                                 continue;
5194                         }
5195                         /*
5196                          * It's written, turn it unwritten.
5197                          * This is better than zeroing it.
5198                          */
5199                         ASSERT(del.br_state == XFS_EXT_NORM);
5200                         ASSERT(xfs_trans_get_block_res(tp) > 0);
5201                         /*
5202                          * If this spans a realtime extent boundary,
5203                          * chop it back to the start of the one we end at.
5204                          */
5205                         if (del.br_blockcount > mod) {
5206                                 del.br_startoff += del.br_blockcount - mod;
5207                                 del.br_startblock += del.br_blockcount - mod;
5208                                 del.br_blockcount = mod;
5209                         }
5210                         del.br_state = XFS_EXT_UNWRITTEN;
5211                         error = xfs_bmap_add_extent_unwritten_real(tp, ip,
5212                                         &lastx, &cur, &del, firstblock, flist,
5213                                         &logflags);
5214                         if (error)
5215                                 goto error0;
5216                         goto nodelete;
5217                 }
5218                 if (isrt && (mod = do_mod(del.br_startblock, mp->m_sb.sb_rextsize))) {
5219                         /*
5220                          * Realtime extent is lined up at the end but not
5221                          * at the front.  We'll get rid of full extents if
5222                          * we can.
5223                          */
5224                         mod = mp->m_sb.sb_rextsize - mod;
5225                         if (del.br_blockcount > mod) {
5226                                 del.br_blockcount -= mod;
5227                                 del.br_startoff += mod;
5228                                 del.br_startblock += mod;
5229                         } else if ((del.br_startoff == start &&
5230                                     (del.br_state == XFS_EXT_UNWRITTEN ||
5231                                      xfs_trans_get_block_res(tp) == 0)) ||
5232                                    !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5233                                 /*
5234                                  * Can't make it unwritten.  There isn't
5235                                  * a full extent here so just skip it.
5236                                  */
5237                                 ASSERT(bno >= del.br_blockcount);
5238                                 bno -= del.br_blockcount;
5239                                 if (got.br_startoff > bno) {
5240                                         if (--lastx >= 0) {
5241                                                 ep = xfs_iext_get_ext(ifp,
5242                                                                       lastx);
5243                                                 xfs_bmbt_get_all(ep, &got);
5244                                         }
5245                                 }
5246                                 continue;
5247                         } else if (del.br_state == XFS_EXT_UNWRITTEN) {
5248                                 /*
5249                                  * This one is already unwritten.
5250                                  * It must have a written left neighbor.
5251                                  * Unwrite the killed part of that one and
5252                                  * try again.
5253                                  */
5254                                 ASSERT(lastx > 0);
5255                                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp,
5256                                                 lastx - 1), &prev);
5257                                 ASSERT(prev.br_state == XFS_EXT_NORM);
5258                                 ASSERT(!isnullstartblock(prev.br_startblock));
5259                                 ASSERT(del.br_startblock ==
5260                                        prev.br_startblock + prev.br_blockcount);
5261                                 if (prev.br_startoff < start) {
5262                                         mod = start - prev.br_startoff;
5263                                         prev.br_blockcount -= mod;
5264                                         prev.br_startblock += mod;
5265                                         prev.br_startoff = start;
5266                                 }
5267                                 prev.br_state = XFS_EXT_UNWRITTEN;
5268                                 lastx--;
5269                                 error = xfs_bmap_add_extent_unwritten_real(tp,
5270                                                 ip, &lastx, &cur, &prev,
5271                                                 firstblock, flist, &logflags);
5272                                 if (error)
5273                                         goto error0;
5274                                 goto nodelete;
5275                         } else {
5276                                 ASSERT(del.br_state == XFS_EXT_NORM);
5277                                 del.br_state = XFS_EXT_UNWRITTEN;
5278                                 error = xfs_bmap_add_extent_unwritten_real(tp,
5279                                                 ip, &lastx, &cur, &del,
5280                                                 firstblock, flist, &logflags);
5281                                 if (error)
5282                                         goto error0;
5283                                 goto nodelete;
5284                         }
5285                 }
5286                 if (wasdel) {
5287                         ASSERT(startblockval(del.br_startblock) > 0);
5288                         /* Update realtime/data freespace, unreserve quota */
5289                         if (isrt) {
5290                                 xfs_filblks_t rtexts;
5291
5292                                 rtexts = XFS_FSB_TO_B(mp, del.br_blockcount);
5293                                 do_div(rtexts, mp->m_sb.sb_rextsize);
5294                                 xfs_mod_incore_sb(mp, XFS_SBS_FREXTENTS,
5295                                                 (int64_t)rtexts, 0);
5296                                 (void)xfs_trans_reserve_quota_nblks(NULL,
5297                                         ip, -((long)del.br_blockcount), 0,
5298                                         XFS_QMOPT_RES_RTBLKS);
5299                         } else {
5300                                 xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
5301                                                 (int64_t)del.br_blockcount, 0);
5302                                 (void)xfs_trans_reserve_quota_nblks(NULL,
5303                                         ip, -((long)del.br_blockcount), 0,
5304                                         XFS_QMOPT_RES_REGBLKS);
5305                         }
5306                         ip->i_delayed_blks -= del.br_blockcount;
5307                         if (cur)
5308                                 cur->bc_private.b.flags |=
5309                                         XFS_BTCUR_BPRV_WASDEL;
5310                 } else if (cur)
5311                         cur->bc_private.b.flags &= ~XFS_BTCUR_BPRV_WASDEL;
5312                 /*
5313                  * If it's the case where the directory code is running
5314                  * with no block reservation, and the deleted block is in
5315                  * the middle of its extent, and the resulting insert
5316                  * of an extent would cause transformation to btree format,
5317                  * then reject it.  The calling code will then swap
5318                  * blocks around instead.
5319                  * We have to do this now, rather than waiting for the
5320                  * conversion to btree format, since the transaction
5321                  * will be dirty.
5322                  */
5323                 if (!wasdel && xfs_trans_get_block_res(tp) == 0 &&
5324                     XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
5325                     XFS_IFORK_NEXTENTS(ip, whichfork) >= ifp->if_ext_max &&
5326                     del.br_startoff > got.br_startoff &&
5327                     del.br_startoff + del.br_blockcount <
5328                     got.br_startoff + got.br_blockcount) {
5329                         error = XFS_ERROR(ENOSPC);
5330                         goto error0;
5331                 }
5332                 error = xfs_bmap_del_extent(ip, tp, &lastx, flist, cur, &del,
5333                                 &tmp_logflags, whichfork);
5334                 logflags |= tmp_logflags;
5335                 if (error)
5336                         goto error0;
5337                 bno = del.br_startoff - 1;
5338 nodelete:
5339                 /*
5340                  * If not done go on to the next (previous) record.
5341                  */
5342                 if (bno != (xfs_fileoff_t)-1 && bno >= start) {
5343                         if (lastx >= 0) {
5344                                 ep = xfs_iext_get_ext(ifp, lastx);
5345                                 if (xfs_bmbt_get_startoff(ep) > bno) {
5346                                         if (--lastx >= 0)
5347                                                 ep = xfs_iext_get_ext(ifp,
5348                                                                       lastx);
5349                                 }
5350                                 xfs_bmbt_get_all(ep, &got);
5351                         }
5352                         extno++;
5353                 }
5354         }
5355         *done = bno == (xfs_fileoff_t)-1 || bno < start || lastx < 0;
5356         ASSERT(ifp->if_ext_max ==
5357                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5358         /*
5359          * Convert to a btree if necessary.
5360          */
5361         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
5362             XFS_IFORK_NEXTENTS(ip, whichfork) > ifp->if_ext_max) {
5363                 ASSERT(cur == NULL);
5364                 error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist,
5365                         &cur, 0, &tmp_logflags, whichfork);
5366                 logflags |= tmp_logflags;
5367                 if (error)
5368                         goto error0;
5369         }
5370         /*
5371          * transform from btree to extents, give it cur
5372          */
5373         else if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE &&
5374                  XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max) {
5375                 ASSERT(cur != NULL);
5376                 error = xfs_bmap_btree_to_extents(tp, ip, cur, &tmp_logflags,
5377                         whichfork);
5378                 logflags |= tmp_logflags;
5379                 if (error)
5380                         goto error0;
5381         }
5382         /*
5383          * transform from extents to local?
5384          */
5385         ASSERT(ifp->if_ext_max ==
5386                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5387         error = 0;
5388 error0:
5389         /*
5390          * Log everything.  Do this after conversion, there's no point in
5391          * logging the extent records if we've converted to btree format.
5392          */
5393         if ((logflags & xfs_ilog_fext(whichfork)) &&
5394             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
5395                 logflags &= ~xfs_ilog_fext(whichfork);
5396         else if ((logflags & xfs_ilog_fbroot(whichfork)) &&
5397                  XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
5398                 logflags &= ~xfs_ilog_fbroot(whichfork);
5399         /*
5400          * Log inode even in the error case, if the transaction
5401          * is dirty we'll need to shut down the filesystem.
5402          */
5403         if (logflags)
5404                 xfs_trans_log_inode(tp, ip, logflags);
5405         if (cur) {
5406                 if (!error) {
5407                         *firstblock = cur->bc_private.b.firstblock;
5408                         cur->bc_private.b.allocated = 0;
5409                 }
5410                 xfs_btree_del_cursor(cur,
5411                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5412         }
5413         return error;
5414 }
5415
5416 /*
5417  * returns 1 for success, 0 if we failed to map the extent.
5418  */
5419 STATIC int
5420 xfs_getbmapx_fix_eof_hole(
5421         xfs_inode_t             *ip,            /* xfs incore inode pointer */
5422         struct getbmapx         *out,           /* output structure */
5423         int                     prealloced,     /* this is a file with
5424                                                  * preallocated data space */
5425         __int64_t               end,            /* last block requested */
5426         xfs_fsblock_t           startblock)
5427 {
5428         __int64_t               fixlen;
5429         xfs_mount_t             *mp;            /* file system mount point */
5430         xfs_ifork_t             *ifp;           /* inode fork pointer */
5431         xfs_extnum_t            lastx;          /* last extent pointer */
5432         xfs_fileoff_t           fileblock;
5433
5434         if (startblock == HOLESTARTBLOCK) {
5435                 mp = ip->i_mount;
5436                 out->bmv_block = -1;
5437                 fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, ip->i_size));
5438                 fixlen -= out->bmv_offset;
5439                 if (prealloced && out->bmv_offset + out->bmv_length == end) {
5440                         /* Came to hole at EOF. Trim it. */
5441                         if (fixlen <= 0)
5442                                 return 0;
5443                         out->bmv_length = fixlen;
5444                 }
5445         } else {
5446                 if (startblock == DELAYSTARTBLOCK)
5447                         out->bmv_block = -2;
5448                 else
5449                         out->bmv_block = xfs_fsb_to_db(ip, startblock);
5450                 fileblock = XFS_BB_TO_FSB(ip->i_mount, out->bmv_offset);
5451                 ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
5452                 if (xfs_iext_bno_to_ext(ifp, fileblock, &lastx) &&
5453                    (lastx == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t))-1))
5454                         out->bmv_oflags |= BMV_OF_LAST;
5455         }
5456
5457         return 1;
5458 }
5459
5460 /*
5461  * Get inode's extents as described in bmv, and format for output.
5462  * Calls formatter to fill the user's buffer until all extents
5463  * are mapped, until the passed-in bmv->bmv_count slots have
5464  * been filled, or until the formatter short-circuits the loop,
5465  * if it is tracking filled-in extents on its own.
5466  */
5467 int                                             /* error code */
5468 xfs_getbmap(
5469         xfs_inode_t             *ip,
5470         struct getbmapx         *bmv,           /* user bmap structure */
5471         xfs_bmap_format_t       formatter,      /* format to user */
5472         void                    *arg)           /* formatter arg */
5473 {
5474         __int64_t               bmvend;         /* last block requested */
5475         int                     error = 0;      /* return value */
5476         __int64_t               fixlen;         /* length for -1 case */
5477         int                     i;              /* extent number */
5478         int                     lock;           /* lock state */
5479         xfs_bmbt_irec_t         *map;           /* buffer for user's data */
5480         xfs_mount_t             *mp;            /* file system mount point */
5481         int                     nex;            /* # of user extents can do */
5482         int                     nexleft;        /* # of user extents left */
5483         int                     subnex;         /* # of bmapi's can do */
5484         int                     nmap;           /* number of map entries */
5485         struct getbmapx         *out;           /* output structure */
5486         int                     whichfork;      /* data or attr fork */
5487         int                     prealloced;     /* this is a file with
5488                                                  * preallocated data space */
5489         int                     iflags;         /* interface flags */
5490         int                     bmapi_flags;    /* flags for xfs_bmapi */
5491         int                     cur_ext = 0;
5492
5493         mp = ip->i_mount;
5494         iflags = bmv->bmv_iflags;
5495         whichfork = iflags & BMV_IF_ATTRFORK ? XFS_ATTR_FORK : XFS_DATA_FORK;
5496
5497         if (whichfork == XFS_ATTR_FORK) {
5498                 if (XFS_IFORK_Q(ip)) {
5499                         if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS &&
5500                             ip->i_d.di_aformat != XFS_DINODE_FMT_BTREE &&
5501                             ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)
5502                                 return XFS_ERROR(EINVAL);
5503                 } else if (unlikely(
5504                            ip->i_d.di_aformat != 0 &&
5505                            ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS)) {
5506                         XFS_ERROR_REPORT("xfs_getbmap", XFS_ERRLEVEL_LOW,
5507                                          ip->i_mount);
5508                         return XFS_ERROR(EFSCORRUPTED);
5509                 }
5510
5511                 prealloced = 0;
5512                 fixlen = 1LL << 32;
5513         } else {
5514                 if (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS &&
5515                     ip->i_d.di_format != XFS_DINODE_FMT_BTREE &&
5516                     ip->i_d.di_format != XFS_DINODE_FMT_LOCAL)
5517                         return XFS_ERROR(EINVAL);
5518
5519                 if (xfs_get_extsz_hint(ip) ||
5520                     ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC|XFS_DIFLAG_APPEND)){
5521                         prealloced = 1;
5522                         fixlen = XFS_MAXIOFFSET(mp);
5523                 } else {
5524                         prealloced = 0;
5525                         fixlen = ip->i_size;
5526                 }
5527         }
5528
5529         if (bmv->bmv_length == -1) {
5530                 fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, fixlen));
5531                 bmv->bmv_length =
5532                         max_t(__int64_t, fixlen - bmv->bmv_offset, 0);
5533         } else if (bmv->bmv_length == 0) {
5534                 bmv->bmv_entries = 0;
5535                 return 0;
5536         } else if (bmv->bmv_length < 0) {
5537                 return XFS_ERROR(EINVAL);
5538         }
5539
5540         nex = bmv->bmv_count - 1;
5541         if (nex <= 0)
5542                 return XFS_ERROR(EINVAL);
5543         bmvend = bmv->bmv_offset + bmv->bmv_length;
5544
5545
5546         if (bmv->bmv_count > ULONG_MAX / sizeof(struct getbmapx))
5547                 return XFS_ERROR(ENOMEM);
5548         out = kmem_zalloc(bmv->bmv_count * sizeof(struct getbmapx), KM_MAYFAIL);
5549         if (!out)
5550                 return XFS_ERROR(ENOMEM);
5551
5552         xfs_ilock(ip, XFS_IOLOCK_SHARED);
5553         if (whichfork == XFS_DATA_FORK && !(iflags & BMV_IF_DELALLOC)) {
5554                 if (ip->i_delayed_blks || ip->i_size > ip->i_d.di_size) {
5555                         error = xfs_flush_pages(ip, 0, -1, 0, FI_REMAPF);
5556                         if (error)
5557                                 goto out_unlock_iolock;
5558                 }
5559                 /*
5560                  * even after flushing the inode, there can still be delalloc
5561                  * blocks on the inode beyond EOF due to speculative
5562                  * preallocation. These are not removed until the release
5563                  * function is called or the inode is inactivated. Hence we
5564                  * cannot assert here that ip->i_delayed_blks == 0.
5565                  */
5566         }
5567
5568         lock = xfs_ilock_map_shared(ip);
5569
5570         /*
5571          * Don't let nex be bigger than the number of extents
5572          * we can have assuming alternating holes and real extents.
5573          */
5574         if (nex > XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1)
5575                 nex = XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1;
5576
5577         bmapi_flags = xfs_bmapi_aflag(whichfork);
5578         if (!(iflags & BMV_IF_PREALLOC))
5579                 bmapi_flags |= XFS_BMAPI_IGSTATE;
5580
5581         /*
5582          * Allocate enough space to handle "subnex" maps at a time.
5583          */
5584         error = ENOMEM;
5585         subnex = 16;
5586         map = kmem_alloc(subnex * sizeof(*map), KM_MAYFAIL | KM_NOFS);
5587         if (!map)
5588                 goto out_unlock_ilock;
5589
5590         bmv->bmv_entries = 0;
5591
5592         if (XFS_IFORK_NEXTENTS(ip, whichfork) == 0 &&
5593             (whichfork == XFS_ATTR_FORK || !(iflags & BMV_IF_DELALLOC))) {
5594                 error = 0;
5595                 goto out_free_map;
5596         }
5597
5598         nexleft = nex;
5599
5600         do {
5601                 nmap = (nexleft > subnex) ? subnex : nexleft;
5602                 error = xfs_bmapi_read(ip, XFS_BB_TO_FSBT(mp, bmv->bmv_offset),
5603                                        XFS_BB_TO_FSB(mp, bmv->bmv_length),
5604                                        map, &nmap, bmapi_flags);
5605                 if (error)
5606                         goto out_free_map;
5607                 ASSERT(nmap <= subnex);
5608
5609                 for (i = 0; i < nmap && nexleft && bmv->bmv_length; i++) {
5610                         out[cur_ext].bmv_oflags = 0;
5611                         if (map[i].br_state == XFS_EXT_UNWRITTEN)
5612                                 out[cur_ext].bmv_oflags |= BMV_OF_PREALLOC;
5613                         else if (map[i].br_startblock == DELAYSTARTBLOCK)
5614                                 out[cur_ext].bmv_oflags |= BMV_OF_DELALLOC;
5615                         out[cur_ext].bmv_offset =
5616                                 XFS_FSB_TO_BB(mp, map[i].br_startoff);
5617                         out[cur_ext].bmv_length =
5618                                 XFS_FSB_TO_BB(mp, map[i].br_blockcount);
5619                         out[cur_ext].bmv_unused1 = 0;
5620                         out[cur_ext].bmv_unused2 = 0;
5621                         ASSERT(((iflags & BMV_IF_DELALLOC) != 0) ||
5622                               (map[i].br_startblock != DELAYSTARTBLOCK));
5623                         if (map[i].br_startblock == HOLESTARTBLOCK &&
5624                             whichfork == XFS_ATTR_FORK) {
5625                                 /* came to the end of attribute fork */
5626                                 out[cur_ext].bmv_oflags |= BMV_OF_LAST;
5627                                 goto out_free_map;
5628                         }
5629
5630                         if (!xfs_getbmapx_fix_eof_hole(ip, &out[cur_ext],
5631                                         prealloced, bmvend,
5632                                         map[i].br_startblock))
5633                                 goto out_free_map;
5634
5635                         bmv->bmv_offset =
5636                                 out[cur_ext].bmv_offset +
5637                                 out[cur_ext].bmv_length;
5638                         bmv->bmv_length =
5639                                 max_t(__int64_t, 0, bmvend - bmv->bmv_offset);
5640
5641                         /*
5642                          * In case we don't want to return the hole,
5643                          * don't increase cur_ext so that we can reuse
5644                          * it in the next loop.
5645                          */
5646                         if ((iflags & BMV_IF_NO_HOLES) &&
5647                             map[i].br_startblock == HOLESTARTBLOCK) {
5648                                 memset(&out[cur_ext], 0, sizeof(out[cur_ext]));
5649                                 continue;
5650                         }
5651
5652                         nexleft--;
5653                         bmv->bmv_entries++;
5654                         cur_ext++;
5655                 }
5656         } while (nmap && nexleft && bmv->bmv_length);
5657
5658  out_free_map:
5659         kmem_free(map);
5660  out_unlock_ilock:
5661         xfs_iunlock_map_shared(ip, lock);
5662  out_unlock_iolock:
5663         xfs_iunlock(ip, XFS_IOLOCK_SHARED);
5664
5665         for (i = 0; i < cur_ext; i++) {
5666                 int full = 0;   /* user array is full */
5667
5668                 /* format results & advance arg */
5669                 error = formatter(&arg, &out[i], &full);
5670                 if (error || full)
5671                         break;
5672         }
5673
5674         kmem_free(out);
5675         return error;
5676 }
5677
5678 #ifdef DEBUG
5679 STATIC struct xfs_buf *
5680 xfs_bmap_get_bp(
5681         struct xfs_btree_cur    *cur,
5682         xfs_fsblock_t           bno)
5683 {
5684         struct xfs_log_item_desc *lidp;
5685         int                     i;
5686
5687         if (!cur)
5688                 return NULL;
5689
5690         for (i = 0; i < XFS_BTREE_MAXLEVELS; i++) {
5691                 if (!cur->bc_bufs[i])
5692                         break;
5693                 if (XFS_BUF_ADDR(cur->bc_bufs[i]) == bno)
5694                         return cur->bc_bufs[i];
5695         }
5696
5697         /* Chase down all the log items to see if the bp is there */
5698         list_for_each_entry(lidp, &cur->bc_tp->t_items, lid_trans) {
5699                 struct xfs_buf_log_item *bip;
5700                 bip = (struct xfs_buf_log_item *)lidp->lid_item;
5701                 if (bip->bli_item.li_type == XFS_LI_BUF &&
5702                     XFS_BUF_ADDR(bip->bli_buf) == bno)
5703                         return bip->bli_buf;
5704         }
5705
5706         return NULL;
5707 }
5708
5709 STATIC void
5710 xfs_check_block(
5711         struct xfs_btree_block  *block,
5712         xfs_mount_t             *mp,
5713         int                     root,
5714         short                   sz)
5715 {
5716         int                     i, j, dmxr;
5717         __be64                  *pp, *thispa;   /* pointer to block address */
5718         xfs_bmbt_key_t          *prevp, *keyp;
5719
5720         ASSERT(be16_to_cpu(block->bb_level) > 0);
5721
5722         prevp = NULL;
5723         for( i = 1; i <= xfs_btree_get_numrecs(block); i++) {
5724                 dmxr = mp->m_bmap_dmxr[0];
5725                 keyp = XFS_BMBT_KEY_ADDR(mp, block, i);
5726
5727                 if (prevp) {
5728                         ASSERT(be64_to_cpu(prevp->br_startoff) <
5729                                be64_to_cpu(keyp->br_startoff));
5730                 }
5731                 prevp = keyp;
5732
5733                 /*
5734                  * Compare the block numbers to see if there are dups.
5735                  */
5736                 if (root)
5737                         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, i, sz);
5738                 else
5739                         pp = XFS_BMBT_PTR_ADDR(mp, block, i, dmxr);
5740
5741                 for (j = i+1; j <= be16_to_cpu(block->bb_numrecs); j++) {
5742                         if (root)
5743                                 thispa = XFS_BMAP_BROOT_PTR_ADDR(mp, block, j, sz);
5744                         else
5745                                 thispa = XFS_BMBT_PTR_ADDR(mp, block, j, dmxr);
5746                         if (*thispa == *pp) {
5747                                 xfs_warn(mp, "%s: thispa(%d) == pp(%d) %Ld",
5748                                         __func__, j, i,
5749                                         (unsigned long long)be64_to_cpu(*thispa));
5750                                 panic("%s: ptrs are equal in node\n",
5751                                         __func__);
5752                         }
5753                 }
5754         }
5755 }
5756
5757 /*
5758  * Check that the extents for the inode ip are in the right order in all
5759  * btree leaves.
5760  */
5761
5762 STATIC void
5763 xfs_bmap_check_leaf_extents(
5764         xfs_btree_cur_t         *cur,   /* btree cursor or null */
5765         xfs_inode_t             *ip,            /* incore inode pointer */
5766         int                     whichfork)      /* data or attr fork */
5767 {
5768         struct xfs_btree_block  *block; /* current btree block */
5769         xfs_fsblock_t           bno;    /* block # of "block" */
5770         xfs_buf_t               *bp;    /* buffer for "block" */
5771         int                     error;  /* error return value */
5772         xfs_extnum_t            i=0, j; /* index into the extents list */
5773         xfs_ifork_t             *ifp;   /* fork structure */
5774         int                     level;  /* btree level, for checking */
5775         xfs_mount_t             *mp;    /* file system mount structure */
5776         __be64                  *pp;    /* pointer to block address */
5777         xfs_bmbt_rec_t          *ep;    /* pointer to current extent */
5778         xfs_bmbt_rec_t          last = {0, 0}; /* last extent in prev block */
5779         xfs_bmbt_rec_t          *nextp; /* pointer to next extent */
5780         int                     bp_release = 0;
5781
5782         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE) {
5783                 return;
5784         }
5785
5786         bno = NULLFSBLOCK;
5787         mp = ip->i_mount;
5788         ifp = XFS_IFORK_PTR(ip, whichfork);
5789         block = ifp->if_broot;
5790         /*
5791          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
5792          */
5793         level = be16_to_cpu(block->bb_level);
5794         ASSERT(level > 0);
5795         xfs_check_block(block, mp, 1, ifp->if_broot_bytes);
5796         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
5797         bno = be64_to_cpu(*pp);
5798
5799         ASSERT(bno != NULLDFSBNO);
5800         ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
5801         ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
5802
5803         /*
5804          * Go down the tree until leaf level is reached, following the first
5805          * pointer (leftmost) at each level.
5806          */
5807         while (level-- > 0) {
5808                 /* See if buf is in cur first */
5809                 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
5810                 if (bp) {
5811                         bp_release = 0;
5812                 } else {
5813                         bp_release = 1;
5814                 }
5815                 if (!bp && (error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
5816                                 XFS_BMAP_BTREE_REF)))
5817                         goto error_norelse;
5818                 block = XFS_BUF_TO_BLOCK(bp);
5819                 XFS_WANT_CORRUPTED_GOTO(
5820                         xfs_bmap_sanity_check(mp, bp, level),
5821                         error0);
5822                 if (level == 0)
5823                         break;
5824
5825                 /*
5826                  * Check this block for basic sanity (increasing keys and
5827                  * no duplicate blocks).
5828                  */
5829
5830                 xfs_check_block(block, mp, 0, 0);
5831                 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
5832                 bno = be64_to_cpu(*pp);
5833                 XFS_WANT_CORRUPTED_GOTO(XFS_FSB_SANITY_CHECK(mp, bno), error0);
5834                 if (bp_release) {
5835                         bp_release = 0;
5836                         xfs_trans_brelse(NULL, bp);
5837                 }
5838         }
5839
5840         /*
5841          * Here with bp and block set to the leftmost leaf node in the tree.
5842          */
5843         i = 0;
5844
5845         /*
5846          * Loop over all leaf nodes checking that all extents are in the right order.
5847          */
5848         for (;;) {
5849                 xfs_fsblock_t   nextbno;
5850                 xfs_extnum_t    num_recs;
5851
5852
5853                 num_recs = xfs_btree_get_numrecs(block);
5854
5855                 /*
5856                  * Read-ahead the next leaf block, if any.
5857                  */
5858
5859                 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
5860
5861                 /*
5862                  * Check all the extents to make sure they are OK.
5863                  * If we had a previous block, the last entry should
5864                  * conform with the first entry in this one.
5865                  */
5866
5867                 ep = XFS_BMBT_REC_ADDR(mp, block, 1);
5868                 if (i) {
5869                         ASSERT(xfs_bmbt_disk_get_startoff(&last) +
5870                                xfs_bmbt_disk_get_blockcount(&last) <=
5871                                xfs_bmbt_disk_get_startoff(ep));
5872                 }
5873                 for (j = 1; j < num_recs; j++) {
5874                         nextp = XFS_BMBT_REC_ADDR(mp, block, j + 1);
5875                         ASSERT(xfs_bmbt_disk_get_startoff(ep) +
5876                                xfs_bmbt_disk_get_blockcount(ep) <=
5877                                xfs_bmbt_disk_get_startoff(nextp));
5878                         ep = nextp;
5879                 }
5880
5881                 last = *ep;
5882                 i += num_recs;
5883                 if (bp_release) {
5884                         bp_release = 0;
5885                         xfs_trans_brelse(NULL, bp);
5886                 }
5887                 bno = nextbno;
5888                 /*
5889                  * If we've reached the end, stop.
5890                  */
5891                 if (bno == NULLFSBLOCK)
5892                         break;
5893
5894                 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
5895                 if (bp) {
5896                         bp_release = 0;
5897                 } else {
5898                         bp_release = 1;
5899                 }
5900                 if (!bp && (error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
5901                                 XFS_BMAP_BTREE_REF)))
5902                         goto error_norelse;
5903                 block = XFS_BUF_TO_BLOCK(bp);
5904         }
5905         if (bp_release) {
5906                 bp_release = 0;
5907                 xfs_trans_brelse(NULL, bp);
5908         }
5909         return;
5910
5911 error0:
5912         xfs_warn(mp, "%s: at error0", __func__);
5913         if (bp_release)
5914                 xfs_trans_brelse(NULL, bp);
5915 error_norelse:
5916         xfs_warn(mp, "%s: BAD after btree leaves for %d extents",
5917                 __func__, i);
5918         panic("%s: CORRUPTED BTREE OR SOMETHING", __func__);
5919         return;
5920 }
5921 #endif
5922
5923 /*
5924  * Count fsblocks of the given fork.
5925  */
5926 int                                             /* error */
5927 xfs_bmap_count_blocks(
5928         xfs_trans_t             *tp,            /* transaction pointer */
5929         xfs_inode_t             *ip,            /* incore inode */
5930         int                     whichfork,      /* data or attr fork */
5931         int                     *count)         /* out: count of blocks */
5932 {
5933         struct xfs_btree_block  *block; /* current btree block */
5934         xfs_fsblock_t           bno;    /* block # of "block" */
5935         xfs_ifork_t             *ifp;   /* fork structure */
5936         int                     level;  /* btree level, for checking */
5937         xfs_mount_t             *mp;    /* file system mount structure */
5938         __be64                  *pp;    /* pointer to block address */
5939
5940         bno = NULLFSBLOCK;
5941         mp = ip->i_mount;
5942         ifp = XFS_IFORK_PTR(ip, whichfork);
5943         if ( XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ) {
5944                 xfs_bmap_count_leaves(ifp, 0,
5945                         ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t),
5946                         count);
5947                 return 0;
5948         }
5949
5950         /*
5951          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
5952          */
5953         block = ifp->if_broot;
5954         level = be16_to_cpu(block->bb_level);
5955         ASSERT(level > 0);
5956         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
5957         bno = be64_to_cpu(*pp);
5958         ASSERT(bno != NULLDFSBNO);
5959         ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
5960         ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
5961
5962         if (unlikely(xfs_bmap_count_tree(mp, tp, ifp, bno, level, count) < 0)) {
5963                 XFS_ERROR_REPORT("xfs_bmap_count_blocks(2)", XFS_ERRLEVEL_LOW,
5964                                  mp);
5965                 return XFS_ERROR(EFSCORRUPTED);
5966         }
5967
5968         return 0;
5969 }
5970
5971 /*
5972  * Recursively walks each level of a btree
5973  * to count total fsblocks is use.
5974  */
5975 STATIC int                                     /* error */
5976 xfs_bmap_count_tree(
5977         xfs_mount_t     *mp,            /* file system mount point */
5978         xfs_trans_t     *tp,            /* transaction pointer */
5979         xfs_ifork_t     *ifp,           /* inode fork pointer */
5980         xfs_fsblock_t   blockno,        /* file system block number */
5981         int             levelin,        /* level in btree */
5982         int             *count)         /* Count of blocks */
5983 {
5984         int                     error;
5985         xfs_buf_t               *bp, *nbp;
5986         int                     level = levelin;
5987         __be64                  *pp;
5988         xfs_fsblock_t           bno = blockno;
5989         xfs_fsblock_t           nextbno;
5990         struct xfs_btree_block  *block, *nextblock;
5991         int                     numrecs;
5992
5993         if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp, XFS_BMAP_BTREE_REF)))
5994                 return error;
5995         *count += 1;
5996         block = XFS_BUF_TO_BLOCK(bp);
5997
5998         if (--level) {
5999                 /* Not at node above leaves, count this level of nodes */
6000                 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
6001                 while (nextbno != NULLFSBLOCK) {
6002                         if ((error = xfs_btree_read_bufl(mp, tp, nextbno,
6003                                 0, &nbp, XFS_BMAP_BTREE_REF)))
6004                                 return error;
6005                         *count += 1;
6006                         nextblock = XFS_BUF_TO_BLOCK(nbp);
6007                         nextbno = be64_to_cpu(nextblock->bb_u.l.bb_rightsib);
6008                         xfs_trans_brelse(tp, nbp);
6009                 }
6010
6011                 /* Dive to the next level */
6012                 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
6013                 bno = be64_to_cpu(*pp);
6014                 if (unlikely((error =
6015                      xfs_bmap_count_tree(mp, tp, ifp, bno, level, count)) < 0)) {
6016                         xfs_trans_brelse(tp, bp);
6017                         XFS_ERROR_REPORT("xfs_bmap_count_tree(1)",
6018                                          XFS_ERRLEVEL_LOW, mp);
6019                         return XFS_ERROR(EFSCORRUPTED);
6020                 }
6021                 xfs_trans_brelse(tp, bp);
6022         } else {
6023                 /* count all level 1 nodes and their leaves */
6024                 for (;;) {
6025                         nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
6026                         numrecs = be16_to_cpu(block->bb_numrecs);
6027                         xfs_bmap_disk_count_leaves(mp, block, numrecs, count);
6028                         xfs_trans_brelse(tp, bp);
6029                         if (nextbno == NULLFSBLOCK)
6030                                 break;
6031                         bno = nextbno;
6032                         if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
6033                                 XFS_BMAP_BTREE_REF)))
6034                                 return error;
6035                         *count += 1;
6036                         block = XFS_BUF_TO_BLOCK(bp);
6037                 }
6038         }
6039         return 0;
6040 }
6041
6042 /*
6043  * Count leaf blocks given a range of extent records.
6044  */
6045 STATIC void
6046 xfs_bmap_count_leaves(
6047         xfs_ifork_t             *ifp,
6048         xfs_extnum_t            idx,
6049         int                     numrecs,
6050         int                     *count)
6051 {
6052         int             b;
6053
6054         for (b = 0; b < numrecs; b++) {
6055                 xfs_bmbt_rec_host_t *frp = xfs_iext_get_ext(ifp, idx + b);
6056                 *count += xfs_bmbt_get_blockcount(frp);
6057         }
6058 }
6059
6060 /*
6061  * Count leaf blocks given a range of extent records originally
6062  * in btree format.
6063  */
6064 STATIC void
6065 xfs_bmap_disk_count_leaves(
6066         struct xfs_mount        *mp,
6067         struct xfs_btree_block  *block,
6068         int                     numrecs,
6069         int                     *count)
6070 {
6071         int             b;
6072         xfs_bmbt_rec_t  *frp;
6073
6074         for (b = 1; b <= numrecs; b++) {
6075                 frp = XFS_BMBT_REC_ADDR(mp, block, b);
6076                 *count += xfs_bmbt_disk_get_blockcount(frp);
6077         }
6078 }
6079
6080 /*
6081  * dead simple method of punching delalyed allocation blocks from a range in
6082  * the inode. Walks a block at a time so will be slow, but is only executed in
6083  * rare error cases so the overhead is not critical. This will alays punch out
6084  * both the start and end blocks, even if the ranges only partially overlap
6085  * them, so it is up to the caller to ensure that partial blocks are not
6086  * passed in.
6087  */
6088 int
6089 xfs_bmap_punch_delalloc_range(
6090         struct xfs_inode        *ip,
6091         xfs_fileoff_t           start_fsb,
6092         xfs_fileoff_t           length)
6093 {
6094         xfs_fileoff_t           remaining = length;
6095         int                     error = 0;
6096
6097         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
6098
6099         do {
6100                 int             done;
6101                 xfs_bmbt_irec_t imap;
6102                 int             nimaps = 1;
6103                 xfs_fsblock_t   firstblock;
6104                 xfs_bmap_free_t flist;
6105
6106                 /*
6107                  * Map the range first and check that it is a delalloc extent
6108                  * before trying to unmap the range. Otherwise we will be
6109                  * trying to remove a real extent (which requires a
6110                  * transaction) or a hole, which is probably a bad idea...
6111                  */
6112                 error = xfs_bmapi_read(ip, start_fsb, 1, &imap, &nimaps,
6113                                        XFS_BMAPI_ENTIRE);
6114
6115                 if (error) {
6116                         /* something screwed, just bail */
6117                         if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
6118                                 xfs_alert(ip->i_mount,
6119                         "Failed delalloc mapping lookup ino %lld fsb %lld.",
6120                                                 ip->i_ino, start_fsb);
6121                         }
6122                         break;
6123                 }
6124                 if (!nimaps) {
6125                         /* nothing there */
6126                         goto next_block;
6127                 }
6128                 if (imap.br_startblock != DELAYSTARTBLOCK) {
6129                         /* been converted, ignore */
6130                         goto next_block;
6131                 }
6132                 WARN_ON(imap.br_blockcount == 0);
6133
6134                 /*
6135                  * Note: while we initialise the firstblock/flist pair, they
6136                  * should never be used because blocks should never be
6137                  * allocated or freed for a delalloc extent and hence we need
6138                  * don't cancel or finish them after the xfs_bunmapi() call.
6139                  */
6140                 xfs_bmap_init(&flist, &firstblock);
6141                 error = xfs_bunmapi(NULL, ip, start_fsb, 1, 0, 1, &firstblock,
6142                                         &flist, &done);
6143                 if (error)
6144                         break;
6145
6146                 ASSERT(!flist.xbf_count && !flist.xbf_first);
6147 next_block:
6148                 start_fsb++;
6149                 remaining--;
6150         } while(remaining > 0);
6151
6152         return error;
6153 }