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_dir2_sf.h"
  33 #include "xfs_dinode.h"
  34 #include "xfs_inode.h"
  35 #include "xfs_btree.h"
  36 #include "xfs_mount.h"
  37 #include "xfs_itable.h"
  38 #include "xfs_dir2_data.h"
  39 #include "xfs_dir2_leaf.h"
  40 #include "xfs_dir2_block.h"
  41 #include "xfs_inode_item.h"
  42 #include "xfs_extfree_item.h"
  43 #include "xfs_alloc.h"
  44 #include "xfs_bmap.h"
  45 #include "xfs_rtalloc.h"
  46 #include "xfs_error.h"
  47 #include "xfs_attr_leaf.h"
  48 #include "xfs_rw.h"
  49 #include "xfs_quota.h"
  50 #include "xfs_trans_space.h"
  51 #include "xfs_buf_item.h"
  52 #include "xfs_filestream.h"
  53 #include "xfs_vnodeops.h"
  54 #include "xfs_trace.h"
  55
  56
  57 #ifdef DEBUG
  58 STATIC void
  59 xfs_bmap_check_leaf_extents(xfs_btree_cur_t *cur, xfs_inode_t *ip, int whichfork);
  60 #endif
  61
  62 kmem_zone_t             *xfs_bmap_free_item_zone;
  63
  64 /*
  65  * Prototypes for internal bmap routines.
  66  */
  67
  68
  69 /*
  70  * Called from xfs_bmap_add_attrfork to handle extents format files.
  71  */
  72 STATIC int                                      /* error */
  73 xfs_bmap_add_attrfork_extents(
  74         xfs_trans_t             *tp,            /* transaction pointer */
  75         xfs_inode_t             *ip,            /* incore inode pointer */
  76         xfs_fsblock_t           *firstblock,    /* first block allocated */
  77         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
  78         int                     *flags);        /* inode logging flags */
  79
  80 /*
  81  * Called from xfs_bmap_add_attrfork to handle local format files.
  82  */
  83 STATIC int                                      /* error */
  84 xfs_bmap_add_attrfork_local(
  85         xfs_trans_t             *tp,            /* transaction pointer */
  86         xfs_inode_t             *ip,            /* incore inode pointer */
  87         xfs_fsblock_t           *firstblock,    /* first block allocated */
  88         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
  89         int                     *flags);        /* inode logging flags */
  90
  91 /*
  92  * Called by xfs_bmapi to update file extent records and the btree
  93  * after allocating space (or doing a delayed allocation).
  94  */
  95 STATIC int                              /* error */
  96 xfs_bmap_add_extent(
  97         xfs_inode_t             *ip,    /* incore inode pointer */
  98         xfs_extnum_t            idx,    /* extent number to update/insert */
  99         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
 100         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
 101         xfs_fsblock_t           *first, /* pointer to firstblock variable */
 102         xfs_bmap_free_t         *flist, /* list of extents to be freed */
 103         int                     *logflagsp, /* inode logging flags */
 104         int                     whichfork, /* data or attr fork */
 105         int                     rsvd);  /* OK to allocate reserved blocks */
 106
 107 /*
 108  * Called by xfs_bmap_add_extent to handle cases converting a delayed
 109  * allocation to a real allocation.
 110  */
 111 STATIC int                              /* error */
 112 xfs_bmap_add_extent_delay_real(
 113         xfs_inode_t             *ip,    /* incore inode pointer */
 114         xfs_extnum_t            idx,    /* extent number to update/insert */
 115         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
 116         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
 117         xfs_filblks_t           *dnew,  /* new delayed-alloc indirect blocks */
 118         xfs_fsblock_t           *first, /* pointer to firstblock variable */
 119         xfs_bmap_free_t         *flist, /* list of extents to be freed */
 120         int                     *logflagsp, /* inode logging flags */
 121         int                     rsvd);  /* OK to allocate reserved blocks */
 122
 123 /*
 124  * Called by xfs_bmap_add_extent to handle cases converting a hole
 125  * to a delayed allocation.
 126  */
 127 STATIC int                              /* error */
 128 xfs_bmap_add_extent_hole_delay(
 129         xfs_inode_t             *ip,    /* incore inode pointer */
 130         xfs_extnum_t            idx,    /* extent number to update/insert */
 131         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
 132         int                     *logflagsp,/* inode logging flags */
 133         int                     rsvd);  /* OK to allocate reserved blocks */
 134
 135 /*
 136  * Called by xfs_bmap_add_extent to handle cases converting a hole
 137  * to a real allocation.
 138  */
 139 STATIC int                              /* error */
 140 xfs_bmap_add_extent_hole_real(
 141         xfs_inode_t             *ip,    /* incore inode pointer */
 142         xfs_extnum_t            idx,    /* extent number to update/insert */
 143         xfs_btree_cur_t         *cur,   /* if null, not a btree */
 144         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
 145         int                     *logflagsp, /* inode logging flags */
 146         int                     whichfork); /* data or attr fork */
 147
 148 /*
 149  * Called by xfs_bmap_add_extent to handle cases converting an unwritten
 150  * allocation to a real allocation or vice versa.
 151  */
 152 STATIC int                              /* error */
 153 xfs_bmap_add_extent_unwritten_real(
 154         xfs_inode_t             *ip,    /* incore inode pointer */
 155         xfs_extnum_t            idx,    /* extent number to update/insert */
 156         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
 157         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
 158         int                     *logflagsp); /* inode logging flags */
 159
 160 /*
 161  * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
 162  * It figures out where to ask the underlying allocator to put the new extent.
 163  */
 164 STATIC int                              /* error */
 165 xfs_bmap_alloc(
 166         xfs_bmalloca_t          *ap);   /* bmap alloc argument struct */
 167
 168 /*
 169  * Transform a btree format file with only one leaf node, where the
 170  * extents list will fit in the inode, into an extents format file.
 171  * Since the file extents are already in-core, all we have to do is
 172  * give up the space for the btree root and pitch the leaf block.
 173  */
 174 STATIC int                              /* error */
 175 xfs_bmap_btree_to_extents(
 176         xfs_trans_t             *tp,    /* transaction pointer */
 177         xfs_inode_t             *ip,    /* incore inode pointer */
 178         xfs_btree_cur_t         *cur,   /* btree cursor */
 179         int                     *logflagsp, /* inode logging flags */
 180         int                     whichfork); /* data or attr fork */
 181
 182 /*
 183  * Called by xfs_bmapi to update file extent records and the btree
 184  * after removing space (or undoing a delayed allocation).
 185  */
 186 STATIC int                              /* error */
 187 xfs_bmap_del_extent(
 188         xfs_inode_t             *ip,    /* incore inode pointer */
 189         xfs_trans_t             *tp,    /* current trans pointer */
 190         xfs_extnum_t            idx,    /* extent number to update/insert */
 191         xfs_bmap_free_t         *flist, /* list of extents to be freed */
 192         xfs_btree_cur_t         *cur,   /* if null, not a btree */
 193         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
 194         int                     *logflagsp,/* inode logging flags */
 195         int                     whichfork, /* data or attr fork */
 196         int                     rsvd);   /* OK to allocate reserved blocks */
 197
 198 /*
 199  * Remove the entry "free" from the free item list.  Prev points to the
 200  * previous entry, unless "free" is the head of the list.
 201  */
 202 STATIC void
 203 xfs_bmap_del_free(
 204         xfs_bmap_free_t         *flist, /* free item list header */
 205         xfs_bmap_free_item_t    *prev,  /* previous item on list, if any */
 206         xfs_bmap_free_item_t    *free); /* list item to be freed */
 207
 208 /*
 209  * Convert an extents-format file into a btree-format file.
 210  * The new file will have a root block (in the inode) and a single child block.
 211  */
 212 STATIC int                                      /* error */
 213 xfs_bmap_extents_to_btree(
 214         xfs_trans_t             *tp,            /* transaction pointer */
 215         xfs_inode_t             *ip,            /* incore inode pointer */
 216         xfs_fsblock_t           *firstblock,    /* first-block-allocated */
 217         xfs_bmap_free_t         *flist,         /* blocks freed in xaction */
 218         xfs_btree_cur_t         **curp,         /* cursor returned to caller */
 219         int                     wasdel,         /* converting a delayed alloc */
 220         int                     *logflagsp,     /* inode logging flags */
 221         int                     whichfork);     /* data or attr fork */
 222
 223 /*
 224  * Convert a local file to an extents file.
 225  * This code is sort of bogus, since the file data needs to get
 226  * logged so it won't be lost.  The bmap-level manipulations are ok, though.
 227  */
 228 STATIC int                              /* error */
 229 xfs_bmap_local_to_extents(
 230         xfs_trans_t     *tp,            /* transaction pointer */
 231         xfs_inode_t     *ip,            /* incore inode pointer */
 232         xfs_fsblock_t   *firstblock,    /* first block allocated in xaction */
 233         xfs_extlen_t    total,          /* total blocks needed by transaction */
 234         int             *logflagsp,     /* inode logging flags */
 235         int             whichfork);     /* data or attr fork */
 236
 237 /*
 238  * Search the extents list for the inode, for the extent containing bno.
 239  * If bno lies in a hole, point to the next entry.  If bno lies past eof,
 240  * *eofp will be set, and *prevp will contain the last entry (null if none).
 241  * Else, *lastxp will be set to the index of the found
 242  * entry; *gotp will contain the entry.
 243  */
 244 STATIC xfs_bmbt_rec_host_t *            /* pointer to found extent entry */
 245 xfs_bmap_search_extents(
 246         xfs_inode_t     *ip,            /* incore inode pointer */
 247         xfs_fileoff_t   bno,            /* block number searched for */
 248         int             whichfork,      /* data or attr fork */
 249         int             *eofp,          /* out: end of file found */
 250         xfs_extnum_t    *lastxp,        /* out: last extent index */
 251         xfs_bmbt_irec_t *gotp,          /* out: extent entry found */
 252         xfs_bmbt_irec_t *prevp);        /* out: previous extent entry found */
 253
 254 /*
 255  * Check the last inode extent to determine whether this allocation will result
 256  * in blocks being allocated at the end of the file. When we allocate new data
 257  * blocks at the end of the file which do not start at the previous data block,
 258  * we will try to align the new blocks at stripe unit boundaries.
 259  */
 260 STATIC int                              /* error */
 261 xfs_bmap_isaeof(
 262         xfs_inode_t     *ip,            /* incore inode pointer */
 263         xfs_fileoff_t   off,            /* file offset in fsblocks */
 264         int             whichfork,      /* data or attribute fork */
 265         char            *aeof);         /* return value */
 266
 267 /*
 268  * Compute the worst-case number of indirect blocks that will be used
 269  * for ip's delayed extent of length "len".
 270  */
 271 STATIC xfs_filblks_t
 272 xfs_bmap_worst_indlen(
 273         xfs_inode_t             *ip,    /* incore inode pointer */
 274         xfs_filblks_t           len);   /* delayed extent length */
 275
 276 #ifdef DEBUG
 277 /*
 278  * Perform various validation checks on the values being returned
 279  * from xfs_bmapi().
 280  */
 281 STATIC void
 282 xfs_bmap_validate_ret(
 283         xfs_fileoff_t           bno,
 284         xfs_filblks_t           len,
 285         int                     flags,
 286         xfs_bmbt_irec_t         *mval,
 287         int                     nmap,
 288         int                     ret_nmap);
 289 #else
 290 #define xfs_bmap_validate_ret(bno,len,flags,mval,onmap,nmap)
 291 #endif /* DEBUG */
 292
 293 STATIC int
 294 xfs_bmap_count_tree(
 295         xfs_mount_t     *mp,
 296         xfs_trans_t     *tp,
 297         xfs_ifork_t     *ifp,
 298         xfs_fsblock_t   blockno,
 299         int             levelin,
 300         int             *count);
 301
 302 STATIC void
 303 xfs_bmap_count_leaves(
 304         xfs_ifork_t             *ifp,
 305         xfs_extnum_t            idx,
 306         int                     numrecs,
 307         int                     *count);
 308
 309 STATIC void
 310 xfs_bmap_disk_count_leaves(
 311         struct xfs_mount        *mp,
 312         struct xfs_btree_block  *block,
 313         int                     numrecs,
 314         int                     *count);
 315
 316 /*
 317  * Bmap internal routines.
 318  */
 319
 320 STATIC int                              /* error */
 321 xfs_bmbt_lookup_eq(
 322         struct xfs_btree_cur    *cur,
 323         xfs_fileoff_t           off,
 324         xfs_fsblock_t           bno,
 325         xfs_filblks_t           len,
 326         int                     *stat)  /* success/failure */
 327 {
 328         cur->bc_rec.b.br_startoff = off;
 329         cur->bc_rec.b.br_startblock = bno;
 330         cur->bc_rec.b.br_blockcount = len;
 331         return xfs_btree_lookup(cur, XFS_LOOKUP_EQ, stat);
 332 }
 333
 334 STATIC int                              /* error */
 335 xfs_bmbt_lookup_ge(
 336         struct xfs_btree_cur    *cur,
 337         xfs_fileoff_t           off,
 338         xfs_fsblock_t           bno,
 339         xfs_filblks_t           len,
 340         int                     *stat)  /* success/failure */
 341 {
 342         cur->bc_rec.b.br_startoff = off;
 343         cur->bc_rec.b.br_startblock = bno;
 344         cur->bc_rec.b.br_blockcount = len;
 345         return xfs_btree_lookup(cur, XFS_LOOKUP_GE, stat);
 346 }
 347
 348 /*
 349 * Update the record referred to by cur to the value given
 350  * by [off, bno, len, state].
 351  * This either works (return 0) or gets an EFSCORRUPTED error.
 352  */
 353 STATIC int
 354 xfs_bmbt_update(
 355         struct xfs_btree_cur    *cur,
 356         xfs_fileoff_t           off,
 357         xfs_fsblock_t           bno,
 358         xfs_filblks_t           len,
 359         xfs_exntst_t            state)
 360 {
 361         union xfs_btree_rec     rec;
 362
 363         xfs_bmbt_disk_set_allf(&rec.bmbt, off, bno, len, state);
 364         return xfs_btree_update(cur, &rec);
 365 }
 366
 367 /*
 368  * Called from xfs_bmap_add_attrfork to handle btree format files.
 369  */
 370 STATIC int                                      /* error */
 371 xfs_bmap_add_attrfork_btree(
 372         xfs_trans_t             *tp,            /* transaction pointer */
 373         xfs_inode_t             *ip,            /* incore inode pointer */
 374         xfs_fsblock_t           *firstblock,    /* first block allocated */
 375         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
 376         int                     *flags)         /* inode logging flags */
 377 {
 378         xfs_btree_cur_t         *cur;           /* btree cursor */
 379         int                     error;          /* error return value */
 380         xfs_mount_t             *mp;            /* file system mount struct */
 381         int                     stat;           /* newroot status */
 382
 383         mp = ip->i_mount;
 384         if (ip->i_df.if_broot_bytes <= XFS_IFORK_DSIZE(ip))
 385                 *flags |= XFS_ILOG_DBROOT;
 386         else {
 387                 cur = xfs_bmbt_init_cursor(mp, tp, ip, XFS_DATA_FORK);
 388                 cur->bc_private.b.flist = flist;
 389                 cur->bc_private.b.firstblock = *firstblock;
 390                 if ((error = xfs_bmbt_lookup_ge(cur, 0, 0, 0, &stat)))
 391                         goto error0;
 392                 /* must be at least one entry */
 393                 XFS_WANT_CORRUPTED_GOTO(stat == 1, error0);
 394                 if ((error = xfs_btree_new_iroot(cur, flags, &stat)))
 395                         goto error0;
 396                 if (stat == 0) {
 397                         xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
 398                         return XFS_ERROR(ENOSPC);
 399                 }
 400                 *firstblock = cur->bc_private.b.firstblock;
 401                 cur->bc_private.b.allocated = 0;
 402                 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
 403         }
 404         return 0;
 405 error0:
 406         xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
 407         return error;
 408 }
 409
 410 /*
 411  * Called from xfs_bmap_add_attrfork to handle extents format files.
 412  */
 413 STATIC int                                      /* error */
 414 xfs_bmap_add_attrfork_extents(
 415         xfs_trans_t             *tp,            /* transaction pointer */
 416         xfs_inode_t             *ip,            /* incore inode pointer */
 417         xfs_fsblock_t           *firstblock,    /* first block allocated */
 418         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
 419         int                     *flags)         /* inode logging flags */
 420 {
 421         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
 422         int                     error;          /* error return value */
 423
 424         if (ip->i_d.di_nextents * sizeof(xfs_bmbt_rec_t) <= XFS_IFORK_DSIZE(ip))
 425                 return 0;
 426         cur = NULL;
 427         error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist, &cur, 0,
 428                 flags, XFS_DATA_FORK);
 429         if (cur) {
 430                 cur->bc_private.b.allocated = 0;
 431                 xfs_btree_del_cursor(cur,
 432                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
 433         }
 434         return error;
 435 }
 436
 437 /*
 438  * Called from xfs_bmap_add_attrfork to handle local format files.
 439  */
 440 STATIC int                                      /* error */
 441 xfs_bmap_add_attrfork_local(
 442         xfs_trans_t             *tp,            /* transaction pointer */
 443         xfs_inode_t             *ip,            /* incore inode pointer */
 444         xfs_fsblock_t           *firstblock,    /* first block allocated */
 445         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
 446         int                     *flags)         /* inode logging flags */
 447 {
 448         xfs_da_args_t           dargs;          /* args for dir/attr code */
 449         int                     error;          /* error return value */
 450         xfs_mount_t             *mp;            /* mount structure pointer */
 451
 452         if (ip->i_df.if_bytes <= XFS_IFORK_DSIZE(ip))
 453                 return 0;
 454         if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) {
 455                 mp = ip->i_mount;
 456                 memset(&dargs, 0, sizeof(dargs));
 457                 dargs.dp = ip;
 458                 dargs.firstblock = firstblock;
 459                 dargs.flist = flist;
 460                 dargs.total = mp->m_dirblkfsbs;
 461                 dargs.whichfork = XFS_DATA_FORK;
 462                 dargs.trans = tp;
 463                 error = xfs_dir2_sf_to_block(&dargs);
 464         } else
 465                 error = xfs_bmap_local_to_extents(tp, ip, firstblock, 1, flags,
 466                         XFS_DATA_FORK);
 467         return error;
 468 }
 469
 470 /*
 471  * Called by xfs_bmapi to update file extent records and the btree
 472  * after allocating space (or doing a delayed allocation).
 473  */
 474 STATIC int                              /* error */
 475 xfs_bmap_add_extent(
 476         xfs_inode_t             *ip,    /* incore inode pointer */
 477         xfs_extnum_t            idx,    /* extent number to update/insert */
 478         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
 479         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
 480         xfs_fsblock_t           *first, /* pointer to firstblock variable */
 481         xfs_bmap_free_t         *flist, /* list of extents to be freed */
 482         int                     *logflagsp, /* inode logging flags */
 483         int                     whichfork, /* data or attr fork */
 484         int                     rsvd)   /* OK to use reserved data blocks */
 485 {
 486         xfs_btree_cur_t         *cur;   /* btree cursor or null */
 487         xfs_filblks_t           da_new; /* new count del alloc blocks used */
 488         xfs_filblks_t           da_old; /* old count del alloc blocks used */
 489         int                     error;  /* error return value */
 490         xfs_ifork_t             *ifp;   /* inode fork ptr */
 491         int                     logflags; /* returned value */
 492         xfs_extnum_t            nextents; /* number of extents in file now */
 493
 494         XFS_STATS_INC(xs_add_exlist);
 495         cur = *curp;
 496         ifp = XFS_IFORK_PTR(ip, whichfork);
 497         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
 498         ASSERT(idx <= nextents);
 499         da_old = da_new = 0;
 500         error = 0;
 501         /*
 502          * This is the first extent added to a new/empty file.
 503          * Special case this one, so other routines get to assume there are
 504          * already extents in the list.
 505          */
 506         if (nextents == 0) {
 507                 xfs_iext_insert(ip, 0, 1, new,
 508                                 whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0);
 509
 510                 ASSERT(cur == NULL);
 511                 ifp->if_lastex = 0;
 512                 if (!isnullstartblock(new->br_startblock)) {
 513                         XFS_IFORK_NEXT_SET(ip, whichfork, 1);
 514                         logflags = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
 515                 } else
 516                         logflags = 0;
 517         }
 518         /*
 519          * Any kind of new delayed allocation goes here.
 520          */
 521         else if (isnullstartblock(new->br_startblock)) {
 522                 if (cur)
 523                         ASSERT((cur->bc_private.b.flags &
 524                                 XFS_BTCUR_BPRV_WASDEL) == 0);
 525                 if ((error = xfs_bmap_add_extent_hole_delay(ip, idx, new,
 526                                 &logflags, rsvd)))
 527                         goto done;
 528         }
 529         /*
 530          * Real allocation off the end of the file.
 531          */
 532         else if (idx == nextents) {
 533                 if (cur)
 534                         ASSERT((cur->bc_private.b.flags &
 535                                 XFS_BTCUR_BPRV_WASDEL) == 0);
 536                 if ((error = xfs_bmap_add_extent_hole_real(ip, idx, cur, new,
 537                                 &logflags, whichfork)))
 538                         goto done;
 539         } else {
 540                 xfs_bmbt_irec_t prev;   /* old extent at offset idx */
 541
 542                 /*
 543                  * Get the record referred to by idx.
 544                  */
 545                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, idx), &prev);
 546                 /*
 547                  * If it's a real allocation record, and the new allocation ends
 548                  * after the start of the referred to record, then we're filling
 549                  * in a delayed or unwritten allocation with a real one, or
 550                  * converting real back to unwritten.
 551                  */
 552                 if (!isnullstartblock(new->br_startblock) &&
 553                     new->br_startoff + new->br_blockcount > prev.br_startoff) {
 554                         if (prev.br_state != XFS_EXT_UNWRITTEN &&
 555                             isnullstartblock(prev.br_startblock)) {
 556                                 da_old = startblockval(prev.br_startblock);
 557                                 if (cur)
 558                                         ASSERT(cur->bc_private.b.flags &
 559                                                 XFS_BTCUR_BPRV_WASDEL);
 560                                 if ((error = xfs_bmap_add_extent_delay_real(ip,
 561                                         idx, &cur, new, &da_new, first, flist,
 562                                         &logflags, rsvd)))
 563                                         goto done;
 564                         } else if (new->br_state == XFS_EXT_NORM) {
 565                                 ASSERT(new->br_state == XFS_EXT_NORM);
 566                                 if ((error = xfs_bmap_add_extent_unwritten_real(
 567                                         ip, idx, &cur, new, &logflags)))
 568                                         goto done;
 569                         } else {
 570                                 ASSERT(new->br_state == XFS_EXT_UNWRITTEN);
 571                                 if ((error = xfs_bmap_add_extent_unwritten_real(
 572                                         ip, idx, &cur, new, &logflags)))
 573                                         goto done;
 574                         }
 575                         ASSERT(*curp == cur || *curp == NULL);
 576                 }
 577                 /*
 578                  * Otherwise we're filling in a hole with an allocation.
 579                  */
 580                 else {
 581                         if (cur)
 582                                 ASSERT((cur->bc_private.b.flags &
 583                                         XFS_BTCUR_BPRV_WASDEL) == 0);
 584                         if ((error = xfs_bmap_add_extent_hole_real(ip, idx, cur,
 585                                         new, &logflags, whichfork)))
 586                                 goto done;
 587                 }
 588         }
 589
 590         ASSERT(*curp == cur || *curp == NULL);
 591         /*
 592          * Convert to a btree if necessary.
 593          */
 594         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
 595             XFS_IFORK_NEXTENTS(ip, whichfork) > ifp->if_ext_max) {
 596                 int     tmp_logflags;   /* partial log flag return val */
 597
 598                 ASSERT(cur == NULL);
 599                 error = xfs_bmap_extents_to_btree(ip->i_transp, ip, first,
 600                         flist, &cur, da_old > 0, &tmp_logflags, whichfork);
 601                 logflags |= tmp_logflags;
 602                 if (error)
 603                         goto done;
 604         }
 605         /*
 606          * Adjust for changes in reserved delayed indirect blocks.
 607          * Nothing to do for disk quotas here.
 608          */
 609         if (da_old || da_new) {
 610                 xfs_filblks_t   nblks;
 611
 612                 nblks = da_new;
 613                 if (cur)
 614                         nblks += cur->bc_private.b.allocated;
 615                 ASSERT(nblks <= da_old);
 616                 if (nblks < da_old)
 617                         xfs_mod_incore_sb(ip->i_mount, XFS_SBS_FDBLOCKS,
 618                                 (int64_t)(da_old - nblks), rsvd);
 619         }
 620         /*
 621          * Clear out the allocated field, done with it now in any case.
 622          */
 623         if (cur) {
 624                 cur->bc_private.b.allocated = 0;
 625                 *curp = cur;
 626         }
 627 done:
 628 #ifdef DEBUG
 629         if (!error)
 630                 xfs_bmap_check_leaf_extents(*curp, ip, whichfork);
 631 #endif
 632         *logflagsp = logflags;
 633         return error;
 634 }
 635
 636 /*
 637  * Called by xfs_bmap_add_extent to handle cases converting a delayed
 638  * allocation to a real allocation.
 639  */
 640 STATIC int                              /* error */
 641 xfs_bmap_add_extent_delay_real(
 642         xfs_inode_t             *ip,    /* incore inode pointer */
 643         xfs_extnum_t            idx,    /* extent number to update/insert */
 644         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
 645         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
 646         xfs_filblks_t           *dnew,  /* new delayed-alloc indirect blocks */
 647         xfs_fsblock_t           *first, /* pointer to firstblock variable */
 648         xfs_bmap_free_t         *flist, /* list of extents to be freed */
 649         int                     *logflagsp, /* inode logging flags */
 650         int                     rsvd)   /* OK to use reserved data block allocation */
 651 {
 652         xfs_btree_cur_t         *cur;   /* btree cursor */
 653         int                     diff;   /* temp value */
 654         xfs_bmbt_rec_host_t     *ep;    /* extent entry for idx */
 655         int                     error;  /* error return value */
 656         int                     i;      /* temp state */
 657         xfs_ifork_t             *ifp;   /* inode fork pointer */
 658         xfs_fileoff_t           new_endoff;     /* end offset of new entry */
 659         xfs_bmbt_irec_t         r[3];   /* neighbor extent entries */
 660                                         /* left is 0, right is 1, prev is 2 */
 661         int                     rval=0; /* return value (logging flags) */
 662         int                     state = 0;/* state bits, accessed thru macros */
 663         xfs_filblks_t           temp=0; /* value for dnew calculations */
 664         xfs_filblks_t           temp2=0;/* value for dnew calculations */
 665         int                     tmp_rval;       /* partial logging flags */
 666
 667 #define LEFT            r[0]
 668 #define RIGHT           r[1]
 669 #define PREV            r[2]
 670
 671         /*
 672          * Set up a bunch of variables to make the tests simpler.
 673          */
 674         cur = *curp;
 675         ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
 676         ep = xfs_iext_get_ext(ifp, idx);
 677         xfs_bmbt_get_all(ep, &PREV);
 678         new_endoff = new->br_startoff + new->br_blockcount;
 679         ASSERT(PREV.br_startoff <= new->br_startoff);
 680         ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
 681
 682         /*
 683          * Set flags determining what part of the previous delayed allocation
 684          * extent is being replaced by a real allocation.
 685          */
 686         if (PREV.br_startoff == new->br_startoff)
 687                 state |= BMAP_LEFT_FILLING;
 688         if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
 689                 state |= BMAP_RIGHT_FILLING;
 690
 691         /*
 692          * Check and set flags if this segment has a left neighbor.
 693          * Don't set contiguous if the combined extent would be too large.
 694          */
 695         if (idx > 0) {
 696                 state |= BMAP_LEFT_VALID;
 697                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, idx - 1), &LEFT);
 698
 699                 if (isnullstartblock(LEFT.br_startblock))
 700                         state |= BMAP_LEFT_DELAY;
 701         }
 702
 703         if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
 704             LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
 705             LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
 706             LEFT.br_state == new->br_state &&
 707             LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
 708                 state |= BMAP_LEFT_CONTIG;
 709
 710         /*
 711          * Check and set flags if this segment has a right neighbor.
 712          * Don't set contiguous if the combined extent would be too large.
 713          * Also check for all-three-contiguous being too large.
 714          */
 715         if (idx < ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1) {
 716                 state |= BMAP_RIGHT_VALID;
 717                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, idx + 1), &RIGHT);
 718
 719                 if (isnullstartblock(RIGHT.br_startblock))
 720                         state |= BMAP_RIGHT_DELAY;
 721         }
 722
 723         if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
 724             new_endoff == RIGHT.br_startoff &&
 725             new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
 726             new->br_state == RIGHT.br_state &&
 727             new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
 728             ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
 729                        BMAP_RIGHT_FILLING)) !=
 730                       (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
 731                        BMAP_RIGHT_FILLING) ||
 732              LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
 733                         <= MAXEXTLEN))
 734                 state |= BMAP_RIGHT_CONTIG;
 735
 736         error = 0;
 737         /*
 738          * Switch out based on the FILLING and CONTIG state bits.
 739          */
 740         switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
 741                          BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
 742         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
 743              BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
 744                 /*
 745                  * Filling in all of a previously delayed allocation extent.
 746                  * The left and right neighbors are both contiguous with new.
 747                  */
 748                 trace_xfs_bmap_pre_update(ip, idx - 1, state, _THIS_IP_);
 749                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
 750                         LEFT.br_blockcount + PREV.br_blockcount +
 751                         RIGHT.br_blockcount);
 752                 trace_xfs_bmap_post_update(ip, idx - 1, state, _THIS_IP_);
 753
 754                 xfs_iext_remove(ip, idx, 2, state);
 755                 ip->i_df.if_lastex = idx - 1;
 756                 ip->i_d.di_nextents--;
 757                 if (cur == NULL)
 758                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
 759                 else {
 760                         rval = XFS_ILOG_CORE;
 761                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
 762                                         RIGHT.br_startblock,
 763                                         RIGHT.br_blockcount, &i)))
 764                                 goto done;
 765                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 766                         if ((error = xfs_btree_delete(cur, &i)))
 767                                 goto done;
 768                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 769                         if ((error = xfs_btree_decrement(cur, 0, &i)))
 770                                 goto done;
 771                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 772                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
 773                                         LEFT.br_startblock,
 774                                         LEFT.br_blockcount +
 775                                         PREV.br_blockcount +
 776                                         RIGHT.br_blockcount, LEFT.br_state)))
 777                                 goto done;
 778                 }
 779                 *dnew = 0;
 780                 break;
 781
 782         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
 783                 /*
 784                  * Filling in all of a previously delayed allocation extent.
 785                  * The left neighbor is contiguous, the right is not.
 786                  */
 787                 trace_xfs_bmap_pre_update(ip, idx - 1, state, _THIS_IP_);
 788                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
 789                         LEFT.br_blockcount + PREV.br_blockcount);
 790                 trace_xfs_bmap_post_update(ip, idx - 1, state, _THIS_IP_);
 791
 792                 ip->i_df.if_lastex = idx - 1;
 793                 xfs_iext_remove(ip, idx, 1, state);
 794                 if (cur == NULL)
 795                         rval = XFS_ILOG_DEXT;
 796                 else {
 797                         rval = 0;
 798                         if ((error = xfs_bmbt_lookup_eq(cur, LEFT.br_startoff,
 799                                         LEFT.br_startblock, LEFT.br_blockcount,
 800                                         &i)))
 801                                 goto done;
 802                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 803                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
 804                                         LEFT.br_startblock,
 805                                         LEFT.br_blockcount +
 806                                         PREV.br_blockcount, LEFT.br_state)))
 807                                 goto done;
 808                 }
 809                 *dnew = 0;
 810                 break;
 811
 812         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
 813                 /*
 814                  * Filling in all of a previously delayed allocation extent.
 815                  * The right neighbor is contiguous, the left is not.
 816                  */
 817                 trace_xfs_bmap_pre_update(ip, idx, state, _THIS_IP_);
 818                 xfs_bmbt_set_startblock(ep, new->br_startblock);
 819                 xfs_bmbt_set_blockcount(ep,
 820                         PREV.br_blockcount + RIGHT.br_blockcount);
 821                 trace_xfs_bmap_post_update(ip, idx, state, _THIS_IP_);
 822
 823                 ip->i_df.if_lastex = idx;
 824                 xfs_iext_remove(ip, idx + 1, 1, state);
 825                 if (cur == NULL)
 826                         rval = XFS_ILOG_DEXT;
 827                 else {
 828                         rval = 0;
 829                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
 830                                         RIGHT.br_startblock,
 831                                         RIGHT.br_blockcount, &i)))
 832                                 goto done;
 833                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 834                         if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
 835                                         new->br_startblock,
 836                                         PREV.br_blockcount +
 837                                         RIGHT.br_blockcount, PREV.br_state)))
 838                                 goto done;
 839                 }
 840                 *dnew = 0;
 841                 break;
 842
 843         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
 844                 /*
 845                  * Filling in all of a previously delayed allocation extent.
 846                  * Neither the left nor right neighbors are contiguous with
 847                  * the new one.
 848                  */
 849                 trace_xfs_bmap_pre_update(ip, idx, state, _THIS_IP_);
 850                 xfs_bmbt_set_startblock(ep, new->br_startblock);
 851                 trace_xfs_bmap_post_update(ip, idx, state, _THIS_IP_);
 852
 853                 ip->i_df.if_lastex = idx;
 854                 ip->i_d.di_nextents++;
 855                 if (cur == NULL)
 856                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
 857                 else {
 858                         rval = XFS_ILOG_CORE;
 859                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
 860                                         new->br_startblock, new->br_blockcount,
 861                                         &i)))
 862                                 goto done;
 863                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
 864                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
 865                         if ((error = xfs_btree_insert(cur, &i)))
 866                                 goto done;
 867                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 868                 }
 869                 *dnew = 0;
 870                 break;
 871
 872         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
 873                 /*
 874                  * Filling in the first part of a previous delayed allocation.
 875                  * The left neighbor is contiguous.
 876                  */
 877                 trace_xfs_bmap_pre_update(ip, idx - 1, state, _THIS_IP_);
 878                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
 879                         LEFT.br_blockcount + new->br_blockcount);
 880                 xfs_bmbt_set_startoff(ep,
 881                         PREV.br_startoff + new->br_blockcount);
 882                 trace_xfs_bmap_post_update(ip, idx - 1, state, _THIS_IP_);
 883
 884                 temp = PREV.br_blockcount - new->br_blockcount;
 885                 trace_xfs_bmap_pre_update(ip, idx, state, _THIS_IP_);
 886                 xfs_bmbt_set_blockcount(ep, temp);
 887                 ip->i_df.if_lastex = idx - 1;
 888                 if (cur == NULL)
 889                         rval = XFS_ILOG_DEXT;
 890                 else {
 891                         rval = 0;
 892                         if ((error = xfs_bmbt_lookup_eq(cur, LEFT.br_startoff,
 893                                         LEFT.br_startblock, LEFT.br_blockcount,
 894                                         &i)))
 895                                 goto done;
 896                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 897                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
 898                                         LEFT.br_startblock,
 899                                         LEFT.br_blockcount +
 900                                         new->br_blockcount,
 901                                         LEFT.br_state)))
 902                                 goto done;
 903                 }
 904                 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
 905                         startblockval(PREV.br_startblock));
 906                 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
 907                 trace_xfs_bmap_post_update(ip, idx, state, _THIS_IP_);
 908                 *dnew = temp;
 909                 break;
 910
 911         case BMAP_LEFT_FILLING:
 912                 /*
 913                  * Filling in the first part of a previous delayed allocation.
 914                  * The left neighbor is not contiguous.
 915                  */
 916                 trace_xfs_bmap_pre_update(ip, idx, state, _THIS_IP_);
 917                 xfs_bmbt_set_startoff(ep, new_endoff);
 918                 temp = PREV.br_blockcount - new->br_blockcount;
 919                 xfs_bmbt_set_blockcount(ep, temp);
 920                 xfs_iext_insert(ip, idx, 1, new, state);
 921                 ip->i_df.if_lastex = idx;
 922                 ip->i_d.di_nextents++;
 923                 if (cur == NULL)
 924                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
 925                 else {
 926                         rval = XFS_ILOG_CORE;
 927                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
 928                                         new->br_startblock, new->br_blockcount,
 929                                         &i)))
 930                                 goto done;
 931                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
 932                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
 933                         if ((error = xfs_btree_insert(cur, &i)))
 934                                 goto done;
 935                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 936                 }
 937                 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
 938                     ip->i_d.di_nextents > ip->i_df.if_ext_max) {
 939                         error = xfs_bmap_extents_to_btree(ip->i_transp, ip,
 940                                         first, flist, &cur, 1, &tmp_rval,
 941                                         XFS_DATA_FORK);
 942                         rval |= tmp_rval;
 943                         if (error)
 944                                 goto done;
 945                 }
 946                 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
 947                         startblockval(PREV.br_startblock) -
 948                         (cur ? cur->bc_private.b.allocated : 0));
 949                 ep = xfs_iext_get_ext(ifp, idx + 1);
 950                 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
 951                 trace_xfs_bmap_post_update(ip, idx + 1, state, _THIS_IP_);
 952                 *dnew = temp;
 953                 break;
 954
 955         case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
 956                 /*
 957                  * Filling in the last part of a previous delayed allocation.
 958                  * The right neighbor is contiguous with the new allocation.
 959                  */
 960                 temp = PREV.br_blockcount - new->br_blockcount;
 961                 trace_xfs_bmap_pre_update(ip, idx, state, _THIS_IP_);
 962                 trace_xfs_bmap_pre_update(ip, idx + 1, state, _THIS_IP_);
 963                 xfs_bmbt_set_blockcount(ep, temp);
 964                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, idx + 1),
 965                         new->br_startoff, new->br_startblock,
 966                         new->br_blockcount + RIGHT.br_blockcount,
 967                         RIGHT.br_state);
 968                 trace_xfs_bmap_post_update(ip, idx + 1, state, _THIS_IP_);
 969                 ip->i_df.if_lastex = idx + 1;
 970                 if (cur == NULL)
 971                         rval = XFS_ILOG_DEXT;
 972                 else {
 973                         rval = 0;
 974                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
 975                                         RIGHT.br_startblock,
 976                                         RIGHT.br_blockcount, &i)))
 977                                 goto done;
 978                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 979                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
 980                                         new->br_startblock,
 981                                         new->br_blockcount +
 982                                         RIGHT.br_blockcount,
 983                                         RIGHT.br_state)))
 984                                 goto done;
 985                 }
 986                 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
 987                         startblockval(PREV.br_startblock));
 988                 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
 989                 trace_xfs_bmap_post_update(ip, idx, state, _THIS_IP_);
 990                 *dnew = temp;
 991                 break;
 992
 993         case BMAP_RIGHT_FILLING:
 994                 /*
 995                  * Filling in the last part of a previous delayed allocation.
 996                  * The right neighbor is not contiguous.
 997                  */
 998                 temp = PREV.br_blockcount - new->br_blockcount;
 999                 trace_xfs_bmap_pre_update(ip, idx, state, _THIS_IP_);
1000                 xfs_bmbt_set_blockcount(ep, temp);
1001                 xfs_iext_insert(ip, idx + 1, 1, new, state);
1002                 ip->i_df.if_lastex = idx + 1;
1003                 ip->i_d.di_nextents++;
1004                 if (cur == NULL)
1005                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1006                 else {
1007                         rval = XFS_ILOG_CORE;
1008                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1009                                         new->br_startblock, new->br_blockcount,
1010                                         &i)))
1011                                 goto done;
1012                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1013                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
1014                         if ((error = xfs_btree_insert(cur, &i)))
1015                                 goto done;
1016                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1017                 }
1018                 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1019                     ip->i_d.di_nextents > ip->i_df.if_ext_max) {
1020                         error = xfs_bmap_extents_to_btree(ip->i_transp, ip,
1021                                 first, flist, &cur, 1, &tmp_rval,
1022                                 XFS_DATA_FORK);
1023                         rval |= tmp_rval;
1024                         if (error)
1025                                 goto done;
1026                 }
1027                 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
1028                         startblockval(PREV.br_startblock) -
1029                         (cur ? cur->bc_private.b.allocated : 0));
1030                 ep = xfs_iext_get_ext(ifp, idx);
1031                 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
1032                 trace_xfs_bmap_post_update(ip, idx, state, _THIS_IP_);
1033                 *dnew = temp;
1034                 break;
1035
1036         case 0:
1037                 /*
1038                  * Filling in the middle part of a previous delayed allocation.
1039                  * Contiguity is impossible here.
1040                  * This case is avoided almost all the time.
1041                  */
1042                 temp = new->br_startoff - PREV.br_startoff;
1043                 trace_xfs_bmap_pre_update(ip, idx, 0, _THIS_IP_);
1044                 xfs_bmbt_set_blockcount(ep, temp);
1045                 r[0] = *new;
1046                 r[1].br_state = PREV.br_state;
1047                 r[1].br_startblock = 0;
1048                 r[1].br_startoff = new_endoff;
1049                 temp2 = PREV.br_startoff + PREV.br_blockcount - new_endoff;
1050                 r[1].br_blockcount = temp2;
1051                 xfs_iext_insert(ip, idx + 1, 2, &r[0], state);
1052                 ip->i_df.if_lastex = idx + 1;
1053                 ip->i_d.di_nextents++;
1054                 if (cur == NULL)
1055                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1056                 else {
1057                         rval = XFS_ILOG_CORE;
1058                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1059                                         new->br_startblock, new->br_blockcount,
1060                                         &i)))
1061                                 goto done;
1062                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1063                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
1064                         if ((error = xfs_btree_insert(cur, &i)))
1065                                 goto done;
1066                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1067                 }
1068                 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1069                     ip->i_d.di_nextents > ip->i_df.if_ext_max) {
1070                         error = xfs_bmap_extents_to_btree(ip->i_transp, ip,
1071                                         first, flist, &cur, 1, &tmp_rval,
1072                                         XFS_DATA_FORK);
1073                         rval |= tmp_rval;
1074                         if (error)
1075                                 goto done;
1076                 }
1077                 temp = xfs_bmap_worst_indlen(ip, temp);
1078                 temp2 = xfs_bmap_worst_indlen(ip, temp2);
1079                 diff = (int)(temp + temp2 - startblockval(PREV.br_startblock) -
1080                         (cur ? cur->bc_private.b.allocated : 0));
1081                 if (diff > 0 &&
1082                     xfs_mod_incore_sb(ip->i_mount, XFS_SBS_FDBLOCKS, -((int64_t)diff), rsvd)) {
1083                         /*
1084                          * Ick gross gag me with a spoon.
1085                          */
1086                         ASSERT(0);      /* want to see if this ever happens! */
1087                         while (diff > 0) {
1088                                 if (temp) {
1089                                         temp--;
1090                                         diff--;
1091                                         if (!diff ||
1092                                             !xfs_mod_incore_sb(ip->i_mount,
1093                                                     XFS_SBS_FDBLOCKS, -((int64_t)diff), rsvd))
1094                                                 break;
1095                                 }
1096                                 if (temp2) {
1097                                         temp2--;
1098                                         diff--;
1099                                         if (!diff ||
1100                                             !xfs_mod_incore_sb(ip->i_mount,
1101                                                     XFS_SBS_FDBLOCKS, -((int64_t)diff), rsvd))
1102                                                 break;
1103                                 }
1104                         }
1105                 }
1106                 ep = xfs_iext_get_ext(ifp, idx);
1107                 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
1108                 trace_xfs_bmap_post_update(ip, idx, state, _THIS_IP_);
1109                 trace_xfs_bmap_pre_update(ip, idx + 2, state, _THIS_IP_);
1110                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, idx + 2),
1111                         nullstartblock((int)temp2));
1112                 trace_xfs_bmap_post_update(ip, idx + 2, state, _THIS_IP_);
1113                 *dnew = temp + temp2;
1114                 break;
1115
1116         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1117         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1118         case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
1119         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1120         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1121         case BMAP_LEFT_CONTIG:
1122         case BMAP_RIGHT_CONTIG:
1123                 /*
1124                  * These cases are all impossible.
1125                  */
1126                 ASSERT(0);
1127         }
1128         *curp = cur;
1129 done:
1130         *logflagsp = rval;
1131         return error;
1132 #undef  LEFT
1133 #undef  RIGHT
1134 #undef  PREV
1135 }
1136
1137 /*
1138  * Called by xfs_bmap_add_extent to handle cases converting an unwritten
1139  * allocation to a real allocation or vice versa.
1140  */
1141 STATIC int                              /* error */
1142 xfs_bmap_add_extent_unwritten_real(
1143         xfs_inode_t             *ip,    /* incore inode pointer */
1144         xfs_extnum_t            idx,    /* extent number to update/insert */
1145         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
1146         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
1147         int                     *logflagsp) /* inode logging flags */
1148 {
1149         xfs_btree_cur_t         *cur;   /* btree cursor */
1150         xfs_bmbt_rec_host_t     *ep;    /* extent entry for idx */
1151         int                     error;  /* error return value */
1152         int                     i;      /* temp state */
1153         xfs_ifork_t             *ifp;   /* inode fork pointer */
1154         xfs_fileoff_t           new_endoff;     /* end offset of new entry */
1155         xfs_exntst_t            newext; /* new extent state */
1156         xfs_exntst_t            oldext; /* old extent state */
1157         xfs_bmbt_irec_t         r[3];   /* neighbor extent entries */
1158                                         /* left is 0, right is 1, prev is 2 */
1159         int                     rval=0; /* return value (logging flags) */
1160         int                     state = 0;/* state bits, accessed thru macros */
1161
1162 #define LEFT            r[0]
1163 #define RIGHT           r[1]
1164 #define PREV            r[2]
1165         /*
1166          * Set up a bunch of variables to make the tests simpler.
1167          */
1168         error = 0;
1169         cur = *curp;
1170         ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
1171         ep = xfs_iext_get_ext(ifp, idx);
1172         xfs_bmbt_get_all(ep, &PREV);
1173         newext = new->br_state;
1174         oldext = (newext == XFS_EXT_UNWRITTEN) ?
1175                 XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
1176         ASSERT(PREV.br_state == oldext);
1177         new_endoff = new->br_startoff + new->br_blockcount;
1178         ASSERT(PREV.br_startoff <= new->br_startoff);
1179         ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
1180
1181         /*
1182          * Set flags determining what part of the previous oldext allocation
1183          * extent is being replaced by a newext allocation.
1184          */
1185         if (PREV.br_startoff == new->br_startoff)
1186                 state |= BMAP_LEFT_FILLING;
1187         if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
1188                 state |= BMAP_RIGHT_FILLING;
1189
1190         /*
1191          * Check and set flags if this segment has a left neighbor.
1192          * Don't set contiguous if the combined extent would be too large.
1193          */
1194         if (idx > 0) {
1195                 state |= BMAP_LEFT_VALID;
1196                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, idx - 1), &LEFT);
1197
1198                 if (isnullstartblock(LEFT.br_startblock))
1199                         state |= BMAP_LEFT_DELAY;
1200         }
1201
1202         if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
1203             LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
1204             LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
1205             LEFT.br_state == newext &&
1206             LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
1207                 state |= BMAP_LEFT_CONTIG;
1208
1209         /*
1210          * Check and set flags if this segment has a right neighbor.
1211          * Don't set contiguous if the combined extent would be too large.
1212          * Also check for all-three-contiguous being too large.
1213          */
1214         if (idx < ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1) {
1215                 state |= BMAP_RIGHT_VALID;
1216                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, idx + 1), &RIGHT);
1217                 if (isnullstartblock(RIGHT.br_startblock))
1218                         state |= BMAP_RIGHT_DELAY;
1219         }
1220
1221         if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
1222             new_endoff == RIGHT.br_startoff &&
1223             new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
1224             newext == RIGHT.br_state &&
1225             new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
1226             ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1227                        BMAP_RIGHT_FILLING)) !=
1228                       (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1229                        BMAP_RIGHT_FILLING) ||
1230              LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
1231                         <= MAXEXTLEN))
1232                 state |= BMAP_RIGHT_CONTIG;
1233
1234         /*
1235          * Switch out based on the FILLING and CONTIG state bits.
1236          */
1237         switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1238                          BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
1239         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1240              BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1241                 /*
1242                  * Setting all of a previous oldext extent to newext.
1243                  * The left and right neighbors are both contiguous with new.
1244                  */
1245                 trace_xfs_bmap_pre_update(ip, idx - 1, state, _THIS_IP_);
1246                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
1247                         LEFT.br_blockcount + PREV.br_blockcount +
1248                         RIGHT.br_blockcount);
1249                 trace_xfs_bmap_post_update(ip, idx - 1, state, _THIS_IP_);
1250
1251                 xfs_iext_remove(ip, idx, 2, state);
1252                 ip->i_df.if_lastex = idx - 1;
1253                 ip->i_d.di_nextents -= 2;
1254                 if (cur == NULL)
1255                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1256                 else {
1257                         rval = XFS_ILOG_CORE;
1258                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
1259                                         RIGHT.br_startblock,
1260                                         RIGHT.br_blockcount, &i)))
1261                                 goto done;
1262                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1263                         if ((error = xfs_btree_delete(cur, &i)))
1264                                 goto done;
1265                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1266                         if ((error = xfs_btree_decrement(cur, 0, &i)))
1267                                 goto done;
1268                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1269                         if ((error = xfs_btree_delete(cur, &i)))
1270                                 goto done;
1271                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1272                         if ((error = xfs_btree_decrement(cur, 0, &i)))
1273                                 goto done;
1274                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1275                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
1276                                 LEFT.br_startblock,
1277                                 LEFT.br_blockcount + PREV.br_blockcount +
1278                                 RIGHT.br_blockcount, LEFT.br_state)))
1279                                 goto done;
1280                 }
1281                 break;
1282
1283         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1284                 /*
1285                  * Setting all of a previous oldext extent to newext.
1286                  * The left neighbor is contiguous, the right is not.
1287                  */
1288                 trace_xfs_bmap_pre_update(ip, idx - 1, state, _THIS_IP_);
1289                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
1290                         LEFT.br_blockcount + PREV.br_blockcount);
1291                 trace_xfs_bmap_post_update(ip, idx - 1, state, _THIS_IP_);
1292
1293                 ip->i_df.if_lastex = idx - 1;
1294                 xfs_iext_remove(ip, idx, 1, state);
1295                 ip->i_d.di_nextents--;
1296                 if (cur == NULL)
1297                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1298                 else {
1299                         rval = XFS_ILOG_CORE;
1300                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1301                                         PREV.br_startblock, PREV.br_blockcount,
1302                                         &i)))
1303                                 goto done;
1304                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1305                         if ((error = xfs_btree_delete(cur, &i)))
1306                                 goto done;
1307                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1308                         if ((error = xfs_btree_decrement(cur, 0, &i)))
1309                                 goto done;
1310                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1311                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
1312                                 LEFT.br_startblock,
1313                                 LEFT.br_blockcount + PREV.br_blockcount,
1314                                 LEFT.br_state)))
1315                                 goto done;
1316                 }
1317                 break;
1318
1319         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1320                 /*
1321                  * Setting all of a previous oldext extent to newext.
1322                  * The right neighbor is contiguous, the left is not.
1323                  */
1324                 trace_xfs_bmap_pre_update(ip, idx, state, _THIS_IP_);
1325                 xfs_bmbt_set_blockcount(ep,
1326                         PREV.br_blockcount + RIGHT.br_blockcount);
1327                 xfs_bmbt_set_state(ep, newext);
1328                 trace_xfs_bmap_post_update(ip, idx, state, _THIS_IP_);
1329                 ip->i_df.if_lastex = idx;
1330                 xfs_iext_remove(ip, idx + 1, 1, state);
1331                 ip->i_d.di_nextents--;
1332                 if (cur == NULL)
1333                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1334                 else {
1335                         rval = XFS_ILOG_CORE;
1336                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
1337                                         RIGHT.br_startblock,
1338                                         RIGHT.br_blockcount, &i)))
1339                                 goto done;
1340                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1341                         if ((error = xfs_btree_delete(cur, &i)))
1342                                 goto done;
1343                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1344                         if ((error = xfs_btree_decrement(cur, 0, &i)))
1345                                 goto done;
1346                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1347                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
1348                                 new->br_startblock,
1349                                 new->br_blockcount + RIGHT.br_blockcount,
1350                                 newext)))
1351                                 goto done;
1352                 }
1353                 break;
1354
1355         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
1356                 /*
1357                  * Setting all of a previous oldext extent to newext.
1358                  * Neither the left nor right neighbors are contiguous with
1359                  * the new one.
1360                  */
1361                 trace_xfs_bmap_pre_update(ip, idx, state, _THIS_IP_);
1362                 xfs_bmbt_set_state(ep, newext);
1363                 trace_xfs_bmap_post_update(ip, idx, state, _THIS_IP_);
1364
1365                 ip->i_df.if_lastex = idx;
1366                 if (cur == NULL)
1367                         rval = XFS_ILOG_DEXT;
1368                 else {
1369                         rval = 0;
1370                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1371                                         new->br_startblock, new->br_blockcount,
1372                                         &i)))
1373                                 goto done;
1374                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1375                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
1376                                 new->br_startblock, new->br_blockcount,
1377                                 newext)))
1378                                 goto done;
1379                 }
1380                 break;
1381
1382         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
1383                 /*
1384                  * Setting the first part of a previous oldext extent to newext.
1385                  * The left neighbor is contiguous.
1386                  */
1387                 trace_xfs_bmap_pre_update(ip, idx - 1, state, _THIS_IP_);
1388                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
1389                         LEFT.br_blockcount + new->br_blockcount);
1390                 xfs_bmbt_set_startoff(ep,
1391                         PREV.br_startoff + new->br_blockcount);
1392                 trace_xfs_bmap_post_update(ip, idx - 1, state, _THIS_IP_);
1393
1394                 trace_xfs_bmap_pre_update(ip, idx, state, _THIS_IP_);
1395                 xfs_bmbt_set_startblock(ep,
1396                         new->br_startblock + new->br_blockcount);
1397                 xfs_bmbt_set_blockcount(ep,
1398                         PREV.br_blockcount - new->br_blockcount);
1399                 trace_xfs_bmap_post_update(ip, idx, state, _THIS_IP_);
1400
1401                 ip->i_df.if_lastex = idx - 1;
1402                 if (cur == NULL)
1403                         rval = XFS_ILOG_DEXT;
1404                 else {
1405                         rval = 0;
1406                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1407                                         PREV.br_startblock, PREV.br_blockcount,
1408                                         &i)))
1409                                 goto done;
1410                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1411                         if ((error = xfs_bmbt_update(cur,
1412                                 PREV.br_startoff + new->br_blockcount,
1413                                 PREV.br_startblock + new->br_blockcount,
1414                                 PREV.br_blockcount - new->br_blockcount,
1415                                 oldext)))
1416                                 goto done;
1417                         if ((error = xfs_btree_decrement(cur, 0, &i)))
1418                                 goto done;
1419                         if (xfs_bmbt_update(cur, LEFT.br_startoff,
1420                                 LEFT.br_startblock,
1421                                 LEFT.br_blockcount + new->br_blockcount,
1422                                 LEFT.br_state))
1423                                 goto done;
1424                 }
1425                 break;
1426
1427         case BMAP_LEFT_FILLING:
1428                 /*
1429                  * Setting the first part of a previous oldext extent to newext.
1430                  * The left neighbor is not contiguous.
1431                  */
1432                 trace_xfs_bmap_pre_update(ip, idx, state, _THIS_IP_);
1433                 ASSERT(ep && xfs_bmbt_get_state(ep) == oldext);
1434                 xfs_bmbt_set_startoff(ep, new_endoff);
1435                 xfs_bmbt_set_blockcount(ep,
1436                         PREV.br_blockcount - new->br_blockcount);
1437                 xfs_bmbt_set_startblock(ep,
1438                         new->br_startblock + new->br_blockcount);
1439                 trace_xfs_bmap_post_update(ip, idx, state, _THIS_IP_);
1440
1441                 xfs_iext_insert(ip, idx, 1, new, state);
1442                 ip->i_df.if_lastex = idx;
1443                 ip->i_d.di_nextents++;
1444                 if (cur == NULL)
1445                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1446                 else {
1447                         rval = XFS_ILOG_CORE;
1448                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1449                                         PREV.br_startblock, PREV.br_blockcount,
1450                                         &i)))
1451                                 goto done;
1452                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1453                         if ((error = xfs_bmbt_update(cur,
1454                                 PREV.br_startoff + new->br_blockcount,
1455                                 PREV.br_startblock + new->br_blockcount,
1456                                 PREV.br_blockcount - new->br_blockcount,
1457                                 oldext)))
1458                                 goto done;
1459                         cur->bc_rec.b = *new;
1460                         if ((error = xfs_btree_insert(cur, &i)))
1461                                 goto done;
1462                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1463                 }
1464                 break;
1465
1466         case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1467                 /*
1468                  * Setting the last part of a previous oldext extent to newext.
1469                  * The right neighbor is contiguous with the new allocation.
1470                  */
1471                 trace_xfs_bmap_pre_update(ip, idx, state, _THIS_IP_);
1472                 trace_xfs_bmap_pre_update(ip, idx + 1, state, _THIS_IP_);
1473                 xfs_bmbt_set_blockcount(ep,
1474                         PREV.br_blockcount - new->br_blockcount);
1475                 trace_xfs_bmap_post_update(ip, idx, state, _THIS_IP_);
1476                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, idx + 1),
1477                         new->br_startoff, new->br_startblock,
1478                         new->br_blockcount + RIGHT.br_blockcount, newext);
1479                 trace_xfs_bmap_post_update(ip, idx + 1, state, _THIS_IP_);
1480
1481                 ip->i_df.if_lastex = idx + 1;
1482                 if (cur == NULL)
1483                         rval = XFS_ILOG_DEXT;
1484                 else {
1485                         rval = 0;
1486                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1487                                         PREV.br_startblock,
1488                                         PREV.br_blockcount, &i)))
1489                                 goto done;
1490                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1491                         if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
1492                                 PREV.br_startblock,
1493                                 PREV.br_blockcount - new->br_blockcount,
1494                                 oldext)))
1495                                 goto done;
1496                         if ((error = xfs_btree_increment(cur, 0, &i)))
1497                                 goto done;
1498                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
1499                                 new->br_startblock,
1500                                 new->br_blockcount + RIGHT.br_blockcount,
1501                                 newext)))
1502                                 goto done;
1503                 }
1504                 break;
1505
1506         case BMAP_RIGHT_FILLING:
1507                 /*
1508                  * Setting the last part of a previous oldext extent to newext.
1509                  * The right neighbor is not contiguous.
1510                  */
1511                 trace_xfs_bmap_pre_update(ip, idx, state, _THIS_IP_);
1512                 xfs_bmbt_set_blockcount(ep,
1513                         PREV.br_blockcount - new->br_blockcount);
1514                 trace_xfs_bmap_post_update(ip, idx, state, _THIS_IP_);
1515
1516                 xfs_iext_insert(ip, idx + 1, 1, new, state);
1517                 ip->i_df.if_lastex = idx + 1;
1518                 ip->i_d.di_nextents++;
1519                 if (cur == NULL)
1520                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1521                 else {
1522                         rval = XFS_ILOG_CORE;
1523                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1524                                         PREV.br_startblock, PREV.br_blockcount,
1525                                         &i)))
1526                                 goto done;
1527                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1528                         if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
1529                                 PREV.br_startblock,
1530                                 PREV.br_blockcount - new->br_blockcount,
1531                                 oldext)))
1532                                 goto done;
1533                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1534                                         new->br_startblock, new->br_blockcount,
1535                                         &i)))
1536                                 goto done;
1537                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1538                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
1539                         if ((error = xfs_btree_insert(cur, &i)))
1540                                 goto done;
1541                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1542                 }
1543                 break;
1544
1545         case 0:
1546                 /*
1547                  * Setting the middle part of a previous oldext extent to
1548                  * newext.  Contiguity is impossible here.
1549                  * One extent becomes three extents.
1550                  */
1551                 trace_xfs_bmap_pre_update(ip, idx, state, _THIS_IP_);
1552                 xfs_bmbt_set_blockcount(ep,
1553                         new->br_startoff - PREV.br_startoff);
1554                 trace_xfs_bmap_post_update(ip, idx, state, _THIS_IP_);
1555
1556                 r[0] = *new;
1557                 r[1].br_startoff = new_endoff;
1558                 r[1].br_blockcount =
1559                         PREV.br_startoff + PREV.br_blockcount - new_endoff;
1560                 r[1].br_startblock = new->br_startblock + new->br_blockcount;
1561                 r[1].br_state = oldext;
1562                 xfs_iext_insert(ip, idx + 1, 2, &r[0], state);
1563                 ip->i_df.if_lastex = idx + 1;
1564                 ip->i_d.di_nextents += 2;
1565                 if (cur == NULL)
1566                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1567                 else {
1568                         rval = XFS_ILOG_CORE;
1569                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1570                                         PREV.br_startblock, PREV.br_blockcount,
1571                                         &i)))
1572                                 goto done;
1573                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1574                         /* new right extent - oldext */
1575                         if ((error = xfs_bmbt_update(cur, r[1].br_startoff,
1576                                 r[1].br_startblock, r[1].br_blockcount,
1577                                 r[1].br_state)))
1578                                 goto done;
1579                         /* new left extent - oldext */
1580                         cur->bc_rec.b = PREV;
1581                         cur->bc_rec.b.br_blockcount =
1582                                 new->br_startoff - PREV.br_startoff;
1583                         if ((error = xfs_btree_insert(cur, &i)))
1584                                 goto done;
1585                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1586                         /*
1587                          * Reset the cursor to the position of the new extent
1588                          * we are about to insert as we can't trust it after
1589                          * the previous insert.
1590                          */
1591                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1592                                         new->br_startblock, new->br_blockcount,
1593                                         &i)))
1594                                 goto done;
1595                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1596                         /* new middle extent - newext */
1597                         cur->bc_rec.b.br_state = new->br_state;
1598                         if ((error = xfs_btree_insert(cur, &i)))
1599                                 goto done;
1600                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1601                 }
1602                 break;
1603
1604         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1605         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1606         case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
1607         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1608         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1609         case BMAP_LEFT_CONTIG:
1610         case BMAP_RIGHT_CONTIG:
1611                 /*
1612                  * These cases are all impossible.
1613                  */
1614                 ASSERT(0);
1615         }
1616         *curp = cur;
1617 done:
1618         *logflagsp = rval;
1619         return error;
1620 #undef  LEFT
1621 #undef  RIGHT
1622 #undef  PREV
1623 }
1624
1625 /*
1626  * Called by xfs_bmap_add_extent to handle cases converting a hole
1627  * to a delayed allocation.
1628  */
1629 /*ARGSUSED*/
1630 STATIC int                              /* error */
1631 xfs_bmap_add_extent_hole_delay(
1632         xfs_inode_t             *ip,    /* incore inode pointer */
1633         xfs_extnum_t            idx,    /* extent number to update/insert */
1634         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
1635         int                     *logflagsp, /* inode logging flags */
1636         int                     rsvd)           /* OK to allocate reserved blocks */
1637 {
1638         xfs_bmbt_rec_host_t     *ep;    /* extent record for idx */
1639         xfs_ifork_t             *ifp;   /* inode fork pointer */
1640         xfs_bmbt_irec_t         left;   /* left neighbor extent entry */
1641         xfs_filblks_t           newlen=0;       /* new indirect size */
1642         xfs_filblks_t           oldlen=0;       /* old indirect size */
1643         xfs_bmbt_irec_t         right;  /* right neighbor extent entry */
1644         int                     state;  /* state bits, accessed thru macros */
1645         xfs_filblks_t           temp=0; /* temp for indirect calculations */
1646
1647         ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
1648         ep = xfs_iext_get_ext(ifp, idx);
1649         state = 0;
1650         ASSERT(isnullstartblock(new->br_startblock));
1651
1652         /*
1653          * Check and set flags if this segment has a left neighbor
1654          */
1655         if (idx > 0) {
1656                 state |= BMAP_LEFT_VALID;
1657                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, idx - 1), &left);
1658
1659                 if (isnullstartblock(left.br_startblock))
1660                         state |= BMAP_LEFT_DELAY;
1661         }
1662
1663         /*
1664          * Check and set flags if the current (right) segment exists.
1665          * If it doesn't exist, we're converting the hole at end-of-file.
1666          */
1667         if (idx < ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t)) {
1668                 state |= BMAP_RIGHT_VALID;
1669                 xfs_bmbt_get_all(ep, &right);
1670
1671                 if (isnullstartblock(right.br_startblock))
1672                         state |= BMAP_RIGHT_DELAY;
1673         }
1674
1675         /*
1676          * Set contiguity flags on the left and right neighbors.
1677          * Don't let extents get too large, even if the pieces are contiguous.
1678          */
1679         if ((state & BMAP_LEFT_VALID) && (state & BMAP_LEFT_DELAY) &&
1680             left.br_startoff + left.br_blockcount == new->br_startoff &&
1681             left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
1682                 state |= BMAP_LEFT_CONTIG;
1683
1684         if ((state & BMAP_RIGHT_VALID) && (state & BMAP_RIGHT_DELAY) &&
1685             new->br_startoff + new->br_blockcount == right.br_startoff &&
1686             new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
1687             (!(state & BMAP_LEFT_CONTIG) ||
1688              (left.br_blockcount + new->br_blockcount +
1689               right.br_blockcount <= MAXEXTLEN)))
1690                 state |= BMAP_RIGHT_CONTIG;
1691
1692         /*
1693          * Switch out based on the contiguity flags.
1694          */
1695         switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
1696         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1697                 /*
1698                  * New allocation is contiguous with delayed allocations
1699                  * on the left and on the right.
1700                  * Merge all three into a single extent record.
1701                  */
1702                 temp = left.br_blockcount + new->br_blockcount +
1703                         right.br_blockcount;
1704
1705                 trace_xfs_bmap_pre_update(ip, idx - 1, state, _THIS_IP_);
1706                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1), temp);
1707                 oldlen = startblockval(left.br_startblock) +
1708                         startblockval(new->br_startblock) +
1709                         startblockval(right.br_startblock);
1710                 newlen = xfs_bmap_worst_indlen(ip, temp);
1711                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, idx - 1),
1712                         nullstartblock((int)newlen));
1713                 trace_xfs_bmap_post_update(ip, idx - 1, state, _THIS_IP_);
1714
1715                 xfs_iext_remove(ip, idx, 1, state);
1716                 ip->i_df.if_lastex = idx - 1;
1717                 break;
1718
1719         case BMAP_LEFT_CONTIG:
1720                 /*
1721                  * New allocation is contiguous with a delayed allocation
1722                  * on the left.
1723                  * Merge the new allocation with the left neighbor.
1724                  */
1725                 temp = left.br_blockcount + new->br_blockcount;
1726                 trace_xfs_bmap_pre_update(ip, idx - 1, state, _THIS_IP_);
1727                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1), temp);
1728                 oldlen = startblockval(left.br_startblock) +
1729                         startblockval(new->br_startblock);
1730                 newlen = xfs_bmap_worst_indlen(ip, temp);
1731                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, idx - 1),
1732                         nullstartblock((int)newlen));
1733                 trace_xfs_bmap_post_update(ip, idx - 1, state, _THIS_IP_);
1734
1735                 ip->i_df.if_lastex = idx - 1;
1736                 break;
1737
1738         case BMAP_RIGHT_CONTIG:
1739                 /*
1740                  * New allocation is contiguous with a delayed allocation
1741                  * on the right.
1742                  * Merge the new allocation with the right neighbor.
1743                  */
1744                 trace_xfs_bmap_pre_update(ip, idx, state, _THIS_IP_);
1745                 temp = new->br_blockcount + right.br_blockcount;
1746                 oldlen = startblockval(new->br_startblock) +
1747                         startblockval(right.br_startblock);
1748                 newlen = xfs_bmap_worst_indlen(ip, temp);
1749                 xfs_bmbt_set_allf(ep, new->br_startoff,
1750                         nullstartblock((int)newlen), temp, right.br_state);
1751                 trace_xfs_bmap_post_update(ip, idx, state, _THIS_IP_);
1752
1753                 ip->i_df.if_lastex = idx;
1754                 break;
1755
1756         case 0:
1757                 /*
1758                  * New allocation is not contiguous with another
1759                  * delayed allocation.
1760                  * Insert a new entry.
1761                  */
1762                 oldlen = newlen = 0;
1763                 xfs_iext_insert(ip, idx, 1, new, state);
1764                 ip->i_df.if_lastex = idx;
1765                 break;
1766         }
1767         if (oldlen != newlen) {
1768                 ASSERT(oldlen > newlen);
1769                 xfs_mod_incore_sb(ip->i_mount, XFS_SBS_FDBLOCKS,
1770                         (int64_t)(oldlen - newlen), rsvd);
1771                 /*
1772                  * Nothing to do for disk quota accounting here.
1773                  */
1774         }
1775         *logflagsp = 0;
1776         return 0;
1777 }
1778
1779 /*
1780  * Called by xfs_bmap_add_extent to handle cases converting a hole
1781  * to a real allocation.
1782  */
1783 STATIC int                              /* error */
1784 xfs_bmap_add_extent_hole_real(
1785         xfs_inode_t             *ip,    /* incore inode pointer */
1786         xfs_extnum_t            idx,    /* extent number to update/insert */
1787         xfs_btree_cur_t         *cur,   /* if null, not a btree */
1788         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
1789         int                     *logflagsp, /* inode logging flags */
1790         int                     whichfork) /* data or attr fork */
1791 {
1792         xfs_bmbt_rec_host_t     *ep;    /* pointer to extent entry ins. point */
1793         int                     error;  /* error return value */
1794         int                     i;      /* temp state */
1795         xfs_ifork_t             *ifp;   /* inode fork pointer */
1796         xfs_bmbt_irec_t         left;   /* left neighbor extent entry */
1797         xfs_bmbt_irec_t         right;  /* right neighbor extent entry */
1798         int                     rval=0; /* return value (logging flags) */
1799         int                     state;  /* state bits, accessed thru macros */
1800
1801         ifp = XFS_IFORK_PTR(ip, whichfork);
1802         ASSERT(idx <= ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t));
1803         ep = xfs_iext_get_ext(ifp, idx);
1804         state = 0;
1805
1806         if (whichfork == XFS_ATTR_FORK)
1807                 state |= BMAP_ATTRFORK;
1808
1809         /*
1810          * Check and set flags if this segment has a left neighbor.
1811          */
1812         if (idx > 0) {
1813                 state |= BMAP_LEFT_VALID;
1814                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, idx - 1), &left);
1815                 if (isnullstartblock(left.br_startblock))
1816                         state |= BMAP_LEFT_DELAY;
1817         }
1818
1819         /*
1820          * Check and set flags if this segment has a current value.
1821          * Not true if we're inserting into the "hole" at eof.
1822          */
1823         if (idx < ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)) {
1824                 state |= BMAP_RIGHT_VALID;
1825                 xfs_bmbt_get_all(ep, &right);
1826                 if (isnullstartblock(right.br_startblock))
1827                         state |= BMAP_RIGHT_DELAY;
1828         }
1829
1830         /*
1831          * We're inserting a real allocation between "left" and "right".
1832          * Set the contiguity flags.  Don't let extents get too large.
1833          */
1834         if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
1835             left.br_startoff + left.br_blockcount == new->br_startoff &&
1836             left.br_startblock + left.br_blockcount == new->br_startblock &&
1837             left.br_state == new->br_state &&
1838             left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
1839                 state |= BMAP_LEFT_CONTIG;
1840
1841         if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
1842             new->br_startoff + new->br_blockcount == right.br_startoff &&
1843             new->br_startblock + new->br_blockcount == right.br_startblock &&
1844             new->br_state == right.br_state &&
1845             new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
1846             (!(state & BMAP_LEFT_CONTIG) ||
1847              left.br_blockcount + new->br_blockcount +
1848              right.br_blockcount <= MAXEXTLEN))
1849                 state |= BMAP_RIGHT_CONTIG;
1850
1851         error = 0;
1852         /*
1853          * Select which case we're in here, and implement it.
1854          */
1855         switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
1856         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1857                 /*
1858                  * New allocation is contiguous with real allocations on the
1859                  * left and on the right.
1860                  * Merge all three into a single extent record.
1861                  */
1862                 trace_xfs_bmap_pre_update(ip, idx - 1, state, _THIS_IP_);
1863                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
1864                         left.br_blockcount + new->br_blockcount +
1865                         right.br_blockcount);
1866                 trace_xfs_bmap_post_update(ip, idx - 1, state, _THIS_IP_);
1867
1868                 xfs_iext_remove(ip, idx, 1, state);
1869                 ifp->if_lastex = idx - 1;
1870                 XFS_IFORK_NEXT_SET(ip, whichfork,
1871                         XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
1872                 if (cur == NULL) {
1873                         rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
1874                 } else {
1875                         rval = XFS_ILOG_CORE;
1876                         if ((error = xfs_bmbt_lookup_eq(cur,
1877                                         right.br_startoff,
1878                                         right.br_startblock,
1879                                         right.br_blockcount, &i)))
1880                                 goto done;
1881                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1882                         if ((error = xfs_btree_delete(cur, &i)))
1883                                 goto done;
1884                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1885                         if ((error = xfs_btree_decrement(cur, 0, &i)))
1886                                 goto done;
1887                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1888                         if ((error = xfs_bmbt_update(cur, left.br_startoff,
1889                                         left.br_startblock,
1890                                         left.br_blockcount +
1891                                                 new->br_blockcount +
1892                                                 right.br_blockcount,
1893                                         left.br_state)))
1894                                 goto done;
1895                 }
1896                 break;
1897
1898         case BMAP_LEFT_CONTIG:
1899                 /*
1900                  * New allocation is contiguous with a real allocation
1901                  * on the left.
1902                  * Merge the new allocation with the left neighbor.
1903                  */
1904                 trace_xfs_bmap_pre_update(ip, idx - 1, state, _THIS_IP_);
1905                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
1906                         left.br_blockcount + new->br_blockcount);
1907                 trace_xfs_bmap_post_update(ip, idx - 1, state, _THIS_IP_);
1908
1909                 ifp->if_lastex = idx - 1;
1910                 if (cur == NULL) {
1911                         rval = xfs_ilog_fext(whichfork);
1912                 } else {
1913                         rval = 0;
1914                         if ((error = xfs_bmbt_lookup_eq(cur,
1915                                         left.br_startoff,
1916                                         left.br_startblock,
1917                                         left.br_blockcount, &i)))
1918                                 goto done;
1919                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1920                         if ((error = xfs_bmbt_update(cur, left.br_startoff,
1921                                         left.br_startblock,
1922                                         left.br_blockcount +
1923                                                 new->br_blockcount,
1924                                         left.br_state)))
1925                                 goto done;
1926                 }
1927                 break;
1928
1929         case BMAP_RIGHT_CONTIG:
1930                 /*
1931                  * New allocation is contiguous with a real allocation
1932                  * on the right.
1933                  * Merge the new allocation with the right neighbor.
1934                  */
1935                 trace_xfs_bmap_pre_update(ip, idx, state, _THIS_IP_);
1936                 xfs_bmbt_set_allf(ep, new->br_startoff, new->br_startblock,
1937                         new->br_blockcount + right.br_blockcount,
1938                         right.br_state);
1939                 trace_xfs_bmap_post_update(ip, idx, state, _THIS_IP_);
1940
1941                 ifp->if_lastex = idx;
1942                 if (cur == NULL) {
1943                         rval = xfs_ilog_fext(whichfork);
1944                 } else {
1945                         rval = 0;
1946                         if ((error = xfs_bmbt_lookup_eq(cur,
1947                                         right.br_startoff,
1948                                         right.br_startblock,
1949                                         right.br_blockcount, &i)))
1950                                 goto done;
1951                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1952                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
1953                                         new->br_startblock,
1954                                         new->br_blockcount +
1955                                                 right.br_blockcount,
1956                                         right.br_state)))
1957                                 goto done;
1958                 }
1959                 break;
1960
1961         case 0:
1962                 /*
1963                  * New allocation is not contiguous with another
1964                  * real allocation.
1965                  * Insert a new entry.
1966                  */
1967                 xfs_iext_insert(ip, idx, 1, new, state);
1968                 ifp->if_lastex = idx;
1969                 XFS_IFORK_NEXT_SET(ip, whichfork,
1970                         XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
1971                 if (cur == NULL) {
1972                         rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
1973                 } else {
1974                         rval = XFS_ILOG_CORE;
1975                         if ((error = xfs_bmbt_lookup_eq(cur,
1976                                         new->br_startoff,
1977                                         new->br_startblock,
1978                                         new->br_blockcount, &i)))
1979                                 goto done;
1980                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1981                         cur->bc_rec.b.br_state = new->br_state;
1982                         if ((error = xfs_btree_insert(cur, &i)))
1983                                 goto done;
1984                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1985                 }
1986                 break;
1987         }
1988 done:
1989         *logflagsp = rval;
1990         return error;
1991 }
1992
1993 /*
1994  * Adjust the size of the new extent based on di_extsize and rt extsize.
1995  */
1996 STATIC int
1997 xfs_bmap_extsize_align(
1998         xfs_mount_t     *mp,
1999         xfs_bmbt_irec_t *gotp,          /* next extent pointer */
2000         xfs_bmbt_irec_t *prevp,         /* previous extent pointer */
2001         xfs_extlen_t    extsz,          /* align to this extent size */
2002         int             rt,             /* is this a realtime inode? */
2003         int             eof,            /* is extent at end-of-file? */
2004         int             delay,          /* creating delalloc extent? */
2005         int             convert,        /* overwriting unwritten extent? */
2006         xfs_fileoff_t   *offp,          /* in/out: aligned offset */
2007         xfs_extlen_t    *lenp)          /* in/out: aligned length */
2008 {
2009         xfs_fileoff_t   orig_off;       /* original offset */
2010         xfs_extlen_t    orig_alen;      /* original length */
2011         xfs_fileoff_t   orig_end;       /* original off+len */
2012         xfs_fileoff_t   nexto;          /* next file offset */
2013         xfs_fileoff_t   prevo;          /* previous file offset */
2014         xfs_fileoff_t   align_off;      /* temp for offset */
2015         xfs_extlen_t    align_alen;     /* temp for length */
2016         xfs_extlen_t    temp;           /* temp for calculations */
2017
2018         if (convert)
2019                 return 0;
2020
2021         orig_off = align_off = *offp;
2022         orig_alen = align_alen = *lenp;
2023         orig_end = orig_off + orig_alen;
2024
2025         /*
2026          * If this request overlaps an existing extent, then don't
2027          * attempt to perform any additional alignment.
2028          */
2029         if (!delay && !eof &&
2030             (orig_off >= gotp->br_startoff) &&
2031             (orig_end <= gotp->br_startoff + gotp->br_blockcount)) {
2032                 return 0;
2033         }
2034
2035         /*
2036          * If the file offset is unaligned vs. the extent size
2037          * we need to align it.  This will be possible unless
2038          * the file was previously written with a kernel that didn't
2039          * perform this alignment, or if a truncate shot us in the
2040          * foot.
2041          */
2042         temp = do_mod(orig_off, extsz);
2043         if (temp) {
2044                 align_alen += temp;
2045                 align_off -= temp;
2046         }
2047         /*
2048          * Same adjustment for the end of the requested area.
2049          */
2050         if ((temp = (align_alen % extsz))) {
2051                 align_alen += extsz - temp;
2052         }
2053         /*
2054          * If the previous block overlaps with this proposed allocation
2055          * then move the start forward without adjusting the length.
2056          */
2057         if (prevp->br_startoff != NULLFILEOFF) {
2058                 if (prevp->br_startblock == HOLESTARTBLOCK)
2059                         prevo = prevp->br_startoff;
2060                 else
2061                         prevo = prevp->br_startoff + prevp->br_blockcount;
2062         } else
2063                 prevo = 0;
2064         if (align_off != orig_off && align_off < prevo)
2065                 align_off = prevo;
2066         /*
2067          * If the next block overlaps with this proposed allocation
2068          * then move the start back without adjusting the length,
2069          * but not before offset 0.
2070          * This may of course make the start overlap previous block,
2071          * and if we hit the offset 0 limit then the next block
2072          * can still overlap too.
2073          */
2074         if (!eof && gotp->br_startoff != NULLFILEOFF) {
2075                 if ((delay && gotp->br_startblock == HOLESTARTBLOCK) ||
2076                     (!delay && gotp->br_startblock == DELAYSTARTBLOCK))
2077                         nexto = gotp->br_startoff + gotp->br_blockcount;
2078                 else
2079                         nexto = gotp->br_startoff;
2080         } else
2081                 nexto = NULLFILEOFF;
2082         if (!eof &&
2083             align_off + align_alen != orig_end &&
2084             align_off + align_alen > nexto)
2085                 align_off = nexto > align_alen ? nexto - align_alen : 0;
2086         /*
2087          * If we're now overlapping the next or previous extent that
2088          * means we can't fit an extsz piece in this hole.  Just move
2089          * the start forward to the first valid spot and set
2090          * the length so we hit the end.
2091          */
2092         if (align_off != orig_off && align_off < prevo)
2093                 align_off = prevo;
2094         if (align_off + align_alen != orig_end &&
2095             align_off + align_alen > nexto &&
2096             nexto != NULLFILEOFF) {
2097                 ASSERT(nexto > prevo);
2098                 align_alen = nexto - align_off;
2099         }
2100
2101         /*
2102          * If realtime, and the result isn't a multiple of the realtime
2103          * extent size we need to remove blocks until it is.
2104          */
2105         if (rt && (temp = (align_alen % mp->m_sb.sb_rextsize))) {
2106                 /*
2107                  * We're not covering the original request, or
2108                  * we won't be able to once we fix the length.
2109                  */
2110                 if (orig_off < align_off ||
2111                     orig_end > align_off + align_alen ||
2112                     align_alen - temp < orig_alen)
2113                         return XFS_ERROR(EINVAL);
2114                 /*
2115                  * Try to fix it by moving the start up.
2116                  */
2117                 if (align_off + temp <= orig_off) {
2118                         align_alen -= temp;
2119                         align_off += temp;
2120                 }
2121                 /*
2122                  * Try to fix it by moving the end in.
2123                  */
2124                 else if (align_off + align_alen - temp >= orig_end)
2125                         align_alen -= temp;
2126                 /*
2127                  * Set the start to the minimum then trim the length.
2128                  */
2129                 else {
2130                         align_alen -= orig_off - align_off;
2131                         align_off = orig_off;
2132                         align_alen -= align_alen % mp->m_sb.sb_rextsize;
2133                 }
2134                 /*
2135                  * Result doesn't cover the request, fail it.
2136                  */
2137                 if (orig_off < align_off || orig_end > align_off + align_alen)
2138                         return XFS_ERROR(EINVAL);
2139         } else {
2140                 ASSERT(orig_off >= align_off);
2141                 ASSERT(orig_end <= align_off + align_alen);
2142         }
2143
2144 #ifdef DEBUG
2145         if (!eof && gotp->br_startoff != NULLFILEOFF)
2146                 ASSERT(align_off + align_alen <= gotp->br_startoff);
2147         if (prevp->br_startoff != NULLFILEOFF)
2148                 ASSERT(align_off >= prevp->br_startoff + prevp->br_blockcount);
2149 #endif
2150
2151         *lenp = align_alen;
2152         *offp = align_off;
2153         return 0;
2154 }
2155
2156 #define XFS_ALLOC_GAP_UNITS     4
2157
2158 STATIC void
2159 xfs_bmap_adjacent(
2160         xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
2161 {
2162         xfs_fsblock_t   adjust;         /* adjustment to block numbers */
2163         xfs_agnumber_t  fb_agno;        /* ag number of ap->firstblock */
2164         xfs_mount_t     *mp;            /* mount point structure */
2165         int             nullfb;         /* true if ap->firstblock isn't set */
2166         int             rt;             /* true if inode is realtime */
2167
2168 #define ISVALID(x,y)    \
2169         (rt ? \
2170                 (x) < mp->m_sb.sb_rblocks : \
2171                 XFS_FSB_TO_AGNO(mp, x) == XFS_FSB_TO_AGNO(mp, y) && \
2172                 XFS_FSB_TO_AGNO(mp, x) < mp->m_sb.sb_agcount && \
2173                 XFS_FSB_TO_AGBNO(mp, x) < mp->m_sb.sb_agblocks)
2174
2175         mp = ap->ip->i_mount;
2176         nullfb = ap->firstblock == NULLFSBLOCK;
2177         rt = XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata;
2178         fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, ap->firstblock);
2179         /*
2180          * If allocating at eof, and there's a previous real block,
2181          * try to use its last block as our starting point.
2182          */
2183         if (ap->eof && ap->prevp->br_startoff != NULLFILEOFF &&
2184             !isnullstartblock(ap->prevp->br_startblock) &&
2185             ISVALID(ap->prevp->br_startblock + ap->prevp->br_blockcount,
2186                     ap->prevp->br_startblock)) {
2187                 ap->rval = ap->prevp->br_startblock + ap->prevp->br_blockcount;
2188                 /*
2189                  * Adjust for the gap between prevp and us.
2190                  */
2191                 adjust = ap->off -
2192                         (ap->prevp->br_startoff + ap->prevp->br_blockcount);
2193                 if (adjust &&
2194                     ISVALID(ap->rval + adjust, ap->prevp->br_startblock))
2195                         ap->rval += adjust;
2196         }
2197         /*
2198          * If not at eof, then compare the two neighbor blocks.
2199          * Figure out whether either one gives us a good starting point,
2200          * and pick the better one.
2201          */
2202         else if (!ap->eof) {
2203                 xfs_fsblock_t   gotbno;         /* right side block number */
2204                 xfs_fsblock_t   gotdiff=0;      /* right side difference */
2205                 xfs_fsblock_t   prevbno;        /* left side block number */
2206                 xfs_fsblock_t   prevdiff=0;     /* left side difference */
2207
2208                 /*
2209                  * If there's a previous (left) block, select a requested
2210                  * start block based on it.
2211                  */
2212                 if (ap->prevp->br_startoff != NULLFILEOFF &&
2213                     !isnullstartblock(ap->prevp->br_startblock) &&
2214                     (prevbno = ap->prevp->br_startblock +
2215                                ap->prevp->br_blockcount) &&
2216                     ISVALID(prevbno, ap->prevp->br_startblock)) {
2217                         /*
2218                          * Calculate gap to end of previous block.
2219                          */
2220                         adjust = prevdiff = ap->off -
2221                                 (ap->prevp->br_startoff +
2222                                  ap->prevp->br_blockcount);
2223                         /*
2224                          * Figure the startblock based on the previous block's
2225                          * end and the gap size.
2226                          * Heuristic!
2227                          * If the gap is large relative to the piece we're
2228                          * allocating, or using it gives us an invalid block
2229                          * number, then just use the end of the previous block.
2230                          */
2231                         if (prevdiff <= XFS_ALLOC_GAP_UNITS * ap->alen &&
2232                             ISVALID(prevbno + prevdiff,
2233                                     ap->prevp->br_startblock))
2234                                 prevbno += adjust;
2235                         else
2236                                 prevdiff += adjust;
2237                         /*
2238                          * If the firstblock forbids it, can't use it,
2239                          * must use default.
2240                          */
2241                         if (!rt && !nullfb &&
2242                             XFS_FSB_TO_AGNO(mp, prevbno) != fb_agno)
2243                                 prevbno = NULLFSBLOCK;
2244                 }
2245                 /*
2246                  * No previous block or can't follow it, just default.
2247                  */
2248                 else
2249                         prevbno = NULLFSBLOCK;
2250                 /*
2251                  * If there's a following (right) block, select a requested
2252                  * start block based on it.
2253                  */
2254                 if (!isnullstartblock(ap->gotp->br_startblock)) {
2255                         /*
2256                          * Calculate gap to start of next block.
2257                          */
2258                         adjust = gotdiff = ap->gotp->br_startoff - ap->off;
2259                         /*
2260                          * Figure the startblock based on the next block's
2261                          * start and the gap size.
2262                          */
2263                         gotbno = ap->gotp->br_startblock;
2264                         /*
2265                          * Heuristic!
2266                          * If the gap is large relative to the piece we're
2267                          * allocating, or using it gives us an invalid block
2268                          * number, then just use the start of the next block
2269                          * offset by our length.
2270                          */
2271                         if (gotdiff <= XFS_ALLOC_GAP_UNITS * ap->alen &&
2272                             ISVALID(gotbno - gotdiff, gotbno))
2273                                 gotbno -= adjust;
2274                         else if (ISVALID(gotbno - ap->alen, gotbno)) {
2275                                 gotbno -= ap->alen;
2276                                 gotdiff += adjust - ap->alen;
2277                         } else
2278                                 gotdiff += adjust;
2279                         /*
2280                          * If the firstblock forbids it, can't use it,
2281                          * must use default.
2282                          */
2283                         if (!rt && !nullfb &&
2284                             XFS_FSB_TO_AGNO(mp, gotbno) != fb_agno)
2285                                 gotbno = NULLFSBLOCK;
2286                 }
2287                 /*
2288                  * No next block, just default.
2289                  */
2290                 else
2291                         gotbno = NULLFSBLOCK;
2292                 /*
2293                  * If both valid, pick the better one, else the only good
2294                  * one, else ap->rval is already set (to 0 or the inode block).
2295                  */
2296                 if (prevbno != NULLFSBLOCK && gotbno != NULLFSBLOCK)
2297                         ap->rval = prevdiff <= gotdiff ? prevbno : gotbno;
2298                 else if (prevbno != NULLFSBLOCK)
2299                         ap->rval = prevbno;
2300                 else if (gotbno != NULLFSBLOCK)
2301                         ap->rval = gotbno;
2302         }
2303 #undef ISVALID
2304 }
2305
2306 STATIC int
2307 xfs_bmap_rtalloc(
2308         xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
2309 {
2310         xfs_alloctype_t atype = 0;      /* type for allocation routines */
2311         int             error;          /* error return value */
2312         xfs_mount_t     *mp;            /* mount point structure */
2313         xfs_extlen_t    prod = 0;       /* product factor for allocators */
2314         xfs_extlen_t    ralen = 0;      /* realtime allocation length */
2315         xfs_extlen_t    align;          /* minimum allocation alignment */
2316         xfs_rtblock_t   rtb;
2317
2318         mp = ap->ip->i_mount;
2319         align = xfs_get_extsz_hint(ap->ip);
2320         prod = align / mp->m_sb.sb_rextsize;
2321         error = xfs_bmap_extsize_align(mp, ap->gotp, ap->prevp,
2322                                         align, 1, ap->eof, 0,
2323                                         ap->conv, &ap->off, &ap->alen);
2324         if (error)
2325                 return error;
2326         ASSERT(ap->alen);
2327         ASSERT(ap->alen % mp->m_sb.sb_rextsize == 0);
2328
2329         /*
2330          * If the offset & length are not perfectly aligned
2331          * then kill prod, it will just get us in trouble.
2332          */
2333         if (do_mod(ap->off, align) || ap->alen % align)
2334                 prod = 1;
2335         /*
2336          * Set ralen to be the actual requested length in rtextents.
2337          */
2338         ralen = ap->alen / mp->m_sb.sb_rextsize;
2339         /*
2340          * If the old value was close enough to MAXEXTLEN that
2341          * we rounded up to it, cut it back so it's valid again.
2342          * Note that if it's a really large request (bigger than
2343          * MAXEXTLEN), we don't hear about that number, and can't
2344          * adjust the starting point to match it.
2345          */
2346         if (ralen * mp->m_sb.sb_rextsize >= MAXEXTLEN)
2347                 ralen = MAXEXTLEN / mp->m_sb.sb_rextsize;
2348         /*
2349          * If it's an allocation to an empty file at offset 0,
2350          * pick an extent that will space things out in the rt area.
2351          */
2352         if (ap->eof && ap->off == 0) {
2353                 xfs_rtblock_t uninitialized_var(rtx); /* realtime extent no */
2354
2355                 error = xfs_rtpick_extent(mp, ap->tp, ralen, &rtx);
2356                 if (error)
2357                         return error;
2358                 ap->rval = rtx * mp->m_sb.sb_rextsize;
2359         } else {
2360                 ap->rval = 0;
2361         }
2362
2363         xfs_bmap_adjacent(ap);
2364
2365         /*
2366          * Realtime allocation, done through xfs_rtallocate_extent.
2367          */
2368         atype = ap->rval == 0 ?  XFS_ALLOCTYPE_ANY_AG : XFS_ALLOCTYPE_NEAR_BNO;
2369         do_div(ap->rval, mp->m_sb.sb_rextsize);
2370         rtb = ap->rval;
2371         ap->alen = ralen;
2372         if ((error = xfs_rtallocate_extent(ap->tp, ap->rval, 1, ap->alen,
2373                                 &ralen, atype, ap->wasdel, prod, &rtb)))
2374                 return error;
2375         if (rtb == NULLFSBLOCK && prod > 1 &&
2376             (error = xfs_rtallocate_extent(ap->tp, ap->rval, 1,
2377                                            ap->alen, &ralen, atype,
2378                                            ap->wasdel, 1, &rtb)))
2379                 return error;
2380         ap->rval = rtb;
2381         if (ap->rval != NULLFSBLOCK) {
2382                 ap->rval *= mp->m_sb.sb_rextsize;
2383                 ralen *= mp->m_sb.sb_rextsize;
2384                 ap->alen = ralen;
2385                 ap->ip->i_d.di_nblocks += ralen;
2386                 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
2387                 if (ap->wasdel)
2388                         ap->ip->i_delayed_blks -= ralen;
2389                 /*
2390                  * Adjust the disk quota also. This was reserved
2391                  * earlier.
2392                  */
2393                 xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
2394                         ap->wasdel ? XFS_TRANS_DQ_DELRTBCOUNT :
2395                                         XFS_TRANS_DQ_RTBCOUNT, (long) ralen);
2396         } else {
2397                 ap->alen = 0;
2398         }
2399         return 0;
2400 }
2401
2402 STATIC int
2403 xfs_bmap_btalloc_nullfb(
2404         struct xfs_bmalloca     *ap,
2405         struct xfs_alloc_arg    *args,
2406         xfs_extlen_t            *blen)
2407 {
2408         struct xfs_mount        *mp = ap->ip->i_mount;
2409         struct xfs_perag        *pag;
2410         xfs_agnumber_t          ag, startag;
2411         int                     notinit = 0;
2412         int                     error;
2413
2414         if (ap->userdata && xfs_inode_is_filestream(ap->ip))
2415                 args->type = XFS_ALLOCTYPE_NEAR_BNO;
2416         else
2417                 args->type = XFS_ALLOCTYPE_START_BNO;
2418         args->total = ap->total;
2419
2420         /*
2421          * Search for an allocation group with a single extent large enough
2422          * for the request.  If one isn't found, then adjust the minimum
2423          * allocation size to the largest space found.
2424          */
2425         startag = ag = XFS_FSB_TO_AGNO(mp, args->fsbno);
2426         if (startag == NULLAGNUMBER)
2427                 startag = ag = 0;
2428
2429         pag = xfs_perag_get(mp, ag);
2430         while (*blen < ap->alen) {
2431                 if (!pag->pagf_init) {
2432                         error = xfs_alloc_pagf_init(mp, args->tp, ag,
2433                                                     XFS_ALLOC_FLAG_TRYLOCK);
2434                         if (error) {
2435                                 xfs_perag_put(pag);
2436                                 return error;
2437                         }
2438                 }
2439
2440                 /*
2441                  * See xfs_alloc_fix_freelist...
2442                  */
2443                 if (pag->pagf_init) {
2444                         xfs_extlen_t    longest;
2445                         longest = xfs_alloc_longest_free_extent(mp, pag);
2446                         if (*blen < longest)
2447                                 *blen = longest;
2448                 } else
2449                         notinit = 1;
2450
2451                 if (xfs_inode_is_filestream(ap->ip)) {
2452                         if (*blen >= ap->alen)
2453                                 break;
2454
2455                         if (ap->userdata) {
2456                                 /*
2457                                  * If startag is an invalid AG, we've
2458                                  * come here once before and
2459                                  * xfs_filestream_new_ag picked the
2460                                  * best currently available.
2461                                  *
2462                                  * Don't continue looping, since we
2463                                  * could loop forever.
2464                                  */
2465                                 if (startag == NULLAGNUMBER)
2466                                         break;
2467
2468                                 error = xfs_filestream_new_ag(ap, &ag);
2469                                 xfs_perag_put(pag);
2470                                 if (error)
2471                                         return error;
2472
2473                                 /* loop again to set 'blen'*/
2474                                 startag = NULLAGNUMBER;
2475                                 pag = xfs_perag_get(mp, ag);
2476                                 continue;
2477                         }
2478                 }
2479                 if (++ag == mp->m_sb.sb_agcount)
2480                         ag = 0;
2481                 if (ag == startag)
2482                         break;
2483                 xfs_perag_put(pag);
2484                 pag = xfs_perag_get(mp, ag);
2485         }
2486         xfs_perag_put(pag);
2487
2488         /*
2489          * Since the above loop did a BUF_TRYLOCK, it is
2490          * possible that there is space for this request.
2491          */
2492         if (notinit || *blen < ap->minlen)
2493                 args->minlen = ap->minlen;
2494         /*
2495          * If the best seen length is less than the request
2496          * length, use the best as the minimum.
2497          */
2498         else if (*blen < ap->alen)
2499                 args->minlen = *blen;
2500         /*
2501          * Otherwise we've seen an extent as big as alen,
2502          * use that as the minimum.
2503          */
2504         else
2505                 args->minlen = ap->alen;
2506
2507         /*
2508          * set the failure fallback case to look in the selected
2509          * AG as the stream may have moved.
2510          */
2511         if (xfs_inode_is_filestream(ap->ip))
2512                 ap->rval = args->fsbno = XFS_AGB_TO_FSB(mp, ag, 0);
2513
2514         return 0;
2515 }
2516
2517 STATIC int
2518 xfs_bmap_btalloc(
2519         xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
2520 {
2521         xfs_mount_t     *mp;            /* mount point structure */
2522         xfs_alloctype_t atype = 0;      /* type for allocation routines */
2523         xfs_extlen_t    align;          /* minimum allocation alignment */
2524         xfs_agnumber_t  fb_agno;        /* ag number of ap->firstblock */
2525         xfs_agnumber_t  ag;
2526         xfs_alloc_arg_t args;
2527         xfs_extlen_t    blen;
2528         xfs_extlen_t    nextminlen = 0;
2529         int             nullfb;         /* true if ap->firstblock isn't set */
2530         int             isaligned;
2531         int             tryagain;
2532         int             error;
2533
2534         mp = ap->ip->i_mount;
2535         align = ap->userdata ? xfs_get_extsz_hint(ap->ip) : 0;
2536         if (unlikely(align)) {
2537                 error = xfs_bmap_extsize_align(mp, ap->gotp, ap->prevp,
2538                                                 align, 0, ap->eof, 0, ap->conv,
2539                                                 &ap->off, &ap->alen);
2540                 ASSERT(!error);
2541                 ASSERT(ap->alen);
2542         }
2543         nullfb = ap->firstblock == NULLFSBLOCK;
2544         fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, ap->firstblock);
2545         if (nullfb) {
2546                 if (ap->userdata && xfs_inode_is_filestream(ap->ip)) {
2547                         ag = xfs_filestream_lookup_ag(ap->ip);
2548                         ag = (ag != NULLAGNUMBER) ? ag : 0;
2549                         ap->rval = XFS_AGB_TO_FSB(mp, ag, 0);
2550                 } else {
2551                         ap->rval = XFS_INO_TO_FSB(mp, ap->ip->i_ino);
2552                 }
2553         } else
2554                 ap->rval = ap->firstblock;
2555
2556         xfs_bmap_adjacent(ap);
2557
2558         /*
2559          * If allowed, use ap->rval; otherwise must use firstblock since
2560          * it's in the right allocation group.
2561          */
2562         if (nullfb || XFS_FSB_TO_AGNO(mp, ap->rval) == fb_agno)
2563                 ;
2564         else
2565                 ap->rval = ap->firstblock;
2566         /*
2567          * Normal allocation, done through xfs_alloc_vextent.
2568          */
2569         tryagain = isaligned = 0;
2570         args.tp = ap->tp;
2571         args.mp = mp;
2572         args.fsbno = ap->rval;
2573         args.maxlen = MIN(ap->alen, mp->m_sb.sb_agblocks);
2574         args.firstblock = ap->firstblock;
2575         blen = 0;
2576         if (nullfb) {
2577                 error = xfs_bmap_btalloc_nullfb(ap, &args, &blen);
2578                 if (error)
2579                         return error;
2580         } else if (ap->low) {
2581                 if (xfs_inode_is_filestream(ap->ip))
2582                         args.type = XFS_ALLOCTYPE_FIRST_AG;
2583                 else
2584                         args.type = XFS_ALLOCTYPE_START_BNO;
2585                 args.total = args.minlen = ap->minlen;
2586         } else {
2587                 args.type = XFS_ALLOCTYPE_NEAR_BNO;
2588                 args.total = ap->total;
2589                 args.minlen = ap->minlen;
2590         }
2591         /* apply extent size hints if obtained earlier */
2592         if (unlikely(align)) {
2593                 args.prod = align;
2594                 if ((args.mod = (xfs_extlen_t)do_mod(ap->off, args.prod)))
2595                         args.mod = (xfs_extlen_t)(args.prod - args.mod);
2596         } else if (mp->m_sb.sb_blocksize >= PAGE_CACHE_SIZE) {
2597                 args.prod = 1;
2598                 args.mod = 0;
2599         } else {
2600                 args.prod = PAGE_CACHE_SIZE >> mp->m_sb.sb_blocklog;
2601                 if ((args.mod = (xfs_extlen_t)(do_mod(ap->off, args.prod))))
2602                         args.mod = (xfs_extlen_t)(args.prod - args.mod);
2603         }
2604         /*
2605          * If we are not low on available data blocks, and the
2606          * underlying logical volume manager is a stripe, and
2607          * the file offset is zero then try to allocate data
2608          * blocks on stripe unit boundary.
2609          * NOTE: ap->aeof is only set if the allocation length
2610          * is >= the stripe unit and the allocation offset is
2611          * at the end of file.
2612          */
2613         if (!ap->low && ap->aeof) {
2614                 if (!ap->off) {
2615                         args.alignment = mp->m_dalign;
2616                         atype = args.type;
2617                         isaligned = 1;
2618                         /*
2619                          * Adjust for alignment
2620                          */
2621                         if (blen > args.alignment && blen <= ap->alen)
2622                                 args.minlen = blen - args.alignment;
2623                         args.minalignslop = 0;
2624                 } else {
2625                         /*
2626                          * First try an exact bno allocation.
2627                          * If it fails then do a near or start bno
2628                          * allocation with alignment turned on.
2629                          */
2630                         atype = args.type;
2631                         tryagain = 1;
2632                         args.type = XFS_ALLOCTYPE_THIS_BNO;
2633                         args.alignment = 1;
2634                         /*
2635                          * Compute the minlen+alignment for the
2636                          * next case.  Set slop so that the value
2637                          * of minlen+alignment+slop doesn't go up
2638                          * between the calls.
2639                          */
2640                         if (blen > mp->m_dalign && blen <= ap->alen)
2641                                 nextminlen = blen - mp->m_dalign;
2642                         else
2643                                 nextminlen = args.minlen;
2644                         if (nextminlen + mp->m_dalign > args.minlen + 1)
2645                                 args.minalignslop =
2646                                         nextminlen + mp->m_dalign -
2647                                         args.minlen - 1;
2648                         else
2649                                 args.minalignslop = 0;
2650                 }
2651         } else {
2652                 args.alignment = 1;
2653                 args.minalignslop = 0;
2654         }
2655         args.minleft = ap->minleft;
2656         args.wasdel = ap->wasdel;
2657         args.isfl = 0;
2658         args.userdata = ap->userdata;
2659         if ((error = xfs_alloc_vextent(&args)))
2660                 return error;
2661         if (tryagain && args.fsbno == NULLFSBLOCK) {
2662                 /*
2663                  * Exact allocation failed. Now try with alignment
2664                  * turned on.
2665                  */
2666                 args.type = atype;
2667                 args.fsbno = ap->rval;
2668                 args.alignment = mp->m_dalign;
2669                 args.minlen = nextminlen;
2670                 args.minalignslop = 0;
2671                 isaligned = 1;
2672                 if ((error = xfs_alloc_vextent(&args)))
2673                         return error;
2674         }
2675         if (isaligned && args.fsbno == NULLFSBLOCK) {
2676                 /*
2677                  * allocation failed, so turn off alignment and
2678                  * try again.
2679                  */
2680                 args.type = atype;
2681                 args.fsbno = ap->rval;
2682                 args.alignment = 0;
2683                 if ((error = xfs_alloc_vextent(&args)))
2684                         return error;
2685         }
2686         if (args.fsbno == NULLFSBLOCK && nullfb &&
2687             args.minlen > ap->minlen) {
2688                 args.minlen = ap->minlen;
2689                 args.type = XFS_ALLOCTYPE_START_BNO;
2690                 args.fsbno = ap->rval;
2691                 if ((error = xfs_alloc_vextent(&args)))
2692                         return error;
2693         }
2694         if (args.fsbno == NULLFSBLOCK && nullfb) {
2695                 args.fsbno = 0;
2696                 args.type = XFS_ALLOCTYPE_FIRST_AG;
2697                 args.total = ap->minlen;
2698                 args.minleft = 0;
2699                 if ((error = xfs_alloc_vextent(&args)))
2700                         return error;
2701                 ap->low = 1;
2702         }
2703         if (args.fsbno != NULLFSBLOCK) {
2704                 ap->firstblock = ap->rval = args.fsbno;
2705                 ASSERT(nullfb || fb_agno == args.agno ||
2706                        (ap->low && fb_agno < args.agno));
2707                 ap->alen = args.len;
2708                 ap->ip->i_d.di_nblocks += args.len;
2709                 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
2710                 if (ap->wasdel)
2711                         ap->ip->i_delayed_blks -= args.len;
2712                 /*
2713                  * Adjust the disk quota also. This was reserved
2714                  * earlier.
2715                  */
2716                 xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
2717                         ap->wasdel ? XFS_TRANS_DQ_DELBCOUNT :
2718                                         XFS_TRANS_DQ_BCOUNT,
2719                         (long) args.len);
2720         } else {
2721                 ap->rval = NULLFSBLOCK;
2722                 ap->alen = 0;
2723         }
2724         return 0;
2725 }
2726
2727 /*
2728  * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
2729  * It figures out where to ask the underlying allocator to put the new extent.
2730  */
2731 STATIC int
2732 xfs_bmap_alloc(
2733         xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
2734 {
2735         if (XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata)
2736                 return xfs_bmap_rtalloc(ap);
2737         return xfs_bmap_btalloc(ap);
2738 }
2739
2740 /*
2741  * Transform a btree format file with only one leaf node, where the
2742  * extents list will fit in the inode, into an extents format file.
2743  * Since the file extents are already in-core, all we have to do is
2744  * give up the space for the btree root and pitch the leaf block.
2745  */
2746 STATIC int                              /* error */
2747 xfs_bmap_btree_to_extents(
2748         xfs_trans_t             *tp,    /* transaction pointer */
2749         xfs_inode_t             *ip,    /* incore inode pointer */
2750         xfs_btree_cur_t         *cur,   /* btree cursor */
2751         int                     *logflagsp, /* inode logging flags */
2752         int                     whichfork)  /* data or attr fork */
2753 {
2754         /* REFERENCED */
2755         struct xfs_btree_block  *cblock;/* child btree block */
2756         xfs_fsblock_t           cbno;   /* child block number */
2757         xfs_buf_t               *cbp;   /* child block's buffer */
2758         int                     error;  /* error return value */
2759         xfs_ifork_t             *ifp;   /* inode fork data */
2760         xfs_mount_t             *mp;    /* mount point structure */
2761         __be64                  *pp;    /* ptr to block address */
2762         struct xfs_btree_block  *rblock;/* root btree block */
2763
2764         mp = ip->i_mount;
2765         ifp = XFS_IFORK_PTR(ip, whichfork);
2766         ASSERT(ifp->if_flags & XFS_IFEXTENTS);
2767         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
2768         rblock = ifp->if_broot;
2769         ASSERT(be16_to_cpu(rblock->bb_level) == 1);
2770         ASSERT(be16_to_cpu(rblock->bb_numrecs) == 1);
2771         ASSERT(xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0) == 1);
2772         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, rblock, 1, ifp->if_broot_bytes);
2773         cbno = be64_to_cpu(*pp);
2774         *logflagsp = 0;
2775 #ifdef DEBUG
2776         if ((error = xfs_btree_check_lptr(cur, cbno, 1)))
2777                 return error;
2778 #endif
2779         if ((error = xfs_btree_read_bufl(mp, tp, cbno, 0, &cbp,
2780                         XFS_BMAP_BTREE_REF)))
2781                 return error;
2782         cblock = XFS_BUF_TO_BLOCK(cbp);
2783         if ((error = xfs_btree_check_block(cur, cblock, 0, cbp)))
2784                 return error;
2785         xfs_bmap_add_free(cbno, 1, cur->bc_private.b.flist, mp);
2786         ip->i_d.di_nblocks--;
2787         xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, -1L);
2788         xfs_trans_binval(tp, cbp);
2789         if (cur->bc_bufs[0] == cbp)
2790                 cur->bc_bufs[0] = NULL;
2791         xfs_iroot_realloc(ip, -1, whichfork);
2792         ASSERT(ifp->if_broot == NULL);
2793         ASSERT((ifp->if_flags & XFS_IFBROOT) == 0);
2794         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
2795         *logflagsp = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
2796         return 0;
2797 }
2798
2799 /*
2800  * Called by xfs_bmapi to update file extent records and the btree
2801  * after removing space (or undoing a delayed allocation).
2802  */
2803 STATIC int                              /* error */
2804 xfs_bmap_del_extent(
2805         xfs_inode_t             *ip,    /* incore inode pointer */
2806         xfs_trans_t             *tp,    /* current transaction pointer */
2807         xfs_extnum_t            idx,    /* extent number to update/delete */
2808         xfs_bmap_free_t         *flist, /* list of extents to be freed */
2809         xfs_btree_cur_t         *cur,   /* if null, not a btree */
2810         xfs_bmbt_irec_t         *del,   /* data to remove from extents */
2811         int                     *logflagsp, /* inode logging flags */
2812         int                     whichfork, /* data or attr fork */
2813         int                     rsvd)   /* OK to allocate reserved blocks */
2814 {
2815         xfs_filblks_t           da_new; /* new delay-alloc indirect blocks */
2816         xfs_filblks_t           da_old; /* old delay-alloc indirect blocks */
2817         xfs_fsblock_t           del_endblock=0; /* first block past del */
2818         xfs_fileoff_t           del_endoff;     /* first offset past del */
2819         int                     delay;  /* current block is delayed allocated */
2820         int                     do_fx;  /* free extent at end of routine */
2821         xfs_bmbt_rec_host_t     *ep;    /* current extent entry pointer */
2822         int                     error;  /* error return value */
2823         int                     flags;  /* inode logging flags */
2824         xfs_bmbt_irec_t         got;    /* current extent entry */
2825         xfs_fileoff_t           got_endoff;     /* first offset past got */
2826         int                     i;      /* temp state */
2827         xfs_ifork_t             *ifp;   /* inode fork pointer */
2828         xfs_mount_t             *mp;    /* mount structure */
2829         xfs_filblks_t           nblks;  /* quota/sb block count */
2830         xfs_bmbt_irec_t         new;    /* new record to be inserted */
2831         /* REFERENCED */
2832         uint                    qfield; /* quota field to update */
2833         xfs_filblks_t           temp;   /* for indirect length calculations */
2834         xfs_filblks_t           temp2;  /* for indirect length calculations */
2835         int                     state = 0;
2836
2837         XFS_STATS_INC(xs_del_exlist);
2838
2839         if (whichfork == XFS_ATTR_FORK)
2840                 state |= BMAP_ATTRFORK;
2841
2842         mp = ip->i_mount;
2843         ifp = XFS_IFORK_PTR(ip, whichfork);
2844         ASSERT((idx >= 0) && (idx < ifp->if_bytes /
2845                 (uint)sizeof(xfs_bmbt_rec_t)));
2846         ASSERT(del->br_blockcount > 0);
2847         ep = xfs_iext_get_ext(ifp, idx);
2848         xfs_bmbt_get_all(ep, &got);
2849         ASSERT(got.br_startoff <= del->br_startoff);
2850         del_endoff = del->br_startoff + del->br_blockcount;
2851         got_endoff = got.br_startoff + got.br_blockcount;
2852         ASSERT(got_endoff >= del_endoff);
2853         delay = isnullstartblock(got.br_startblock);
2854         ASSERT(isnullstartblock(del->br_startblock) == delay);
2855         flags = 0;
2856         qfield = 0;
2857         error = 0;
2858         /*
2859          * If deleting a real allocation, must free up the disk space.
2860          */
2861         if (!delay) {
2862                 flags = XFS_ILOG_CORE;
2863                 /*
2864                  * Realtime allocation.  Free it and record di_nblocks update.
2865                  */
2866                 if (whichfork == XFS_DATA_FORK && XFS_IS_REALTIME_INODE(ip)) {
2867                         xfs_fsblock_t   bno;
2868                         xfs_filblks_t   len;
2869
2870                         ASSERT(do_mod(del->br_blockcount,
2871                                       mp->m_sb.sb_rextsize) == 0);
2872                         ASSERT(do_mod(del->br_startblock,
2873                                       mp->m_sb.sb_rextsize) == 0);
2874                         bno = del->br_startblock;
2875                         len = del->br_blockcount;
2876                         do_div(bno, mp->m_sb.sb_rextsize);
2877                         do_div(len, mp->m_sb.sb_rextsize);
2878                         if ((error = xfs_rtfree_extent(ip->i_transp, bno,
2879                                         (xfs_extlen_t)len)))
2880                                 goto done;
2881                         do_fx = 0;
2882                         nblks = len * mp->m_sb.sb_rextsize;
2883                         qfield = XFS_TRANS_DQ_RTBCOUNT;
2884                 }
2885                 /*
2886                  * Ordinary allocation.
2887                  */
2888                 else {
2889                         do_fx = 1;
2890                         nblks = del->br_blockcount;
2891                         qfield = XFS_TRANS_DQ_BCOUNT;
2892                 }
2893                 /*
2894                  * Set up del_endblock and cur for later.
2895                  */
2896                 del_endblock = del->br_startblock + del->br_blockcount;
2897                 if (cur) {
2898                         if ((error = xfs_bmbt_lookup_eq(cur, got.br_startoff,
2899                                         got.br_startblock, got.br_blockcount,
2900                                         &i)))
2901                                 goto done;
2902                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2903                 }
2904                 da_old = da_new = 0;
2905         } else {
2906                 da_old = startblockval(got.br_startblock);
2907                 da_new = 0;
2908                 nblks = 0;
2909                 do_fx = 0;
2910         }
2911         /*
2912          * Set flag value to use in switch statement.
2913          * Left-contig is 2, right-contig is 1.
2914          */
2915         switch (((got.br_startoff == del->br_startoff) << 1) |
2916                 (got_endoff == del_endoff)) {
2917         case 3:
2918                 /*
2919                  * Matches the whole extent.  Delete the entry.
2920                  */
2921                 xfs_iext_remove(ip, idx, 1,
2922                                 whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0);
2923                 ifp->if_lastex = idx;
2924                 if (delay)
2925                         break;
2926                 XFS_IFORK_NEXT_SET(ip, whichfork,
2927                         XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
2928                 flags |= XFS_ILOG_CORE;
2929                 if (!cur) {
2930                         flags |= xfs_ilog_fext(whichfork);
2931                         break;
2932                 }
2933                 if ((error = xfs_btree_delete(cur, &i)))
2934                         goto done;
2935                 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2936                 break;
2937
2938         case 2:
2939                 /*
2940                  * Deleting the first part of the extent.
2941                  */
2942                 trace_xfs_bmap_pre_update(ip, idx, state, _THIS_IP_);
2943                 xfs_bmbt_set_startoff(ep, del_endoff);
2944                 temp = got.br_blockcount - del->br_blockcount;
2945                 xfs_bmbt_set_blockcount(ep, temp);
2946                 ifp->if_lastex = idx;
2947                 if (delay) {
2948                         temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2949                                 da_old);
2950                         xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
2951                         trace_xfs_bmap_post_update(ip, idx, state, _THIS_IP_);
2952                         da_new = temp;
2953                         break;
2954                 }
2955                 xfs_bmbt_set_startblock(ep, del_endblock);
2956                 trace_xfs_bmap_post_update(ip, idx, state, _THIS_IP_);
2957                 if (!cur) {
2958                         flags |= xfs_ilog_fext(whichfork);
2959                         break;
2960                 }
2961                 if ((error = xfs_bmbt_update(cur, del_endoff, del_endblock,
2962                                 got.br_blockcount - del->br_blockcount,
2963                                 got.br_state)))
2964                         goto done;
2965                 break;
2966
2967         case 1:
2968                 /*
2969                  * Deleting the last part of the extent.
2970                  */
2971                 temp = got.br_blockcount - del->br_blockcount;
2972                 trace_xfs_bmap_pre_update(ip, idx, state, _THIS_IP_);
2973                 xfs_bmbt_set_blockcount(ep, temp);
2974                 ifp->if_lastex = idx;
2975                 if (delay) {
2976                         temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2977                                 da_old);
2978                         xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
2979                         trace_xfs_bmap_post_update(ip, idx, state, _THIS_IP_);
2980                         da_new = temp;
2981                         break;
2982                 }
2983                 trace_xfs_bmap_post_update(ip, idx, state, _THIS_IP_);
2984                 if (!cur) {
2985                         flags |= xfs_ilog_fext(whichfork);
2986                         break;
2987                 }
2988                 if ((error = xfs_bmbt_update(cur, got.br_startoff,
2989                                 got.br_startblock,
2990                                 got.br_blockcount - del->br_blockcount,
2991                                 got.br_state)))
2992                         goto done;
2993                 break;
2994
2995         case 0:
2996                 /*
2997                  * Deleting the middle of the extent.
2998                  */
2999                 temp = del->br_startoff - got.br_startoff;
3000                 trace_xfs_bmap_pre_update(ip, idx, state, _THIS_IP_);
3001                 xfs_bmbt_set_blockcount(ep, temp);
3002                 new.br_startoff = del_endoff;
3003                 temp2 = got_endoff - del_endoff;
3004                 new.br_blockcount = temp2;
3005                 new.br_state = got.br_state;
3006                 if (!delay) {
3007                         new.br_startblock = del_endblock;
3008                         flags |= XFS_ILOG_CORE;
3009                         if (cur) {
3010                                 if ((error = xfs_bmbt_update(cur,
3011                                                 got.br_startoff,
3012                                                 got.br_startblock, temp,
3013                                                 got.br_state)))
3014                                         goto done;
3015                                 if ((error = xfs_btree_increment(cur, 0, &i)))
3016                                         goto done;
3017                                 cur->bc_rec.b = new;
3018                                 error = xfs_btree_insert(cur, &i);
3019                                 if (error && error != ENOSPC)
3020                                         goto done;
3021                                 /*
3022                                  * If get no-space back from btree insert,
3023                                  * it tried a split, and we have a zero
3024                                  * block reservation.
3025                                  * Fix up our state and return the error.
3026                                  */
3027                                 if (error == ENOSPC) {
3028                                         /*
3029                                          * Reset the cursor, don't trust
3030                                          * it after any insert operation.
3031                                          */
3032                                         if ((error = xfs_bmbt_lookup_eq(cur,
3033                                                         got.br_startoff,
3034                                                         got.br_startblock,
3035                                                         temp, &i)))
3036                                                 goto done;
3037                                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3038                                         /*
3039                                          * Update the btree record back
3040                                          * to the original value.
3041                                          */
3042                                         if ((error = xfs_bmbt_update(cur,
3043                                                         got.br_startoff,
3044                                                         got.br_startblock,
3045                                                         got.br_blockcount,
3046                                                         got.br_state)))
3047                                                 goto done;
3048                                         /*
3049                                          * Reset the extent record back
3050                                          * to the original value.
3051                                          */
3052                                         xfs_bmbt_set_blockcount(ep,
3053                                                 got.br_blockcount);
3054                                         flags = 0;
3055                                         error = XFS_ERROR(ENOSPC);
3056                                         goto done;
3057                                 }
3058                                 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3059                         } else
3060                                 flags |= xfs_ilog_fext(whichfork);
3061                         XFS_IFORK_NEXT_SET(ip, whichfork,
3062                                 XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
3063                 } else {
3064                         ASSERT(whichfork == XFS_DATA_FORK);
3065                         temp = xfs_bmap_worst_indlen(ip, temp);
3066                         xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
3067                         temp2 = xfs_bmap_worst_indlen(ip, temp2);
3068                         new.br_startblock = nullstartblock((int)temp2);
3069                         da_new = temp + temp2;
3070                         while (da_new > da_old) {
3071                                 if (temp) {
3072                                         temp--;
3073                                         da_new--;
3074                                         xfs_bmbt_set_startblock(ep,
3075                                                 nullstartblock((int)temp));
3076                                 }
3077                                 if (da_new == da_old)
3078                                         break;
3079                                 if (temp2) {
3080                                         temp2--;
3081                                         da_new--;
3082                                         new.br_startblock =
3083                                                 nullstartblock((int)temp2);
3084                                 }
3085                         }
3086                 }
3087                 trace_xfs_bmap_post_update(ip, idx, state, _THIS_IP_);
3088                 xfs_iext_insert(ip, idx + 1, 1, &new, state);
3089                 ifp->if_lastex = idx + 1;
3090                 break;
3091         }
3092         /*
3093          * If we need to, add to list of extents to delete.
3094          */
3095         if (do_fx)
3096                 xfs_bmap_add_free(del->br_startblock, del->br_blockcount, flist,
3097                         mp);
3098         /*
3099          * Adjust inode # blocks in the file.
3100          */
3101         if (nblks)
3102                 ip->i_d.di_nblocks -= nblks;
3103         /*
3104          * Adjust quota data.
3105          */
3106         if (qfield)
3107                 xfs_trans_mod_dquot_byino(tp, ip, qfield, (long)-nblks);
3108
3109         /*
3110          * Account for change in delayed indirect blocks.
3111          * Nothing to do for disk quota accounting here.
3112          */
3113         ASSERT(da_old >= da_new);
3114         if (da_old > da_new)
3115                 xfs_mod_incore_sb(mp, XFS_SBS_FDBLOCKS, (int64_t)(da_old - da_new),
3116                         rsvd);
3117 done:
3118         *logflagsp = flags;
3119         return error;
3120 }
3121
3122 /*
3123  * Remove the entry "free" from the free item list.  Prev points to the
3124  * previous entry, unless "free" is the head of the list.
3125  */
3126 STATIC void
3127 xfs_bmap_del_free(
3128         xfs_bmap_free_t         *flist, /* free item list header */
3129         xfs_bmap_free_item_t    *prev,  /* previous item on list, if any */
3130         xfs_bmap_free_item_t    *free)  /* list item to be freed */
3131 {
3132         if (prev)
3133                 prev->xbfi_next = free->xbfi_next;
3134         else
3135                 flist->xbf_first = free->xbfi_next;
3136         flist->xbf_count--;
3137         kmem_zone_free(xfs_bmap_free_item_zone, free);
3138 }
3139
3140 /*
3141  * Convert an extents-format file into a btree-format file.
3142  * The new file will have a root block (in the inode) and a single child block.
3143  */
3144 STATIC int                                      /* error */
3145 xfs_bmap_extents_to_btree(
3146         xfs_trans_t             *tp,            /* transaction pointer */
3147         xfs_inode_t             *ip,            /* incore inode pointer */
3148         xfs_fsblock_t           *firstblock,    /* first-block-allocated */
3149         xfs_bmap_free_t         *flist,         /* blocks freed in xaction */
3150         xfs_btree_cur_t         **curp,         /* cursor returned to caller */
3151         int                     wasdel,         /* converting a delayed alloc */
3152         int                     *logflagsp,     /* inode logging flags */
3153         int                     whichfork)      /* data or attr fork */
3154 {
3155         struct xfs_btree_block  *ablock;        /* allocated (child) bt block */
3156         xfs_buf_t               *abp;           /* buffer for ablock */
3157         xfs_alloc_arg_t         args;           /* allocation arguments */
3158         xfs_bmbt_rec_t          *arp;           /* child record pointer */
3159         struct xfs_btree_block  *block;         /* btree root block */
3160         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
3161         xfs_bmbt_rec_host_t     *ep;            /* extent record pointer */
3162         int                     error;          /* error return value */
3163         xfs_extnum_t            i, cnt;         /* extent record index */
3164         xfs_ifork_t             *ifp;           /* inode fork pointer */
3165         xfs_bmbt_key_t          *kp;            /* root block key pointer */
3166         xfs_mount_t             *mp;            /* mount structure */
3167         xfs_extnum_t            nextents;       /* number of file extents */
3168         xfs_bmbt_ptr_t          *pp;            /* root block address pointer */
3169
3170         ifp = XFS_IFORK_PTR(ip, whichfork);
3171         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS);
3172         ASSERT(ifp->if_ext_max ==
3173                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
3174         /*
3175          * Make space in the inode incore.
3176          */
3177         xfs_iroot_realloc(ip, 1, whichfork);
3178         ifp->if_flags |= XFS_IFBROOT;
3179
3180         /*
3181          * Fill in the root.
3182          */
3183         block = ifp->if_broot;
3184         block->bb_magic = cpu_to_be32(XFS_BMAP_MAGIC);
3185         block->bb_level = cpu_to_be16(1);
3186         block->bb_numrecs = cpu_to_be16(1);
3187         block->bb_u.l.bb_leftsib = cpu_to_be64(NULLDFSBNO);
3188         block->bb_u.l.bb_rightsib = cpu_to_be64(NULLDFSBNO);
3189
3190         /*
3191          * Need a cursor.  Can't allocate until bb_level is filled in.
3192          */
3193         mp = ip->i_mount;
3194         cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
3195         cur->bc_private.b.firstblock = *firstblock;
3196         cur->bc_private.b.flist = flist;
3197         cur->bc_private.b.flags = wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
3198         /*
3199          * Convert to a btree with two levels, one record in root.
3200          */
3201         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_BTREE);
3202         args.tp = tp;
3203         args.mp = mp;
3204         args.firstblock = *firstblock;
3205         if (*firstblock == NULLFSBLOCK) {
3206                 args.type = XFS_ALLOCTYPE_START_BNO;
3207                 args.fsbno = XFS_INO_TO_FSB(mp, ip->i_ino);
3208         } else if (flist->xbf_low) {
3209                 args.type = XFS_ALLOCTYPE_START_BNO;
3210                 args.fsbno = *firstblock;
3211         } else {
3212                 args.type = XFS_ALLOCTYPE_NEAR_BNO;
3213                 args.fsbno = *firstblock;
3214         }
3215         args.minlen = args.maxlen = args.prod = 1;
3216         args.total = args.minleft = args.alignment = args.mod = args.isfl =
3217                 args.minalignslop = 0;
3218         args.wasdel = wasdel;
3219         *logflagsp = 0;
3220         if ((error = xfs_alloc_vextent(&args))) {
3221                 xfs_iroot_realloc(ip, -1, whichfork);
3222                 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
3223                 return error;
3224         }
3225         /*
3226          * Allocation can't fail, the space was reserved.
3227          */
3228         ASSERT(args.fsbno != NULLFSBLOCK);
3229         ASSERT(*firstblock == NULLFSBLOCK ||
3230                args.agno == XFS_FSB_TO_AGNO(mp, *firstblock) ||
3231                (flist->xbf_low &&
3232                 args.agno > XFS_FSB_TO_AGNO(mp, *firstblock)));
3233         *firstblock = cur->bc_private.b.firstblock = args.fsbno;
3234         cur->bc_private.b.allocated++;
3235         ip->i_d.di_nblocks++;
3236         xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, 1L);
3237         abp = xfs_btree_get_bufl(mp, tp, args.fsbno, 0);
3238         /*
3239          * Fill in the child block.
3240          */
3241         ablock = XFS_BUF_TO_BLOCK(abp);
3242         ablock->bb_magic = cpu_to_be32(XFS_BMAP_MAGIC);
3243         ablock->bb_level = 0;
3244         ablock->bb_u.l.bb_leftsib = cpu_to_be64(NULLDFSBNO);
3245         ablock->bb_u.l.bb_rightsib = cpu_to_be64(NULLDFSBNO);
3246         arp = XFS_BMBT_REC_ADDR(mp, ablock, 1);
3247         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3248         for (cnt = i = 0; i < nextents; i++) {
3249                 ep = xfs_iext_get_ext(ifp, i);
3250                 if (!isnullstartblock(xfs_bmbt_get_startblock(ep))) {
3251                         arp->l0 = cpu_to_be64(ep->l0);
3252                         arp->l1 = cpu_to_be64(ep->l1);
3253                         arp++; cnt++;
3254                 }
3255         }
3256         ASSERT(cnt == XFS_IFORK_NEXTENTS(ip, whichfork));
3257         xfs_btree_set_numrecs(ablock, cnt);
3258
3259         /*
3260          * Fill in the root key and pointer.
3261          */
3262         kp = XFS_BMBT_KEY_ADDR(mp, block, 1);
3263         arp = XFS_BMBT_REC_ADDR(mp, ablock, 1);
3264         kp->br_startoff = cpu_to_be64(xfs_bmbt_disk_get_startoff(arp));
3265         pp = XFS_BMBT_PTR_ADDR(mp, block, 1, xfs_bmbt_get_maxrecs(cur,
3266                                                 be16_to_cpu(block->bb_level)));
3267         *pp = cpu_to_be64(args.fsbno);
3268
3269         /*
3270          * Do all this logging at the end so that
3271          * the root is at the right level.
3272          */
3273         xfs_btree_log_block(cur, abp, XFS_BB_ALL_BITS);
3274         xfs_btree_log_recs(cur, abp, 1, be16_to_cpu(ablock->bb_numrecs));
3275         ASSERT(*curp == NULL);
3276         *curp = cur;
3277         *logflagsp = XFS_ILOG_CORE | xfs_ilog_fbroot(whichfork);
3278         return 0;
3279 }
3280
3281 /*
3282  * Calculate the default attribute fork offset for newly created inodes.
3283  */
3284 uint
3285 xfs_default_attroffset(
3286         struct xfs_inode        *ip)
3287 {
3288         struct xfs_mount        *mp = ip->i_mount;
3289         uint                    offset;
3290
3291         if (mp->m_sb.sb_inodesize == 256) {
3292                 offset = XFS_LITINO(mp) -
3293                                 XFS_BMDR_SPACE_CALC(MINABTPTRS);
3294         } else {
3295                 offset = XFS_BMDR_SPACE_CALC(6 * MINABTPTRS);
3296         }
3297
3298         ASSERT(offset < XFS_LITINO(mp));
3299         return offset;
3300 }
3301
3302 /*
3303  * Helper routine to reset inode di_forkoff field when switching
3304  * attribute fork from local to extent format - we reset it where
3305  * possible to make space available for inline data fork extents.
3306  */
3307 STATIC void
3308 xfs_bmap_forkoff_reset(
3309         xfs_mount_t     *mp,
3310         xfs_inode_t     *ip,
3311         int             whichfork)
3312 {
3313         if (whichfork == XFS_ATTR_FORK &&
3314             ip->i_d.di_format != XFS_DINODE_FMT_DEV &&
3315             ip->i_d.di_format != XFS_DINODE_FMT_UUID &&
3316             ip->i_d.di_format != XFS_DINODE_FMT_BTREE) {
3317                 uint    dfl_forkoff = xfs_default_attroffset(ip) >> 3;
3318
3319                 if (dfl_forkoff > ip->i_d.di_forkoff) {
3320                         ip->i_d.di_forkoff = dfl_forkoff;
3321                         ip->i_df.if_ext_max =
3322                                 XFS_IFORK_DSIZE(ip) / sizeof(xfs_bmbt_rec_t);
3323                         ip->i_afp->if_ext_max =
3324                                 XFS_IFORK_ASIZE(ip) / sizeof(xfs_bmbt_rec_t);
3325                 }
3326         }
3327 }
3328
3329 /*
3330  * Convert a local file to an extents file.
3331  * This code is out of bounds for data forks of regular files,
3332  * since the file data needs to get logged so things will stay consistent.
3333  * (The bmap-level manipulations are ok, though).
3334  */
3335 STATIC int                              /* error */
3336 xfs_bmap_local_to_extents(
3337         xfs_trans_t     *tp,            /* transaction pointer */
3338         xfs_inode_t     *ip,            /* incore inode pointer */
3339         xfs_fsblock_t   *firstblock,    /* first block allocated in xaction */
3340         xfs_extlen_t    total,          /* total blocks needed by transaction */
3341         int             *logflagsp,     /* inode logging flags */
3342         int             whichfork)      /* data or attr fork */
3343 {
3344         int             error;          /* error return value */
3345         int             flags;          /* logging flags returned */
3346         xfs_ifork_t     *ifp;           /* inode fork pointer */
3347
3348         /*
3349          * We don't want to deal with the case of keeping inode data inline yet.
3350          * So sending the data fork of a regular inode is invalid.
3351          */
3352         ASSERT(!((ip->i_d.di_mode & S_IFMT) == S_IFREG &&
3353                  whichfork == XFS_DATA_FORK));
3354         ifp = XFS_IFORK_PTR(ip, whichfork);
3355         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
3356         flags = 0;
3357         error = 0;
3358         if (ifp->if_bytes) {
3359                 xfs_alloc_arg_t args;   /* allocation arguments */
3360                 xfs_buf_t       *bp;    /* buffer for extent block */
3361                 xfs_bmbt_rec_host_t *ep;/* extent record pointer */
3362
3363                 args.tp = tp;
3364                 args.mp = ip->i_mount;
3365                 args.firstblock = *firstblock;
3366                 ASSERT((ifp->if_flags &
3367                         (XFS_IFINLINE|XFS_IFEXTENTS|XFS_IFEXTIREC)) == XFS_IFINLINE);
3368                 /*
3369                  * Allocate a block.  We know we need only one, since the
3370                  * file currently fits in an inode.
3371                  */
3372                 if (*firstblock == NULLFSBLOCK) {
3373                         args.fsbno = XFS_INO_TO_FSB(args.mp, ip->i_ino);
3374                         args.type = XFS_ALLOCTYPE_START_BNO;
3375                 } else {
3376                         args.fsbno = *firstblock;
3377                         args.type = XFS_ALLOCTYPE_NEAR_BNO;
3378                 }
3379                 args.total = total;
3380                 args.mod = args.minleft = args.alignment = args.wasdel =
3381                         args.isfl = args.minalignslop = 0;
3382                 args.minlen = args.maxlen = args.prod = 1;
3383                 if ((error = xfs_alloc_vextent(&args)))
3384                         goto done;
3385                 /*
3386                  * Can't fail, the space was reserved.
3387                  */
3388                 ASSERT(args.fsbno != NULLFSBLOCK);
3389                 ASSERT(args.len == 1);
3390                 *firstblock = args.fsbno;
3391                 bp = xfs_btree_get_bufl(args.mp, tp, args.fsbno, 0);
3392                 memcpy((char *)XFS_BUF_PTR(bp), ifp->if_u1.if_data,
3393                         ifp->if_bytes);
3394                 xfs_trans_log_buf(tp, bp, 0, ifp->if_bytes - 1);
3395                 xfs_bmap_forkoff_reset(args.mp, ip, whichfork);
3396                 xfs_idata_realloc(ip, -ifp->if_bytes, whichfork);
3397                 xfs_iext_add(ifp, 0, 1);
3398                 ep = xfs_iext_get_ext(ifp, 0);
3399                 xfs_bmbt_set_allf(ep, 0, args.fsbno, 1, XFS_EXT_NORM);
3400                 trace_xfs_bmap_post_update(ip, 0,
3401                                 whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0,
3402                                 _THIS_IP_);
3403                 XFS_IFORK_NEXT_SET(ip, whichfork, 1);
3404                 ip->i_d.di_nblocks = 1;
3405                 xfs_trans_mod_dquot_byino(tp, ip,
3406                         XFS_TRANS_DQ_BCOUNT, 1L);
3407                 flags |= xfs_ilog_fext(whichfork);
3408         } else {
3409                 ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) == 0);
3410                 xfs_bmap_forkoff_reset(ip->i_mount, ip, whichfork);
3411         }
3412         ifp->if_flags &= ~XFS_IFINLINE;
3413         ifp->if_flags |= XFS_IFEXTENTS;
3414         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
3415         flags |= XFS_ILOG_CORE;
3416 done:
3417         *logflagsp = flags;
3418         return error;
3419 }
3420
3421 /*
3422  * Search the extent records for the entry containing block bno.
3423  * If bno lies in a hole, point to the next entry.  If bno lies
3424  * past eof, *eofp will be set, and *prevp will contain the last
3425  * entry (null if none).  Else, *lastxp will be set to the index
3426  * of the found entry; *gotp will contain the entry.
3427  */
3428 STATIC xfs_bmbt_rec_host_t *            /* pointer to found extent entry */
3429 xfs_bmap_search_multi_extents(
3430         xfs_ifork_t     *ifp,           /* inode fork pointer */
3431         xfs_fileoff_t   bno,            /* block number searched for */
3432         int             *eofp,          /* out: end of file found */
3433         xfs_extnum_t    *lastxp,        /* out: last extent index */
3434         xfs_bmbt_irec_t *gotp,          /* out: extent entry found */
3435         xfs_bmbt_irec_t *prevp)         /* out: previous extent entry found */
3436 {
3437         xfs_bmbt_rec_host_t *ep;                /* extent record pointer */
3438         xfs_extnum_t    lastx;          /* last extent index */
3439
3440         /*
3441          * Initialize the extent entry structure to catch access to
3442          * uninitialized br_startblock field.
3443          */
3444         gotp->br_startoff = 0xffa5a5a5a5a5a5a5LL;
3445         gotp->br_blockcount = 0xa55a5a5a5a5a5a5aLL;
3446         gotp->br_state = XFS_EXT_INVALID;
3447 #if XFS_BIG_BLKNOS
3448         gotp->br_startblock = 0xffffa5a5a5a5a5a5LL;
3449 #else
3450         gotp->br_startblock = 0xffffa5a5;
3451 #endif
3452         prevp->br_startoff = NULLFILEOFF;
3453
3454         ep = xfs_iext_bno_to_ext(ifp, bno, &lastx);
3455         if (lastx > 0) {
3456                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx - 1), prevp);
3457         }
3458         if (lastx < (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t))) {
3459                 xfs_bmbt_get_all(ep, gotp);
3460                 *eofp = 0;
3461         } else {
3462                 if (lastx > 0) {
3463                         *gotp = *prevp;
3464                 }
3465                 *eofp = 1;
3466                 ep = NULL;
3467         }
3468         *lastxp = lastx;
3469         return ep;
3470 }
3471
3472 /*
3473  * Search the extents list for the inode, for the extent containing bno.
3474  * If bno lies in a hole, point to the next entry.  If bno lies past eof,
3475  * *eofp will be set, and *prevp will contain the last entry (null if none).
3476  * Else, *lastxp will be set to the index of the found
3477  * entry; *gotp will contain the entry.
3478  */
3479 STATIC xfs_bmbt_rec_host_t *                 /* pointer to found extent entry */
3480 xfs_bmap_search_extents(
3481         xfs_inode_t     *ip,            /* incore inode pointer */
3482         xfs_fileoff_t   bno,            /* block number searched for */
3483         int             fork,           /* data or attr fork */
3484         int             *eofp,          /* out: end of file found */
3485         xfs_extnum_t    *lastxp,        /* out: last extent index */
3486         xfs_bmbt_irec_t *gotp,          /* out: extent entry found */
3487         xfs_bmbt_irec_t *prevp)         /* out: previous extent entry found */
3488 {
3489         xfs_ifork_t     *ifp;           /* inode fork pointer */
3490         xfs_bmbt_rec_host_t  *ep;            /* extent record pointer */
3491
3492         XFS_STATS_INC(xs_look_exlist);
3493         ifp = XFS_IFORK_PTR(ip, fork);
3494
3495         ep = xfs_bmap_search_multi_extents(ifp, bno, eofp, lastxp, gotp, prevp);
3496
3497         if (unlikely(!(gotp->br_startblock) && (*lastxp != NULLEXTNUM) &&
3498                      !(XFS_IS_REALTIME_INODE(ip) && fork == XFS_DATA_FORK))) {
3499                 xfs_cmn_err(XFS_PTAG_FSBLOCK_ZERO, CE_ALERT, ip->i_mount,
3500                                 "Access to block zero in inode %llu "
3501                                 "start_block: %llx start_off: %llx "
3502                                 "blkcnt: %llx extent-state: %x lastx: %x\n",
3503                         (unsigned long long)ip->i_ino,
3504                         (unsigned long long)gotp->br_startblock,
3505                         (unsigned long long)gotp->br_startoff,
3506                         (unsigned long long)gotp->br_blockcount,
3507                         gotp->br_state, *lastxp);
3508                 *lastxp = NULLEXTNUM;
3509                 *eofp = 1;
3510                 return NULL;
3511         }
3512         return ep;
3513 }
3514
3515 /*
3516  * Compute the worst-case number of indirect blocks that will be used
3517  * for ip's delayed extent of length "len".
3518  */
3519 STATIC xfs_filblks_t
3520 xfs_bmap_worst_indlen(
3521         xfs_inode_t     *ip,            /* incore inode pointer */
3522         xfs_filblks_t   len)            /* delayed extent length */
3523 {
3524         int             level;          /* btree level number */
3525         int             maxrecs;        /* maximum record count at this level */
3526         xfs_mount_t     *mp;            /* mount structure */
3527         xfs_filblks_t   rval;           /* return value */
3528
3529         mp = ip->i_mount;
3530         maxrecs = mp->m_bmap_dmxr[0];
3531         for (level = 0, rval = 0;
3532              level < XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK);
3533              level++) {
3534                 len += maxrecs - 1;
3535                 do_div(len, maxrecs);
3536                 rval += len;
3537                 if (len == 1)
3538                         return rval + XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK) -
3539                                 level - 1;
3540                 if (level == 0)
3541                         maxrecs = mp->m_bmap_dmxr[1];
3542         }
3543         return rval;
3544 }
3545
3546 /*
3547  * Convert inode from non-attributed to attributed.
3548  * Must not be in a transaction, ip must not be locked.
3549  */
3550 int                                             /* error code */
3551 xfs_bmap_add_attrfork(
3552         xfs_inode_t             *ip,            /* incore inode pointer */
3553         int                     size,           /* space new attribute needs */
3554         int                     rsvd)           /* xact may use reserved blks */
3555 {
3556         xfs_fsblock_t           firstblock;     /* 1st block/ag allocated */
3557         xfs_bmap_free_t         flist;          /* freed extent records */
3558         xfs_mount_t             *mp;            /* mount structure */
3559         xfs_trans_t             *tp;            /* transaction pointer */
3560         int                     blks;           /* space reservation */
3561         int                     version = 1;    /* superblock attr version */
3562         int                     committed;      /* xaction was committed */
3563         int                     logflags;       /* logging flags */
3564         int                     error;          /* error return value */
3565
3566         ASSERT(XFS_IFORK_Q(ip) == 0);
3567         ASSERT(ip->i_df.if_ext_max ==
3568                XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t));
3569
3570         mp = ip->i_mount;
3571         ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
3572         tp = xfs_trans_alloc(mp, XFS_TRANS_ADDAFORK);
3573         blks = XFS_ADDAFORK_SPACE_RES(mp);
3574         if (rsvd)
3575                 tp->t_flags |= XFS_TRANS_RESERVE;
3576         if ((error = xfs_trans_reserve(tp, blks, XFS_ADDAFORK_LOG_RES(mp), 0,
3577                         XFS_TRANS_PERM_LOG_RES, XFS_ADDAFORK_LOG_COUNT)))
3578                 goto error0;
3579         xfs_ilock(ip, XFS_ILOCK_EXCL);
3580         error = xfs_trans_reserve_quota_nblks(tp, ip, blks, 0, rsvd ?
3581                         XFS_QMOPT_RES_REGBLKS | XFS_QMOPT_FORCE_RES :
3582                         XFS_QMOPT_RES_REGBLKS);
3583         if (error) {
3584                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3585                 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES);
3586                 return error;
3587         }
3588         if (XFS_IFORK_Q(ip))
3589                 goto error1;
3590         if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS) {
3591                 /*
3592                  * For inodes coming from pre-6.2 filesystems.
3593                  */
3594                 ASSERT(ip->i_d.di_aformat == 0);
3595                 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
3596         }
3597         ASSERT(ip->i_d.di_anextents == 0);
3598
3599         xfs_trans_ijoin_ref(tp, ip, XFS_ILOCK_EXCL);
3600         xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
3601
3602         switch (ip->i_d.di_format) {
3603         case XFS_DINODE_FMT_DEV:
3604                 ip->i_d.di_forkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
3605                 break;
3606         case XFS_DINODE_FMT_UUID:
3607                 ip->i_d.di_forkoff = roundup(sizeof(uuid_t), 8) >> 3;
3608                 break;
3609         case XFS_DINODE_FMT_LOCAL:
3610         case XFS_DINODE_FMT_EXTENTS:
3611         case XFS_DINODE_FMT_BTREE:
3612                 ip->i_d.di_forkoff = xfs_attr_shortform_bytesfit(ip, size);
3613                 if (!ip->i_d.di_forkoff)
3614                         ip->i_d.di_forkoff = xfs_default_attroffset(ip) >> 3;
3615                 else if (mp->m_flags & XFS_MOUNT_ATTR2)
3616                         version = 2;
3617                 break;
3618         default:
3619                 ASSERT(0);
3620                 error = XFS_ERROR(EINVAL);
3621                 goto error1;
3622         }
3623         ip->i_df.if_ext_max =
3624                 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
3625         ASSERT(ip->i_afp == NULL);
3626         ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP);
3627         ip->i_afp->if_ext_max =
3628                 XFS_IFORK_ASIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
3629         ip->i_afp->if_flags = XFS_IFEXTENTS;
3630         logflags = 0;
3631         xfs_bmap_init(&flist, &firstblock);
3632         switch (ip->i_d.di_format) {
3633         case XFS_DINODE_FMT_LOCAL:
3634                 error = xfs_bmap_add_attrfork_local(tp, ip, &firstblock, &flist,
3635                         &logflags);
3636                 break;
3637         case XFS_DINODE_FMT_EXTENTS:
3638                 error = xfs_bmap_add_attrfork_extents(tp, ip, &firstblock,
3639                         &flist, &logflags);
3640                 break;
3641         case XFS_DINODE_FMT_BTREE:
3642                 error = xfs_bmap_add_attrfork_btree(tp, ip, &firstblock, &flist,
3643                         &logflags);
3644                 break;
3645         default:
3646                 error = 0;
3647                 break;
3648         }
3649         if (logflags)
3650                 xfs_trans_log_inode(tp, ip, logflags);
3651         if (error)
3652                 goto error2;
3653         if (!xfs_sb_version_hasattr(&mp->m_sb) ||
3654            (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2)) {
3655                 __int64_t sbfields = 0;
3656
3657                 spin_lock(&mp->m_sb_lock);
3658                 if (!xfs_sb_version_hasattr(&mp->m_sb)) {
3659                         xfs_sb_version_addattr(&mp->m_sb);
3660                         sbfields |= XFS_SB_VERSIONNUM;
3661                 }
3662                 if (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2) {
3663                         xfs_sb_version_addattr2(&mp->m_sb);
3664                         sbfields |= (XFS_SB_VERSIONNUM | XFS_SB_FEATURES2);
3665                 }
3666                 if (sbfields) {
3667                         spin_unlock(&mp->m_sb_lock);
3668                         xfs_mod_sb(tp, sbfields);
3669                 } else
3670                         spin_unlock(&mp->m_sb_lock);
3671         }
3672         if ((error = xfs_bmap_finish(&tp, &flist, &committed)))
3673                 goto error2;
3674         error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
3675         ASSERT(ip->i_df.if_ext_max ==
3676                XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t));
3677         return error;
3678 error2:
3679         xfs_bmap_cancel(&flist);
3680 error1:
3681         xfs_iunlock(ip, XFS_ILOCK_EXCL);
3682 error0:
3683         xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
3684         ASSERT(ip->i_df.if_ext_max ==
3685                XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t));
3686         return error;
3687 }
3688
3689 /*
3690  * Add the extent to the list of extents to be free at transaction end.
3691  * The list is maintained sorted (by block number).
3692  */
3693 /* ARGSUSED */
3694 void
3695 xfs_bmap_add_free(
3696         xfs_fsblock_t           bno,            /* fs block number of extent */
3697         xfs_filblks_t           len,            /* length of extent */
3698         xfs_bmap_free_t         *flist,         /* list of extents */
3699         xfs_mount_t             *mp)            /* mount point structure */
3700 {
3701         xfs_bmap_free_item_t    *cur;           /* current (next) element */
3702         xfs_bmap_free_item_t    *new;           /* new element */
3703         xfs_bmap_free_item_t    *prev;          /* previous element */
3704 #ifdef DEBUG
3705         xfs_agnumber_t          agno;
3706         xfs_agblock_t           agbno;
3707
3708         ASSERT(bno != NULLFSBLOCK);
3709         ASSERT(len > 0);
3710         ASSERT(len <= MAXEXTLEN);
3711         ASSERT(!isnullstartblock(bno));
3712         agno = XFS_FSB_TO_AGNO(mp, bno);
3713         agbno = XFS_FSB_TO_AGBNO(mp, bno);
3714         ASSERT(agno < mp->m_sb.sb_agcount);
3715         ASSERT(agbno < mp->m_sb.sb_agblocks);
3716         ASSERT(len < mp->m_sb.sb_agblocks);
3717         ASSERT(agbno + len <= mp->m_sb.sb_agblocks);
3718 #endif
3719         ASSERT(xfs_bmap_free_item_zone != NULL);
3720         new = kmem_zone_alloc(xfs_bmap_free_item_zone, KM_SLEEP);
3721         new->xbfi_startblock = bno;
3722         new->xbfi_blockcount = (xfs_extlen_t)len;
3723         for (prev = NULL, cur = flist->xbf_first;
3724              cur != NULL;
3725              prev = cur, cur = cur->xbfi_next) {
3726                 if (cur->xbfi_startblock >= bno)
3727                         break;
3728         }
3729         if (prev)
3730                 prev->xbfi_next = new;
3731         else
3732                 flist->xbf_first = new;
3733         new->xbfi_next = cur;
3734         flist->xbf_count++;
3735 }
3736
3737 /*
3738  * Compute and fill in the value of the maximum depth of a bmap btree
3739  * in this filesystem.  Done once, during mount.
3740  */
3741 void
3742 xfs_bmap_compute_maxlevels(
3743         xfs_mount_t     *mp,            /* file system mount structure */
3744         int             whichfork)      /* data or attr fork */
3745 {
3746         int             level;          /* btree level */
3747         uint            maxblocks;      /* max blocks at this level */
3748         uint            maxleafents;    /* max leaf entries possible */
3749         int             maxrootrecs;    /* max records in root block */
3750         int             minleafrecs;    /* min records in leaf block */
3751         int             minnoderecs;    /* min records in node block */
3752         int             sz;             /* root block size */
3753
3754         /*
3755          * The maximum number of extents in a file, hence the maximum
3756          * number of leaf entries, is controlled by the type of di_nextents
3757          * (a signed 32-bit number, xfs_extnum_t), or by di_anextents
3758          * (a signed 16-bit number, xfs_aextnum_t).
3759          *
3760          * Note that we can no longer assume that if we are in ATTR1 that
3761          * the fork offset of all the inodes will be
3762          * (xfs_default_attroffset(ip) >> 3) because we could have mounted
3763          * with ATTR2 and then mounted back with ATTR1, keeping the
3764          * di_forkoff's fixed but probably at various positions. Therefore,
3765          * for both ATTR1 and ATTR2 we have to assume the worst case scenario
3766          * of a minimum size available.
3767          */
3768         if (whichfork == XFS_DATA_FORK) {
3769                 maxleafents = MAXEXTNUM;
3770                 sz = XFS_BMDR_SPACE_CALC(MINDBTPTRS);
3771         } else {
3772                 maxleafents = MAXAEXTNUM;
3773                 sz = XFS_BMDR_SPACE_CALC(MINABTPTRS);
3774         }
3775         maxrootrecs = xfs_bmdr_maxrecs(mp, sz, 0);
3776         minleafrecs = mp->m_bmap_dmnr[0];
3777         minnoderecs = mp->m_bmap_dmnr[1];
3778         maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs;
3779         for (level = 1; maxblocks > 1; level++) {
3780                 if (maxblocks <= maxrootrecs)
3781                         maxblocks = 1;
3782                 else
3783                         maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs;
3784         }
3785         mp->m_bm_maxlevels[whichfork] = level;
3786 }
3787
3788 /*
3789  * Routine to be called at transaction's end by xfs_bmapi, xfs_bunmapi
3790  * caller.  Frees all the extents that need freeing, which must be done
3791  * last due to locking considerations.  We never free any extents in
3792  * the first transaction.  This is to allow the caller to make the first
3793  * transaction a synchronous one so that the pointers to the data being
3794  * broken in this transaction will be permanent before the data is actually
3795  * freed.  This is necessary to prevent blocks from being reallocated
3796  * and written to before the free and reallocation are actually permanent.
3797  * We do not just make the first transaction synchronous here, because
3798  * there are more efficient ways to gain the same protection in some cases
3799  * (see the file truncation code).
3800  *
3801  * Return 1 if the given transaction was committed and a new one
3802  * started, and 0 otherwise in the committed parameter.
3803  */
3804 /*ARGSUSED*/
3805 int                                             /* error */
3806 xfs_bmap_finish(
3807         xfs_trans_t             **tp,           /* transaction pointer addr */
3808         xfs_bmap_free_t         *flist,         /* i/o: list extents to free */
3809         int                     *committed)     /* xact committed or not */
3810 {
3811         xfs_efd_log_item_t      *efd;           /* extent free data */
3812         xfs_efi_log_item_t      *efi;           /* extent free intention */
3813         int                     error;          /* error return value */
3814         xfs_bmap_free_item_t    *free;          /* free extent item */
3815         unsigned int            logres;         /* new log reservation */
3816         unsigned int            logcount;       /* new log count */
3817         xfs_mount_t             *mp;            /* filesystem mount structure */
3818         xfs_bmap_free_item_t    *next;          /* next item on free list */
3819         xfs_trans_t             *ntp;           /* new transaction pointer */
3820
3821         ASSERT((*tp)->t_flags & XFS_TRANS_PERM_LOG_RES);
3822         if (flist->xbf_count == 0) {
3823                 *committed = 0;
3824                 return 0;
3825         }
3826         ntp = *tp;
3827         efi = xfs_trans_get_efi(ntp, flist->xbf_count);
3828         for (free = flist->xbf_first; free; free = free->xbfi_next)
3829                 xfs_trans_log_efi_extent(ntp, efi, free->xbfi_startblock,
3830                         free->xbfi_blockcount);
3831         logres = ntp->t_log_res;
3832         logcount = ntp->t_log_count;
3833         ntp = xfs_trans_dup(*tp);
3834         error = xfs_trans_commit(*tp, 0);
3835         *tp = ntp;
3836         *committed = 1;
3837         /*
3838          * We have a new transaction, so we should return committed=1,
3839          * even though we're returning an error.
3840          */
3841         if (error)
3842                 return error;
3843
3844         /*
3845          * transaction commit worked ok so we can drop the extra ticket
3846          * reference that we gained in xfs_trans_dup()
3847          */
3848         xfs_log_ticket_put(ntp->t_ticket);
3849
3850         if ((error = xfs_trans_reserve(ntp, 0, logres, 0, XFS_TRANS_PERM_LOG_RES,
3851                         logcount)))
3852                 return error;
3853         efd = xfs_trans_get_efd(ntp, efi, flist->xbf_count);
3854         for (free = flist->xbf_first; free != NULL; free = next) {
3855                 next = free->xbfi_next;
3856                 if ((error = xfs_free_extent(ntp, free->xbfi_startblock,
3857                                 free->xbfi_blockcount))) {
3858                         /*
3859                          * The bmap free list will be cleaned up at a
3860                          * higher level.  The EFI will be canceled when
3861                          * this transaction is aborted.
3862                          * Need to force shutdown here to make sure it
3863                          * happens, since this transaction may not be
3864                          * dirty yet.
3865                          */
3866                         mp = ntp->t_mountp;
3867                         if (!XFS_FORCED_SHUTDOWN(mp))
3868                                 xfs_force_shutdown(mp,
3869                                                    (error == EFSCORRUPTED) ?
3870                                                    SHUTDOWN_CORRUPT_INCORE :
3871                                                    SHUTDOWN_META_IO_ERROR);
3872                         return error;
3873                 }
3874                 xfs_trans_log_efd_extent(ntp, efd, free->xbfi_startblock,
3875                         free->xbfi_blockcount);
3876                 xfs_bmap_del_free(flist, NULL, free);
3877         }
3878         return 0;
3879 }
3880
3881 /*
3882  * Free up any items left in the list.
3883  */
3884 void
3885 xfs_bmap_cancel(
3886         xfs_bmap_free_t         *flist) /* list of bmap_free_items */
3887 {
3888         xfs_bmap_free_item_t    *free;  /* free list item */
3889         xfs_bmap_free_item_t    *next;
3890
3891         if (flist->xbf_count == 0)
3892                 return;
3893         ASSERT(flist->xbf_first != NULL);
3894         for (free = flist->xbf_first; free; free = next) {
3895                 next = free->xbfi_next;
3896                 xfs_bmap_del_free(flist, NULL, free);
3897         }
3898         ASSERT(flist->xbf_count == 0);
3899 }
3900
3901 /*
3902  * Returns the file-relative block number of the first unused block(s)
3903  * in the file with at least "len" logically contiguous blocks free.
3904  * This is the lowest-address hole if the file has holes, else the first block
3905  * past the end of file.
3906  * Return 0 if the file is currently local (in-inode).
3907  */
3908 int                                             /* error */
3909 xfs_bmap_first_unused(
3910         xfs_trans_t     *tp,                    /* transaction pointer */
3911         xfs_inode_t     *ip,                    /* incore inode */
3912         xfs_extlen_t    len,                    /* size of hole to find */
3913         xfs_fileoff_t   *first_unused,          /* unused block */
3914         int             whichfork)              /* data or attr fork */
3915 {
3916         int             error;                  /* error return value */
3917         int             idx;                    /* extent record index */
3918         xfs_ifork_t     *ifp;                   /* inode fork pointer */
3919         xfs_fileoff_t   lastaddr;               /* last block number seen */
3920         xfs_fileoff_t   lowest;                 /* lowest useful block */
3921         xfs_fileoff_t   max;                    /* starting useful block */
3922         xfs_fileoff_t   off;                    /* offset for this block */
3923         xfs_extnum_t    nextents;               /* number of extent entries */
3924
3925         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE ||
3926                XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ||
3927                XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
3928         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
3929                 *first_unused = 0;
3930                 return 0;
3931         }
3932         ifp = XFS_IFORK_PTR(ip, whichfork);
3933         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
3934             (error = xfs_iread_extents(tp, ip, whichfork)))
3935                 return error;
3936         lowest = *first_unused;
3937         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3938         for (idx = 0, lastaddr = 0, max = lowest; idx < nextents; idx++) {
3939                 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, idx);
3940                 off = xfs_bmbt_get_startoff(ep);
3941                 /*
3942                  * See if the hole before this extent will work.
3943                  */
3944                 if (off >= lowest + len && off - max >= len) {
3945                         *first_unused = max;
3946                         return 0;
3947                 }
3948                 lastaddr = off + xfs_bmbt_get_blockcount(ep);
3949                 max = XFS_FILEOFF_MAX(lastaddr, lowest);
3950         }
3951         *first_unused = max;
3952         return 0;
3953 }
3954
3955 /*
3956  * Returns the file-relative block number of the last block + 1 before
3957  * last_block (input value) in the file.
3958  * This is not based on i_size, it is based on the extent records.
3959  * Returns 0 for local files, as they do not have extent records.
3960  */
3961 int                                             /* error */
3962 xfs_bmap_last_before(
3963         xfs_trans_t     *tp,                    /* transaction pointer */
3964         xfs_inode_t     *ip,                    /* incore inode */
3965         xfs_fileoff_t   *last_block,            /* last block */
3966         int             whichfork)              /* data or attr fork */
3967 {
3968         xfs_fileoff_t   bno;                    /* input file offset */
3969         int             eof;                    /* hit end of file */
3970         xfs_bmbt_rec_host_t *ep;                /* pointer to last extent */
3971         int             error;                  /* error return value */
3972         xfs_bmbt_irec_t got;                    /* current extent value */
3973         xfs_ifork_t     *ifp;                   /* inode fork pointer */
3974         xfs_extnum_t    lastx;                  /* last extent used */
3975         xfs_bmbt_irec_t prev;                   /* previous extent value */
3976
3977         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
3978             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
3979             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL)
3980                return XFS_ERROR(EIO);
3981         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
3982                 *last_block = 0;
3983                 return 0;
3984         }
3985         ifp = XFS_IFORK_PTR(ip, whichfork);
3986         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
3987             (error = xfs_iread_extents(tp, ip, whichfork)))
3988                 return error;
3989         bno = *last_block - 1;
3990         ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
3991                 &prev);
3992         if (eof || xfs_bmbt_get_startoff(ep) > bno) {
3993                 if (prev.br_startoff == NULLFILEOFF)
3994                         *last_block = 0;
3995                 else
3996                         *last_block = prev.br_startoff + prev.br_blockcount;
3997         }
3998         /*
3999          * Otherwise *last_block is already the right answer.
4000          */
4001         return 0;
4002 }
4003
4004 /*
4005  * Returns the file-relative block number of the first block past eof in
4006  * the file.  This is not based on i_size, it is based on the extent records.
4007  * Returns 0 for local files, as they do not have extent records.
4008  */
4009 int                                             /* error */
4010 xfs_bmap_last_offset(
4011         xfs_trans_t     *tp,                    /* transaction pointer */
4012         xfs_inode_t     *ip,                    /* incore inode */
4013         xfs_fileoff_t   *last_block,            /* last block */
4014         int             whichfork)              /* data or attr fork */
4015 {
4016         xfs_bmbt_rec_host_t *ep;                /* pointer to last extent */
4017         int             error;                  /* error return value */
4018         xfs_ifork_t     *ifp;                   /* inode fork pointer */
4019         xfs_extnum_t    nextents;               /* number of extent entries */
4020
4021         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
4022             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4023             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL)
4024                return XFS_ERROR(EIO);
4025         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
4026                 *last_block = 0;
4027                 return 0;
4028         }
4029         ifp = XFS_IFORK_PTR(ip, whichfork);
4030         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
4031             (error = xfs_iread_extents(tp, ip, whichfork)))
4032                 return error;
4033         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4034         if (!nextents) {
4035                 *last_block = 0;
4036                 return 0;
4037         }
4038         ep = xfs_iext_get_ext(ifp, nextents - 1);
4039         *last_block = xfs_bmbt_get_startoff(ep) + xfs_bmbt_get_blockcount(ep);
4040         return 0;
4041 }
4042
4043 /*
4044  * Returns whether the selected fork of the inode has exactly one
4045  * block or not.  For the data fork we check this matches di_size,
4046  * implying the file's range is 0..bsize-1.
4047  */
4048 int                                     /* 1=>1 block, 0=>otherwise */
4049 xfs_bmap_one_block(
4050         xfs_inode_t     *ip,            /* incore inode */
4051         int             whichfork)      /* data or attr fork */
4052 {
4053         xfs_bmbt_rec_host_t *ep;        /* ptr to fork's extent */
4054         xfs_ifork_t     *ifp;           /* inode fork pointer */
4055         int             rval;           /* return value */
4056         xfs_bmbt_irec_t s;              /* internal version of extent */
4057
4058 #ifndef DEBUG
4059         if (whichfork == XFS_DATA_FORK) {
4060                 return ((ip->i_d.di_mode & S_IFMT) == S_IFREG) ?
4061                         (ip->i_size == ip->i_mount->m_sb.sb_blocksize) :
4062                         (ip->i_d.di_size == ip->i_mount->m_sb.sb_blocksize);
4063         }
4064 #endif  /* !DEBUG */
4065         if (XFS_IFORK_NEXTENTS(ip, whichfork) != 1)
4066                 return 0;
4067         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
4068                 return 0;
4069         ifp = XFS_IFORK_PTR(ip, whichfork);
4070         ASSERT(ifp->if_flags & XFS_IFEXTENTS);
4071         ep = xfs_iext_get_ext(ifp, 0);
4072         xfs_bmbt_get_all(ep, &s);
4073         rval = s.br_startoff == 0 && s.br_blockcount == 1;
4074         if (rval && whichfork == XFS_DATA_FORK)
4075                 ASSERT(ip->i_size == ip->i_mount->m_sb.sb_blocksize);
4076         return rval;
4077 }
4078
4079 STATIC int
4080 xfs_bmap_sanity_check(
4081         struct xfs_mount        *mp,
4082         struct xfs_buf          *bp,
4083         int                     level)
4084 {
4085         struct xfs_btree_block  *block = XFS_BUF_TO_BLOCK(bp);
4086
4087         if (be32_to_cpu(block->bb_magic) != XFS_BMAP_MAGIC ||
4088             be16_to_cpu(block->bb_level) != level ||
4089             be16_to_cpu(block->bb_numrecs) == 0 ||
4090             be16_to_cpu(block->bb_numrecs) > mp->m_bmap_dmxr[level != 0])
4091                 return 0;
4092         return 1;
4093 }
4094
4095 /*
4096  * Read in the extents to if_extents.
4097  * All inode fields are set up by caller, we just traverse the btree
4098  * and copy the records in. If the file system cannot contain unwritten
4099  * extents, the records are checked for no "state" flags.
4100  */
4101 int                                     /* error */
4102 xfs_bmap_read_extents(
4103         xfs_trans_t             *tp,    /* transaction pointer */
4104         xfs_inode_t             *ip,    /* incore inode */
4105         int                     whichfork) /* data or attr fork */
4106 {
4107         struct xfs_btree_block  *block; /* current btree block */
4108         xfs_fsblock_t           bno;    /* block # of "block" */
4109         xfs_buf_t               *bp;    /* buffer for "block" */
4110         int                     error;  /* error return value */
4111         xfs_exntfmt_t           exntf;  /* XFS_EXTFMT_NOSTATE, if checking */
4112         xfs_extnum_t            i, j;   /* index into the extents list */
4113         xfs_ifork_t             *ifp;   /* fork structure */
4114         int                     level;  /* btree level, for checking */
4115         xfs_mount_t             *mp;    /* file system mount structure */
4116         __be64                  *pp;    /* pointer to block address */
4117         /* REFERENCED */
4118         xfs_extnum_t            room;   /* number of entries there's room for */
4119
4120         bno = NULLFSBLOCK;
4121         mp = ip->i_mount;
4122         ifp = XFS_IFORK_PTR(ip, whichfork);
4123         exntf = (whichfork != XFS_DATA_FORK) ? XFS_EXTFMT_NOSTATE :
4124                                         XFS_EXTFMT_INODE(ip);
4125         block = ifp->if_broot;
4126         /*
4127          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
4128          */
4129         level = be16_to_cpu(block->bb_level);
4130         ASSERT(level > 0);
4131         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
4132         bno = be64_to_cpu(*pp);
4133         ASSERT(bno != NULLDFSBNO);
4134         ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
4135         ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
4136         /*
4137          * Go down the tree until leaf level is reached, following the first
4138          * pointer (leftmost) at each level.
4139          */
4140         while (level-- > 0) {
4141                 if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
4142                                 XFS_BMAP_BTREE_REF)))
4143                         return error;
4144                 block = XFS_BUF_TO_BLOCK(bp);
4145                 XFS_WANT_CORRUPTED_GOTO(
4146                         xfs_bmap_sanity_check(mp, bp, level),
4147                         error0);
4148                 if (level == 0)
4149                         break;
4150                 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
4151                 bno = be64_to_cpu(*pp);
4152                 XFS_WANT_CORRUPTED_GOTO(XFS_FSB_SANITY_CHECK(mp, bno), error0);
4153                 xfs_trans_brelse(tp, bp);
4154         }
4155         /*
4156          * Here with bp and block set to the leftmost leaf node in the tree.
4157          */
4158         room = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4159         i = 0;
4160         /*
4161          * Loop over all leaf nodes.  Copy information to the extent records.
4162          */
4163         for (;;) {
4164                 xfs_bmbt_rec_t  *frp;
4165                 xfs_fsblock_t   nextbno;
4166                 xfs_extnum_t    num_recs;
4167                 xfs_extnum_t    start;
4168
4169
4170                 num_recs = xfs_btree_get_numrecs(block);
4171                 if (unlikely(i + num_recs > room)) {
4172                         ASSERT(i + num_recs <= room);
4173                         xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
4174                                 "corrupt dinode %Lu, (btree extents).",
4175                                 (unsigned long long) ip->i_ino);
4176                         XFS_ERROR_REPORT("xfs_bmap_read_extents(1)",
4177                                          XFS_ERRLEVEL_LOW,
4178                                         ip->i_mount);
4179                         goto error0;
4180                 }
4181                 XFS_WANT_CORRUPTED_GOTO(
4182                         xfs_bmap_sanity_check(mp, bp, 0),
4183                         error0);
4184                 /*
4185                  * Read-ahead the next leaf block, if any.
4186                  */
4187                 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
4188                 if (nextbno != NULLFSBLOCK)
4189                         xfs_btree_reada_bufl(mp, nextbno, 1);
4190                 /*
4191                  * Copy records into the extent records.
4192                  */
4193                 frp = XFS_BMBT_REC_ADDR(mp, block, 1);
4194                 start = i;
4195                 for (j = 0; j < num_recs; j++, i++, frp++) {
4196                         xfs_bmbt_rec_host_t *trp = xfs_iext_get_ext(ifp, i);
4197                         trp->l0 = be64_to_cpu(frp->l0);
4198                         trp->l1 = be64_to_cpu(frp->l1);
4199                 }
4200                 if (exntf == XFS_EXTFMT_NOSTATE) {
4201                         /*
4202                          * Check all attribute bmap btree records and
4203                          * any "older" data bmap btree records for a
4204                          * set bit in the "extent flag" position.
4205                          */
4206                         if (unlikely(xfs_check_nostate_extents(ifp,
4207                                         start, num_recs))) {
4208                                 XFS_ERROR_REPORT("xfs_bmap_read_extents(2)",
4209                                                  XFS_ERRLEVEL_LOW,
4210                                                  ip->i_mount);
4211                                 goto error0;
4212                         }
4213                 }
4214                 xfs_trans_brelse(tp, bp);
4215                 bno = nextbno;
4216                 /*
4217                  * If we've reached the end, stop.
4218                  */
4219                 if (bno == NULLFSBLOCK)
4220                         break;
4221                 if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
4222                                 XFS_BMAP_BTREE_REF)))
4223                         return error;
4224                 block = XFS_BUF_TO_BLOCK(bp);
4225         }
4226         ASSERT(i == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)));
4227         ASSERT(i == XFS_IFORK_NEXTENTS(ip, whichfork));
4228         XFS_BMAP_TRACE_EXLIST(ip, i, whichfork);
4229         return 0;
4230 error0:
4231         xfs_trans_brelse(tp, bp);
4232         return XFS_ERROR(EFSCORRUPTED);
4233 }
4234
4235 #ifdef DEBUG
4236 /*
4237  * Add bmap trace insert entries for all the contents of the extent records.
4238  */
4239 void
4240 xfs_bmap_trace_exlist(
4241         xfs_inode_t     *ip,            /* incore inode pointer */
4242         xfs_extnum_t    cnt,            /* count of entries in the list */
4243         int             whichfork,      /* data or attr fork */
4244         unsigned long   caller_ip)
4245 {
4246         xfs_extnum_t    idx;            /* extent record index */
4247         xfs_ifork_t     *ifp;           /* inode fork pointer */
4248         int             state = 0;
4249
4250         if (whichfork == XFS_ATTR_FORK)
4251                 state |= BMAP_ATTRFORK;
4252
4253         ifp = XFS_IFORK_PTR(ip, whichfork);
4254         ASSERT(cnt == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)));
4255         for (idx = 0; idx < cnt; idx++)
4256                 trace_xfs_extlist(ip, idx, whichfork, caller_ip);
4257 }
4258
4259 /*
4260  * Validate that the bmbt_irecs being returned from bmapi are valid
4261  * given the callers original parameters.  Specifically check the
4262  * ranges of the returned irecs to ensure that they only extent beyond
4263  * the given parameters if the XFS_BMAPI_ENTIRE flag was set.
4264  */
4265 STATIC void
4266 xfs_bmap_validate_ret(
4267         xfs_fileoff_t           bno,
4268         xfs_filblks_t           len,
4269         int                     flags,
4270         xfs_bmbt_irec_t         *mval,
4271         int                     nmap,
4272         int                     ret_nmap)
4273 {
4274         int                     i;              /* index to map values */
4275
4276         ASSERT(ret_nmap <= nmap);
4277
4278         for (i = 0; i < ret_nmap; i++) {
4279                 ASSERT(mval[i].br_blockcount > 0);
4280                 if (!(flags & XFS_BMAPI_ENTIRE)) {
4281                         ASSERT(mval[i].br_startoff >= bno);
4282                         ASSERT(mval[i].br_blockcount <= len);
4283                         ASSERT(mval[i].br_startoff + mval[i].br_blockcount <=
4284                                bno + len);
4285                 } else {
4286                         ASSERT(mval[i].br_startoff < bno + len);
4287                         ASSERT(mval[i].br_startoff + mval[i].br_blockcount >
4288                                bno);
4289                 }
4290                 ASSERT(i == 0 ||
4291                        mval[i - 1].br_startoff + mval[i - 1].br_blockcount ==
4292                        mval[i].br_startoff);
4293                 if ((flags & XFS_BMAPI_WRITE) && !(flags & XFS_BMAPI_DELAY))
4294                         ASSERT(mval[i].br_startblock != DELAYSTARTBLOCK &&
4295                                mval[i].br_startblock != HOLESTARTBLOCK);
4296                 ASSERT(mval[i].br_state == XFS_EXT_NORM ||
4297                        mval[i].br_state == XFS_EXT_UNWRITTEN);
4298         }
4299 }
4300 #endif /* DEBUG */
4301
4302
4303 /*
4304  * Map file blocks to filesystem blocks.
4305  * File range is given by the bno/len pair.
4306  * Adds blocks to file if a write ("flags & XFS_BMAPI_WRITE" set)
4307  * into a hole or past eof.
4308  * Only allocates blocks from a single allocation group,
4309  * to avoid locking problems.
4310  * The returned value in "firstblock" from the first call in a transaction
4311  * must be remembered and presented to subsequent calls in "firstblock".
4312  * An upper bound for the number of blocks to be allocated is supplied to
4313  * the first call in "total"; if no allocation group has that many free
4314  * blocks then the call will fail (return NULLFSBLOCK in "firstblock").
4315  */
4316 int                                     /* error */
4317 xfs_bmapi(
4318         xfs_trans_t     *tp,            /* transaction pointer */
4319         xfs_inode_t     *ip,            /* incore inode */
4320         xfs_fileoff_t   bno,            /* starting file offs. mapped */
4321         xfs_filblks_t   len,            /* length to map in file */
4322         int             flags,          /* XFS_BMAPI_... */
4323         xfs_fsblock_t   *firstblock,    /* first allocated block
4324                                            controls a.g. for allocs */
4325         xfs_extlen_t    total,          /* total blocks needed */
4326         xfs_bmbt_irec_t *mval,          /* output: map values */
4327         int             *nmap,          /* i/o: mval size/count */
4328         xfs_bmap_free_t *flist)         /* i/o: list extents to free */
4329 {
4330         xfs_fsblock_t   abno;           /* allocated block number */
4331         xfs_extlen_t    alen;           /* allocated extent length */
4332         xfs_fileoff_t   aoff;           /* allocated file offset */
4333         xfs_bmalloca_t  bma = { 0 };    /* args for xfs_bmap_alloc */
4334         xfs_btree_cur_t *cur;           /* bmap btree cursor */
4335         xfs_fileoff_t   end;            /* end of mapped file region */
4336         int             eof;            /* we've hit the end of extents */
4337         xfs_bmbt_rec_host_t *ep;        /* extent record pointer */
4338         int             error;          /* error return */
4339         xfs_bmbt_irec_t got;            /* current file extent record */
4340         xfs_ifork_t     *ifp;           /* inode fork pointer */
4341         xfs_extlen_t    indlen;         /* indirect blocks length */
4342         xfs_extnum_t    lastx;          /* last useful extent number */
4343         int             logflags;       /* flags for transaction logging */
4344         xfs_extlen_t    minleft;        /* min blocks left after allocation */
4345         xfs_extlen_t    minlen;         /* min allocation size */
4346         xfs_mount_t     *mp;            /* xfs mount structure */
4347         int             n;              /* current extent index */
4348         int             nallocs;        /* number of extents alloc'd */
4349         xfs_extnum_t    nextents;       /* number of extents in file */
4350         xfs_fileoff_t   obno;           /* old block number (offset) */
4351         xfs_bmbt_irec_t prev;           /* previous file extent record */
4352         int             tmp_logflags;   /* temp flags holder */
4353         int             whichfork;      /* data or attr fork */
4354         char            inhole;         /* current location is hole in file */
4355         char            wasdelay;       /* old extent was delayed */
4356         char            wr;             /* this is a write request */
4357         char            rt;             /* this is a realtime file */
4358 #ifdef DEBUG
4359         xfs_fileoff_t   orig_bno;       /* original block number value */
4360         int             orig_flags;     /* original flags arg value */
4361         xfs_filblks_t   orig_len;       /* original value of len arg */
4362         xfs_bmbt_irec_t *orig_mval;     /* original value of mval */
4363         int             orig_nmap;      /* original value of *nmap */
4364
4365         orig_bno = bno;
4366         orig_len = len;
4367         orig_flags = flags;
4368         orig_mval = mval;
4369         orig_nmap = *nmap;
4370 #endif
4371         ASSERT(*nmap >= 1);
4372         ASSERT(*nmap <= XFS_BMAP_MAX_NMAP || !(flags & XFS_BMAPI_WRITE));
4373         whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4374                 XFS_ATTR_FORK : XFS_DATA_FORK;
4375         mp = ip->i_mount;
4376         if (unlikely(XFS_TEST_ERROR(
4377             (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4378              XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
4379              XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL),
4380              mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4381                 XFS_ERROR_REPORT("xfs_bmapi", XFS_ERRLEVEL_LOW, mp);
4382                 return XFS_ERROR(EFSCORRUPTED);
4383         }
4384         if (XFS_FORCED_SHUTDOWN(mp))
4385                 return XFS_ERROR(EIO);
4386         rt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
4387         ifp = XFS_IFORK_PTR(ip, whichfork);
4388         ASSERT(ifp->if_ext_max ==
4389                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
4390         if ((wr = (flags & XFS_BMAPI_WRITE)) != 0)
4391                 XFS_STATS_INC(xs_blk_mapw);
4392         else
4393                 XFS_STATS_INC(xs_blk_mapr);
4394         /*
4395          * IGSTATE flag is used to combine extents which
4396          * differ only due to the state of the extents.
4397          * This technique is used from xfs_getbmap()
4398          * when the caller does not wish to see the
4399          * separation (which is the default).
4400          *
4401          * This technique is also used when writing a
4402          * buffer which has been partially written,
4403          * (usually by being flushed during a chunkread),
4404          * to ensure one write takes place. This also
4405          * prevents a change in the xfs inode extents at
4406          * this time, intentionally. This change occurs
4407          * on completion of the write operation, in
4408          * xfs_strat_comp(), where the xfs_bmapi() call
4409          * is transactioned, and the extents combined.
4410          */
4411         if ((flags & XFS_BMAPI_IGSTATE) && wr)  /* if writing unwritten space */
4412                 wr = 0;                         /* no allocations are allowed */
4413         ASSERT(wr || !(flags & XFS_BMAPI_DELAY));
4414         logflags = 0;
4415         nallocs = 0;
4416         cur = NULL;
4417         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
4418                 ASSERT(wr && tp);
4419                 if ((error = xfs_bmap_local_to_extents(tp, ip,
4420                                 firstblock, total, &logflags, whichfork)))
4421                         goto error0;
4422         }
4423         if (wr && *firstblock == NULLFSBLOCK) {
4424                 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE)
4425                         minleft = be16_to_cpu(ifp->if_broot->bb_level) + 1;
4426                 else
4427                         minleft = 1;
4428         } else
4429                 minleft = 0;
4430         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
4431             (error = xfs_iread_extents(tp, ip, whichfork)))
4432                 goto error0;
4433         ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
4434                 &prev);
4435         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4436         n = 0;
4437         end = bno + len;
4438         obno = bno;
4439         bma.ip = NULL;
4440
4441         while (bno < end && n < *nmap) {
4442                 /*
4443                  * Reading past eof, act as though there's a hole
4444                  * up to end.
4445                  */
4446                 if (eof && !wr)
4447                         got.br_startoff = end;
4448                 inhole = eof || got.br_startoff > bno;
4449                 wasdelay = wr && !inhole && !(flags & XFS_BMAPI_DELAY) &&
4450                         isnullstartblock(got.br_startblock);
4451                 /*
4452                  * First, deal with the hole before the allocated space
4453                  * that we found, if any.
4454                  */
4455                 if (wr && (inhole || wasdelay)) {
4456                         /*
4457                          * For the wasdelay case, we could also just
4458                          * allocate the stuff asked for in this bmap call
4459                          * but that wouldn't be as good.
4460                          */
4461                         if (wasdelay) {
4462                                 alen = (xfs_extlen_t)got.br_blockcount;
4463                                 aoff = got.br_startoff;
4464                                 if (lastx != NULLEXTNUM && lastx) {
4465                                         ep = xfs_iext_get_ext(ifp, lastx - 1);
4466                                         xfs_bmbt_get_all(ep, &prev);
4467                                 }
4468                         } else {
4469                                 alen = (xfs_extlen_t)
4470                                         XFS_FILBLKS_MIN(len, MAXEXTLEN);
4471                                 if (!eof)
4472                                         alen = (xfs_extlen_t)
4473                                                 XFS_FILBLKS_MIN(alen,
4474                                                         got.br_startoff - bno);
4475                                 aoff = bno;
4476                         }
4477                         minlen = (flags & XFS_BMAPI_CONTIG) ? alen : 1;
4478                         if (flags & XFS_BMAPI_DELAY) {
4479                                 xfs_extlen_t    extsz;
4480
4481                                 /* Figure out the extent size, adjust alen */
4482                                 extsz = xfs_get_extsz_hint(ip);
4483                                 if (extsz) {
4484                                         error = xfs_bmap_extsize_align(mp,
4485                                                         &got, &prev, extsz,
4486                                                         rt, eof,
4487                                                         flags&XFS_BMAPI_DELAY,
4488                                                         flags&XFS_BMAPI_CONVERT,
4489                                                         &aoff, &alen);
4490                                         ASSERT(!error);
4491                                 }
4492
4493                                 if (rt)
4494                                         extsz = alen / mp->m_sb.sb_rextsize;
4495
4496                                 /*
4497                                  * Make a transaction-less quota reservation for
4498                                  * delayed allocation blocks. This number gets
4499                                  * adjusted later.  We return if we haven't
4500                                  * allocated blocks already inside this loop.
4501                                  */
4502                                 error = xfs_trans_reserve_quota_nblks(
4503                                                 NULL, ip, (long)alen, 0,
4504                                                 rt ? XFS_QMOPT_RES_RTBLKS :
4505                                                      XFS_QMOPT_RES_REGBLKS);
4506                                 if (error) {
4507                                         if (n == 0) {
4508                                                 *nmap = 0;
4509                                                 ASSERT(cur == NULL);
4510                                                 return error;
4511                                         }
4512                                         break;
4513                                 }
4514
4515                                 /*
4516                                  * Split changing sb for alen and indlen since
4517                                  * they could be coming from different places.
4518                                  */
4519                                 indlen = (xfs_extlen_t)
4520                                         xfs_bmap_worst_indlen(ip, alen);
4521                                 ASSERT(indlen > 0);
4522
4523                                 if (rt) {
4524                                         error = xfs_mod_incore_sb(mp,
4525                                                         XFS_SBS_FREXTENTS,
4526                                                         -((int64_t)extsz), (flags &
4527                                                         XFS_BMAPI_RSVBLOCKS));
4528                                 } else {
4529                                         error = xfs_mod_incore_sb(mp,
4530                                                         XFS_SBS_FDBLOCKS,
4531                                                         -((int64_t)alen), (flags &
4532                                                         XFS_BMAPI_RSVBLOCKS));
4533                                 }
4534                                 if (!error) {
4535                                         error = xfs_mod_incore_sb(mp,
4536                                                         XFS_SBS_FDBLOCKS,
4537                                                         -((int64_t)indlen), (flags &
4538                                                         XFS_BMAPI_RSVBLOCKS));
4539                                         if (error && rt)
4540                                                 xfs_mod_incore_sb(mp,
4541                                                         XFS_SBS_FREXTENTS,
4542                                                         (int64_t)extsz, (flags &
4543                                                         XFS_BMAPI_RSVBLOCKS));
4544                                         else if (error)
4545                                                 xfs_mod_incore_sb(mp,
4546                                                         XFS_SBS_FDBLOCKS,
4547                                                         (int64_t)alen, (flags &
4548                                                         XFS_BMAPI_RSVBLOCKS));
4549                                 }
4550
4551                                 if (error) {
4552                                         if (XFS_IS_QUOTA_ON(mp))
4553                                                 /* unreserve the blocks now */
4554                                                 (void)
4555                                                 xfs_trans_unreserve_quota_nblks(
4556                                                         NULL, ip,
4557                                                         (long)alen, 0, rt ?
4558                                                         XFS_QMOPT_RES_RTBLKS :
4559                                                         XFS_QMOPT_RES_REGBLKS);
4560                                         break;
4561                                 }
4562
4563                                 ip->i_delayed_blks += alen;
4564                                 abno = nullstartblock(indlen);
4565                         } else {
4566                                 /*
4567                                  * If first time, allocate and fill in
4568                                  * once-only bma fields.
4569                                  */
4570                                 if (bma.ip == NULL) {
4571                                         bma.tp = tp;
4572                                         bma.ip = ip;
4573                                         bma.prevp = &prev;
4574                                         bma.gotp = &got;
4575                                         bma.total = total;
4576                                         bma.userdata = 0;
4577                                 }
4578                                 /* Indicate if this is the first user data
4579                                  * in the file, or just any user data.
4580                                  */
4581                                 if (!(flags & XFS_BMAPI_METADATA)) {
4582                                         bma.userdata = (aoff == 0) ?
4583                                                 XFS_ALLOC_INITIAL_USER_DATA :
4584                                                 XFS_ALLOC_USERDATA;
4585                                 }
4586                                 /*
4587                                  * Fill in changeable bma fields.
4588                                  */
4589                                 bma.eof = eof;
4590                                 bma.firstblock = *firstblock;
4591                                 bma.alen = alen;
4592                                 bma.off = aoff;
4593                                 bma.conv = !!(flags & XFS_BMAPI_CONVERT);
4594                                 bma.wasdel = wasdelay;
4595                                 bma.minlen = minlen;
4596                                 bma.low = flist->xbf_low;
4597                                 bma.minleft = minleft;
4598                                 /*
4599                                  * Only want to do the alignment at the
4600                                  * eof if it is userdata and allocation length
4601                                  * is larger than a stripe unit.
4602                                  */
4603                                 if (mp->m_dalign && alen >= mp->m_dalign &&
4604                                     (!(flags & XFS_BMAPI_METADATA)) &&
4605                                     (whichfork == XFS_DATA_FORK)) {
4606                                         if ((error = xfs_bmap_isaeof(ip, aoff,
4607                                                         whichfork, &bma.aeof)))
4608                                                 goto error0;
4609                                 } else
4610                                         bma.aeof = 0;
4611                                 /*
4612                                  * Call allocator.
4613                                  */
4614                                 if ((error = xfs_bmap_alloc(&bma)))
4615                                         goto error0;
4616                                 /*
4617                                  * Copy out result fields.
4618                                  */
4619                                 abno = bma.rval;
4620                                 if ((flist->xbf_low = bma.low))
4621                                         minleft = 0;
4622                                 alen = bma.alen;
4623                                 aoff = bma.off;
4624                                 ASSERT(*firstblock == NULLFSBLOCK ||
4625                                        XFS_FSB_TO_AGNO(mp, *firstblock) ==
4626                                        XFS_FSB_TO_AGNO(mp, bma.firstblock) ||
4627                                        (flist->xbf_low &&
4628                                         XFS_FSB_TO_AGNO(mp, *firstblock) <
4629                                         XFS_FSB_TO_AGNO(mp, bma.firstblock)));
4630                                 *firstblock = bma.firstblock;
4631                                 if (cur)
4632                                         cur->bc_private.b.firstblock =
4633                                                 *firstblock;
4634                                 if (abno == NULLFSBLOCK)
4635                                         break;
4636                                 if ((ifp->if_flags & XFS_IFBROOT) && !cur) {
4637                                         cur = xfs_bmbt_init_cursor(mp, tp,
4638                                                 ip, whichfork);
4639                                         cur->bc_private.b.firstblock =
4640                                                 *firstblock;
4641                                         cur->bc_private.b.flist = flist;
4642                                 }
4643                                 /*
4644                                  * Bump the number of extents we've allocated
4645                                  * in this call.
4646                                  */
4647                                 nallocs++;
4648                         }
4649                         if (cur)
4650                                 cur->bc_private.b.flags =
4651                                         wasdelay ? XFS_BTCUR_BPRV_WASDEL : 0;
4652                         got.br_startoff = aoff;
4653                         got.br_startblock = abno;
4654                         got.br_blockcount = alen;
4655                         got.br_state = XFS_EXT_NORM;    /* assume normal */
4656                         /*
4657                          * Determine state of extent, and the filesystem.
4658                          * A wasdelay extent has been initialized, so
4659                          * shouldn't be flagged as unwritten.
4660                          */
4661                         if (wr && xfs_sb_version_hasextflgbit(&mp->m_sb)) {
4662                                 if (!wasdelay && (flags & XFS_BMAPI_PREALLOC))
4663                                         got.br_state = XFS_EXT_UNWRITTEN;
4664                         }
4665                         error = xfs_bmap_add_extent(ip, lastx, &cur, &got,
4666                                 firstblock, flist, &tmp_logflags,
4667                                 whichfork, (flags & XFS_BMAPI_RSVBLOCKS));
4668                         logflags |= tmp_logflags;
4669                         if (error)
4670                                 goto error0;
4671                         lastx = ifp->if_lastex;
4672                         ep = xfs_iext_get_ext(ifp, lastx);
4673                         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4674                         xfs_bmbt_get_all(ep, &got);
4675                         ASSERT(got.br_startoff <= aoff);
4676                         ASSERT(got.br_startoff + got.br_blockcount >=
4677                                 aoff + alen);
4678 #ifdef DEBUG
4679                         if (flags & XFS_BMAPI_DELAY) {
4680                                 ASSERT(isnullstartblock(got.br_startblock));
4681                                 ASSERT(startblockval(got.br_startblock) > 0);
4682                         }
4683                         ASSERT(got.br_state == XFS_EXT_NORM ||
4684                                got.br_state == XFS_EXT_UNWRITTEN);
4685 #endif
4686                         /*
4687                          * Fall down into the found allocated space case.
4688                          */
4689                 } else if (inhole) {
4690                         /*
4691                          * Reading in a hole.
4692                          */
4693                         mval->br_startoff = bno;
4694                         mval->br_startblock = HOLESTARTBLOCK;
4695                         mval->br_blockcount =
4696                                 XFS_FILBLKS_MIN(len, got.br_startoff - bno);
4697                         mval->br_state = XFS_EXT_NORM;
4698                         bno += mval->br_blockcount;
4699                         len -= mval->br_blockcount;
4700                         mval++;
4701                         n++;
4702                         continue;
4703                 }
4704                 /*
4705                  * Then deal with the allocated space we found.
4706                  */
4707                 ASSERT(ep != NULL);
4708                 if (!(flags & XFS_BMAPI_ENTIRE) &&
4709                     (got.br_startoff + got.br_blockcount > obno)) {
4710                         if (obno > bno)
4711                                 bno = obno;
4712                         ASSERT((bno >= obno) || (n == 0));
4713                         ASSERT(bno < end);
4714                         mval->br_startoff = bno;
4715                         if (isnullstartblock(got.br_startblock)) {
4716                                 ASSERT(!wr || (flags & XFS_BMAPI_DELAY));
4717                                 mval->br_startblock = DELAYSTARTBLOCK;
4718                         } else
4719                                 mval->br_startblock =
4720                                         got.br_startblock +
4721                                         (bno - got.br_startoff);
4722                         /*
4723                          * Return the minimum of what we got and what we
4724                          * asked for for the length.  We can use the len
4725                          * variable here because it is modified below
4726                          * and we could have been there before coming
4727                          * here if the first part of the allocation
4728                          * didn't overlap what was asked for.
4729                          */
4730                         mval->br_blockcount =
4731                                 XFS_FILBLKS_MIN(end - bno, got.br_blockcount -
4732                                         (bno - got.br_startoff));
4733                         mval->br_state = got.br_state;
4734                         ASSERT(mval->br_blockcount <= len);
4735                 } else {
4736                         *mval = got;
4737                         if (isnullstartblock(mval->br_startblock)) {
4738                                 ASSERT(!wr || (flags & XFS_BMAPI_DELAY));
4739                                 mval->br_startblock = DELAYSTARTBLOCK;
4740                         }
4741                 }
4742
4743                 /*
4744                  * Check if writing previously allocated but
4745                  * unwritten extents.
4746                  */
4747                 if (wr && mval->br_state == XFS_EXT_UNWRITTEN &&
4748                     ((flags & (XFS_BMAPI_PREALLOC|XFS_BMAPI_DELAY)) == 0)) {
4749                         /*
4750                          * Modify (by adding) the state flag, if writing.
4751                          */
4752                         ASSERT(mval->br_blockcount <= len);
4753                         if ((ifp->if_flags & XFS_IFBROOT) && !cur) {
4754                                 cur = xfs_bmbt_init_cursor(mp,
4755                                         tp, ip, whichfork);
4756                                 cur->bc_private.b.firstblock =
4757                                         *firstblock;
4758                                 cur->bc_private.b.flist = flist;
4759                         }
4760                         mval->br_state = XFS_EXT_NORM;
4761                         error = xfs_bmap_add_extent(ip, lastx, &cur, mval,
4762                                 firstblock, flist, &tmp_logflags,
4763                                 whichfork, (flags & XFS_BMAPI_RSVBLOCKS));
4764                         logflags |= tmp_logflags;
4765                         if (error)
4766                                 goto error0;
4767                         lastx = ifp->if_lastex;
4768                         ep = xfs_iext_get_ext(ifp, lastx);
4769                         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4770                         xfs_bmbt_get_all(ep, &got);
4771                         /*
4772                          * We may have combined previously unwritten
4773                          * space with written space, so generate
4774                          * another request.
4775                          */
4776                         if (mval->br_blockcount < len)
4777                                 continue;
4778                 }
4779
4780                 ASSERT((flags & XFS_BMAPI_ENTIRE) ||
4781                        ((mval->br_startoff + mval->br_blockcount) <= end));
4782                 ASSERT((flags & XFS_BMAPI_ENTIRE) ||
4783                        (mval->br_blockcount <= len) ||
4784                        (mval->br_startoff < obno));
4785                 bno = mval->br_startoff + mval->br_blockcount;
4786                 len = end - bno;
4787                 if (n > 0 && mval->br_startoff == mval[-1].br_startoff) {
4788                         ASSERT(mval->br_startblock == mval[-1].br_startblock);
4789                         ASSERT(mval->br_blockcount > mval[-1].br_blockcount);
4790                         ASSERT(mval->br_state == mval[-1].br_state);
4791                         mval[-1].br_blockcount = mval->br_blockcount;
4792                         mval[-1].br_state = mval->br_state;
4793                 } else if (n > 0 && mval->br_startblock != DELAYSTARTBLOCK &&
4794                            mval[-1].br_startblock != DELAYSTARTBLOCK &&
4795                            mval[-1].br_startblock != HOLESTARTBLOCK &&
4796                            mval->br_startblock ==
4797                            mval[-1].br_startblock + mval[-1].br_blockcount &&
4798                            ((flags & XFS_BMAPI_IGSTATE) ||
4799                                 mval[-1].br_state == mval->br_state)) {
4800                         ASSERT(mval->br_startoff ==
4801                                mval[-1].br_startoff + mval[-1].br_blockcount);
4802                         mval[-1].br_blockcount += mval->br_blockcount;
4803                 } else if (n > 0 &&
4804                            mval->br_startblock == DELAYSTARTBLOCK &&
4805                            mval[-1].br_startblock == DELAYSTARTBLOCK &&
4806                            mval->br_startoff ==
4807                            mval[-1].br_startoff + mval[-1].br_blockcount) {
4808                         mval[-1].br_blockcount += mval->br_blockcount;
4809                         mval[-1].br_state = mval->br_state;
4810                 } else if (!((n == 0) &&
4811                              ((mval->br_startoff + mval->br_blockcount) <=
4812                               obno))) {
4813                         mval++;
4814                         n++;
4815                 }
4816                 /*
4817                  * If we're done, stop now.  Stop when we've allocated
4818                  * XFS_BMAP_MAX_NMAP extents no matter what.  Otherwise
4819                  * the transaction may get too big.
4820                  */
4821                 if (bno >= end || n >= *nmap || nallocs >= *nmap)
4822                         break;
4823                 /*
4824                  * Else go on to the next record.
4825                  */
4826                 ep = xfs_iext_get_ext(ifp, ++lastx);
4827                 prev = got;
4828                 if (lastx >= nextents)
4829                         eof = 1;
4830                 else
4831                         xfs_bmbt_get_all(ep, &got);
4832         }
4833         ifp->if_lastex = lastx;
4834         *nmap = n;
4835         /*
4836          * Transform from btree to extents, give it cur.
4837          */
4838         if (tp && XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE &&
4839             XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max) {
4840                 ASSERT(wr && cur);
4841                 error = xfs_bmap_btree_to_extents(tp, ip, cur,
4842                         &tmp_logflags, whichfork);
4843                 logflags |= tmp_logflags;
4844                 if (error)
4845                         goto error0;
4846         }
4847         ASSERT(ifp->if_ext_max ==
4848                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
4849         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE ||
4850                XFS_IFORK_NEXTENTS(ip, whichfork) > ifp->if_ext_max);
4851         error = 0;
4852 error0:
4853         /*
4854          * Log everything.  Do this after conversion, there's no point in
4855          * logging the extent records if we've converted to btree format.
4856          */
4857         if ((logflags & xfs_ilog_fext(whichfork)) &&
4858             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
4859                 logflags &= ~xfs_ilog_fext(whichfork);
4860         else if ((logflags & xfs_ilog_fbroot(whichfork)) &&
4861                  XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
4862                 logflags &= ~xfs_ilog_fbroot(whichfork);
4863         /*
4864          * Log whatever the flags say, even if error.  Otherwise we might miss
4865          * detecting a case where the data is changed, there's an error,
4866          * and it's not logged so we don't shutdown when we should.
4867          */
4868         if (logflags) {
4869                 ASSERT(tp && wr);
4870                 xfs_trans_log_inode(tp, ip, logflags);
4871         }
4872         if (cur) {
4873                 if (!error) {
4874                         ASSERT(*firstblock == NULLFSBLOCK ||
4875                                XFS_FSB_TO_AGNO(mp, *firstblock) ==
4876                                XFS_FSB_TO_AGNO(mp,
4877                                        cur->bc_private.b.firstblock) ||
4878                                (flist->xbf_low &&
4879                                 XFS_FSB_TO_AGNO(mp, *firstblock) <
4880                                 XFS_FSB_TO_AGNO(mp,
4881                                         cur->bc_private.b.firstblock)));
4882                         *firstblock = cur->bc_private.b.firstblock;
4883                 }
4884                 xfs_btree_del_cursor(cur,
4885                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
4886         }
4887         if (!error)
4888                 xfs_bmap_validate_ret(orig_bno, orig_len, orig_flags, orig_mval,
4889                         orig_nmap, *nmap);
4890         return error;
4891 }
4892
4893 /*
4894  * Map file blocks to filesystem blocks, simple version.
4895  * One block (extent) only, read-only.
4896  * For flags, only the XFS_BMAPI_ATTRFORK flag is examined.
4897  * For the other flag values, the effect is as if XFS_BMAPI_METADATA
4898  * was set and all the others were clear.
4899  */
4900 int                                             /* error */
4901 xfs_bmapi_single(
4902         xfs_trans_t     *tp,            /* transaction pointer */
4903         xfs_inode_t     *ip,            /* incore inode */
4904         int             whichfork,      /* data or attr fork */
4905         xfs_fsblock_t   *fsb,           /* output: mapped block */
4906         xfs_fileoff_t   bno)            /* starting file offs. mapped */
4907 {
4908         int             eof;            /* we've hit the end of extents */
4909         int             error;          /* error return */
4910         xfs_bmbt_irec_t got;            /* current file extent record */
4911         xfs_ifork_t     *ifp;           /* inode fork pointer */
4912         xfs_extnum_t    lastx;          /* last useful extent number */
4913         xfs_bmbt_irec_t prev;           /* previous file extent record */
4914
4915         ifp = XFS_IFORK_PTR(ip, whichfork);
4916         if (unlikely(
4917             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
4918             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)) {
4919                XFS_ERROR_REPORT("xfs_bmapi_single", XFS_ERRLEVEL_LOW,
4920                                 ip->i_mount);
4921                return XFS_ERROR(EFSCORRUPTED);
4922         }
4923         if (XFS_FORCED_SHUTDOWN(ip->i_mount))
4924                 return XFS_ERROR(EIO);
4925         XFS_STATS_INC(xs_blk_mapr);
4926         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
4927             (error = xfs_iread_extents(tp, ip, whichfork)))
4928                 return error;
4929         (void)xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
4930                 &prev);
4931         /*
4932          * Reading past eof, act as though there's a hole
4933          * up to end.
4934          */
4935         if (eof || got.br_startoff > bno) {
4936                 *fsb = NULLFSBLOCK;
4937                 return 0;
4938         }
4939         ASSERT(!isnullstartblock(got.br_startblock));
4940         ASSERT(bno < got.br_startoff + got.br_blockcount);
4941         *fsb = got.br_startblock + (bno - got.br_startoff);
4942         ifp->if_lastex = lastx;
4943         return 0;
4944 }
4945
4946 /*
4947  * Unmap (remove) blocks from a file.
4948  * If nexts is nonzero then the number of extents to remove is limited to
4949  * that value.  If not all extents in the block range can be removed then
4950  * *done is set.
4951  */
4952 int                                             /* error */
4953 xfs_bunmapi(
4954         xfs_trans_t             *tp,            /* transaction pointer */
4955         struct xfs_inode        *ip,            /* incore inode */
4956         xfs_fileoff_t           bno,            /* starting offset to unmap */
4957         xfs_filblks_t           len,            /* length to unmap in file */
4958         int                     flags,          /* misc flags */
4959         xfs_extnum_t            nexts,          /* number of extents max */
4960         xfs_fsblock_t           *firstblock,    /* first allocated block
4961                                                    controls a.g. for allocs */
4962         xfs_bmap_free_t         *flist,         /* i/o: list extents to free */
4963         int                     *done)          /* set if not done yet */
4964 {
4965         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
4966         xfs_bmbt_irec_t         del;            /* extent being deleted */
4967         int                     eof;            /* is deleting at eof */
4968         xfs_bmbt_rec_host_t     *ep;            /* extent record pointer */
4969         int                     error;          /* error return value */
4970         xfs_extnum_t            extno;          /* extent number in list */
4971         xfs_bmbt_irec_t         got;            /* current extent record */
4972         xfs_ifork_t             *ifp;           /* inode fork pointer */
4973         int                     isrt;           /* freeing in rt area */
4974         xfs_extnum_t            lastx;          /* last extent index used */
4975         int                     logflags;       /* transaction logging flags */
4976         xfs_extlen_t            mod;            /* rt extent offset */
4977         xfs_mount_t             *mp;            /* mount structure */
4978         xfs_extnum_t            nextents;       /* number of file extents */
4979         xfs_bmbt_irec_t         prev;           /* previous extent record */
4980         xfs_fileoff_t           start;          /* first file offset deleted */
4981         int                     tmp_logflags;   /* partial logging flags */
4982         int                     wasdel;         /* was a delayed alloc extent */
4983         int                     whichfork;      /* data or attribute fork */
4984         int                     rsvd;           /* OK to allocate reserved blocks */
4985         xfs_fsblock_t           sum;
4986
4987         trace_xfs_bunmap(ip, bno, len, flags, _RET_IP_);
4988
4989         whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4990                 XFS_ATTR_FORK : XFS_DATA_FORK;
4991         ifp = XFS_IFORK_PTR(ip, whichfork);
4992         if (unlikely(
4993             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4994             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
4995                 XFS_ERROR_REPORT("xfs_bunmapi", XFS_ERRLEVEL_LOW,
4996                                  ip->i_mount);
4997                 return XFS_ERROR(EFSCORRUPTED);
4998         }
4999         mp = ip->i_mount;
5000         if (XFS_FORCED_SHUTDOWN(mp))
5001                 return XFS_ERROR(EIO);
5002         rsvd = (flags & XFS_BMAPI_RSVBLOCKS) != 0;
5003         ASSERT(len > 0);
5004         ASSERT(nexts >= 0);
5005         ASSERT(ifp->if_ext_max ==
5006                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5007         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5008             (error = xfs_iread_extents(tp, ip, whichfork)))
5009                 return error;
5010         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5011         if (nextents == 0) {
5012                 *done = 1;
5013                 return 0;
5014         }
5015         XFS_STATS_INC(xs_blk_unmap);
5016         isrt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
5017         start = bno;
5018         bno = start + len - 1;
5019         ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
5020                 &prev);
5021
5022         /*
5023          * Check to see if the given block number is past the end of the
5024          * file, back up to the last block if so...
5025          */
5026         if (eof) {
5027                 ep = xfs_iext_get_ext(ifp, --lastx);
5028                 xfs_bmbt_get_all(ep, &got);
5029                 bno = got.br_startoff + got.br_blockcount - 1;
5030         }
5031         logflags = 0;
5032         if (ifp->if_flags & XFS_IFBROOT) {
5033                 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
5034                 cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5035                 cur->bc_private.b.firstblock = *firstblock;
5036                 cur->bc_private.b.flist = flist;
5037                 cur->bc_private.b.flags = 0;
5038         } else
5039                 cur = NULL;
5040         extno = 0;
5041         while (bno != (xfs_fileoff_t)-1 && bno >= start && lastx >= 0 &&
5042                (nexts == 0 || extno < nexts)) {
5043                 /*
5044                  * Is the found extent after a hole in which bno lives?
5045                  * Just back up to the previous extent, if so.
5046                  */
5047                 if (got.br_startoff > bno) {
5048                         if (--lastx < 0)
5049                                 break;
5050                         ep = xfs_iext_get_ext(ifp, lastx);
5051                         xfs_bmbt_get_all(ep, &got);
5052                 }
5053                 /*
5054                  * Is the last block of this extent before the range
5055                  * we're supposed to delete?  If so, we're done.
5056                  */
5057                 bno = XFS_FILEOFF_MIN(bno,
5058                         got.br_startoff + got.br_blockcount - 1);
5059                 if (bno < start)
5060                         break;
5061                 /*
5062                  * Then deal with the (possibly delayed) allocated space
5063                  * we found.
5064                  */
5065                 ASSERT(ep != NULL);
5066                 del = got;
5067                 wasdel = isnullstartblock(del.br_startblock);
5068                 if (got.br_startoff < start) {
5069                         del.br_startoff = start;
5070                         del.br_blockcount -= start - got.br_startoff;
5071                         if (!wasdel)
5072                                 del.br_startblock += start - got.br_startoff;
5073                 }
5074                 if (del.br_startoff + del.br_blockcount > bno + 1)
5075                         del.br_blockcount = bno + 1 - del.br_startoff;
5076                 sum = del.br_startblock + del.br_blockcount;
5077                 if (isrt &&
5078                     (mod = do_mod(sum, mp->m_sb.sb_rextsize))) {
5079                         /*
5080                          * Realtime extent not lined up at the end.
5081                          * The extent could have been split into written
5082                          * and unwritten pieces, or we could just be
5083                          * unmapping part of it.  But we can't really
5084                          * get rid of part of a realtime extent.
5085                          */
5086                         if (del.br_state == XFS_EXT_UNWRITTEN ||
5087                             !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5088                                 /*
5089                                  * This piece is unwritten, or we're not
5090                                  * using unwritten extents.  Skip over it.
5091                                  */
5092                                 ASSERT(bno >= mod);
5093                                 bno -= mod > del.br_blockcount ?
5094                                         del.br_blockcount : mod;
5095                                 if (bno < got.br_startoff) {
5096                                         if (--lastx >= 0)
5097                                                 xfs_bmbt_get_all(xfs_iext_get_ext(
5098                                                         ifp, lastx), &got);
5099                                 }
5100                                 continue;
5101                         }
5102                         /*
5103                          * It's written, turn it unwritten.
5104                          * This is better than zeroing it.
5105                          */
5106                         ASSERT(del.br_state == XFS_EXT_NORM);
5107                         ASSERT(xfs_trans_get_block_res(tp) > 0);
5108                         /*
5109                          * If this spans a realtime extent boundary,
5110                          * chop it back to the start of the one we end at.
5111                          */
5112                         if (del.br_blockcount > mod) {
5113                                 del.br_startoff += del.br_blockcount - mod;
5114                                 del.br_startblock += del.br_blockcount - mod;
5115                                 del.br_blockcount = mod;
5116                         }
5117                         del.br_state = XFS_EXT_UNWRITTEN;
5118                         error = xfs_bmap_add_extent(ip, lastx, &cur, &del,
5119                                 firstblock, flist, &logflags,
5120                                 XFS_DATA_FORK, 0);
5121                         if (error)
5122                                 goto error0;
5123                         goto nodelete;
5124                 }
5125                 if (isrt && (mod = do_mod(del.br_startblock, mp->m_sb.sb_rextsize))) {
5126                         /*
5127                          * Realtime extent is lined up at the end but not
5128                          * at the front.  We'll get rid of full extents if
5129                          * we can.
5130                          */
5131                         mod = mp->m_sb.sb_rextsize - mod;
5132                         if (del.br_blockcount > mod) {
5133                                 del.br_blockcount -= mod;
5134                                 del.br_startoff += mod;
5135                                 del.br_startblock += mod;
5136                         } else if ((del.br_startoff == start &&
5137                                     (del.br_state == XFS_EXT_UNWRITTEN ||
5138                                      xfs_trans_get_block_res(tp) == 0)) ||
5139                                    !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5140                                 /*
5141                                  * Can't make it unwritten.  There isn't
5142                                  * a full extent here so just skip it.
5143                                  */
5144                                 ASSERT(bno >= del.br_blockcount);
5145                                 bno -= del.br_blockcount;
5146                                 if (bno < got.br_startoff) {
5147                                         if (--lastx >= 0)
5148                                                 xfs_bmbt_get_all(--ep, &got);
5149                                 }
5150                                 continue;
5151                         } else if (del.br_state == XFS_EXT_UNWRITTEN) {
5152                                 /*
5153                                  * This one is already unwritten.
5154                                  * It must have a written left neighbor.
5155                                  * Unwrite the killed part of that one and
5156                                  * try again.
5157                                  */
5158                                 ASSERT(lastx > 0);
5159                                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp,
5160                                                 lastx - 1), &prev);
5161                                 ASSERT(prev.br_state == XFS_EXT_NORM);
5162                                 ASSERT(!isnullstartblock(prev.br_startblock));
5163                                 ASSERT(del.br_startblock ==
5164                                        prev.br_startblock + prev.br_blockcount);
5165                                 if (prev.br_startoff < start) {
5166                                         mod = start - prev.br_startoff;
5167                                         prev.br_blockcount -= mod;
5168                                         prev.br_startblock += mod;
5169                                         prev.br_startoff = start;
5170                                 }
5171                                 prev.br_state = XFS_EXT_UNWRITTEN;
5172                                 error = xfs_bmap_add_extent(ip, lastx - 1, &cur,
5173                                         &prev, firstblock, flist, &logflags,
5174                                         XFS_DATA_FORK, 0);
5175                                 if (error)
5176                                         goto error0;
5177                                 goto nodelete;
5178                         } else {
5179                                 ASSERT(del.br_state == XFS_EXT_NORM);
5180                                 del.br_state = XFS_EXT_UNWRITTEN;
5181                                 error = xfs_bmap_add_extent(ip, lastx, &cur,
5182                                         &del, firstblock, flist, &logflags,
5183                                         XFS_DATA_FORK, 0);
5184                                 if (error)
5185                                         goto error0;
5186                                 goto nodelete;
5187                         }
5188                 }
5189                 if (wasdel) {
5190                         ASSERT(startblockval(del.br_startblock) > 0);
5191                         /* Update realtime/data freespace, unreserve quota */
5192                         if (isrt) {
5193                                 xfs_filblks_t rtexts;
5194
5195                                 rtexts = XFS_FSB_TO_B(mp, del.br_blockcount);
5196                                 do_div(rtexts, mp->m_sb.sb_rextsize);
5197                                 xfs_mod_incore_sb(mp, XFS_SBS_FREXTENTS,
5198                                                 (int64_t)rtexts, rsvd);
5199                                 (void)xfs_trans_reserve_quota_nblks(NULL,
5200                                         ip, -((long)del.br_blockcount), 0,
5201                                         XFS_QMOPT_RES_RTBLKS);
5202                         } else {
5203                                 xfs_mod_incore_sb(mp, XFS_SBS_FDBLOCKS,
5204                                                 (int64_t)del.br_blockcount, rsvd);
5205                                 (void)xfs_trans_reserve_quota_nblks(NULL,
5206                                         ip, -((long)del.br_blockcount), 0,
5207                                         XFS_QMOPT_RES_REGBLKS);
5208                         }
5209                         ip->i_delayed_blks -= del.br_blockcount;
5210                         if (cur)
5211                                 cur->bc_private.b.flags |=
5212                                         XFS_BTCUR_BPRV_WASDEL;
5213                 } else if (cur)
5214                         cur->bc_private.b.flags &= ~XFS_BTCUR_BPRV_WASDEL;
5215                 /*
5216                  * If it's the case where the directory code is running
5217                  * with no block reservation, and the deleted block is in
5218                  * the middle of its extent, and the resulting insert
5219                  * of an extent would cause transformation to btree format,
5220                  * then reject it.  The calling code will then swap
5221                  * blocks around instead.
5222                  * We have to do this now, rather than waiting for the
5223                  * conversion to btree format, since the transaction
5224                  * will be dirty.
5225                  */
5226                 if (!wasdel && xfs_trans_get_block_res(tp) == 0 &&
5227                     XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
5228                     XFS_IFORK_NEXTENTS(ip, whichfork) >= ifp->if_ext_max &&
5229                     del.br_startoff > got.br_startoff &&
5230                     del.br_startoff + del.br_blockcount <
5231                     got.br_startoff + got.br_blockcount) {
5232                         error = XFS_ERROR(ENOSPC);
5233                         goto error0;
5234                 }
5235                 error = xfs_bmap_del_extent(ip, tp, lastx, flist, cur, &del,
5236                                 &tmp_logflags, whichfork, rsvd);
5237                 logflags |= tmp_logflags;
5238                 if (error)
5239                         goto error0;
5240                 bno = del.br_startoff - 1;
5241 nodelete:
5242                 lastx = ifp->if_lastex;
5243                 /*
5244                  * If not done go on to the next (previous) record.
5245                  * Reset ep in case the extents array was re-alloced.
5246                  */
5247                 ep = xfs_iext_get_ext(ifp, lastx);
5248                 if (bno != (xfs_fileoff_t)-1 && bno >= start) {
5249                         if (lastx >= XFS_IFORK_NEXTENTS(ip, whichfork) ||
5250                             xfs_bmbt_get_startoff(ep) > bno) {
5251                                 if (--lastx >= 0)
5252                                         ep = xfs_iext_get_ext(ifp, lastx);
5253                         }
5254                         if (lastx >= 0)
5255                                 xfs_bmbt_get_all(ep, &got);
5256                         extno++;
5257                 }
5258         }
5259         ifp->if_lastex = lastx;
5260         *done = bno == (xfs_fileoff_t)-1 || bno < start || lastx < 0;
5261         ASSERT(ifp->if_ext_max ==
5262                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5263         /*
5264          * Convert to a btree if necessary.
5265          */
5266         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
5267             XFS_IFORK_NEXTENTS(ip, whichfork) > ifp->if_ext_max) {
5268                 ASSERT(cur == NULL);
5269                 error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist,
5270                         &cur, 0, &tmp_logflags, whichfork);
5271                 logflags |= tmp_logflags;
5272                 if (error)
5273                         goto error0;
5274         }
5275         /*
5276          * transform from btree to extents, give it cur
5277          */
5278         else if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE &&
5279                  XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max) {
5280                 ASSERT(cur != NULL);
5281                 error = xfs_bmap_btree_to_extents(tp, ip, cur, &tmp_logflags,
5282                         whichfork);
5283                 logflags |= tmp_logflags;
5284                 if (error)
5285                         goto error0;
5286         }
5287         /*
5288          * transform from extents to local?
5289          */
5290         ASSERT(ifp->if_ext_max ==
5291                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5292         error = 0;
5293 error0:
5294         /*
5295          * Log everything.  Do this after conversion, there's no point in
5296          * logging the extent records if we've converted to btree format.
5297          */
5298         if ((logflags & xfs_ilog_fext(whichfork)) &&
5299             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
5300                 logflags &= ~xfs_ilog_fext(whichfork);
5301         else if ((logflags & xfs_ilog_fbroot(whichfork)) &&
5302                  XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
5303                 logflags &= ~xfs_ilog_fbroot(whichfork);
5304         /*
5305          * Log inode even in the error case, if the transaction
5306          * is dirty we'll need to shut down the filesystem.
5307          */
5308         if (logflags)
5309                 xfs_trans_log_inode(tp, ip, logflags);
5310         if (cur) {
5311                 if (!error) {
5312                         *firstblock = cur->bc_private.b.firstblock;
5313                         cur->bc_private.b.allocated = 0;
5314                 }
5315                 xfs_btree_del_cursor(cur,
5316                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5317         }
5318         return error;
5319 }
5320
5321 /*
5322  * returns 1 for success, 0 if we failed to map the extent.
5323  */
5324 STATIC int
5325 xfs_getbmapx_fix_eof_hole(
5326         xfs_inode_t             *ip,            /* xfs incore inode pointer */
5327         struct getbmapx         *out,           /* output structure */
5328         int                     prealloced,     /* this is a file with
5329                                                  * preallocated data space */
5330         __int64_t               end,            /* last block requested */
5331         xfs_fsblock_t           startblock)
5332 {
5333         __int64_t               fixlen;
5334         xfs_mount_t             *mp;            /* file system mount point */
5335         xfs_ifork_t             *ifp;           /* inode fork pointer */
5336         xfs_extnum_t            lastx;          /* last extent pointer */
5337         xfs_fileoff_t           fileblock;
5338
5339         if (startblock == HOLESTARTBLOCK) {
5340                 mp = ip->i_mount;
5341                 out->bmv_block = -1;
5342                 fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, ip->i_size));
5343                 fixlen -= out->bmv_offset;
5344                 if (prealloced && out->bmv_offset + out->bmv_length == end) {
5345                         /* Came to hole at EOF. Trim it. */
5346                         if (fixlen <= 0)
5347                                 return 0;
5348                         out->bmv_length = fixlen;
5349                 }
5350         } else {
5351                 if (startblock == DELAYSTARTBLOCK)
5352                         out->bmv_block = -2;
5353                 else
5354                         out->bmv_block = xfs_fsb_to_db(ip, startblock);
5355                 fileblock = XFS_BB_TO_FSB(ip->i_mount, out->bmv_offset);
5356                 ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
5357                 if (xfs_iext_bno_to_ext(ifp, fileblock, &lastx) &&
5358                    (lastx == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t))-1))
5359                         out->bmv_oflags |= BMV_OF_LAST;
5360         }
5361
5362         return 1;
5363 }
5364
5365 /*
5366  * Get inode's extents as described in bmv, and format for output.
5367  * Calls formatter to fill the user's buffer until all extents
5368  * are mapped, until the passed-in bmv->bmv_count slots have
5369  * been filled, or until the formatter short-circuits the loop,
5370  * if it is tracking filled-in extents on its own.
5371  */
5372 int                                             /* error code */
5373 xfs_getbmap(
5374         xfs_inode_t             *ip,
5375         struct getbmapx         *bmv,           /* user bmap structure */
5376         xfs_bmap_format_t       formatter,      /* format to user */
5377         void                    *arg)           /* formatter arg */
5378 {
5379         __int64_t               bmvend;         /* last block requested */
5380         int                     error = 0;      /* return value */
5381         __int64_t               fixlen;         /* length for -1 case */
5382         int                     i;              /* extent number */
5383         int                     lock;           /* lock state */
5384         xfs_bmbt_irec_t         *map;           /* buffer for user's data */
5385         xfs_mount_t             *mp;            /* file system mount point */
5386         int                     nex;            /* # of user extents can do */
5387         int                     nexleft;        /* # of user extents left */
5388         int                     subnex;         /* # of bmapi's can do */
5389         int                     nmap;           /* number of map entries */
5390         struct getbmapx         *out;           /* output structure */
5391         int                     whichfork;      /* data or attr fork */
5392         int                     prealloced;     /* this is a file with
5393                                                  * preallocated data space */
5394         int                     iflags;         /* interface flags */
5395         int                     bmapi_flags;    /* flags for xfs_bmapi */
5396         int                     cur_ext = 0;
5397
5398         mp = ip->i_mount;
5399         iflags = bmv->bmv_iflags;
5400         whichfork = iflags & BMV_IF_ATTRFORK ? XFS_ATTR_FORK : XFS_DATA_FORK;
5401
5402         if (whichfork == XFS_ATTR_FORK) {
5403                 if (XFS_IFORK_Q(ip)) {
5404                         if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS &&
5405                             ip->i_d.di_aformat != XFS_DINODE_FMT_BTREE &&
5406                             ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)
5407                                 return XFS_ERROR(EINVAL);
5408                 } else if (unlikely(
5409                            ip->i_d.di_aformat != 0 &&
5410                            ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS)) {
5411                         XFS_ERROR_REPORT("xfs_getbmap", XFS_ERRLEVEL_LOW,
5412                                          ip->i_mount);
5413                         return XFS_ERROR(EFSCORRUPTED);
5414                 }
5415
5416                 prealloced = 0;
5417                 fixlen = 1LL << 32;
5418         } else {
5419                 if (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS &&
5420                     ip->i_d.di_format != XFS_DINODE_FMT_BTREE &&
5421                     ip->i_d.di_format != XFS_DINODE_FMT_LOCAL)
5422                         return XFS_ERROR(EINVAL);
5423
5424                 if (xfs_get_extsz_hint(ip) ||
5425                     ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC|XFS_DIFLAG_APPEND)){
5426                         prealloced = 1;
5427                         fixlen = XFS_MAXIOFFSET(mp);
5428                 } else {
5429                         prealloced = 0;
5430                         fixlen = ip->i_size;
5431                 }
5432         }
5433
5434         if (bmv->bmv_length == -1) {
5435                 fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, fixlen));
5436                 bmv->bmv_length =
5437                         max_t(__int64_t, fixlen - bmv->bmv_offset, 0);
5438         } else if (bmv->bmv_length == 0) {
5439                 bmv->bmv_entries = 0;
5440                 return 0;
5441         } else if (bmv->bmv_length < 0) {
5442                 return XFS_ERROR(EINVAL);
5443         }
5444
5445         nex = bmv->bmv_count - 1;
5446         if (nex <= 0)
5447                 return XFS_ERROR(EINVAL);
5448         bmvend = bmv->bmv_offset + bmv->bmv_length;
5449
5450
5451         if (bmv->bmv_count > ULONG_MAX / sizeof(struct getbmapx))
5452                 return XFS_ERROR(ENOMEM);
5453         out = kmem_zalloc(bmv->bmv_count * sizeof(struct getbmapx), KM_MAYFAIL);
5454         if (!out)
5455                 return XFS_ERROR(ENOMEM);
5456
5457         xfs_ilock(ip, XFS_IOLOCK_SHARED);
5458         if (whichfork == XFS_DATA_FORK && !(iflags & BMV_IF_DELALLOC)) {
5459                 if (ip->i_delayed_blks || ip->i_size > ip->i_d.di_size) {
5460                         error = xfs_flush_pages(ip, 0, -1, 0, FI_REMAPF);
5461                         if (error)
5462                                 goto out_unlock_iolock;
5463                 }
5464
5465                 ASSERT(ip->i_delayed_blks == 0);
5466         }
5467
5468         lock = xfs_ilock_map_shared(ip);
5469
5470         /*
5471          * Don't let nex be bigger than the number of extents
5472          * we can have assuming alternating holes and real extents.
5473          */
5474         if (nex > XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1)
5475                 nex = XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1;
5476
5477         bmapi_flags = xfs_bmapi_aflag(whichfork);
5478         if (!(iflags & BMV_IF_PREALLOC))
5479                 bmapi_flags |= XFS_BMAPI_IGSTATE;
5480
5481         /*
5482          * Allocate enough space to handle "subnex" maps at a time.
5483          */
5484         error = ENOMEM;
5485         subnex = 16;
5486         map = kmem_alloc(subnex * sizeof(*map), KM_MAYFAIL | KM_NOFS);
5487         if (!map)
5488                 goto out_unlock_ilock;
5489
5490         bmv->bmv_entries = 0;
5491
5492         if (XFS_IFORK_NEXTENTS(ip, whichfork) == 0 &&
5493             (whichfork == XFS_ATTR_FORK || !(iflags & BMV_IF_DELALLOC))) {
5494                 error = 0;
5495                 goto out_free_map;
5496         }
5497
5498         nexleft = nex;
5499
5500         do {
5501                 nmap = (nexleft > subnex) ? subnex : nexleft;
5502                 error = xfs_bmapi(NULL, ip, XFS_BB_TO_FSBT(mp, bmv->bmv_offset),
5503                                   XFS_BB_TO_FSB(mp, bmv->bmv_length),
5504                                   bmapi_flags, NULL, 0, map, &nmap,
5505                                   NULL);
5506                 if (error)
5507                         goto out_free_map;
5508                 ASSERT(nmap <= subnex);
5509
5510                 for (i = 0; i < nmap && nexleft && bmv->bmv_length; i++) {
5511                         out[cur_ext].bmv_oflags = 0;
5512                         if (map[i].br_state == XFS_EXT_UNWRITTEN)
5513                                 out[cur_ext].bmv_oflags |= BMV_OF_PREALLOC;
5514                         else if (map[i].br_startblock == DELAYSTARTBLOCK)
5515                                 out[cur_ext].bmv_oflags |= BMV_OF_DELALLOC;
5516                         out[cur_ext].bmv_offset =
5517                                 XFS_FSB_TO_BB(mp, map[i].br_startoff);
5518                         out[cur_ext].bmv_length =
5519                                 XFS_FSB_TO_BB(mp, map[i].br_blockcount);
5520                         out[cur_ext].bmv_unused1 = 0;
5521                         out[cur_ext].bmv_unused2 = 0;
5522                         ASSERT(((iflags & BMV_IF_DELALLOC) != 0) ||
5523                               (map[i].br_startblock != DELAYSTARTBLOCK));
5524                         if (map[i].br_startblock == HOLESTARTBLOCK &&
5525                             whichfork == XFS_ATTR_FORK) {
5526                                 /* came to the end of attribute fork */
5527                                 out[cur_ext].bmv_oflags |= BMV_OF_LAST;
5528                                 goto out_free_map;
5529                         }
5530
5531                         if (!xfs_getbmapx_fix_eof_hole(ip, &out[cur_ext],
5532                                         prealloced, bmvend,
5533                                         map[i].br_startblock))
5534                                 goto out_free_map;
5535
5536                         bmv->bmv_offset =
5537                                 out[cur_ext].bmv_offset +
5538                                 out[cur_ext].bmv_length;
5539                         bmv->bmv_length =
5540                                 max_t(__int64_t, 0, bmvend - bmv->bmv_offset);
5541
5542                         /*
5543                          * In case we don't want to return the hole,
5544                          * don't increase cur_ext so that we can reuse
5545                          * it in the next loop.
5546                          */
5547                         if ((iflags & BMV_IF_NO_HOLES) &&
5548                             map[i].br_startblock == HOLESTARTBLOCK) {
5549                                 memset(&out[cur_ext], 0, sizeof(out[cur_ext]));
5550                                 continue;
5551                         }
5552
5553                         nexleft--;
5554                         bmv->bmv_entries++;
5555                         cur_ext++;
5556                 }
5557         } while (nmap && nexleft && bmv->bmv_length);
5558
5559  out_free_map:
5560         kmem_free(map);
5561  out_unlock_ilock:
5562         xfs_iunlock_map_shared(ip, lock);
5563  out_unlock_iolock:
5564         xfs_iunlock(ip, XFS_IOLOCK_SHARED);
5565
5566         for (i = 0; i < cur_ext; i++) {
5567                 int full = 0;   /* user array is full */
5568
5569                 /* format results & advance arg */
5570                 error = formatter(&arg, &out[i], &full);
5571                 if (error || full)
5572                         break;
5573         }
5574
5575         kmem_free(out);
5576         return error;
5577 }
5578
5579 /*
5580  * Check the last inode extent to determine whether this allocation will result
5581  * in blocks being allocated at the end of the file. When we allocate new data
5582  * blocks at the end of the file which do not start at the previous data block,
5583  * we will try to align the new blocks at stripe unit boundaries.
5584  */
5585 STATIC int                              /* error */
5586 xfs_bmap_isaeof(
5587         xfs_inode_t     *ip,            /* incore inode pointer */
5588         xfs_fileoff_t   off,            /* file offset in fsblocks */
5589         int             whichfork,      /* data or attribute fork */
5590         char            *aeof)          /* return value */
5591 {
5592         int             error;          /* error return value */
5593         xfs_ifork_t     *ifp;           /* inode fork pointer */
5594         xfs_bmbt_rec_host_t *lastrec;   /* extent record pointer */
5595         xfs_extnum_t    nextents;       /* number of file extents */
5596         xfs_bmbt_irec_t s;              /* expanded extent record */
5597
5598         ASSERT(whichfork == XFS_DATA_FORK);
5599         ifp = XFS_IFORK_PTR(ip, whichfork);
5600         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5601             (error = xfs_iread_extents(NULL, ip, whichfork)))
5602                 return error;
5603         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5604         if (nextents == 0) {
5605                 *aeof = 1;
5606                 return 0;
5607         }
5608         /*
5609          * Go to the last extent
5610          */
5611         lastrec = xfs_iext_get_ext(ifp, nextents - 1);
5612         xfs_bmbt_get_all(lastrec, &s);
5613         /*
5614          * Check we are allocating in the last extent (for delayed allocations)
5615          * or past the last extent for non-delayed allocations.
5616          */
5617         *aeof = (off >= s.br_startoff &&
5618                  off < s.br_startoff + s.br_blockcount &&
5619                  isnullstartblock(s.br_startblock)) ||
5620                 off >= s.br_startoff + s.br_blockcount;
5621         return 0;
5622 }
5623
5624 /*
5625  * Check if the endoff is outside the last extent. If so the caller will grow
5626  * the allocation to a stripe unit boundary.
5627  */
5628 int                                     /* error */
5629 xfs_bmap_eof(
5630         xfs_inode_t     *ip,            /* incore inode pointer */
5631         xfs_fileoff_t   endoff,         /* file offset in fsblocks */
5632         int             whichfork,      /* data or attribute fork */
5633         int             *eof)           /* result value */
5634 {
5635         xfs_fsblock_t   blockcount;     /* extent block count */
5636         int             error;          /* error return value */
5637         xfs_ifork_t     *ifp;           /* inode fork pointer */
5638         xfs_bmbt_rec_host_t *lastrec;   /* extent record pointer */
5639         xfs_extnum_t    nextents;       /* number of file extents */
5640         xfs_fileoff_t   startoff;       /* extent starting file offset */
5641
5642         ASSERT(whichfork == XFS_DATA_FORK);
5643         ifp = XFS_IFORK_PTR(ip, whichfork);
5644         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5645             (error = xfs_iread_extents(NULL, ip, whichfork)))
5646                 return error;
5647         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5648         if (nextents == 0) {
5649                 *eof = 1;
5650                 return 0;
5651         }
5652         /*
5653          * Go to the last extent
5654          */
5655         lastrec = xfs_iext_get_ext(ifp, nextents - 1);
5656         startoff = xfs_bmbt_get_startoff(lastrec);
5657         blockcount = xfs_bmbt_get_blockcount(lastrec);
5658         *eof = endoff >= startoff + blockcount;
5659         return 0;
5660 }
5661
5662 #ifdef DEBUG
5663 STATIC struct xfs_buf *
5664 xfs_bmap_get_bp(
5665         struct xfs_btree_cur    *cur,
5666         xfs_fsblock_t           bno)
5667 {
5668         struct xfs_log_item_desc *lidp;
5669         int                     i;
5670
5671         if (!cur)
5672                 return NULL;
5673
5674         for (i = 0; i < XFS_BTREE_MAXLEVELS; i++) {
5675                 if (!cur->bc_bufs[i])
5676                         break;
5677                 if (XFS_BUF_ADDR(cur->bc_bufs[i]) == bno)
5678                         return cur->bc_bufs[i];
5679         }
5680
5681         /* Chase down all the log items to see if the bp is there */
5682         list_for_each_entry(lidp, &cur->bc_tp->t_items, lid_trans) {
5683                 struct xfs_buf_log_item *bip;
5684                 bip = (struct xfs_buf_log_item *)lidp->lid_item;
5685                 if (bip->bli_item.li_type == XFS_LI_BUF &&
5686                     XFS_BUF_ADDR(bip->bli_buf) == bno)
5687                         return bip->bli_buf;
5688         }
5689
5690         return NULL;
5691 }
5692
5693 STATIC void
5694 xfs_check_block(
5695         struct xfs_btree_block  *block,
5696         xfs_mount_t             *mp,
5697         int                     root,
5698         short                   sz)
5699 {
5700         int                     i, j, dmxr;
5701         __be64                  *pp, *thispa;   /* pointer to block address */
5702         xfs_bmbt_key_t          *prevp, *keyp;
5703
5704         ASSERT(be16_to_cpu(block->bb_level) > 0);
5705
5706         prevp = NULL;
5707         for( i = 1; i <= xfs_btree_get_numrecs(block); i++) {
5708                 dmxr = mp->m_bmap_dmxr[0];
5709                 keyp = XFS_BMBT_KEY_ADDR(mp, block, i);
5710
5711                 if (prevp) {
5712                         ASSERT(be64_to_cpu(prevp->br_startoff) <
5713                                be64_to_cpu(keyp->br_startoff));
5714                 }
5715                 prevp = keyp;
5716
5717                 /*
5718                  * Compare the block numbers to see if there are dups.
5719                  */
5720                 if (root)
5721                         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, i, sz);
5722                 else
5723                         pp = XFS_BMBT_PTR_ADDR(mp, block, i, dmxr);
5724
5725                 for (j = i+1; j <= be16_to_cpu(block->bb_numrecs); j++) {
5726                         if (root)
5727                                 thispa = XFS_BMAP_BROOT_PTR_ADDR(mp, block, j, sz);
5728                         else
5729                                 thispa = XFS_BMBT_PTR_ADDR(mp, block, j, dmxr);
5730                         if (*thispa == *pp) {
5731                                 cmn_err(CE_WARN, "%s: thispa(%d) == pp(%d) %Ld",
5732                                         __func__, j, i,
5733                                         (unsigned long long)be64_to_cpu(*thispa));
5734                                 panic("%s: ptrs are equal in node\n",
5735                                         __func__);
5736                         }
5737                 }
5738         }
5739 }
5740
5741 /*
5742  * Check that the extents for the inode ip are in the right order in all
5743  * btree leaves.
5744  */
5745
5746 STATIC void
5747 xfs_bmap_check_leaf_extents(
5748         xfs_btree_cur_t         *cur,   /* btree cursor or null */
5749         xfs_inode_t             *ip,            /* incore inode pointer */
5750         int                     whichfork)      /* data or attr fork */
5751 {
5752         struct xfs_btree_block  *block; /* current btree block */
5753         xfs_fsblock_t           bno;    /* block # of "block" */
5754         xfs_buf_t               *bp;    /* buffer for "block" */
5755         int                     error;  /* error return value */
5756         xfs_extnum_t            i=0, j; /* index into the extents list */
5757         xfs_ifork_t             *ifp;   /* fork structure */
5758         int                     level;  /* btree level, for checking */
5759         xfs_mount_t             *mp;    /* file system mount structure */
5760         __be64                  *pp;    /* pointer to block address */
5761         xfs_bmbt_rec_t          *ep;    /* pointer to current extent */
5762         xfs_bmbt_rec_t          last = {0, 0}; /* last extent in prev block */
5763         xfs_bmbt_rec_t          *nextp; /* pointer to next extent */
5764         int                     bp_release = 0;
5765
5766         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE) {
5767                 return;
5768         }
5769
5770         bno = NULLFSBLOCK;
5771         mp = ip->i_mount;
5772         ifp = XFS_IFORK_PTR(ip, whichfork);
5773         block = ifp->if_broot;
5774         /*
5775          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
5776          */
5777         level = be16_to_cpu(block->bb_level);
5778         ASSERT(level > 0);
5779         xfs_check_block(block, mp, 1, ifp->if_broot_bytes);
5780         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
5781         bno = be64_to_cpu(*pp);
5782
5783         ASSERT(bno != NULLDFSBNO);
5784         ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
5785         ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
5786
5787         /*
5788          * Go down the tree until leaf level is reached, following the first
5789          * pointer (leftmost) at each level.
5790          */
5791         while (level-- > 0) {
5792                 /* See if buf is in cur first */
5793                 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
5794                 if (bp) {
5795                         bp_release = 0;
5796                 } else {
5797                         bp_release = 1;
5798                 }
5799                 if (!bp && (error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
5800                                 XFS_BMAP_BTREE_REF)))
5801                         goto error_norelse;
5802                 block = XFS_BUF_TO_BLOCK(bp);
5803                 XFS_WANT_CORRUPTED_GOTO(
5804                         xfs_bmap_sanity_check(mp, bp, level),
5805                         error0);
5806                 if (level == 0)
5807                         break;
5808
5809                 /*
5810                  * Check this block for basic sanity (increasing keys and
5811                  * no duplicate blocks).
5812                  */
5813
5814                 xfs_check_block(block, mp, 0, 0);
5815                 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
5816                 bno = be64_to_cpu(*pp);
5817                 XFS_WANT_CORRUPTED_GOTO(XFS_FSB_SANITY_CHECK(mp, bno), error0);
5818                 if (bp_release) {
5819                         bp_release = 0;
5820                         xfs_trans_brelse(NULL, bp);
5821                 }
5822         }
5823
5824         /*
5825          * Here with bp and block set to the leftmost leaf node in the tree.
5826          */
5827         i = 0;
5828
5829         /*
5830          * Loop over all leaf nodes checking that all extents are in the right order.
5831          */
5832         for (;;) {
5833                 xfs_fsblock_t   nextbno;
5834                 xfs_extnum_t    num_recs;
5835
5836
5837                 num_recs = xfs_btree_get_numrecs(block);
5838
5839                 /*
5840                  * Read-ahead the next leaf block, if any.
5841                  */
5842
5843                 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
5844
5845                 /*
5846                  * Check all the extents to make sure they are OK.
5847                  * If we had a previous block, the last entry should
5848                  * conform with the first entry in this one.
5849                  */
5850
5851                 ep = XFS_BMBT_REC_ADDR(mp, block, 1);
5852                 if (i) {
5853                         ASSERT(xfs_bmbt_disk_get_startoff(&last) +
5854                                xfs_bmbt_disk_get_blockcount(&last) <=
5855                                xfs_bmbt_disk_get_startoff(ep));
5856                 }
5857                 for (j = 1; j < num_recs; j++) {
5858                         nextp = XFS_BMBT_REC_ADDR(mp, block, j + 1);
5859                         ASSERT(xfs_bmbt_disk_get_startoff(ep) +
5860                                xfs_bmbt_disk_get_blockcount(ep) <=
5861                                xfs_bmbt_disk_get_startoff(nextp));
5862                         ep = nextp;
5863                 }
5864
5865                 last = *ep;
5866                 i += num_recs;
5867                 if (bp_release) {
5868                         bp_release = 0;
5869                         xfs_trans_brelse(NULL, bp);
5870                 }
5871                 bno = nextbno;
5872                 /*
5873                  * If we've reached the end, stop.
5874                  */
5875                 if (bno == NULLFSBLOCK)
5876                         break;
5877
5878                 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
5879                 if (bp) {
5880                         bp_release = 0;
5881                 } else {
5882                         bp_release = 1;
5883                 }
5884                 if (!bp && (error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
5885                                 XFS_BMAP_BTREE_REF)))
5886                         goto error_norelse;
5887                 block = XFS_BUF_TO_BLOCK(bp);
5888         }
5889         if (bp_release) {
5890                 bp_release = 0;
5891                 xfs_trans_brelse(NULL, bp);
5892         }
5893         return;
5894
5895 error0:
5896         cmn_err(CE_WARN, "%s: at error0", __func__);
5897         if (bp_release)
5898                 xfs_trans_brelse(NULL, bp);
5899 error_norelse:
5900         cmn_err(CE_WARN, "%s: BAD after btree leaves for %d extents",
5901                 __func__, i);
5902         panic("%s: CORRUPTED BTREE OR SOMETHING", __func__);
5903         return;
5904 }
5905 #endif
5906
5907 /*
5908  * Count fsblocks of the given fork.
5909  */
5910 int                                             /* error */
5911 xfs_bmap_count_blocks(
5912         xfs_trans_t             *tp,            /* transaction pointer */
5913         xfs_inode_t             *ip,            /* incore inode */
5914         int                     whichfork,      /* data or attr fork */
5915         int                     *count)         /* out: count of blocks */
5916 {
5917         struct xfs_btree_block  *block; /* current btree block */
5918         xfs_fsblock_t           bno;    /* block # of "block" */
5919         xfs_ifork_t             *ifp;   /* fork structure */
5920         int                     level;  /* btree level, for checking */
5921         xfs_mount_t             *mp;    /* file system mount structure */
5922         __be64                  *pp;    /* pointer to block address */
5923
5924         bno = NULLFSBLOCK;
5925         mp = ip->i_mount;
5926         ifp = XFS_IFORK_PTR(ip, whichfork);
5927         if ( XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ) {
5928                 xfs_bmap_count_leaves(ifp, 0,
5929                         ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t),
5930                         count);
5931                 return 0;
5932         }
5933
5934         /*
5935          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
5936          */
5937         block = ifp->if_broot;
5938         level = be16_to_cpu(block->bb_level);
5939         ASSERT(level > 0);
5940         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
5941         bno = be64_to_cpu(*pp);
5942         ASSERT(bno != NULLDFSBNO);
5943         ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
5944         ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
5945
5946         if (unlikely(xfs_bmap_count_tree(mp, tp, ifp, bno, level, count) < 0)) {
5947                 XFS_ERROR_REPORT("xfs_bmap_count_blocks(2)", XFS_ERRLEVEL_LOW,
5948                                  mp);
5949                 return XFS_ERROR(EFSCORRUPTED);
5950         }
5951
5952         return 0;
5953 }
5954
5955 /*
5956  * Recursively walks each level of a btree
5957  * to count total fsblocks is use.
5958  */
5959 STATIC int                                     /* error */
5960 xfs_bmap_count_tree(
5961         xfs_mount_t     *mp,            /* file system mount point */
5962         xfs_trans_t     *tp,            /* transaction pointer */
5963         xfs_ifork_t     *ifp,           /* inode fork pointer */
5964         xfs_fsblock_t   blockno,        /* file system block number */
5965         int             levelin,        /* level in btree */
5966         int             *count)         /* Count of blocks */
5967 {
5968         int                     error;
5969         xfs_buf_t               *bp, *nbp;
5970         int                     level = levelin;
5971         __be64                  *pp;
5972         xfs_fsblock_t           bno = blockno;
5973         xfs_fsblock_t           nextbno;
5974         struct xfs_btree_block  *block, *nextblock;
5975         int                     numrecs;
5976
5977         if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp, XFS_BMAP_BTREE_REF)))
5978                 return error;
5979         *count += 1;
5980         block = XFS_BUF_TO_BLOCK(bp);
5981
5982         if (--level) {
5983                 /* Not at node above leaves, count this level of nodes */
5984                 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
5985                 while (nextbno != NULLFSBLOCK) {
5986                         if ((error = xfs_btree_read_bufl(mp, tp, nextbno,
5987                                 0, &nbp, XFS_BMAP_BTREE_REF)))
5988                                 return error;
5989                         *count += 1;
5990                         nextblock = XFS_BUF_TO_BLOCK(nbp);
5991                         nextbno = be64_to_cpu(nextblock->bb_u.l.bb_rightsib);
5992                         xfs_trans_brelse(tp, nbp);
5993                 }
5994
5995                 /* Dive to the next level */
5996                 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
5997                 bno = be64_to_cpu(*pp);
5998                 if (unlikely((error =
5999                      xfs_bmap_count_tree(mp, tp, ifp, bno, level, count)) < 0)) {
6000                         xfs_trans_brelse(tp, bp);
6001                         XFS_ERROR_REPORT("xfs_bmap_count_tree(1)",
6002                                          XFS_ERRLEVEL_LOW, mp);
6003                         return XFS_ERROR(EFSCORRUPTED);
6004                 }
6005                 xfs_trans_brelse(tp, bp);
6006         } else {
6007                 /* count all level 1 nodes and their leaves */
6008                 for (;;) {
6009                         nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
6010                         numrecs = be16_to_cpu(block->bb_numrecs);
6011                         xfs_bmap_disk_count_leaves(mp, block, numrecs, count);
6012                         xfs_trans_brelse(tp, bp);
6013                         if (nextbno == NULLFSBLOCK)
6014                                 break;
6015                         bno = nextbno;
6016                         if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
6017                                 XFS_BMAP_BTREE_REF)))
6018                                 return error;
6019                         *count += 1;
6020                         block = XFS_BUF_TO_BLOCK(bp);
6021                 }
6022         }
6023         return 0;
6024 }
6025
6026 /*
6027  * Count leaf blocks given a range of extent records.
6028  */
6029 STATIC void
6030 xfs_bmap_count_leaves(
6031         xfs_ifork_t             *ifp,
6032         xfs_extnum_t            idx,
6033         int                     numrecs,
6034         int                     *count)
6035 {
6036         int             b;
6037
6038         for (b = 0; b < numrecs; b++) {
6039                 xfs_bmbt_rec_host_t *frp = xfs_iext_get_ext(ifp, idx + b);
6040                 *count += xfs_bmbt_get_blockcount(frp);
6041         }
6042 }
6043
6044 /*
6045  * Count leaf blocks given a range of extent records originally
6046  * in btree format.
6047  */
6048 STATIC void
6049 xfs_bmap_disk_count_leaves(
6050         struct xfs_mount        *mp,
6051         struct xfs_btree_block  *block,
6052         int                     numrecs,
6053         int                     *count)
6054 {
6055         int             b;
6056         xfs_bmbt_rec_t  *frp;
6057
6058         for (b = 1; b <= numrecs; b++) {
6059                 frp = XFS_BMBT_REC_ADDR(mp, block, b);
6060                 *count += xfs_bmbt_disk_get_blockcount(frp);
6061         }
6062 }