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_icsb_modify_counters(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                  * We start with a delayed allocation:
1043                  *
1044                  * +ddddddddddddddddddddddddddddddddddddddddddddddddddddddd+
1045                  *  PREV @ idx
1046                  *
1047                  * and we are allocating:
1048                  *                     +rrrrrrrrrrrrrrrrr+
1049                  *                            new
1050                  *
1051                  * and we set it up for insertion as:
1052                  * +ddddddddddddddddddd+rrrrrrrrrrrrrrrrr+ddddddddddddddddd+
1053                  *                            new
1054                  *  PREV @ idx          LEFT              RIGHT
1055                  *                      inserted at idx + 1
1056                  */
1057                 temp = new->br_startoff - PREV.br_startoff;
1058                 temp2 = PREV.br_startoff + PREV.br_blockcount - new_endoff;
1059                 trace_xfs_bmap_pre_update(ip, idx, 0, _THIS_IP_);
1060                 xfs_bmbt_set_blockcount(ep, temp);      /* truncate PREV */
1061                 LEFT = *new;
1062                 RIGHT.br_state = PREV.br_state;
1063                 RIGHT.br_startblock = nullstartblock(
1064                                 (int)xfs_bmap_worst_indlen(ip, temp2));
1065                 RIGHT.br_startoff = new_endoff;
1066                 RIGHT.br_blockcount = temp2;
1067                 /* insert LEFT (r[0]) and RIGHT (r[1]) at the same time */
1068                 xfs_iext_insert(ip, idx + 1, 2, &LEFT, state);
1069                 ip->i_df.if_lastex = idx + 1;
1070                 ip->i_d.di_nextents++;
1071                 if (cur == NULL)
1072                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1073                 else {
1074                         rval = XFS_ILOG_CORE;
1075                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1076                                         new->br_startblock, new->br_blockcount,
1077                                         &i)))
1078                                 goto done;
1079                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1080                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
1081                         if ((error = xfs_btree_insert(cur, &i)))
1082                                 goto done;
1083                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1084                 }
1085                 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1086                     ip->i_d.di_nextents > ip->i_df.if_ext_max) {
1087                         error = xfs_bmap_extents_to_btree(ip->i_transp, ip,
1088                                         first, flist, &cur, 1, &tmp_rval,
1089                                         XFS_DATA_FORK);
1090                         rval |= tmp_rval;
1091                         if (error)
1092                                 goto done;
1093                 }
1094                 temp = xfs_bmap_worst_indlen(ip, temp);
1095                 temp2 = xfs_bmap_worst_indlen(ip, temp2);
1096                 diff = (int)(temp + temp2 - startblockval(PREV.br_startblock) -
1097                         (cur ? cur->bc_private.b.allocated : 0));
1098                 if (diff > 0 &&
1099                     xfs_icsb_modify_counters(ip->i_mount, XFS_SBS_FDBLOCKS,
1100                                              -((int64_t)diff), rsvd)) {
1101                         /*
1102                          * Ick gross gag me with a spoon.
1103                          */
1104                         ASSERT(0);      /* want to see if this ever happens! */
1105                         while (diff > 0) {
1106                                 if (temp) {
1107                                         temp--;
1108                                         diff--;
1109                                         if (!diff ||
1110                                             !xfs_icsb_modify_counters(ip->i_mount,
1111                                                     XFS_SBS_FDBLOCKS,
1112                                                     -((int64_t)diff), rsvd))
1113                                                 break;
1114                                 }
1115                                 if (temp2) {
1116                                         temp2--;
1117                                         diff--;
1118                                         if (!diff ||
1119                                             !xfs_icsb_modify_counters(ip->i_mount,
1120                                                     XFS_SBS_FDBLOCKS,
1121                                                     -((int64_t)diff), rsvd))
1122                                                 break;
1123                                 }
1124                         }
1125                 }
1126                 ep = xfs_iext_get_ext(ifp, idx);
1127                 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
1128                 trace_xfs_bmap_post_update(ip, idx, state, _THIS_IP_);
1129                 trace_xfs_bmap_pre_update(ip, idx + 2, state, _THIS_IP_);
1130                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, idx + 2),
1131                         nullstartblock((int)temp2));
1132                 trace_xfs_bmap_post_update(ip, idx + 2, state, _THIS_IP_);
1133                 *dnew = temp + temp2;
1134                 break;
1135
1136         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1137         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1138         case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
1139         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1140         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1141         case BMAP_LEFT_CONTIG:
1142         case BMAP_RIGHT_CONTIG:
1143                 /*
1144                  * These cases are all impossible.
1145                  */
1146                 ASSERT(0);
1147         }
1148         *curp = cur;
1149 done:
1150         *logflagsp = rval;
1151         return error;
1152 #undef  LEFT
1153 #undef  RIGHT
1154 #undef  PREV
1155 }
1156
1157 /*
1158  * Called by xfs_bmap_add_extent to handle cases converting an unwritten
1159  * allocation to a real allocation or vice versa.
1160  */
1161 STATIC int                              /* error */
1162 xfs_bmap_add_extent_unwritten_real(
1163         xfs_inode_t             *ip,    /* incore inode pointer */
1164         xfs_extnum_t            idx,    /* extent number to update/insert */
1165         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
1166         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
1167         int                     *logflagsp) /* inode logging flags */
1168 {
1169         xfs_btree_cur_t         *cur;   /* btree cursor */
1170         xfs_bmbt_rec_host_t     *ep;    /* extent entry for idx */
1171         int                     error;  /* error return value */
1172         int                     i;      /* temp state */
1173         xfs_ifork_t             *ifp;   /* inode fork pointer */
1174         xfs_fileoff_t           new_endoff;     /* end offset of new entry */
1175         xfs_exntst_t            newext; /* new extent state */
1176         xfs_exntst_t            oldext; /* old extent state */
1177         xfs_bmbt_irec_t         r[3];   /* neighbor extent entries */
1178                                         /* left is 0, right is 1, prev is 2 */
1179         int                     rval=0; /* return value (logging flags) */
1180         int                     state = 0;/* state bits, accessed thru macros */
1181
1182 #define LEFT            r[0]
1183 #define RIGHT           r[1]
1184 #define PREV            r[2]
1185         /*
1186          * Set up a bunch of variables to make the tests simpler.
1187          */
1188         error = 0;
1189         cur = *curp;
1190         ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
1191         ep = xfs_iext_get_ext(ifp, idx);
1192         xfs_bmbt_get_all(ep, &PREV);
1193         newext = new->br_state;
1194         oldext = (newext == XFS_EXT_UNWRITTEN) ?
1195                 XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
1196         ASSERT(PREV.br_state == oldext);
1197         new_endoff = new->br_startoff + new->br_blockcount;
1198         ASSERT(PREV.br_startoff <= new->br_startoff);
1199         ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
1200
1201         /*
1202          * Set flags determining what part of the previous oldext allocation
1203          * extent is being replaced by a newext allocation.
1204          */
1205         if (PREV.br_startoff == new->br_startoff)
1206                 state |= BMAP_LEFT_FILLING;
1207         if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
1208                 state |= BMAP_RIGHT_FILLING;
1209
1210         /*
1211          * Check and set flags if this segment has a left neighbor.
1212          * Don't set contiguous if the combined extent would be too large.
1213          */
1214         if (idx > 0) {
1215                 state |= BMAP_LEFT_VALID;
1216                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, idx - 1), &LEFT);
1217
1218                 if (isnullstartblock(LEFT.br_startblock))
1219                         state |= BMAP_LEFT_DELAY;
1220         }
1221
1222         if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
1223             LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
1224             LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
1225             LEFT.br_state == newext &&
1226             LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
1227                 state |= BMAP_LEFT_CONTIG;
1228
1229         /*
1230          * Check and set flags if this segment has a right neighbor.
1231          * Don't set contiguous if the combined extent would be too large.
1232          * Also check for all-three-contiguous being too large.
1233          */
1234         if (idx < ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1) {
1235                 state |= BMAP_RIGHT_VALID;
1236                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, idx + 1), &RIGHT);
1237                 if (isnullstartblock(RIGHT.br_startblock))
1238                         state |= BMAP_RIGHT_DELAY;
1239         }
1240
1241         if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
1242             new_endoff == RIGHT.br_startoff &&
1243             new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
1244             newext == RIGHT.br_state &&
1245             new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
1246             ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1247                        BMAP_RIGHT_FILLING)) !=
1248                       (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1249                        BMAP_RIGHT_FILLING) ||
1250              LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
1251                         <= MAXEXTLEN))
1252                 state |= BMAP_RIGHT_CONTIG;
1253
1254         /*
1255          * Switch out based on the FILLING and CONTIG state bits.
1256          */
1257         switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1258                          BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
1259         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1260              BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1261                 /*
1262                  * Setting all of a previous oldext extent to newext.
1263                  * The left and right neighbors are both contiguous with new.
1264                  */
1265                 trace_xfs_bmap_pre_update(ip, idx - 1, state, _THIS_IP_);
1266                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
1267                         LEFT.br_blockcount + PREV.br_blockcount +
1268                         RIGHT.br_blockcount);
1269                 trace_xfs_bmap_post_update(ip, idx - 1, state, _THIS_IP_);
1270
1271                 xfs_iext_remove(ip, idx, 2, state);
1272                 ip->i_df.if_lastex = idx - 1;
1273                 ip->i_d.di_nextents -= 2;
1274                 if (cur == NULL)
1275                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1276                 else {
1277                         rval = XFS_ILOG_CORE;
1278                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
1279                                         RIGHT.br_startblock,
1280                                         RIGHT.br_blockcount, &i)))
1281                                 goto done;
1282                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1283                         if ((error = xfs_btree_delete(cur, &i)))
1284                                 goto done;
1285                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1286                         if ((error = xfs_btree_decrement(cur, 0, &i)))
1287                                 goto done;
1288                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1289                         if ((error = xfs_btree_delete(cur, &i)))
1290                                 goto done;
1291                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1292                         if ((error = xfs_btree_decrement(cur, 0, &i)))
1293                                 goto done;
1294                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1295                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
1296                                 LEFT.br_startblock,
1297                                 LEFT.br_blockcount + PREV.br_blockcount +
1298                                 RIGHT.br_blockcount, LEFT.br_state)))
1299                                 goto done;
1300                 }
1301                 break;
1302
1303         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1304                 /*
1305                  * Setting all of a previous oldext extent to newext.
1306                  * The left neighbor is contiguous, the right is not.
1307                  */
1308                 trace_xfs_bmap_pre_update(ip, idx - 1, state, _THIS_IP_);
1309                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
1310                         LEFT.br_blockcount + PREV.br_blockcount);
1311                 trace_xfs_bmap_post_update(ip, idx - 1, state, _THIS_IP_);
1312
1313                 ip->i_df.if_lastex = idx - 1;
1314                 xfs_iext_remove(ip, idx, 1, state);
1315                 ip->i_d.di_nextents--;
1316                 if (cur == NULL)
1317                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1318                 else {
1319                         rval = XFS_ILOG_CORE;
1320                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1321                                         PREV.br_startblock, PREV.br_blockcount,
1322                                         &i)))
1323                                 goto done;
1324                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1325                         if ((error = xfs_btree_delete(cur, &i)))
1326                                 goto done;
1327                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1328                         if ((error = xfs_btree_decrement(cur, 0, &i)))
1329                                 goto done;
1330                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1331                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
1332                                 LEFT.br_startblock,
1333                                 LEFT.br_blockcount + PREV.br_blockcount,
1334                                 LEFT.br_state)))
1335                                 goto done;
1336                 }
1337                 break;
1338
1339         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1340                 /*
1341                  * Setting all of a previous oldext extent to newext.
1342                  * The right neighbor is contiguous, the left is not.
1343                  */
1344                 trace_xfs_bmap_pre_update(ip, idx, state, _THIS_IP_);
1345                 xfs_bmbt_set_blockcount(ep,
1346                         PREV.br_blockcount + RIGHT.br_blockcount);
1347                 xfs_bmbt_set_state(ep, newext);
1348                 trace_xfs_bmap_post_update(ip, idx, state, _THIS_IP_);
1349                 ip->i_df.if_lastex = idx;
1350                 xfs_iext_remove(ip, idx + 1, 1, state);
1351                 ip->i_d.di_nextents--;
1352                 if (cur == NULL)
1353                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1354                 else {
1355                         rval = XFS_ILOG_CORE;
1356                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
1357                                         RIGHT.br_startblock,
1358                                         RIGHT.br_blockcount, &i)))
1359                                 goto done;
1360                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1361                         if ((error = xfs_btree_delete(cur, &i)))
1362                                 goto done;
1363                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1364                         if ((error = xfs_btree_decrement(cur, 0, &i)))
1365                                 goto done;
1366                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1367                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
1368                                 new->br_startblock,
1369                                 new->br_blockcount + RIGHT.br_blockcount,
1370                                 newext)))
1371                                 goto done;
1372                 }
1373                 break;
1374
1375         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
1376                 /*
1377                  * Setting all of a previous oldext extent to newext.
1378                  * Neither the left nor right neighbors are contiguous with
1379                  * the new one.
1380                  */
1381                 trace_xfs_bmap_pre_update(ip, idx, state, _THIS_IP_);
1382                 xfs_bmbt_set_state(ep, newext);
1383                 trace_xfs_bmap_post_update(ip, idx, state, _THIS_IP_);
1384
1385                 ip->i_df.if_lastex = idx;
1386                 if (cur == NULL)
1387                         rval = XFS_ILOG_DEXT;
1388                 else {
1389                         rval = 0;
1390                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1391                                         new->br_startblock, new->br_blockcount,
1392                                         &i)))
1393                                 goto done;
1394                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1395                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
1396                                 new->br_startblock, new->br_blockcount,
1397                                 newext)))
1398                                 goto done;
1399                 }
1400                 break;
1401
1402         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
1403                 /*
1404                  * Setting the first part of a previous oldext extent to newext.
1405                  * The left neighbor is contiguous.
1406                  */
1407                 trace_xfs_bmap_pre_update(ip, idx - 1, state, _THIS_IP_);
1408                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
1409                         LEFT.br_blockcount + new->br_blockcount);
1410                 xfs_bmbt_set_startoff(ep,
1411                         PREV.br_startoff + new->br_blockcount);
1412                 trace_xfs_bmap_post_update(ip, idx - 1, state, _THIS_IP_);
1413
1414                 trace_xfs_bmap_pre_update(ip, idx, state, _THIS_IP_);
1415                 xfs_bmbt_set_startblock(ep,
1416                         new->br_startblock + new->br_blockcount);
1417                 xfs_bmbt_set_blockcount(ep,
1418                         PREV.br_blockcount - new->br_blockcount);
1419                 trace_xfs_bmap_post_update(ip, idx, state, _THIS_IP_);
1420
1421                 ip->i_df.if_lastex = idx - 1;
1422                 if (cur == NULL)
1423                         rval = XFS_ILOG_DEXT;
1424                 else {
1425                         rval = 0;
1426                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1427                                         PREV.br_startblock, PREV.br_blockcount,
1428                                         &i)))
1429                                 goto done;
1430                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1431                         if ((error = xfs_bmbt_update(cur,
1432                                 PREV.br_startoff + new->br_blockcount,
1433                                 PREV.br_startblock + new->br_blockcount,
1434                                 PREV.br_blockcount - new->br_blockcount,
1435                                 oldext)))
1436                                 goto done;
1437                         if ((error = xfs_btree_decrement(cur, 0, &i)))
1438                                 goto done;
1439                         if (xfs_bmbt_update(cur, LEFT.br_startoff,
1440                                 LEFT.br_startblock,
1441                                 LEFT.br_blockcount + new->br_blockcount,
1442                                 LEFT.br_state))
1443                                 goto done;
1444                 }
1445                 break;
1446
1447         case BMAP_LEFT_FILLING:
1448                 /*
1449                  * Setting the first part of a previous oldext extent to newext.
1450                  * The left neighbor is not contiguous.
1451                  */
1452                 trace_xfs_bmap_pre_update(ip, idx, state, _THIS_IP_);
1453                 ASSERT(ep && xfs_bmbt_get_state(ep) == oldext);
1454                 xfs_bmbt_set_startoff(ep, new_endoff);
1455                 xfs_bmbt_set_blockcount(ep,
1456                         PREV.br_blockcount - new->br_blockcount);
1457                 xfs_bmbt_set_startblock(ep,
1458                         new->br_startblock + new->br_blockcount);
1459                 trace_xfs_bmap_post_update(ip, idx, state, _THIS_IP_);
1460
1461                 xfs_iext_insert(ip, idx, 1, new, state);
1462                 ip->i_df.if_lastex = idx;
1463                 ip->i_d.di_nextents++;
1464                 if (cur == NULL)
1465                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1466                 else {
1467                         rval = XFS_ILOG_CORE;
1468                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1469                                         PREV.br_startblock, PREV.br_blockcount,
1470                                         &i)))
1471                                 goto done;
1472                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1473                         if ((error = xfs_bmbt_update(cur,
1474                                 PREV.br_startoff + new->br_blockcount,
1475                                 PREV.br_startblock + new->br_blockcount,
1476                                 PREV.br_blockcount - new->br_blockcount,
1477                                 oldext)))
1478                                 goto done;
1479                         cur->bc_rec.b = *new;
1480                         if ((error = xfs_btree_insert(cur, &i)))
1481                                 goto done;
1482                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1483                 }
1484                 break;
1485
1486         case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1487                 /*
1488                  * Setting the last part of a previous oldext extent to newext.
1489                  * The right neighbor is contiguous with the new allocation.
1490                  */
1491                 trace_xfs_bmap_pre_update(ip, idx, state, _THIS_IP_);
1492                 trace_xfs_bmap_pre_update(ip, idx + 1, state, _THIS_IP_);
1493                 xfs_bmbt_set_blockcount(ep,
1494                         PREV.br_blockcount - new->br_blockcount);
1495                 trace_xfs_bmap_post_update(ip, idx, state, _THIS_IP_);
1496                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, idx + 1),
1497                         new->br_startoff, new->br_startblock,
1498                         new->br_blockcount + RIGHT.br_blockcount, newext);
1499                 trace_xfs_bmap_post_update(ip, idx + 1, state, _THIS_IP_);
1500
1501                 ip->i_df.if_lastex = idx + 1;
1502                 if (cur == NULL)
1503                         rval = XFS_ILOG_DEXT;
1504                 else {
1505                         rval = 0;
1506                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1507                                         PREV.br_startblock,
1508                                         PREV.br_blockcount, &i)))
1509                                 goto done;
1510                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1511                         if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
1512                                 PREV.br_startblock,
1513                                 PREV.br_blockcount - new->br_blockcount,
1514                                 oldext)))
1515                                 goto done;
1516                         if ((error = xfs_btree_increment(cur, 0, &i)))
1517                                 goto done;
1518                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
1519                                 new->br_startblock,
1520                                 new->br_blockcount + RIGHT.br_blockcount,
1521                                 newext)))
1522                                 goto done;
1523                 }
1524                 break;
1525
1526         case BMAP_RIGHT_FILLING:
1527                 /*
1528                  * Setting the last part of a previous oldext extent to newext.
1529                  * The right neighbor is not contiguous.
1530                  */
1531                 trace_xfs_bmap_pre_update(ip, idx, state, _THIS_IP_);
1532                 xfs_bmbt_set_blockcount(ep,
1533                         PREV.br_blockcount - new->br_blockcount);
1534                 trace_xfs_bmap_post_update(ip, idx, state, _THIS_IP_);
1535
1536                 xfs_iext_insert(ip, idx + 1, 1, new, state);
1537                 ip->i_df.if_lastex = idx + 1;
1538                 ip->i_d.di_nextents++;
1539                 if (cur == NULL)
1540                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1541                 else {
1542                         rval = XFS_ILOG_CORE;
1543                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1544                                         PREV.br_startblock, PREV.br_blockcount,
1545                                         &i)))
1546                                 goto done;
1547                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1548                         if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
1549                                 PREV.br_startblock,
1550                                 PREV.br_blockcount - new->br_blockcount,
1551                                 oldext)))
1552                                 goto done;
1553                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1554                                         new->br_startblock, new->br_blockcount,
1555                                         &i)))
1556                                 goto done;
1557                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1558                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
1559                         if ((error = xfs_btree_insert(cur, &i)))
1560                                 goto done;
1561                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1562                 }
1563                 break;
1564
1565         case 0:
1566                 /*
1567                  * Setting the middle part of a previous oldext extent to
1568                  * newext.  Contiguity is impossible here.
1569                  * One extent becomes three extents.
1570                  */
1571                 trace_xfs_bmap_pre_update(ip, idx, state, _THIS_IP_);
1572                 xfs_bmbt_set_blockcount(ep,
1573                         new->br_startoff - PREV.br_startoff);
1574                 trace_xfs_bmap_post_update(ip, idx, state, _THIS_IP_);
1575
1576                 r[0] = *new;
1577                 r[1].br_startoff = new_endoff;
1578                 r[1].br_blockcount =
1579                         PREV.br_startoff + PREV.br_blockcount - new_endoff;
1580                 r[1].br_startblock = new->br_startblock + new->br_blockcount;
1581                 r[1].br_state = oldext;
1582                 xfs_iext_insert(ip, idx + 1, 2, &r[0], state);
1583                 ip->i_df.if_lastex = idx + 1;
1584                 ip->i_d.di_nextents += 2;
1585                 if (cur == NULL)
1586                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1587                 else {
1588                         rval = XFS_ILOG_CORE;
1589                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1590                                         PREV.br_startblock, PREV.br_blockcount,
1591                                         &i)))
1592                                 goto done;
1593                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1594                         /* new right extent - oldext */
1595                         if ((error = xfs_bmbt_update(cur, r[1].br_startoff,
1596                                 r[1].br_startblock, r[1].br_blockcount,
1597                                 r[1].br_state)))
1598                                 goto done;
1599                         /* new left extent - oldext */
1600                         cur->bc_rec.b = PREV;
1601                         cur->bc_rec.b.br_blockcount =
1602                                 new->br_startoff - PREV.br_startoff;
1603                         if ((error = xfs_btree_insert(cur, &i)))
1604                                 goto done;
1605                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1606                         /*
1607                          * Reset the cursor to the position of the new extent
1608                          * we are about to insert as we can't trust it after
1609                          * the previous insert.
1610                          */
1611                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1612                                         new->br_startblock, new->br_blockcount,
1613                                         &i)))
1614                                 goto done;
1615                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1616                         /* new middle extent - newext */
1617                         cur->bc_rec.b.br_state = new->br_state;
1618                         if ((error = xfs_btree_insert(cur, &i)))
1619                                 goto done;
1620                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1621                 }
1622                 break;
1623
1624         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1625         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1626         case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
1627         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1628         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1629         case BMAP_LEFT_CONTIG:
1630         case BMAP_RIGHT_CONTIG:
1631                 /*
1632                  * These cases are all impossible.
1633                  */
1634                 ASSERT(0);
1635         }
1636         *curp = cur;
1637 done:
1638         *logflagsp = rval;
1639         return error;
1640 #undef  LEFT
1641 #undef  RIGHT
1642 #undef  PREV
1643 }
1644
1645 /*
1646  * Called by xfs_bmap_add_extent to handle cases converting a hole
1647  * to a delayed allocation.
1648  */
1649 /*ARGSUSED*/
1650 STATIC int                              /* error */
1651 xfs_bmap_add_extent_hole_delay(
1652         xfs_inode_t             *ip,    /* incore inode pointer */
1653         xfs_extnum_t            idx,    /* extent number to update/insert */
1654         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
1655         int                     *logflagsp, /* inode logging flags */
1656         int                     rsvd)           /* OK to allocate reserved blocks */
1657 {
1658         xfs_bmbt_rec_host_t     *ep;    /* extent record for idx */
1659         xfs_ifork_t             *ifp;   /* inode fork pointer */
1660         xfs_bmbt_irec_t         left;   /* left neighbor extent entry */
1661         xfs_filblks_t           newlen=0;       /* new indirect size */
1662         xfs_filblks_t           oldlen=0;       /* old indirect size */
1663         xfs_bmbt_irec_t         right;  /* right neighbor extent entry */
1664         int                     state;  /* state bits, accessed thru macros */
1665         xfs_filblks_t           temp=0; /* temp for indirect calculations */
1666
1667         ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
1668         ep = xfs_iext_get_ext(ifp, idx);
1669         state = 0;
1670         ASSERT(isnullstartblock(new->br_startblock));
1671
1672         /*
1673          * Check and set flags if this segment has a left neighbor
1674          */
1675         if (idx > 0) {
1676                 state |= BMAP_LEFT_VALID;
1677                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, idx - 1), &left);
1678
1679                 if (isnullstartblock(left.br_startblock))
1680                         state |= BMAP_LEFT_DELAY;
1681         }
1682
1683         /*
1684          * Check and set flags if the current (right) segment exists.
1685          * If it doesn't exist, we're converting the hole at end-of-file.
1686          */
1687         if (idx < ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t)) {
1688                 state |= BMAP_RIGHT_VALID;
1689                 xfs_bmbt_get_all(ep, &right);
1690
1691                 if (isnullstartblock(right.br_startblock))
1692                         state |= BMAP_RIGHT_DELAY;
1693         }
1694
1695         /*
1696          * Set contiguity flags on the left and right neighbors.
1697          * Don't let extents get too large, even if the pieces are contiguous.
1698          */
1699         if ((state & BMAP_LEFT_VALID) && (state & BMAP_LEFT_DELAY) &&
1700             left.br_startoff + left.br_blockcount == new->br_startoff &&
1701             left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
1702                 state |= BMAP_LEFT_CONTIG;
1703
1704         if ((state & BMAP_RIGHT_VALID) && (state & BMAP_RIGHT_DELAY) &&
1705             new->br_startoff + new->br_blockcount == right.br_startoff &&
1706             new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
1707             (!(state & BMAP_LEFT_CONTIG) ||
1708              (left.br_blockcount + new->br_blockcount +
1709               right.br_blockcount <= MAXEXTLEN)))
1710                 state |= BMAP_RIGHT_CONTIG;
1711
1712         /*
1713          * Switch out based on the contiguity flags.
1714          */
1715         switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
1716         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1717                 /*
1718                  * New allocation is contiguous with delayed allocations
1719                  * on the left and on the right.
1720                  * Merge all three into a single extent record.
1721                  */
1722                 temp = left.br_blockcount + new->br_blockcount +
1723                         right.br_blockcount;
1724
1725                 trace_xfs_bmap_pre_update(ip, idx - 1, state, _THIS_IP_);
1726                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1), temp);
1727                 oldlen = startblockval(left.br_startblock) +
1728                         startblockval(new->br_startblock) +
1729                         startblockval(right.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                 xfs_iext_remove(ip, idx, 1, state);
1736                 ip->i_df.if_lastex = idx - 1;
1737                 break;
1738
1739         case BMAP_LEFT_CONTIG:
1740                 /*
1741                  * New allocation is contiguous with a delayed allocation
1742                  * on the left.
1743                  * Merge the new allocation with the left neighbor.
1744                  */
1745                 temp = left.br_blockcount + new->br_blockcount;
1746                 trace_xfs_bmap_pre_update(ip, idx - 1, state, _THIS_IP_);
1747                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1), temp);
1748                 oldlen = startblockval(left.br_startblock) +
1749                         startblockval(new->br_startblock);
1750                 newlen = xfs_bmap_worst_indlen(ip, temp);
1751                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, idx - 1),
1752                         nullstartblock((int)newlen));
1753                 trace_xfs_bmap_post_update(ip, idx - 1, state, _THIS_IP_);
1754
1755                 ip->i_df.if_lastex = idx - 1;
1756                 break;
1757
1758         case BMAP_RIGHT_CONTIG:
1759                 /*
1760                  * New allocation is contiguous with a delayed allocation
1761                  * on the right.
1762                  * Merge the new allocation with the right neighbor.
1763                  */
1764                 trace_xfs_bmap_pre_update(ip, idx, state, _THIS_IP_);
1765                 temp = new->br_blockcount + right.br_blockcount;
1766                 oldlen = startblockval(new->br_startblock) +
1767                         startblockval(right.br_startblock);
1768                 newlen = xfs_bmap_worst_indlen(ip, temp);
1769                 xfs_bmbt_set_allf(ep, new->br_startoff,
1770                         nullstartblock((int)newlen), temp, right.br_state);
1771                 trace_xfs_bmap_post_update(ip, idx, state, _THIS_IP_);
1772
1773                 ip->i_df.if_lastex = idx;
1774                 break;
1775
1776         case 0:
1777                 /*
1778                  * New allocation is not contiguous with another
1779                  * delayed allocation.
1780                  * Insert a new entry.
1781                  */
1782                 oldlen = newlen = 0;
1783                 xfs_iext_insert(ip, idx, 1, new, state);
1784                 ip->i_df.if_lastex = idx;
1785                 break;
1786         }
1787         if (oldlen != newlen) {
1788                 ASSERT(oldlen > newlen);
1789                 xfs_icsb_modify_counters(ip->i_mount, XFS_SBS_FDBLOCKS,
1790                         (int64_t)(oldlen - newlen), rsvd);
1791                 /*
1792                  * Nothing to do for disk quota accounting here.
1793                  */
1794         }
1795         *logflagsp = 0;
1796         return 0;
1797 }
1798
1799 /*
1800  * Called by xfs_bmap_add_extent to handle cases converting a hole
1801  * to a real allocation.
1802  */
1803 STATIC int                              /* error */
1804 xfs_bmap_add_extent_hole_real(
1805         xfs_inode_t             *ip,    /* incore inode pointer */
1806         xfs_extnum_t            idx,    /* extent number to update/insert */
1807         xfs_btree_cur_t         *cur,   /* if null, not a btree */
1808         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
1809         int                     *logflagsp, /* inode logging flags */
1810         int                     whichfork) /* data or attr fork */
1811 {
1812         xfs_bmbt_rec_host_t     *ep;    /* pointer to extent entry ins. point */
1813         int                     error;  /* error return value */
1814         int                     i;      /* temp state */
1815         xfs_ifork_t             *ifp;   /* inode fork pointer */
1816         xfs_bmbt_irec_t         left;   /* left neighbor extent entry */
1817         xfs_bmbt_irec_t         right;  /* right neighbor extent entry */
1818         int                     rval=0; /* return value (logging flags) */
1819         int                     state;  /* state bits, accessed thru macros */
1820
1821         ifp = XFS_IFORK_PTR(ip, whichfork);
1822         ASSERT(idx <= ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t));
1823         ep = xfs_iext_get_ext(ifp, idx);
1824         state = 0;
1825
1826         if (whichfork == XFS_ATTR_FORK)
1827                 state |= BMAP_ATTRFORK;
1828
1829         /*
1830          * Check and set flags if this segment has a left neighbor.
1831          */
1832         if (idx > 0) {
1833                 state |= BMAP_LEFT_VALID;
1834                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, idx - 1), &left);
1835                 if (isnullstartblock(left.br_startblock))
1836                         state |= BMAP_LEFT_DELAY;
1837         }
1838
1839         /*
1840          * Check and set flags if this segment has a current value.
1841          * Not true if we're inserting into the "hole" at eof.
1842          */
1843         if (idx < ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)) {
1844                 state |= BMAP_RIGHT_VALID;
1845                 xfs_bmbt_get_all(ep, &right);
1846                 if (isnullstartblock(right.br_startblock))
1847                         state |= BMAP_RIGHT_DELAY;
1848         }
1849
1850         /*
1851          * We're inserting a real allocation between "left" and "right".
1852          * Set the contiguity flags.  Don't let extents get too large.
1853          */
1854         if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
1855             left.br_startoff + left.br_blockcount == new->br_startoff &&
1856             left.br_startblock + left.br_blockcount == new->br_startblock &&
1857             left.br_state == new->br_state &&
1858             left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
1859                 state |= BMAP_LEFT_CONTIG;
1860
1861         if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
1862             new->br_startoff + new->br_blockcount == right.br_startoff &&
1863             new->br_startblock + new->br_blockcount == right.br_startblock &&
1864             new->br_state == right.br_state &&
1865             new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
1866             (!(state & BMAP_LEFT_CONTIG) ||
1867              left.br_blockcount + new->br_blockcount +
1868              right.br_blockcount <= MAXEXTLEN))
1869                 state |= BMAP_RIGHT_CONTIG;
1870
1871         error = 0;
1872         /*
1873          * Select which case we're in here, and implement it.
1874          */
1875         switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
1876         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1877                 /*
1878                  * New allocation is contiguous with real allocations on the
1879                  * left and on the right.
1880                  * Merge all three into a single extent record.
1881                  */
1882                 trace_xfs_bmap_pre_update(ip, idx - 1, state, _THIS_IP_);
1883                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
1884                         left.br_blockcount + new->br_blockcount +
1885                         right.br_blockcount);
1886                 trace_xfs_bmap_post_update(ip, idx - 1, state, _THIS_IP_);
1887
1888                 xfs_iext_remove(ip, idx, 1, state);
1889                 ifp->if_lastex = idx - 1;
1890                 XFS_IFORK_NEXT_SET(ip, whichfork,
1891                         XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
1892                 if (cur == NULL) {
1893                         rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
1894                 } else {
1895                         rval = XFS_ILOG_CORE;
1896                         if ((error = xfs_bmbt_lookup_eq(cur,
1897                                         right.br_startoff,
1898                                         right.br_startblock,
1899                                         right.br_blockcount, &i)))
1900                                 goto done;
1901                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1902                         if ((error = xfs_btree_delete(cur, &i)))
1903                                 goto done;
1904                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1905                         if ((error = xfs_btree_decrement(cur, 0, &i)))
1906                                 goto done;
1907                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1908                         if ((error = xfs_bmbt_update(cur, left.br_startoff,
1909                                         left.br_startblock,
1910                                         left.br_blockcount +
1911                                                 new->br_blockcount +
1912                                                 right.br_blockcount,
1913                                         left.br_state)))
1914                                 goto done;
1915                 }
1916                 break;
1917
1918         case BMAP_LEFT_CONTIG:
1919                 /*
1920                  * New allocation is contiguous with a real allocation
1921                  * on the left.
1922                  * Merge the new allocation with the left neighbor.
1923                  */
1924                 trace_xfs_bmap_pre_update(ip, idx - 1, state, _THIS_IP_);
1925                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
1926                         left.br_blockcount + new->br_blockcount);
1927                 trace_xfs_bmap_post_update(ip, idx - 1, state, _THIS_IP_);
1928
1929                 ifp->if_lastex = idx - 1;
1930                 if (cur == NULL) {
1931                         rval = xfs_ilog_fext(whichfork);
1932                 } else {
1933                         rval = 0;
1934                         if ((error = xfs_bmbt_lookup_eq(cur,
1935                                         left.br_startoff,
1936                                         left.br_startblock,
1937                                         left.br_blockcount, &i)))
1938                                 goto done;
1939                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1940                         if ((error = xfs_bmbt_update(cur, left.br_startoff,
1941                                         left.br_startblock,
1942                                         left.br_blockcount +
1943                                                 new->br_blockcount,
1944                                         left.br_state)))
1945                                 goto done;
1946                 }
1947                 break;
1948
1949         case BMAP_RIGHT_CONTIG:
1950                 /*
1951                  * New allocation is contiguous with a real allocation
1952                  * on the right.
1953                  * Merge the new allocation with the right neighbor.
1954                  */
1955                 trace_xfs_bmap_pre_update(ip, idx, state, _THIS_IP_);
1956                 xfs_bmbt_set_allf(ep, new->br_startoff, new->br_startblock,
1957                         new->br_blockcount + right.br_blockcount,
1958                         right.br_state);
1959                 trace_xfs_bmap_post_update(ip, idx, state, _THIS_IP_);
1960
1961                 ifp->if_lastex = idx;
1962                 if (cur == NULL) {
1963                         rval = xfs_ilog_fext(whichfork);
1964                 } else {
1965                         rval = 0;
1966                         if ((error = xfs_bmbt_lookup_eq(cur,
1967                                         right.br_startoff,
1968                                         right.br_startblock,
1969                                         right.br_blockcount, &i)))
1970                                 goto done;
1971                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1972                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
1973                                         new->br_startblock,
1974                                         new->br_blockcount +
1975                                                 right.br_blockcount,
1976                                         right.br_state)))
1977                                 goto done;
1978                 }
1979                 break;
1980
1981         case 0:
1982                 /*
1983                  * New allocation is not contiguous with another
1984                  * real allocation.
1985                  * Insert a new entry.
1986                  */
1987                 xfs_iext_insert(ip, idx, 1, new, state);
1988                 ifp->if_lastex = idx;
1989                 XFS_IFORK_NEXT_SET(ip, whichfork,
1990                         XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
1991                 if (cur == NULL) {
1992                         rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
1993                 } else {
1994                         rval = XFS_ILOG_CORE;
1995                         if ((error = xfs_bmbt_lookup_eq(cur,
1996                                         new->br_startoff,
1997                                         new->br_startblock,
1998                                         new->br_blockcount, &i)))
1999                                 goto done;
2000                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
2001                         cur->bc_rec.b.br_state = new->br_state;
2002                         if ((error = xfs_btree_insert(cur, &i)))
2003                                 goto done;
2004                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2005                 }
2006                 break;
2007         }
2008 done:
2009         *logflagsp = rval;
2010         return error;
2011 }
2012
2013 /*
2014  * Adjust the size of the new extent based on di_extsize and rt extsize.
2015  */
2016 STATIC int
2017 xfs_bmap_extsize_align(
2018         xfs_mount_t     *mp,
2019         xfs_bmbt_irec_t *gotp,          /* next extent pointer */
2020         xfs_bmbt_irec_t *prevp,         /* previous extent pointer */
2021         xfs_extlen_t    extsz,          /* align to this extent size */
2022         int             rt,             /* is this a realtime inode? */
2023         int             eof,            /* is extent at end-of-file? */
2024         int             delay,          /* creating delalloc extent? */
2025         int             convert,        /* overwriting unwritten extent? */
2026         xfs_fileoff_t   *offp,          /* in/out: aligned offset */
2027         xfs_extlen_t    *lenp)          /* in/out: aligned length */
2028 {
2029         xfs_fileoff_t   orig_off;       /* original offset */
2030         xfs_extlen_t    orig_alen;      /* original length */
2031         xfs_fileoff_t   orig_end;       /* original off+len */
2032         xfs_fileoff_t   nexto;          /* next file offset */
2033         xfs_fileoff_t   prevo;          /* previous file offset */
2034         xfs_fileoff_t   align_off;      /* temp for offset */
2035         xfs_extlen_t    align_alen;     /* temp for length */
2036         xfs_extlen_t    temp;           /* temp for calculations */
2037
2038         if (convert)
2039                 return 0;
2040
2041         orig_off = align_off = *offp;
2042         orig_alen = align_alen = *lenp;
2043         orig_end = orig_off + orig_alen;
2044
2045         /*
2046          * If this request overlaps an existing extent, then don't
2047          * attempt to perform any additional alignment.
2048          */
2049         if (!delay && !eof &&
2050             (orig_off >= gotp->br_startoff) &&
2051             (orig_end <= gotp->br_startoff + gotp->br_blockcount)) {
2052                 return 0;
2053         }
2054
2055         /*
2056          * If the file offset is unaligned vs. the extent size
2057          * we need to align it.  This will be possible unless
2058          * the file was previously written with a kernel that didn't
2059          * perform this alignment, or if a truncate shot us in the
2060          * foot.
2061          */
2062         temp = do_mod(orig_off, extsz);
2063         if (temp) {
2064                 align_alen += temp;
2065                 align_off -= temp;
2066         }
2067         /*
2068          * Same adjustment for the end of the requested area.
2069          */
2070         if ((temp = (align_alen % extsz))) {
2071                 align_alen += extsz - temp;
2072         }
2073         /*
2074          * If the previous block overlaps with this proposed allocation
2075          * then move the start forward without adjusting the length.
2076          */
2077         if (prevp->br_startoff != NULLFILEOFF) {
2078                 if (prevp->br_startblock == HOLESTARTBLOCK)
2079                         prevo = prevp->br_startoff;
2080                 else
2081                         prevo = prevp->br_startoff + prevp->br_blockcount;
2082         } else
2083                 prevo = 0;
2084         if (align_off != orig_off && align_off < prevo)
2085                 align_off = prevo;
2086         /*
2087          * If the next block overlaps with this proposed allocation
2088          * then move the start back without adjusting the length,
2089          * but not before offset 0.
2090          * This may of course make the start overlap previous block,
2091          * and if we hit the offset 0 limit then the next block
2092          * can still overlap too.
2093          */
2094         if (!eof && gotp->br_startoff != NULLFILEOFF) {
2095                 if ((delay && gotp->br_startblock == HOLESTARTBLOCK) ||
2096                     (!delay && gotp->br_startblock == DELAYSTARTBLOCK))
2097                         nexto = gotp->br_startoff + gotp->br_blockcount;
2098                 else
2099                         nexto = gotp->br_startoff;
2100         } else
2101                 nexto = NULLFILEOFF;
2102         if (!eof &&
2103             align_off + align_alen != orig_end &&
2104             align_off + align_alen > nexto)
2105                 align_off = nexto > align_alen ? nexto - align_alen : 0;
2106         /*
2107          * If we're now overlapping the next or previous extent that
2108          * means we can't fit an extsz piece in this hole.  Just move
2109          * the start forward to the first valid spot and set
2110          * the length so we hit the end.
2111          */
2112         if (align_off != orig_off && align_off < prevo)
2113                 align_off = prevo;
2114         if (align_off + align_alen != orig_end &&
2115             align_off + align_alen > nexto &&
2116             nexto != NULLFILEOFF) {
2117                 ASSERT(nexto > prevo);
2118                 align_alen = nexto - align_off;
2119         }
2120
2121         /*
2122          * If realtime, and the result isn't a multiple of the realtime
2123          * extent size we need to remove blocks until it is.
2124          */
2125         if (rt && (temp = (align_alen % mp->m_sb.sb_rextsize))) {
2126                 /*
2127                  * We're not covering the original request, or
2128                  * we won't be able to once we fix the length.
2129                  */
2130                 if (orig_off < align_off ||
2131                     orig_end > align_off + align_alen ||
2132                     align_alen - temp < orig_alen)
2133                         return XFS_ERROR(EINVAL);
2134                 /*
2135                  * Try to fix it by moving the start up.
2136                  */
2137                 if (align_off + temp <= orig_off) {
2138                         align_alen -= temp;
2139                         align_off += temp;
2140                 }
2141                 /*
2142                  * Try to fix it by moving the end in.
2143                  */
2144                 else if (align_off + align_alen - temp >= orig_end)
2145                         align_alen -= temp;
2146                 /*
2147                  * Set the start to the minimum then trim the length.
2148                  */
2149                 else {
2150                         align_alen -= orig_off - align_off;
2151                         align_off = orig_off;
2152                         align_alen -= align_alen % mp->m_sb.sb_rextsize;
2153                 }
2154                 /*
2155                  * Result doesn't cover the request, fail it.
2156                  */
2157                 if (orig_off < align_off || orig_end > align_off + align_alen)
2158                         return XFS_ERROR(EINVAL);
2159         } else {
2160                 ASSERT(orig_off >= align_off);
2161                 ASSERT(orig_end <= align_off + align_alen);
2162         }
2163
2164 #ifdef DEBUG
2165         if (!eof && gotp->br_startoff != NULLFILEOFF)
2166                 ASSERT(align_off + align_alen <= gotp->br_startoff);
2167         if (prevp->br_startoff != NULLFILEOFF)
2168                 ASSERT(align_off >= prevp->br_startoff + prevp->br_blockcount);
2169 #endif
2170
2171         *lenp = align_alen;
2172         *offp = align_off;
2173         return 0;
2174 }
2175
2176 #define XFS_ALLOC_GAP_UNITS     4
2177
2178 STATIC void
2179 xfs_bmap_adjacent(
2180         xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
2181 {
2182         xfs_fsblock_t   adjust;         /* adjustment to block numbers */
2183         xfs_agnumber_t  fb_agno;        /* ag number of ap->firstblock */
2184         xfs_mount_t     *mp;            /* mount point structure */
2185         int             nullfb;         /* true if ap->firstblock isn't set */
2186         int             rt;             /* true if inode is realtime */
2187
2188 #define ISVALID(x,y)    \
2189         (rt ? \
2190                 (x) < mp->m_sb.sb_rblocks : \
2191                 XFS_FSB_TO_AGNO(mp, x) == XFS_FSB_TO_AGNO(mp, y) && \
2192                 XFS_FSB_TO_AGNO(mp, x) < mp->m_sb.sb_agcount && \
2193                 XFS_FSB_TO_AGBNO(mp, x) < mp->m_sb.sb_agblocks)
2194
2195         mp = ap->ip->i_mount;
2196         nullfb = ap->firstblock == NULLFSBLOCK;
2197         rt = XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata;
2198         fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, ap->firstblock);
2199         /*
2200          * If allocating at eof, and there's a previous real block,
2201          * try to use its last block as our starting point.
2202          */
2203         if (ap->eof && ap->prevp->br_startoff != NULLFILEOFF &&
2204             !isnullstartblock(ap->prevp->br_startblock) &&
2205             ISVALID(ap->prevp->br_startblock + ap->prevp->br_blockcount,
2206                     ap->prevp->br_startblock)) {
2207                 ap->rval = ap->prevp->br_startblock + ap->prevp->br_blockcount;
2208                 /*
2209                  * Adjust for the gap between prevp and us.
2210                  */
2211                 adjust = ap->off -
2212                         (ap->prevp->br_startoff + ap->prevp->br_blockcount);
2213                 if (adjust &&
2214                     ISVALID(ap->rval + adjust, ap->prevp->br_startblock))
2215                         ap->rval += adjust;
2216         }
2217         /*
2218          * If not at eof, then compare the two neighbor blocks.
2219          * Figure out whether either one gives us a good starting point,
2220          * and pick the better one.
2221          */
2222         else if (!ap->eof) {
2223                 xfs_fsblock_t   gotbno;         /* right side block number */
2224                 xfs_fsblock_t   gotdiff=0;      /* right side difference */
2225                 xfs_fsblock_t   prevbno;        /* left side block number */
2226                 xfs_fsblock_t   prevdiff=0;     /* left side difference */
2227
2228                 /*
2229                  * If there's a previous (left) block, select a requested
2230                  * start block based on it.
2231                  */
2232                 if (ap->prevp->br_startoff != NULLFILEOFF &&
2233                     !isnullstartblock(ap->prevp->br_startblock) &&
2234                     (prevbno = ap->prevp->br_startblock +
2235                                ap->prevp->br_blockcount) &&
2236                     ISVALID(prevbno, ap->prevp->br_startblock)) {
2237                         /*
2238                          * Calculate gap to end of previous block.
2239                          */
2240                         adjust = prevdiff = ap->off -
2241                                 (ap->prevp->br_startoff +
2242                                  ap->prevp->br_blockcount);
2243                         /*
2244                          * Figure the startblock based on the previous block's
2245                          * end and the gap size.
2246                          * Heuristic!
2247                          * If the gap is large relative to the piece we're
2248                          * allocating, or using it gives us an invalid block
2249                          * number, then just use the end of the previous block.
2250                          */
2251                         if (prevdiff <= XFS_ALLOC_GAP_UNITS * ap->alen &&
2252                             ISVALID(prevbno + prevdiff,
2253                                     ap->prevp->br_startblock))
2254                                 prevbno += adjust;
2255                         else
2256                                 prevdiff += adjust;
2257                         /*
2258                          * If the firstblock forbids it, can't use it,
2259                          * must use default.
2260                          */
2261                         if (!rt && !nullfb &&
2262                             XFS_FSB_TO_AGNO(mp, prevbno) != fb_agno)
2263                                 prevbno = NULLFSBLOCK;
2264                 }
2265                 /*
2266                  * No previous block or can't follow it, just default.
2267                  */
2268                 else
2269                         prevbno = NULLFSBLOCK;
2270                 /*
2271                  * If there's a following (right) block, select a requested
2272                  * start block based on it.
2273                  */
2274                 if (!isnullstartblock(ap->gotp->br_startblock)) {
2275                         /*
2276                          * Calculate gap to start of next block.
2277                          */
2278                         adjust = gotdiff = ap->gotp->br_startoff - ap->off;
2279                         /*
2280                          * Figure the startblock based on the next block's
2281                          * start and the gap size.
2282                          */
2283                         gotbno = ap->gotp->br_startblock;
2284                         /*
2285                          * Heuristic!
2286                          * If the gap is large relative to the piece we're
2287                          * allocating, or using it gives us an invalid block
2288                          * number, then just use the start of the next block
2289                          * offset by our length.
2290                          */
2291                         if (gotdiff <= XFS_ALLOC_GAP_UNITS * ap->alen &&
2292                             ISVALID(gotbno - gotdiff, gotbno))
2293                                 gotbno -= adjust;
2294                         else if (ISVALID(gotbno - ap->alen, gotbno)) {
2295                                 gotbno -= ap->alen;
2296                                 gotdiff += adjust - ap->alen;
2297                         } else
2298                                 gotdiff += adjust;
2299                         /*
2300                          * If the firstblock forbids it, can't use it,
2301                          * must use default.
2302                          */
2303                         if (!rt && !nullfb &&
2304                             XFS_FSB_TO_AGNO(mp, gotbno) != fb_agno)
2305                                 gotbno = NULLFSBLOCK;
2306                 }
2307                 /*
2308                  * No next block, just default.
2309                  */
2310                 else
2311                         gotbno = NULLFSBLOCK;
2312                 /*
2313                  * If both valid, pick the better one, else the only good
2314                  * one, else ap->rval is already set (to 0 or the inode block).
2315                  */
2316                 if (prevbno != NULLFSBLOCK && gotbno != NULLFSBLOCK)
2317                         ap->rval = prevdiff <= gotdiff ? prevbno : gotbno;
2318                 else if (prevbno != NULLFSBLOCK)
2319                         ap->rval = prevbno;
2320                 else if (gotbno != NULLFSBLOCK)
2321                         ap->rval = gotbno;
2322         }
2323 #undef ISVALID
2324 }
2325
2326 STATIC int
2327 xfs_bmap_rtalloc(
2328         xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
2329 {
2330         xfs_alloctype_t atype = 0;      /* type for allocation routines */
2331         int             error;          /* error return value */
2332         xfs_mount_t     *mp;            /* mount point structure */
2333         xfs_extlen_t    prod = 0;       /* product factor for allocators */
2334         xfs_extlen_t    ralen = 0;      /* realtime allocation length */
2335         xfs_extlen_t    align;          /* minimum allocation alignment */
2336         xfs_rtblock_t   rtb;
2337
2338         mp = ap->ip->i_mount;
2339         align = xfs_get_extsz_hint(ap->ip);
2340         prod = align / mp->m_sb.sb_rextsize;
2341         error = xfs_bmap_extsize_align(mp, ap->gotp, ap->prevp,
2342                                         align, 1, ap->eof, 0,
2343                                         ap->conv, &ap->off, &ap->alen);
2344         if (error)
2345                 return error;
2346         ASSERT(ap->alen);
2347         ASSERT(ap->alen % mp->m_sb.sb_rextsize == 0);
2348
2349         /*
2350          * If the offset & length are not perfectly aligned
2351          * then kill prod, it will just get us in trouble.
2352          */
2353         if (do_mod(ap->off, align) || ap->alen % align)
2354                 prod = 1;
2355         /*
2356          * Set ralen to be the actual requested length in rtextents.
2357          */
2358         ralen = ap->alen / mp->m_sb.sb_rextsize;
2359         /*
2360          * If the old value was close enough to MAXEXTLEN that
2361          * we rounded up to it, cut it back so it's valid again.
2362          * Note that if it's a really large request (bigger than
2363          * MAXEXTLEN), we don't hear about that number, and can't
2364          * adjust the starting point to match it.
2365          */
2366         if (ralen * mp->m_sb.sb_rextsize >= MAXEXTLEN)
2367                 ralen = MAXEXTLEN / mp->m_sb.sb_rextsize;
2368
2369         /*
2370          * Lock out other modifications to the RT bitmap inode.
2371          */
2372         xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL);
2373         xfs_trans_ijoin_ref(ap->tp, mp->m_rbmip, XFS_ILOCK_EXCL);
2374
2375         /*
2376          * If it's an allocation to an empty file at offset 0,
2377          * pick an extent that will space things out in the rt area.
2378          */
2379         if (ap->eof && ap->off == 0) {
2380                 xfs_rtblock_t uninitialized_var(rtx); /* realtime extent no */
2381
2382                 error = xfs_rtpick_extent(mp, ap->tp, ralen, &rtx);
2383                 if (error)
2384                         return error;
2385                 ap->rval = rtx * mp->m_sb.sb_rextsize;
2386         } else {
2387                 ap->rval = 0;
2388         }
2389
2390         xfs_bmap_adjacent(ap);
2391
2392         /*
2393          * Realtime allocation, done through xfs_rtallocate_extent.
2394          */
2395         atype = ap->rval == 0 ?  XFS_ALLOCTYPE_ANY_AG : XFS_ALLOCTYPE_NEAR_BNO;
2396         do_div(ap->rval, mp->m_sb.sb_rextsize);
2397         rtb = ap->rval;
2398         ap->alen = ralen;
2399         if ((error = xfs_rtallocate_extent(ap->tp, ap->rval, 1, ap->alen,
2400                                 &ralen, atype, ap->wasdel, prod, &rtb)))
2401                 return error;
2402         if (rtb == NULLFSBLOCK && prod > 1 &&
2403             (error = xfs_rtallocate_extent(ap->tp, ap->rval, 1,
2404                                            ap->alen, &ralen, atype,
2405                                            ap->wasdel, 1, &rtb)))
2406                 return error;
2407         ap->rval = rtb;
2408         if (ap->rval != NULLFSBLOCK) {
2409                 ap->rval *= mp->m_sb.sb_rextsize;
2410                 ralen *= mp->m_sb.sb_rextsize;
2411                 ap->alen = ralen;
2412                 ap->ip->i_d.di_nblocks += ralen;
2413                 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
2414                 if (ap->wasdel)
2415                         ap->ip->i_delayed_blks -= ralen;
2416                 /*
2417                  * Adjust the disk quota also. This was reserved
2418                  * earlier.
2419                  */
2420                 xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
2421                         ap->wasdel ? XFS_TRANS_DQ_DELRTBCOUNT :
2422                                         XFS_TRANS_DQ_RTBCOUNT, (long) ralen);
2423         } else {
2424                 ap->alen = 0;
2425         }
2426         return 0;
2427 }
2428
2429 STATIC int
2430 xfs_bmap_btalloc_nullfb(
2431         struct xfs_bmalloca     *ap,
2432         struct xfs_alloc_arg    *args,
2433         xfs_extlen_t            *blen)
2434 {
2435         struct xfs_mount        *mp = ap->ip->i_mount;
2436         struct xfs_perag        *pag;
2437         xfs_agnumber_t          ag, startag;
2438         int                     notinit = 0;
2439         int                     error;
2440
2441         if (ap->userdata && xfs_inode_is_filestream(ap->ip))
2442                 args->type = XFS_ALLOCTYPE_NEAR_BNO;
2443         else
2444                 args->type = XFS_ALLOCTYPE_START_BNO;
2445         args->total = ap->total;
2446
2447         /*
2448          * Search for an allocation group with a single extent large enough
2449          * for the request.  If one isn't found, then adjust the minimum
2450          * allocation size to the largest space found.
2451          */
2452         startag = ag = XFS_FSB_TO_AGNO(mp, args->fsbno);
2453         if (startag == NULLAGNUMBER)
2454                 startag = ag = 0;
2455
2456         pag = xfs_perag_get(mp, ag);
2457         while (*blen < args->maxlen) {
2458                 if (!pag->pagf_init) {
2459                         error = xfs_alloc_pagf_init(mp, args->tp, ag,
2460                                                     XFS_ALLOC_FLAG_TRYLOCK);
2461                         if (error) {
2462                                 xfs_perag_put(pag);
2463                                 return error;
2464                         }
2465                 }
2466
2467                 /*
2468                  * See xfs_alloc_fix_freelist...
2469                  */
2470                 if (pag->pagf_init) {
2471                         xfs_extlen_t    longest;
2472                         longest = xfs_alloc_longest_free_extent(mp, pag);
2473                         if (*blen < longest)
2474                                 *blen = longest;
2475                 } else
2476                         notinit = 1;
2477
2478                 if (xfs_inode_is_filestream(ap->ip)) {
2479                         if (*blen >= args->maxlen)
2480                                 break;
2481
2482                         if (ap->userdata) {
2483                                 /*
2484                                  * If startag is an invalid AG, we've
2485                                  * come here once before and
2486                                  * xfs_filestream_new_ag picked the
2487                                  * best currently available.
2488                                  *
2489                                  * Don't continue looping, since we
2490                                  * could loop forever.
2491                                  */
2492                                 if (startag == NULLAGNUMBER)
2493                                         break;
2494
2495                                 error = xfs_filestream_new_ag(ap, &ag);
2496                                 xfs_perag_put(pag);
2497                                 if (error)
2498                                         return error;
2499
2500                                 /* loop again to set 'blen'*/
2501                                 startag = NULLAGNUMBER;
2502                                 pag = xfs_perag_get(mp, ag);
2503                                 continue;
2504                         }
2505                 }
2506                 if (++ag == mp->m_sb.sb_agcount)
2507                         ag = 0;
2508                 if (ag == startag)
2509                         break;
2510                 xfs_perag_put(pag);
2511                 pag = xfs_perag_get(mp, ag);
2512         }
2513         xfs_perag_put(pag);
2514
2515         /*
2516          * Since the above loop did a BUF_TRYLOCK, it is
2517          * possible that there is space for this request.
2518          */
2519         if (notinit || *blen < ap->minlen)
2520                 args->minlen = ap->minlen;
2521         /*
2522          * If the best seen length is less than the request
2523          * length, use the best as the minimum.
2524          */
2525         else if (*blen < args->maxlen)
2526                 args->minlen = *blen;
2527         /*
2528          * Otherwise we've seen an extent as big as maxlen,
2529          * use that as the minimum.
2530          */
2531         else
2532                 args->minlen = args->maxlen;
2533
2534         /*
2535          * set the failure fallback case to look in the selected
2536          * AG as the stream may have moved.
2537          */
2538         if (xfs_inode_is_filestream(ap->ip))
2539                 ap->rval = args->fsbno = XFS_AGB_TO_FSB(mp, ag, 0);
2540
2541         return 0;
2542 }
2543
2544 STATIC int
2545 xfs_bmap_btalloc(
2546         xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
2547 {
2548         xfs_mount_t     *mp;            /* mount point structure */
2549         xfs_alloctype_t atype = 0;      /* type for allocation routines */
2550         xfs_extlen_t    align;          /* minimum allocation alignment */
2551         xfs_agnumber_t  fb_agno;        /* ag number of ap->firstblock */
2552         xfs_agnumber_t  ag;
2553         xfs_alloc_arg_t args;
2554         xfs_extlen_t    blen;
2555         xfs_extlen_t    nextminlen = 0;
2556         int             nullfb;         /* true if ap->firstblock isn't set */
2557         int             isaligned;
2558         int             tryagain;
2559         int             error;
2560
2561         mp = ap->ip->i_mount;
2562         align = ap->userdata ? xfs_get_extsz_hint(ap->ip) : 0;
2563         if (unlikely(align)) {
2564                 error = xfs_bmap_extsize_align(mp, ap->gotp, ap->prevp,
2565                                                 align, 0, ap->eof, 0, ap->conv,
2566                                                 &ap->off, &ap->alen);
2567                 ASSERT(!error);
2568                 ASSERT(ap->alen);
2569         }
2570         nullfb = ap->firstblock == NULLFSBLOCK;
2571         fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, ap->firstblock);
2572         if (nullfb) {
2573                 if (ap->userdata && xfs_inode_is_filestream(ap->ip)) {
2574                         ag = xfs_filestream_lookup_ag(ap->ip);
2575                         ag = (ag != NULLAGNUMBER) ? ag : 0;
2576                         ap->rval = XFS_AGB_TO_FSB(mp, ag, 0);
2577                 } else {
2578                         ap->rval = XFS_INO_TO_FSB(mp, ap->ip->i_ino);
2579                 }
2580         } else
2581                 ap->rval = ap->firstblock;
2582
2583         xfs_bmap_adjacent(ap);
2584
2585         /*
2586          * If allowed, use ap->rval; otherwise must use firstblock since
2587          * it's in the right allocation group.
2588          */
2589         if (nullfb || XFS_FSB_TO_AGNO(mp, ap->rval) == fb_agno)
2590                 ;
2591         else
2592                 ap->rval = ap->firstblock;
2593         /*
2594          * Normal allocation, done through xfs_alloc_vextent.
2595          */
2596         tryagain = isaligned = 0;
2597         args.tp = ap->tp;
2598         args.mp = mp;
2599         args.fsbno = ap->rval;
2600
2601         /* Trim the allocation back to the maximum an AG can fit. */
2602         args.maxlen = MIN(ap->alen, XFS_ALLOC_AG_MAX_USABLE(mp));
2603         args.firstblock = ap->firstblock;
2604         blen = 0;
2605         if (nullfb) {
2606                 error = xfs_bmap_btalloc_nullfb(ap, &args, &blen);
2607                 if (error)
2608                         return error;
2609         } else if (ap->low) {
2610                 if (xfs_inode_is_filestream(ap->ip))
2611                         args.type = XFS_ALLOCTYPE_FIRST_AG;
2612                 else
2613                         args.type = XFS_ALLOCTYPE_START_BNO;
2614                 args.total = args.minlen = ap->minlen;
2615         } else {
2616                 args.type = XFS_ALLOCTYPE_NEAR_BNO;
2617                 args.total = ap->total;
2618                 args.minlen = ap->minlen;
2619         }
2620         /* apply extent size hints if obtained earlier */
2621         if (unlikely(align)) {
2622                 args.prod = align;
2623                 if ((args.mod = (xfs_extlen_t)do_mod(ap->off, args.prod)))
2624                         args.mod = (xfs_extlen_t)(args.prod - args.mod);
2625         } else if (mp->m_sb.sb_blocksize >= PAGE_CACHE_SIZE) {
2626                 args.prod = 1;
2627                 args.mod = 0;
2628         } else {
2629                 args.prod = PAGE_CACHE_SIZE >> mp->m_sb.sb_blocklog;
2630                 if ((args.mod = (xfs_extlen_t)(do_mod(ap->off, args.prod))))
2631                         args.mod = (xfs_extlen_t)(args.prod - args.mod);
2632         }
2633         /*
2634          * If we are not low on available data blocks, and the
2635          * underlying logical volume manager is a stripe, and
2636          * the file offset is zero then try to allocate data
2637          * blocks on stripe unit boundary.
2638          * NOTE: ap->aeof is only set if the allocation length
2639          * is >= the stripe unit and the allocation offset is
2640          * at the end of file.
2641          */
2642         if (!ap->low && ap->aeof) {
2643                 if (!ap->off) {
2644                         args.alignment = mp->m_dalign;
2645                         atype = args.type;
2646                         isaligned = 1;
2647                         /*
2648                          * Adjust for alignment
2649                          */
2650                         if (blen > args.alignment && blen <= args.maxlen)
2651                                 args.minlen = blen - args.alignment;
2652                         args.minalignslop = 0;
2653                 } else {
2654                         /*
2655                          * First try an exact bno allocation.
2656                          * If it fails then do a near or start bno
2657                          * allocation with alignment turned on.
2658                          */
2659                         atype = args.type;
2660                         tryagain = 1;
2661                         args.type = XFS_ALLOCTYPE_THIS_BNO;
2662                         args.alignment = 1;
2663                         /*
2664                          * Compute the minlen+alignment for the
2665                          * next case.  Set slop so that the value
2666                          * of minlen+alignment+slop doesn't go up
2667                          * between the calls.
2668                          */
2669                         if (blen > mp->m_dalign && blen <= args.maxlen)
2670                                 nextminlen = blen - mp->m_dalign;
2671                         else
2672                                 nextminlen = args.minlen;
2673                         if (nextminlen + mp->m_dalign > args.minlen + 1)
2674                                 args.minalignslop =
2675                                         nextminlen + mp->m_dalign -
2676                                         args.minlen - 1;
2677                         else
2678                                 args.minalignslop = 0;
2679                 }
2680         } else {
2681                 args.alignment = 1;
2682                 args.minalignslop = 0;
2683         }
2684         args.minleft = ap->minleft;
2685         args.wasdel = ap->wasdel;
2686         args.isfl = 0;
2687         args.userdata = ap->userdata;
2688         if ((error = xfs_alloc_vextent(&args)))
2689                 return error;
2690         if (tryagain && args.fsbno == NULLFSBLOCK) {
2691                 /*
2692                  * Exact allocation failed. Now try with alignment
2693                  * turned on.
2694                  */
2695                 args.type = atype;
2696                 args.fsbno = ap->rval;
2697                 args.alignment = mp->m_dalign;
2698                 args.minlen = nextminlen;
2699                 args.minalignslop = 0;
2700                 isaligned = 1;
2701                 if ((error = xfs_alloc_vextent(&args)))
2702                         return error;
2703         }
2704         if (isaligned && args.fsbno == NULLFSBLOCK) {
2705                 /*
2706                  * allocation failed, so turn off alignment and
2707                  * try again.
2708                  */
2709                 args.type = atype;
2710                 args.fsbno = ap->rval;
2711                 args.alignment = 0;
2712                 if ((error = xfs_alloc_vextent(&args)))
2713                         return error;
2714         }
2715         if (args.fsbno == NULLFSBLOCK && nullfb &&
2716             args.minlen > ap->minlen) {
2717                 args.minlen = ap->minlen;
2718                 args.type = XFS_ALLOCTYPE_START_BNO;
2719                 args.fsbno = ap->rval;
2720                 if ((error = xfs_alloc_vextent(&args)))
2721                         return error;
2722         }
2723         if (args.fsbno == NULLFSBLOCK && nullfb) {
2724                 args.fsbno = 0;
2725                 args.type = XFS_ALLOCTYPE_FIRST_AG;
2726                 args.total = ap->minlen;
2727                 args.minleft = 0;
2728                 if ((error = xfs_alloc_vextent(&args)))
2729                         return error;
2730                 ap->low = 1;
2731         }
2732         if (args.fsbno != NULLFSBLOCK) {
2733                 ap->firstblock = ap->rval = args.fsbno;
2734                 ASSERT(nullfb || fb_agno == args.agno ||
2735                        (ap->low && fb_agno < args.agno));
2736                 ap->alen = args.len;
2737                 ap->ip->i_d.di_nblocks += args.len;
2738                 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
2739                 if (ap->wasdel)
2740                         ap->ip->i_delayed_blks -= args.len;
2741                 /*
2742                  * Adjust the disk quota also. This was reserved
2743                  * earlier.
2744                  */
2745                 xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
2746                         ap->wasdel ? XFS_TRANS_DQ_DELBCOUNT :
2747                                         XFS_TRANS_DQ_BCOUNT,
2748                         (long) args.len);
2749         } else {
2750                 ap->rval = NULLFSBLOCK;
2751                 ap->alen = 0;
2752         }
2753         return 0;
2754 }
2755
2756 /*
2757  * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
2758  * It figures out where to ask the underlying allocator to put the new extent.
2759  */
2760 STATIC int
2761 xfs_bmap_alloc(
2762         xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
2763 {
2764         if (XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata)
2765                 return xfs_bmap_rtalloc(ap);
2766         return xfs_bmap_btalloc(ap);
2767 }
2768
2769 /*
2770  * Transform a btree format file with only one leaf node, where the
2771  * extents list will fit in the inode, into an extents format file.
2772  * Since the file extents are already in-core, all we have to do is
2773  * give up the space for the btree root and pitch the leaf block.
2774  */
2775 STATIC int                              /* error */
2776 xfs_bmap_btree_to_extents(
2777         xfs_trans_t             *tp,    /* transaction pointer */
2778         xfs_inode_t             *ip,    /* incore inode pointer */
2779         xfs_btree_cur_t         *cur,   /* btree cursor */
2780         int                     *logflagsp, /* inode logging flags */
2781         int                     whichfork)  /* data or attr fork */
2782 {
2783         /* REFERENCED */
2784         struct xfs_btree_block  *cblock;/* child btree block */
2785         xfs_fsblock_t           cbno;   /* child block number */
2786         xfs_buf_t               *cbp;   /* child block's buffer */
2787         int                     error;  /* error return value */
2788         xfs_ifork_t             *ifp;   /* inode fork data */
2789         xfs_mount_t             *mp;    /* mount point structure */
2790         __be64                  *pp;    /* ptr to block address */
2791         struct xfs_btree_block  *rblock;/* root btree block */
2792
2793         mp = ip->i_mount;
2794         ifp = XFS_IFORK_PTR(ip, whichfork);
2795         ASSERT(ifp->if_flags & XFS_IFEXTENTS);
2796         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
2797         rblock = ifp->if_broot;
2798         ASSERT(be16_to_cpu(rblock->bb_level) == 1);
2799         ASSERT(be16_to_cpu(rblock->bb_numrecs) == 1);
2800         ASSERT(xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0) == 1);
2801         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, rblock, 1, ifp->if_broot_bytes);
2802         cbno = be64_to_cpu(*pp);
2803         *logflagsp = 0;
2804 #ifdef DEBUG
2805         if ((error = xfs_btree_check_lptr(cur, cbno, 1)))
2806                 return error;
2807 #endif
2808         if ((error = xfs_btree_read_bufl(mp, tp, cbno, 0, &cbp,
2809                         XFS_BMAP_BTREE_REF)))
2810                 return error;
2811         cblock = XFS_BUF_TO_BLOCK(cbp);
2812         if ((error = xfs_btree_check_block(cur, cblock, 0, cbp)))
2813                 return error;
2814         xfs_bmap_add_free(cbno, 1, cur->bc_private.b.flist, mp);
2815         ip->i_d.di_nblocks--;
2816         xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, -1L);
2817         xfs_trans_binval(tp, cbp);
2818         if (cur->bc_bufs[0] == cbp)
2819                 cur->bc_bufs[0] = NULL;
2820         xfs_iroot_realloc(ip, -1, whichfork);
2821         ASSERT(ifp->if_broot == NULL);
2822         ASSERT((ifp->if_flags & XFS_IFBROOT) == 0);
2823         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
2824         *logflagsp = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
2825         return 0;
2826 }
2827
2828 /*
2829  * Called by xfs_bmapi to update file extent records and the btree
2830  * after removing space (or undoing a delayed allocation).
2831  */
2832 STATIC int                              /* error */
2833 xfs_bmap_del_extent(
2834         xfs_inode_t             *ip,    /* incore inode pointer */
2835         xfs_trans_t             *tp,    /* current transaction pointer */
2836         xfs_extnum_t            idx,    /* extent number to update/delete */
2837         xfs_bmap_free_t         *flist, /* list of extents to be freed */
2838         xfs_btree_cur_t         *cur,   /* if null, not a btree */
2839         xfs_bmbt_irec_t         *del,   /* data to remove from extents */
2840         int                     *logflagsp, /* inode logging flags */
2841         int                     whichfork, /* data or attr fork */
2842         int                     rsvd)   /* OK to allocate reserved blocks */
2843 {
2844         xfs_filblks_t           da_new; /* new delay-alloc indirect blocks */
2845         xfs_filblks_t           da_old; /* old delay-alloc indirect blocks */
2846         xfs_fsblock_t           del_endblock=0; /* first block past del */
2847         xfs_fileoff_t           del_endoff;     /* first offset past del */
2848         int                     delay;  /* current block is delayed allocated */
2849         int                     do_fx;  /* free extent at end of routine */
2850         xfs_bmbt_rec_host_t     *ep;    /* current extent entry pointer */
2851         int                     error;  /* error return value */
2852         int                     flags;  /* inode logging flags */
2853         xfs_bmbt_irec_t         got;    /* current extent entry */
2854         xfs_fileoff_t           got_endoff;     /* first offset past got */
2855         int                     i;      /* temp state */
2856         xfs_ifork_t             *ifp;   /* inode fork pointer */
2857         xfs_mount_t             *mp;    /* mount structure */
2858         xfs_filblks_t           nblks;  /* quota/sb block count */
2859         xfs_bmbt_irec_t         new;    /* new record to be inserted */
2860         /* REFERENCED */
2861         uint                    qfield; /* quota field to update */
2862         xfs_filblks_t           temp;   /* for indirect length calculations */
2863         xfs_filblks_t           temp2;  /* for indirect length calculations */
2864         int                     state = 0;
2865
2866         XFS_STATS_INC(xs_del_exlist);
2867
2868         if (whichfork == XFS_ATTR_FORK)
2869                 state |= BMAP_ATTRFORK;
2870
2871         mp = ip->i_mount;
2872         ifp = XFS_IFORK_PTR(ip, whichfork);
2873         ASSERT((idx >= 0) && (idx < ifp->if_bytes /
2874                 (uint)sizeof(xfs_bmbt_rec_t)));
2875         ASSERT(del->br_blockcount > 0);
2876         ep = xfs_iext_get_ext(ifp, idx);
2877         xfs_bmbt_get_all(ep, &got);
2878         ASSERT(got.br_startoff <= del->br_startoff);
2879         del_endoff = del->br_startoff + del->br_blockcount;
2880         got_endoff = got.br_startoff + got.br_blockcount;
2881         ASSERT(got_endoff >= del_endoff);
2882         delay = isnullstartblock(got.br_startblock);
2883         ASSERT(isnullstartblock(del->br_startblock) == delay);
2884         flags = 0;
2885         qfield = 0;
2886         error = 0;
2887         /*
2888          * If deleting a real allocation, must free up the disk space.
2889          */
2890         if (!delay) {
2891                 flags = XFS_ILOG_CORE;
2892                 /*
2893                  * Realtime allocation.  Free it and record di_nblocks update.
2894                  */
2895                 if (whichfork == XFS_DATA_FORK && XFS_IS_REALTIME_INODE(ip)) {
2896                         xfs_fsblock_t   bno;
2897                         xfs_filblks_t   len;
2898
2899                         ASSERT(do_mod(del->br_blockcount,
2900                                       mp->m_sb.sb_rextsize) == 0);
2901                         ASSERT(do_mod(del->br_startblock,
2902                                       mp->m_sb.sb_rextsize) == 0);
2903                         bno = del->br_startblock;
2904                         len = del->br_blockcount;
2905                         do_div(bno, mp->m_sb.sb_rextsize);
2906                         do_div(len, mp->m_sb.sb_rextsize);
2907                         if ((error = xfs_rtfree_extent(ip->i_transp, bno,
2908                                         (xfs_extlen_t)len)))
2909                                 goto done;
2910                         do_fx = 0;
2911                         nblks = len * mp->m_sb.sb_rextsize;
2912                         qfield = XFS_TRANS_DQ_RTBCOUNT;
2913                 }
2914                 /*
2915                  * Ordinary allocation.
2916                  */
2917                 else {
2918                         do_fx = 1;
2919                         nblks = del->br_blockcount;
2920                         qfield = XFS_TRANS_DQ_BCOUNT;
2921                 }
2922                 /*
2923                  * Set up del_endblock and cur for later.
2924                  */
2925                 del_endblock = del->br_startblock + del->br_blockcount;
2926                 if (cur) {
2927                         if ((error = xfs_bmbt_lookup_eq(cur, got.br_startoff,
2928                                         got.br_startblock, got.br_blockcount,
2929                                         &i)))
2930                                 goto done;
2931                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2932                 }
2933                 da_old = da_new = 0;
2934         } else {
2935                 da_old = startblockval(got.br_startblock);
2936                 da_new = 0;
2937                 nblks = 0;
2938                 do_fx = 0;
2939         }
2940         /*
2941          * Set flag value to use in switch statement.
2942          * Left-contig is 2, right-contig is 1.
2943          */
2944         switch (((got.br_startoff == del->br_startoff) << 1) |
2945                 (got_endoff == del_endoff)) {
2946         case 3:
2947                 /*
2948                  * Matches the whole extent.  Delete the entry.
2949                  */
2950                 xfs_iext_remove(ip, idx, 1,
2951                                 whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0);
2952                 ifp->if_lastex = idx;
2953                 if (delay)
2954                         break;
2955                 XFS_IFORK_NEXT_SET(ip, whichfork,
2956                         XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
2957                 flags |= XFS_ILOG_CORE;
2958                 if (!cur) {
2959                         flags |= xfs_ilog_fext(whichfork);
2960                         break;
2961                 }
2962                 if ((error = xfs_btree_delete(cur, &i)))
2963                         goto done;
2964                 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2965                 break;
2966
2967         case 2:
2968                 /*
2969                  * Deleting the first part of the extent.
2970                  */
2971                 trace_xfs_bmap_pre_update(ip, idx, state, _THIS_IP_);
2972                 xfs_bmbt_set_startoff(ep, del_endoff);
2973                 temp = got.br_blockcount - del->br_blockcount;
2974                 xfs_bmbt_set_blockcount(ep, temp);
2975                 ifp->if_lastex = idx;
2976                 if (delay) {
2977                         temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2978                                 da_old);
2979                         xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
2980                         trace_xfs_bmap_post_update(ip, idx, state, _THIS_IP_);
2981                         da_new = temp;
2982                         break;
2983                 }
2984                 xfs_bmbt_set_startblock(ep, del_endblock);
2985                 trace_xfs_bmap_post_update(ip, idx, state, _THIS_IP_);
2986                 if (!cur) {
2987                         flags |= xfs_ilog_fext(whichfork);
2988                         break;
2989                 }
2990                 if ((error = xfs_bmbt_update(cur, del_endoff, del_endblock,
2991                                 got.br_blockcount - del->br_blockcount,
2992                                 got.br_state)))
2993                         goto done;
2994                 break;
2995
2996         case 1:
2997                 /*
2998                  * Deleting the last part of the extent.
2999                  */
3000                 temp = got.br_blockcount - del->br_blockcount;
3001                 trace_xfs_bmap_pre_update(ip, idx, state, _THIS_IP_);
3002                 xfs_bmbt_set_blockcount(ep, temp);
3003                 ifp->if_lastex = idx;
3004                 if (delay) {
3005                         temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
3006                                 da_old);
3007                         xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
3008                         trace_xfs_bmap_post_update(ip, idx, state, _THIS_IP_);
3009                         da_new = temp;
3010                         break;
3011                 }
3012                 trace_xfs_bmap_post_update(ip, idx, state, _THIS_IP_);
3013                 if (!cur) {
3014                         flags |= xfs_ilog_fext(whichfork);
3015                         break;
3016                 }
3017                 if ((error = xfs_bmbt_update(cur, got.br_startoff,
3018                                 got.br_startblock,
3019                                 got.br_blockcount - del->br_blockcount,
3020                                 got.br_state)))
3021                         goto done;
3022                 break;
3023
3024         case 0:
3025                 /*
3026                  * Deleting the middle of the extent.
3027                  */
3028                 temp = del->br_startoff - got.br_startoff;
3029                 trace_xfs_bmap_pre_update(ip, idx, state, _THIS_IP_);
3030                 xfs_bmbt_set_blockcount(ep, temp);
3031                 new.br_startoff = del_endoff;
3032                 temp2 = got_endoff - del_endoff;
3033                 new.br_blockcount = temp2;
3034                 new.br_state = got.br_state;
3035                 if (!delay) {
3036                         new.br_startblock = del_endblock;
3037                         flags |= XFS_ILOG_CORE;
3038                         if (cur) {
3039                                 if ((error = xfs_bmbt_update(cur,
3040                                                 got.br_startoff,
3041                                                 got.br_startblock, temp,
3042                                                 got.br_state)))
3043                                         goto done;
3044                                 if ((error = xfs_btree_increment(cur, 0, &i)))
3045                                         goto done;
3046                                 cur->bc_rec.b = new;
3047                                 error = xfs_btree_insert(cur, &i);
3048                                 if (error && error != ENOSPC)
3049                                         goto done;
3050                                 /*
3051                                  * If get no-space back from btree insert,
3052                                  * it tried a split, and we have a zero
3053                                  * block reservation.
3054                                  * Fix up our state and return the error.
3055                                  */
3056                                 if (error == ENOSPC) {
3057                                         /*
3058                                          * Reset the cursor, don't trust
3059                                          * it after any insert operation.
3060                                          */
3061                                         if ((error = xfs_bmbt_lookup_eq(cur,
3062                                                         got.br_startoff,
3063                                                         got.br_startblock,
3064                                                         temp, &i)))
3065                                                 goto done;
3066                                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3067                                         /*
3068                                          * Update the btree record back
3069                                          * to the original value.
3070                                          */
3071                                         if ((error = xfs_bmbt_update(cur,
3072                                                         got.br_startoff,
3073                                                         got.br_startblock,
3074                                                         got.br_blockcount,
3075                                                         got.br_state)))
3076                                                 goto done;
3077                                         /*
3078                                          * Reset the extent record back
3079                                          * to the original value.
3080                                          */
3081                                         xfs_bmbt_set_blockcount(ep,
3082                                                 got.br_blockcount);
3083                                         flags = 0;
3084                                         error = XFS_ERROR(ENOSPC);
3085                                         goto done;
3086                                 }
3087                                 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3088                         } else
3089                                 flags |= xfs_ilog_fext(whichfork);
3090                         XFS_IFORK_NEXT_SET(ip, whichfork,
3091                                 XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
3092                 } else {
3093                         ASSERT(whichfork == XFS_DATA_FORK);
3094                         temp = xfs_bmap_worst_indlen(ip, temp);
3095                         xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
3096                         temp2 = xfs_bmap_worst_indlen(ip, temp2);
3097                         new.br_startblock = nullstartblock((int)temp2);
3098                         da_new = temp + temp2;
3099                         while (da_new > da_old) {
3100                                 if (temp) {
3101                                         temp--;
3102                                         da_new--;
3103                                         xfs_bmbt_set_startblock(ep,
3104                                                 nullstartblock((int)temp));
3105                                 }
3106                                 if (da_new == da_old)
3107                                         break;
3108                                 if (temp2) {
3109                                         temp2--;
3110                                         da_new--;
3111                                         new.br_startblock =
3112                                                 nullstartblock((int)temp2);
3113                                 }
3114                         }
3115                 }
3116                 trace_xfs_bmap_post_update(ip, idx, state, _THIS_IP_);
3117                 xfs_iext_insert(ip, idx + 1, 1, &new, state);
3118                 ifp->if_lastex = idx + 1;
3119                 break;
3120         }
3121         /*
3122          * If we need to, add to list of extents to delete.
3123          */
3124         if (do_fx)
3125                 xfs_bmap_add_free(del->br_startblock, del->br_blockcount, flist,
3126                         mp);
3127         /*
3128          * Adjust inode # blocks in the file.
3129          */
3130         if (nblks)
3131                 ip->i_d.di_nblocks -= nblks;
3132         /*
3133          * Adjust quota data.
3134          */
3135         if (qfield)
3136                 xfs_trans_mod_dquot_byino(tp, ip, qfield, (long)-nblks);
3137
3138         /*
3139          * Account for change in delayed indirect blocks.
3140          * Nothing to do for disk quota accounting here.
3141          */
3142         ASSERT(da_old >= da_new);
3143         if (da_old > da_new) {
3144                 xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
3145                         (int64_t)(da_old - da_new), rsvd);
3146         }
3147 done:
3148         *logflagsp = flags;
3149         return error;
3150 }
3151
3152 /*
3153  * Remove the entry "free" from the free item list.  Prev points to the
3154  * previous entry, unless "free" is the head of the list.
3155  */
3156 STATIC void
3157 xfs_bmap_del_free(
3158         xfs_bmap_free_t         *flist, /* free item list header */
3159         xfs_bmap_free_item_t    *prev,  /* previous item on list, if any */
3160         xfs_bmap_free_item_t    *free)  /* list item to be freed */
3161 {
3162         if (prev)
3163                 prev->xbfi_next = free->xbfi_next;
3164         else
3165                 flist->xbf_first = free->xbfi_next;
3166         flist->xbf_count--;
3167         kmem_zone_free(xfs_bmap_free_item_zone, free);
3168 }
3169
3170 /*
3171  * Convert an extents-format file into a btree-format file.
3172  * The new file will have a root block (in the inode) and a single child block.
3173  */
3174 STATIC int                                      /* error */
3175 xfs_bmap_extents_to_btree(
3176         xfs_trans_t             *tp,            /* transaction pointer */
3177         xfs_inode_t             *ip,            /* incore inode pointer */
3178         xfs_fsblock_t           *firstblock,    /* first-block-allocated */
3179         xfs_bmap_free_t         *flist,         /* blocks freed in xaction */
3180         xfs_btree_cur_t         **curp,         /* cursor returned to caller */
3181         int                     wasdel,         /* converting a delayed alloc */
3182         int                     *logflagsp,     /* inode logging flags */
3183         int                     whichfork)      /* data or attr fork */
3184 {
3185         struct xfs_btree_block  *ablock;        /* allocated (child) bt block */
3186         xfs_buf_t               *abp;           /* buffer for ablock */
3187         xfs_alloc_arg_t         args;           /* allocation arguments */
3188         xfs_bmbt_rec_t          *arp;           /* child record pointer */
3189         struct xfs_btree_block  *block;         /* btree root block */
3190         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
3191         xfs_bmbt_rec_host_t     *ep;            /* extent record pointer */
3192         int                     error;          /* error return value */
3193         xfs_extnum_t            i, cnt;         /* extent record index */
3194         xfs_ifork_t             *ifp;           /* inode fork pointer */
3195         xfs_bmbt_key_t          *kp;            /* root block key pointer */
3196         xfs_mount_t             *mp;            /* mount structure */
3197         xfs_extnum_t            nextents;       /* number of file extents */
3198         xfs_bmbt_ptr_t          *pp;            /* root block address pointer */
3199
3200         ifp = XFS_IFORK_PTR(ip, whichfork);
3201         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS);
3202         ASSERT(ifp->if_ext_max ==
3203                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
3204         /*
3205          * Make space in the inode incore.
3206          */
3207         xfs_iroot_realloc(ip, 1, whichfork);
3208         ifp->if_flags |= XFS_IFBROOT;
3209
3210         /*
3211          * Fill in the root.
3212          */
3213         block = ifp->if_broot;
3214         block->bb_magic = cpu_to_be32(XFS_BMAP_MAGIC);
3215         block->bb_level = cpu_to_be16(1);
3216         block->bb_numrecs = cpu_to_be16(1);
3217         block->bb_u.l.bb_leftsib = cpu_to_be64(NULLDFSBNO);
3218         block->bb_u.l.bb_rightsib = cpu_to_be64(NULLDFSBNO);
3219
3220         /*
3221          * Need a cursor.  Can't allocate until bb_level is filled in.
3222          */
3223         mp = ip->i_mount;
3224         cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
3225         cur->bc_private.b.firstblock = *firstblock;
3226         cur->bc_private.b.flist = flist;
3227         cur->bc_private.b.flags = wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
3228         /*
3229          * Convert to a btree with two levels, one record in root.
3230          */
3231         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_BTREE);
3232         args.tp = tp;
3233         args.mp = mp;
3234         args.firstblock = *firstblock;
3235         if (*firstblock == NULLFSBLOCK) {
3236                 args.type = XFS_ALLOCTYPE_START_BNO;
3237                 args.fsbno = XFS_INO_TO_FSB(mp, ip->i_ino);
3238         } else if (flist->xbf_low) {
3239                 args.type = XFS_ALLOCTYPE_START_BNO;
3240                 args.fsbno = *firstblock;
3241         } else {
3242                 args.type = XFS_ALLOCTYPE_NEAR_BNO;
3243                 args.fsbno = *firstblock;
3244         }
3245         args.minlen = args.maxlen = args.prod = 1;
3246         args.total = args.minleft = args.alignment = args.mod = args.isfl =
3247                 args.minalignslop = 0;
3248         args.wasdel = wasdel;
3249         *logflagsp = 0;
3250         if ((error = xfs_alloc_vextent(&args))) {
3251                 xfs_iroot_realloc(ip, -1, whichfork);
3252                 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
3253                 return error;
3254         }
3255         /*
3256          * Allocation can't fail, the space was reserved.
3257          */
3258         ASSERT(args.fsbno != NULLFSBLOCK);
3259         ASSERT(*firstblock == NULLFSBLOCK ||
3260                args.agno == XFS_FSB_TO_AGNO(mp, *firstblock) ||
3261                (flist->xbf_low &&
3262                 args.agno > XFS_FSB_TO_AGNO(mp, *firstblock)));
3263         *firstblock = cur->bc_private.b.firstblock = args.fsbno;
3264         cur->bc_private.b.allocated++;
3265         ip->i_d.di_nblocks++;
3266         xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, 1L);
3267         abp = xfs_btree_get_bufl(mp, tp, args.fsbno, 0);
3268         /*
3269          * Fill in the child block.
3270          */
3271         ablock = XFS_BUF_TO_BLOCK(abp);
3272         ablock->bb_magic = cpu_to_be32(XFS_BMAP_MAGIC);
3273         ablock->bb_level = 0;
3274         ablock->bb_u.l.bb_leftsib = cpu_to_be64(NULLDFSBNO);
3275         ablock->bb_u.l.bb_rightsib = cpu_to_be64(NULLDFSBNO);
3276         arp = XFS_BMBT_REC_ADDR(mp, ablock, 1);
3277         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3278         for (cnt = i = 0; i < nextents; i++) {
3279                 ep = xfs_iext_get_ext(ifp, i);
3280                 if (!isnullstartblock(xfs_bmbt_get_startblock(ep))) {
3281                         arp->l0 = cpu_to_be64(ep->l0);
3282                         arp->l1 = cpu_to_be64(ep->l1);
3283                         arp++; cnt++;
3284                 }
3285         }
3286         ASSERT(cnt == XFS_IFORK_NEXTENTS(ip, whichfork));
3287         xfs_btree_set_numrecs(ablock, cnt);
3288
3289         /*
3290          * Fill in the root key and pointer.
3291          */
3292         kp = XFS_BMBT_KEY_ADDR(mp, block, 1);
3293         arp = XFS_BMBT_REC_ADDR(mp, ablock, 1);
3294         kp->br_startoff = cpu_to_be64(xfs_bmbt_disk_get_startoff(arp));
3295         pp = XFS_BMBT_PTR_ADDR(mp, block, 1, xfs_bmbt_get_maxrecs(cur,
3296                                                 be16_to_cpu(block->bb_level)));
3297         *pp = cpu_to_be64(args.fsbno);
3298
3299         /*
3300          * Do all this logging at the end so that
3301          * the root is at the right level.
3302          */
3303         xfs_btree_log_block(cur, abp, XFS_BB_ALL_BITS);
3304         xfs_btree_log_recs(cur, abp, 1, be16_to_cpu(ablock->bb_numrecs));
3305         ASSERT(*curp == NULL);
3306         *curp = cur;
3307         *logflagsp = XFS_ILOG_CORE | xfs_ilog_fbroot(whichfork);
3308         return 0;
3309 }
3310
3311 /*
3312  * Calculate the default attribute fork offset for newly created inodes.
3313  */
3314 uint
3315 xfs_default_attroffset(
3316         struct xfs_inode        *ip)
3317 {
3318         struct xfs_mount        *mp = ip->i_mount;
3319         uint                    offset;
3320
3321         if (mp->m_sb.sb_inodesize == 256) {
3322                 offset = XFS_LITINO(mp) -
3323                                 XFS_BMDR_SPACE_CALC(MINABTPTRS);
3324         } else {
3325                 offset = XFS_BMDR_SPACE_CALC(6 * MINABTPTRS);
3326         }
3327
3328         ASSERT(offset < XFS_LITINO(mp));
3329         return offset;
3330 }
3331
3332 /*
3333  * Helper routine to reset inode di_forkoff field when switching
3334  * attribute fork from local to extent format - we reset it where
3335  * possible to make space available for inline data fork extents.
3336  */
3337 STATIC void
3338 xfs_bmap_forkoff_reset(
3339         xfs_mount_t     *mp,
3340         xfs_inode_t     *ip,
3341         int             whichfork)
3342 {
3343         if (whichfork == XFS_ATTR_FORK &&
3344             ip->i_d.di_format != XFS_DINODE_FMT_DEV &&
3345             ip->i_d.di_format != XFS_DINODE_FMT_UUID &&
3346             ip->i_d.di_format != XFS_DINODE_FMT_BTREE) {
3347                 uint    dfl_forkoff = xfs_default_attroffset(ip) >> 3;
3348
3349                 if (dfl_forkoff > ip->i_d.di_forkoff) {
3350                         ip->i_d.di_forkoff = dfl_forkoff;
3351                         ip->i_df.if_ext_max =
3352                                 XFS_IFORK_DSIZE(ip) / sizeof(xfs_bmbt_rec_t);
3353                         ip->i_afp->if_ext_max =
3354                                 XFS_IFORK_ASIZE(ip) / sizeof(xfs_bmbt_rec_t);
3355                 }
3356         }
3357 }
3358
3359 /*
3360  * Convert a local file to an extents file.
3361  * This code is out of bounds for data forks of regular files,
3362  * since the file data needs to get logged so things will stay consistent.
3363  * (The bmap-level manipulations are ok, though).
3364  */
3365 STATIC int                              /* error */
3366 xfs_bmap_local_to_extents(
3367         xfs_trans_t     *tp,            /* transaction pointer */
3368         xfs_inode_t     *ip,            /* incore inode pointer */
3369         xfs_fsblock_t   *firstblock,    /* first block allocated in xaction */
3370         xfs_extlen_t    total,          /* total blocks needed by transaction */
3371         int             *logflagsp,     /* inode logging flags */
3372         int             whichfork)      /* data or attr fork */
3373 {
3374         int             error;          /* error return value */
3375         int             flags;          /* logging flags returned */
3376         xfs_ifork_t     *ifp;           /* inode fork pointer */
3377
3378         /*
3379          * We don't want to deal with the case of keeping inode data inline yet.
3380          * So sending the data fork of a regular inode is invalid.
3381          */
3382         ASSERT(!((ip->i_d.di_mode & S_IFMT) == S_IFREG &&
3383                  whichfork == XFS_DATA_FORK));
3384         ifp = XFS_IFORK_PTR(ip, whichfork);
3385         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
3386         flags = 0;
3387         error = 0;
3388         if (ifp->if_bytes) {
3389                 xfs_alloc_arg_t args;   /* allocation arguments */
3390                 xfs_buf_t       *bp;    /* buffer for extent block */
3391                 xfs_bmbt_rec_host_t *ep;/* extent record pointer */
3392
3393                 args.tp = tp;
3394                 args.mp = ip->i_mount;
3395                 args.firstblock = *firstblock;
3396                 ASSERT((ifp->if_flags &
3397                         (XFS_IFINLINE|XFS_IFEXTENTS|XFS_IFEXTIREC)) == XFS_IFINLINE);
3398                 /*
3399                  * Allocate a block.  We know we need only one, since the
3400                  * file currently fits in an inode.
3401                  */
3402                 if (*firstblock == NULLFSBLOCK) {
3403                         args.fsbno = XFS_INO_TO_FSB(args.mp, ip->i_ino);
3404                         args.type = XFS_ALLOCTYPE_START_BNO;
3405                 } else {
3406                         args.fsbno = *firstblock;
3407                         args.type = XFS_ALLOCTYPE_NEAR_BNO;
3408                 }
3409                 args.total = total;
3410                 args.mod = args.minleft = args.alignment = args.wasdel =
3411                         args.isfl = args.minalignslop = 0;
3412                 args.minlen = args.maxlen = args.prod = 1;
3413                 if ((error = xfs_alloc_vextent(&args)))
3414                         goto done;
3415                 /*
3416                  * Can't fail, the space was reserved.
3417                  */
3418                 ASSERT(args.fsbno != NULLFSBLOCK);
3419                 ASSERT(args.len == 1);
3420                 *firstblock = args.fsbno;
3421                 bp = xfs_btree_get_bufl(args.mp, tp, args.fsbno, 0);
3422                 memcpy((char *)XFS_BUF_PTR(bp), ifp->if_u1.if_data,
3423                         ifp->if_bytes);
3424                 xfs_trans_log_buf(tp, bp, 0, ifp->if_bytes - 1);
3425                 xfs_bmap_forkoff_reset(args.mp, ip, whichfork);
3426                 xfs_idata_realloc(ip, -ifp->if_bytes, whichfork);
3427                 xfs_iext_add(ifp, 0, 1);
3428                 ep = xfs_iext_get_ext(ifp, 0);
3429                 xfs_bmbt_set_allf(ep, 0, args.fsbno, 1, XFS_EXT_NORM);
3430                 trace_xfs_bmap_post_update(ip, 0,
3431                                 whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0,
3432                                 _THIS_IP_);
3433                 XFS_IFORK_NEXT_SET(ip, whichfork, 1);
3434                 ip->i_d.di_nblocks = 1;
3435                 xfs_trans_mod_dquot_byino(tp, ip,
3436                         XFS_TRANS_DQ_BCOUNT, 1L);
3437                 flags |= xfs_ilog_fext(whichfork);
3438         } else {
3439                 ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) == 0);
3440                 xfs_bmap_forkoff_reset(ip->i_mount, ip, whichfork);
3441         }
3442         ifp->if_flags &= ~XFS_IFINLINE;
3443         ifp->if_flags |= XFS_IFEXTENTS;
3444         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
3445         flags |= XFS_ILOG_CORE;
3446 done:
3447         *logflagsp = flags;
3448         return error;
3449 }
3450
3451 /*
3452  * Search the extent records for the entry containing block bno.
3453  * If bno lies in a hole, point to the next entry.  If bno lies
3454  * past eof, *eofp will be set, and *prevp will contain the last
3455  * entry (null if none).  Else, *lastxp will be set to the index
3456  * of the found entry; *gotp will contain the entry.
3457  */
3458 STATIC xfs_bmbt_rec_host_t *            /* pointer to found extent entry */
3459 xfs_bmap_search_multi_extents(
3460         xfs_ifork_t     *ifp,           /* inode fork pointer */
3461         xfs_fileoff_t   bno,            /* block number searched for */
3462         int             *eofp,          /* out: end of file found */
3463         xfs_extnum_t    *lastxp,        /* out: last extent index */
3464         xfs_bmbt_irec_t *gotp,          /* out: extent entry found */
3465         xfs_bmbt_irec_t *prevp)         /* out: previous extent entry found */
3466 {
3467         xfs_bmbt_rec_host_t *ep;                /* extent record pointer */
3468         xfs_extnum_t    lastx;          /* last extent index */
3469
3470         /*
3471          * Initialize the extent entry structure to catch access to
3472          * uninitialized br_startblock field.
3473          */
3474         gotp->br_startoff = 0xffa5a5a5a5a5a5a5LL;
3475         gotp->br_blockcount = 0xa55a5a5a5a5a5a5aLL;
3476         gotp->br_state = XFS_EXT_INVALID;
3477 #if XFS_BIG_BLKNOS
3478         gotp->br_startblock = 0xffffa5a5a5a5a5a5LL;
3479 #else
3480         gotp->br_startblock = 0xffffa5a5;
3481 #endif
3482         prevp->br_startoff = NULLFILEOFF;
3483
3484         ep = xfs_iext_bno_to_ext(ifp, bno, &lastx);
3485         if (lastx > 0) {
3486                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx - 1), prevp);
3487         }
3488         if (lastx < (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t))) {
3489                 xfs_bmbt_get_all(ep, gotp);
3490                 *eofp = 0;
3491         } else {
3492                 if (lastx > 0) {
3493                         *gotp = *prevp;
3494                 }
3495                 *eofp = 1;
3496                 ep = NULL;
3497         }
3498         *lastxp = lastx;
3499         return ep;
3500 }
3501
3502 /*
3503  * Search the extents list for the inode, for the extent containing bno.
3504  * If bno lies in a hole, point to the next entry.  If bno lies past eof,
3505  * *eofp will be set, and *prevp will contain the last entry (null if none).
3506  * Else, *lastxp will be set to the index of the found
3507  * entry; *gotp will contain the entry.
3508  */
3509 STATIC xfs_bmbt_rec_host_t *                 /* pointer to found extent entry */
3510 xfs_bmap_search_extents(
3511         xfs_inode_t     *ip,            /* incore inode pointer */
3512         xfs_fileoff_t   bno,            /* block number searched for */
3513         int             fork,           /* data or attr fork */
3514         int             *eofp,          /* out: end of file found */
3515         xfs_extnum_t    *lastxp,        /* out: last extent index */
3516         xfs_bmbt_irec_t *gotp,          /* out: extent entry found */
3517         xfs_bmbt_irec_t *prevp)         /* out: previous extent entry found */
3518 {
3519         xfs_ifork_t     *ifp;           /* inode fork pointer */
3520         xfs_bmbt_rec_host_t  *ep;            /* extent record pointer */
3521
3522         XFS_STATS_INC(xs_look_exlist);
3523         ifp = XFS_IFORK_PTR(ip, fork);
3524
3525         ep = xfs_bmap_search_multi_extents(ifp, bno, eofp, lastxp, gotp, prevp);
3526
3527         if (unlikely(!(gotp->br_startblock) && (*lastxp != NULLEXTNUM) &&
3528                      !(XFS_IS_REALTIME_INODE(ip) && fork == XFS_DATA_FORK))) {
3529                 xfs_alert_tag(ip->i_mount, XFS_PTAG_FSBLOCK_ZERO,
3530                                 "Access to block zero in inode %llu "
3531                                 "start_block: %llx start_off: %llx "
3532                                 "blkcnt: %llx extent-state: %x lastx: %x\n",
3533                         (unsigned long long)ip->i_ino,
3534                         (unsigned long long)gotp->br_startblock,
3535                         (unsigned long long)gotp->br_startoff,
3536                         (unsigned long long)gotp->br_blockcount,
3537                         gotp->br_state, *lastxp);
3538                 *lastxp = NULLEXTNUM;
3539                 *eofp = 1;
3540                 return NULL;
3541         }
3542         return ep;
3543 }
3544
3545 /*
3546  * Compute the worst-case number of indirect blocks that will be used
3547  * for ip's delayed extent of length "len".
3548  */
3549 STATIC xfs_filblks_t
3550 xfs_bmap_worst_indlen(
3551         xfs_inode_t     *ip,            /* incore inode pointer */
3552         xfs_filblks_t   len)            /* delayed extent length */
3553 {
3554         int             level;          /* btree level number */
3555         int             maxrecs;        /* maximum record count at this level */
3556         xfs_mount_t     *mp;            /* mount structure */
3557         xfs_filblks_t   rval;           /* return value */
3558
3559         mp = ip->i_mount;
3560         maxrecs = mp->m_bmap_dmxr[0];
3561         for (level = 0, rval = 0;
3562              level < XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK);
3563              level++) {
3564                 len += maxrecs - 1;
3565                 do_div(len, maxrecs);
3566                 rval += len;
3567                 if (len == 1)
3568                         return rval + XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK) -
3569                                 level - 1;
3570                 if (level == 0)
3571                         maxrecs = mp->m_bmap_dmxr[1];
3572         }
3573         return rval;
3574 }
3575
3576 /*
3577  * Convert inode from non-attributed to attributed.
3578  * Must not be in a transaction, ip must not be locked.
3579  */
3580 int                                             /* error code */
3581 xfs_bmap_add_attrfork(
3582         xfs_inode_t             *ip,            /* incore inode pointer */
3583         int                     size,           /* space new attribute needs */
3584         int                     rsvd)           /* xact may use reserved blks */
3585 {
3586         xfs_fsblock_t           firstblock;     /* 1st block/ag allocated */
3587         xfs_bmap_free_t         flist;          /* freed extent records */
3588         xfs_mount_t             *mp;            /* mount structure */
3589         xfs_trans_t             *tp;            /* transaction pointer */
3590         int                     blks;           /* space reservation */
3591         int                     version = 1;    /* superblock attr version */
3592         int                     committed;      /* xaction was committed */
3593         int                     logflags;       /* logging flags */
3594         int                     error;          /* error return value */
3595
3596         ASSERT(XFS_IFORK_Q(ip) == 0);
3597         ASSERT(ip->i_df.if_ext_max ==
3598                XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t));
3599
3600         mp = ip->i_mount;
3601         ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
3602         tp = xfs_trans_alloc(mp, XFS_TRANS_ADDAFORK);
3603         blks = XFS_ADDAFORK_SPACE_RES(mp);
3604         if (rsvd)
3605                 tp->t_flags |= XFS_TRANS_RESERVE;
3606         if ((error = xfs_trans_reserve(tp, blks, XFS_ADDAFORK_LOG_RES(mp), 0,
3607                         XFS_TRANS_PERM_LOG_RES, XFS_ADDAFORK_LOG_COUNT)))
3608                 goto error0;
3609         xfs_ilock(ip, XFS_ILOCK_EXCL);
3610         error = xfs_trans_reserve_quota_nblks(tp, ip, blks, 0, rsvd ?
3611                         XFS_QMOPT_RES_REGBLKS | XFS_QMOPT_FORCE_RES :
3612                         XFS_QMOPT_RES_REGBLKS);
3613         if (error) {
3614                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3615                 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES);
3616                 return error;
3617         }
3618         if (XFS_IFORK_Q(ip))
3619                 goto error1;
3620         if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS) {
3621                 /*
3622                  * For inodes coming from pre-6.2 filesystems.
3623                  */
3624                 ASSERT(ip->i_d.di_aformat == 0);
3625                 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
3626         }
3627         ASSERT(ip->i_d.di_anextents == 0);
3628
3629         xfs_trans_ijoin_ref(tp, ip, XFS_ILOCK_EXCL);
3630         xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
3631
3632         switch (ip->i_d.di_format) {
3633         case XFS_DINODE_FMT_DEV:
3634                 ip->i_d.di_forkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
3635                 break;
3636         case XFS_DINODE_FMT_UUID:
3637                 ip->i_d.di_forkoff = roundup(sizeof(uuid_t), 8) >> 3;
3638                 break;
3639         case XFS_DINODE_FMT_LOCAL:
3640         case XFS_DINODE_FMT_EXTENTS:
3641         case XFS_DINODE_FMT_BTREE:
3642                 ip->i_d.di_forkoff = xfs_attr_shortform_bytesfit(ip, size);
3643                 if (!ip->i_d.di_forkoff)
3644                         ip->i_d.di_forkoff = xfs_default_attroffset(ip) >> 3;
3645                 else if (mp->m_flags & XFS_MOUNT_ATTR2)
3646                         version = 2;
3647                 break;
3648         default:
3649                 ASSERT(0);
3650                 error = XFS_ERROR(EINVAL);
3651                 goto error1;
3652         }
3653         ip->i_df.if_ext_max =
3654                 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
3655         ASSERT(ip->i_afp == NULL);
3656         ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP);
3657         ip->i_afp->if_ext_max =
3658                 XFS_IFORK_ASIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
3659         ip->i_afp->if_flags = XFS_IFEXTENTS;
3660         logflags = 0;
3661         xfs_bmap_init(&flist, &firstblock);
3662         switch (ip->i_d.di_format) {
3663         case XFS_DINODE_FMT_LOCAL:
3664                 error = xfs_bmap_add_attrfork_local(tp, ip, &firstblock, &flist,
3665                         &logflags);
3666                 break;
3667         case XFS_DINODE_FMT_EXTENTS:
3668                 error = xfs_bmap_add_attrfork_extents(tp, ip, &firstblock,
3669                         &flist, &logflags);
3670                 break;
3671         case XFS_DINODE_FMT_BTREE:
3672                 error = xfs_bmap_add_attrfork_btree(tp, ip, &firstblock, &flist,
3673                         &logflags);
3674                 break;
3675         default:
3676                 error = 0;
3677                 break;
3678         }
3679         if (logflags)
3680                 xfs_trans_log_inode(tp, ip, logflags);
3681         if (error)
3682                 goto error2;
3683         if (!xfs_sb_version_hasattr(&mp->m_sb) ||
3684            (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2)) {
3685                 __int64_t sbfields = 0;
3686
3687                 spin_lock(&mp->m_sb_lock);
3688                 if (!xfs_sb_version_hasattr(&mp->m_sb)) {
3689                         xfs_sb_version_addattr(&mp->m_sb);
3690                         sbfields |= XFS_SB_VERSIONNUM;
3691                 }
3692                 if (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2) {
3693                         xfs_sb_version_addattr2(&mp->m_sb);
3694                         sbfields |= (XFS_SB_VERSIONNUM | XFS_SB_FEATURES2);
3695                 }
3696                 if (sbfields) {
3697                         spin_unlock(&mp->m_sb_lock);
3698                         xfs_mod_sb(tp, sbfields);
3699                 } else
3700                         spin_unlock(&mp->m_sb_lock);
3701         }
3702         if ((error = xfs_bmap_finish(&tp, &flist, &committed)))
3703                 goto error2;
3704         error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
3705         ASSERT(ip->i_df.if_ext_max ==
3706                XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t));
3707         return error;
3708 error2:
3709         xfs_bmap_cancel(&flist);
3710 error1:
3711         xfs_iunlock(ip, XFS_ILOCK_EXCL);
3712 error0:
3713         xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
3714         ASSERT(ip->i_df.if_ext_max ==
3715                XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t));
3716         return error;
3717 }
3718
3719 /*
3720  * Add the extent to the list of extents to be free at transaction end.
3721  * The list is maintained sorted (by block number).
3722  */
3723 /* ARGSUSED */
3724 void
3725 xfs_bmap_add_free(
3726         xfs_fsblock_t           bno,            /* fs block number of extent */
3727         xfs_filblks_t           len,            /* length of extent */
3728         xfs_bmap_free_t         *flist,         /* list of extents */
3729         xfs_mount_t             *mp)            /* mount point structure */
3730 {
3731         xfs_bmap_free_item_t    *cur;           /* current (next) element */
3732         xfs_bmap_free_item_t    *new;           /* new element */
3733         xfs_bmap_free_item_t    *prev;          /* previous element */
3734 #ifdef DEBUG
3735         xfs_agnumber_t          agno;
3736         xfs_agblock_t           agbno;
3737
3738         ASSERT(bno != NULLFSBLOCK);
3739         ASSERT(len > 0);
3740         ASSERT(len <= MAXEXTLEN);
3741         ASSERT(!isnullstartblock(bno));
3742         agno = XFS_FSB_TO_AGNO(mp, bno);
3743         agbno = XFS_FSB_TO_AGBNO(mp, bno);
3744         ASSERT(agno < mp->m_sb.sb_agcount);
3745         ASSERT(agbno < mp->m_sb.sb_agblocks);
3746         ASSERT(len < mp->m_sb.sb_agblocks);
3747         ASSERT(agbno + len <= mp->m_sb.sb_agblocks);
3748 #endif
3749         ASSERT(xfs_bmap_free_item_zone != NULL);
3750         new = kmem_zone_alloc(xfs_bmap_free_item_zone, KM_SLEEP);
3751         new->xbfi_startblock = bno;
3752         new->xbfi_blockcount = (xfs_extlen_t)len;
3753         for (prev = NULL, cur = flist->xbf_first;
3754              cur != NULL;
3755              prev = cur, cur = cur->xbfi_next) {
3756                 if (cur->xbfi_startblock >= bno)
3757                         break;
3758         }
3759         if (prev)
3760                 prev->xbfi_next = new;
3761         else
3762                 flist->xbf_first = new;
3763         new->xbfi_next = cur;
3764         flist->xbf_count++;
3765 }
3766
3767 /*
3768  * Compute and fill in the value of the maximum depth of a bmap btree
3769  * in this filesystem.  Done once, during mount.
3770  */
3771 void
3772 xfs_bmap_compute_maxlevels(
3773         xfs_mount_t     *mp,            /* file system mount structure */
3774         int             whichfork)      /* data or attr fork */
3775 {
3776         int             level;          /* btree level */
3777         uint            maxblocks;      /* max blocks at this level */
3778         uint            maxleafents;    /* max leaf entries possible */
3779         int             maxrootrecs;    /* max records in root block */
3780         int             minleafrecs;    /* min records in leaf block */
3781         int             minnoderecs;    /* min records in node block */
3782         int             sz;             /* root block size */
3783
3784         /*
3785          * The maximum number of extents in a file, hence the maximum
3786          * number of leaf entries, is controlled by the type of di_nextents
3787          * (a signed 32-bit number, xfs_extnum_t), or by di_anextents
3788          * (a signed 16-bit number, xfs_aextnum_t).
3789          *
3790          * Note that we can no longer assume that if we are in ATTR1 that
3791          * the fork offset of all the inodes will be
3792          * (xfs_default_attroffset(ip) >> 3) because we could have mounted
3793          * with ATTR2 and then mounted back with ATTR1, keeping the
3794          * di_forkoff's fixed but probably at various positions. Therefore,
3795          * for both ATTR1 and ATTR2 we have to assume the worst case scenario
3796          * of a minimum size available.
3797          */
3798         if (whichfork == XFS_DATA_FORK) {
3799                 maxleafents = MAXEXTNUM;
3800                 sz = XFS_BMDR_SPACE_CALC(MINDBTPTRS);
3801         } else {
3802                 maxleafents = MAXAEXTNUM;
3803                 sz = XFS_BMDR_SPACE_CALC(MINABTPTRS);
3804         }
3805         maxrootrecs = xfs_bmdr_maxrecs(mp, sz, 0);
3806         minleafrecs = mp->m_bmap_dmnr[0];
3807         minnoderecs = mp->m_bmap_dmnr[1];
3808         maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs;
3809         for (level = 1; maxblocks > 1; level++) {
3810                 if (maxblocks <= maxrootrecs)
3811                         maxblocks = 1;
3812                 else
3813                         maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs;
3814         }
3815         mp->m_bm_maxlevels[whichfork] = level;
3816 }
3817
3818 /*
3819  * Routine to be called at transaction's end by xfs_bmapi, xfs_bunmapi
3820  * caller.  Frees all the extents that need freeing, which must be done
3821  * last due to locking considerations.  We never free any extents in
3822  * the first transaction.  This is to allow the caller to make the first
3823  * transaction a synchronous one so that the pointers to the data being
3824  * broken in this transaction will be permanent before the data is actually
3825  * freed.  This is necessary to prevent blocks from being reallocated
3826  * and written to before the free and reallocation are actually permanent.
3827  * We do not just make the first transaction synchronous here, because
3828  * there are more efficient ways to gain the same protection in some cases
3829  * (see the file truncation code).
3830  *
3831  * Return 1 if the given transaction was committed and a new one
3832  * started, and 0 otherwise in the committed parameter.
3833  */
3834 /*ARGSUSED*/
3835 int                                             /* error */
3836 xfs_bmap_finish(
3837         xfs_trans_t             **tp,           /* transaction pointer addr */
3838         xfs_bmap_free_t         *flist,         /* i/o: list extents to free */
3839         int                     *committed)     /* xact committed or not */
3840 {
3841         xfs_efd_log_item_t      *efd;           /* extent free data */
3842         xfs_efi_log_item_t      *efi;           /* extent free intention */
3843         int                     error;          /* error return value */
3844         xfs_bmap_free_item_t    *free;          /* free extent item */
3845         unsigned int            logres;         /* new log reservation */
3846         unsigned int            logcount;       /* new log count */
3847         xfs_mount_t             *mp;            /* filesystem mount structure */
3848         xfs_bmap_free_item_t    *next;          /* next item on free list */
3849         xfs_trans_t             *ntp;           /* new transaction pointer */
3850
3851         ASSERT((*tp)->t_flags & XFS_TRANS_PERM_LOG_RES);
3852         if (flist->xbf_count == 0) {
3853                 *committed = 0;
3854                 return 0;
3855         }
3856         ntp = *tp;
3857         efi = xfs_trans_get_efi(ntp, flist->xbf_count);
3858         for (free = flist->xbf_first; free; free = free->xbfi_next)
3859                 xfs_trans_log_efi_extent(ntp, efi, free->xbfi_startblock,
3860                         free->xbfi_blockcount);
3861         logres = ntp->t_log_res;
3862         logcount = ntp->t_log_count;
3863         ntp = xfs_trans_dup(*tp);
3864         error = xfs_trans_commit(*tp, 0);
3865         *tp = ntp;
3866         *committed = 1;
3867         /*
3868          * We have a new transaction, so we should return committed=1,
3869          * even though we're returning an error.
3870          */
3871         if (error)
3872                 return error;
3873
3874         /*
3875          * transaction commit worked ok so we can drop the extra ticket
3876          * reference that we gained in xfs_trans_dup()
3877          */
3878         xfs_log_ticket_put(ntp->t_ticket);
3879
3880         if ((error = xfs_trans_reserve(ntp, 0, logres, 0, XFS_TRANS_PERM_LOG_RES,
3881                         logcount)))
3882                 return error;
3883         efd = xfs_trans_get_efd(ntp, efi, flist->xbf_count);
3884         for (free = flist->xbf_first; free != NULL; free = next) {
3885                 next = free->xbfi_next;
3886                 if ((error = xfs_free_extent(ntp, free->xbfi_startblock,
3887                                 free->xbfi_blockcount))) {
3888                         /*
3889                          * The bmap free list will be cleaned up at a
3890                          * higher level.  The EFI will be canceled when
3891                          * this transaction is aborted.
3892                          * Need to force shutdown here to make sure it
3893                          * happens, since this transaction may not be
3894                          * dirty yet.
3895                          */
3896                         mp = ntp->t_mountp;
3897                         if (!XFS_FORCED_SHUTDOWN(mp))
3898                                 xfs_force_shutdown(mp,
3899                                                    (error == EFSCORRUPTED) ?
3900                                                    SHUTDOWN_CORRUPT_INCORE :
3901                                                    SHUTDOWN_META_IO_ERROR);
3902                         return error;
3903                 }
3904                 xfs_trans_log_efd_extent(ntp, efd, free->xbfi_startblock,
3905                         free->xbfi_blockcount);
3906                 xfs_bmap_del_free(flist, NULL, free);
3907         }
3908         return 0;
3909 }
3910
3911 /*
3912  * Free up any items left in the list.
3913  */
3914 void
3915 xfs_bmap_cancel(
3916         xfs_bmap_free_t         *flist) /* list of bmap_free_items */
3917 {
3918         xfs_bmap_free_item_t    *free;  /* free list item */
3919         xfs_bmap_free_item_t    *next;
3920
3921         if (flist->xbf_count == 0)
3922                 return;
3923         ASSERT(flist->xbf_first != NULL);
3924         for (free = flist->xbf_first; free; free = next) {
3925                 next = free->xbfi_next;
3926                 xfs_bmap_del_free(flist, NULL, free);
3927         }
3928         ASSERT(flist->xbf_count == 0);
3929 }
3930
3931 /*
3932  * Returns the file-relative block number of the first unused block(s)
3933  * in the file with at least "len" logically contiguous blocks free.
3934  * This is the lowest-address hole if the file has holes, else the first block
3935  * past the end of file.
3936  * Return 0 if the file is currently local (in-inode).
3937  */
3938 int                                             /* error */
3939 xfs_bmap_first_unused(
3940         xfs_trans_t     *tp,                    /* transaction pointer */
3941         xfs_inode_t     *ip,                    /* incore inode */
3942         xfs_extlen_t    len,                    /* size of hole to find */
3943         xfs_fileoff_t   *first_unused,          /* unused block */
3944         int             whichfork)              /* data or attr fork */
3945 {
3946         int             error;                  /* error return value */
3947         int             idx;                    /* extent record index */
3948         xfs_ifork_t     *ifp;                   /* inode fork pointer */
3949         xfs_fileoff_t   lastaddr;               /* last block number seen */
3950         xfs_fileoff_t   lowest;                 /* lowest useful block */
3951         xfs_fileoff_t   max;                    /* starting useful block */
3952         xfs_fileoff_t   off;                    /* offset for this block */
3953         xfs_extnum_t    nextents;               /* number of extent entries */
3954
3955         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE ||
3956                XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ||
3957                XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
3958         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
3959                 *first_unused = 0;
3960                 return 0;
3961         }
3962         ifp = XFS_IFORK_PTR(ip, whichfork);
3963         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
3964             (error = xfs_iread_extents(tp, ip, whichfork)))
3965                 return error;
3966         lowest = *first_unused;
3967         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3968         for (idx = 0, lastaddr = 0, max = lowest; idx < nextents; idx++) {
3969                 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, idx);
3970                 off = xfs_bmbt_get_startoff(ep);
3971                 /*
3972                  * See if the hole before this extent will work.
3973                  */
3974                 if (off >= lowest + len && off - max >= len) {
3975                         *first_unused = max;
3976                         return 0;
3977                 }
3978                 lastaddr = off + xfs_bmbt_get_blockcount(ep);
3979                 max = XFS_FILEOFF_MAX(lastaddr, lowest);
3980         }
3981         *first_unused = max;
3982         return 0;
3983 }
3984
3985 /*
3986  * Returns the file-relative block number of the last block + 1 before
3987  * last_block (input value) in the file.
3988  * This is not based on i_size, it is based on the extent records.
3989  * Returns 0 for local files, as they do not have extent records.
3990  */
3991 int                                             /* error */
3992 xfs_bmap_last_before(
3993         xfs_trans_t     *tp,                    /* transaction pointer */
3994         xfs_inode_t     *ip,                    /* incore inode */
3995         xfs_fileoff_t   *last_block,            /* last block */
3996         int             whichfork)              /* data or attr fork */
3997 {
3998         xfs_fileoff_t   bno;                    /* input file offset */
3999         int             eof;                    /* hit end of file */
4000         xfs_bmbt_rec_host_t *ep;                /* pointer to last extent */
4001         int             error;                  /* error return value */
4002         xfs_bmbt_irec_t got;                    /* current extent value */
4003         xfs_ifork_t     *ifp;                   /* inode fork pointer */
4004         xfs_extnum_t    lastx;                  /* last extent used */
4005         xfs_bmbt_irec_t prev;                   /* previous extent value */
4006
4007         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
4008             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4009             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL)
4010                return XFS_ERROR(EIO);
4011         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
4012                 *last_block = 0;
4013                 return 0;
4014         }
4015         ifp = XFS_IFORK_PTR(ip, whichfork);
4016         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
4017             (error = xfs_iread_extents(tp, ip, whichfork)))
4018                 return error;
4019         bno = *last_block - 1;
4020         ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
4021                 &prev);
4022         if (eof || xfs_bmbt_get_startoff(ep) > bno) {
4023                 if (prev.br_startoff == NULLFILEOFF)
4024                         *last_block = 0;
4025                 else
4026                         *last_block = prev.br_startoff + prev.br_blockcount;
4027         }
4028         /*
4029          * Otherwise *last_block is already the right answer.
4030          */
4031         return 0;
4032 }
4033
4034 /*
4035  * Returns the file-relative block number of the first block past eof in
4036  * the file.  This is not based on i_size, it is based on the extent records.
4037  * Returns 0 for local files, as they do not have extent records.
4038  */
4039 int                                             /* error */
4040 xfs_bmap_last_offset(
4041         xfs_trans_t     *tp,                    /* transaction pointer */
4042         xfs_inode_t     *ip,                    /* incore inode */
4043         xfs_fileoff_t   *last_block,            /* last block */
4044         int             whichfork)              /* data or attr fork */
4045 {
4046         xfs_bmbt_rec_host_t *ep;                /* pointer to last extent */
4047         int             error;                  /* error return value */
4048         xfs_ifork_t     *ifp;                   /* inode fork pointer */
4049         xfs_extnum_t    nextents;               /* number of extent entries */
4050
4051         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
4052             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4053             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL)
4054                return XFS_ERROR(EIO);
4055         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
4056                 *last_block = 0;
4057                 return 0;
4058         }
4059         ifp = XFS_IFORK_PTR(ip, whichfork);
4060         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
4061             (error = xfs_iread_extents(tp, ip, whichfork)))
4062                 return error;
4063         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4064         if (!nextents) {
4065                 *last_block = 0;
4066                 return 0;
4067         }
4068         ep = xfs_iext_get_ext(ifp, nextents - 1);
4069         *last_block = xfs_bmbt_get_startoff(ep) + xfs_bmbt_get_blockcount(ep);
4070         return 0;
4071 }
4072
4073 /*
4074  * Returns whether the selected fork of the inode has exactly one
4075  * block or not.  For the data fork we check this matches di_size,
4076  * implying the file's range is 0..bsize-1.
4077  */
4078 int                                     /* 1=>1 block, 0=>otherwise */
4079 xfs_bmap_one_block(
4080         xfs_inode_t     *ip,            /* incore inode */
4081         int             whichfork)      /* data or attr fork */
4082 {
4083         xfs_bmbt_rec_host_t *ep;        /* ptr to fork's extent */
4084         xfs_ifork_t     *ifp;           /* inode fork pointer */
4085         int             rval;           /* return value */
4086         xfs_bmbt_irec_t s;              /* internal version of extent */
4087
4088 #ifndef DEBUG
4089         if (whichfork == XFS_DATA_FORK) {
4090                 return ((ip->i_d.di_mode & S_IFMT) == S_IFREG) ?
4091                         (ip->i_size == ip->i_mount->m_sb.sb_blocksize) :
4092                         (ip->i_d.di_size == ip->i_mount->m_sb.sb_blocksize);
4093         }
4094 #endif  /* !DEBUG */
4095         if (XFS_IFORK_NEXTENTS(ip, whichfork) != 1)
4096                 return 0;
4097         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
4098                 return 0;
4099         ifp = XFS_IFORK_PTR(ip, whichfork);
4100         ASSERT(ifp->if_flags & XFS_IFEXTENTS);
4101         ep = xfs_iext_get_ext(ifp, 0);
4102         xfs_bmbt_get_all(ep, &s);
4103         rval = s.br_startoff == 0 && s.br_blockcount == 1;
4104         if (rval && whichfork == XFS_DATA_FORK)
4105                 ASSERT(ip->i_size == ip->i_mount->m_sb.sb_blocksize);
4106         return rval;
4107 }
4108
4109 STATIC int
4110 xfs_bmap_sanity_check(
4111         struct xfs_mount        *mp,
4112         struct xfs_buf          *bp,
4113         int                     level)
4114 {
4115         struct xfs_btree_block  *block = XFS_BUF_TO_BLOCK(bp);
4116
4117         if (be32_to_cpu(block->bb_magic) != XFS_BMAP_MAGIC ||
4118             be16_to_cpu(block->bb_level) != level ||
4119             be16_to_cpu(block->bb_numrecs) == 0 ||
4120             be16_to_cpu(block->bb_numrecs) > mp->m_bmap_dmxr[level != 0])
4121                 return 0;
4122         return 1;
4123 }
4124
4125 /*
4126  * Read in the extents to if_extents.
4127  * All inode fields are set up by caller, we just traverse the btree
4128  * and copy the records in. If the file system cannot contain unwritten
4129  * extents, the records are checked for no "state" flags.
4130  */
4131 int                                     /* error */
4132 xfs_bmap_read_extents(
4133         xfs_trans_t             *tp,    /* transaction pointer */
4134         xfs_inode_t             *ip,    /* incore inode */
4135         int                     whichfork) /* data or attr fork */
4136 {
4137         struct xfs_btree_block  *block; /* current btree block */
4138         xfs_fsblock_t           bno;    /* block # of "block" */
4139         xfs_buf_t               *bp;    /* buffer for "block" */
4140         int                     error;  /* error return value */
4141         xfs_exntfmt_t           exntf;  /* XFS_EXTFMT_NOSTATE, if checking */
4142         xfs_extnum_t            i, j;   /* index into the extents list */
4143         xfs_ifork_t             *ifp;   /* fork structure */
4144         int                     level;  /* btree level, for checking */
4145         xfs_mount_t             *mp;    /* file system mount structure */
4146         __be64                  *pp;    /* pointer to block address */
4147         /* REFERENCED */
4148         xfs_extnum_t            room;   /* number of entries there's room for */
4149
4150         bno = NULLFSBLOCK;
4151         mp = ip->i_mount;
4152         ifp = XFS_IFORK_PTR(ip, whichfork);
4153         exntf = (whichfork != XFS_DATA_FORK) ? XFS_EXTFMT_NOSTATE :
4154                                         XFS_EXTFMT_INODE(ip);
4155         block = ifp->if_broot;
4156         /*
4157          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
4158          */
4159         level = be16_to_cpu(block->bb_level);
4160         ASSERT(level > 0);
4161         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
4162         bno = be64_to_cpu(*pp);
4163         ASSERT(bno != NULLDFSBNO);
4164         ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
4165         ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
4166         /*
4167          * Go down the tree until leaf level is reached, following the first
4168          * pointer (leftmost) at each level.
4169          */
4170         while (level-- > 0) {
4171                 if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
4172                                 XFS_BMAP_BTREE_REF)))
4173                         return error;
4174                 block = XFS_BUF_TO_BLOCK(bp);
4175                 XFS_WANT_CORRUPTED_GOTO(
4176                         xfs_bmap_sanity_check(mp, bp, level),
4177                         error0);
4178                 if (level == 0)
4179                         break;
4180                 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
4181                 bno = be64_to_cpu(*pp);
4182                 XFS_WANT_CORRUPTED_GOTO(XFS_FSB_SANITY_CHECK(mp, bno), error0);
4183                 xfs_trans_brelse(tp, bp);
4184         }
4185         /*
4186          * Here with bp and block set to the leftmost leaf node in the tree.
4187          */
4188         room = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4189         i = 0;
4190         /*
4191          * Loop over all leaf nodes.  Copy information to the extent records.
4192          */
4193         for (;;) {
4194                 xfs_bmbt_rec_t  *frp;
4195                 xfs_fsblock_t   nextbno;
4196                 xfs_extnum_t    num_recs;
4197                 xfs_extnum_t    start;
4198
4199
4200                 num_recs = xfs_btree_get_numrecs(block);
4201                 if (unlikely(i + num_recs > room)) {
4202                         ASSERT(i + num_recs <= room);
4203                         xfs_warn(ip->i_mount,
4204                                 "corrupt dinode %Lu, (btree extents).",
4205                                 (unsigned long long) ip->i_ino);
4206                         XFS_CORRUPTION_ERROR("xfs_bmap_read_extents(1)",
4207                                 XFS_ERRLEVEL_LOW, ip->i_mount, block);
4208                         goto error0;
4209                 }
4210                 XFS_WANT_CORRUPTED_GOTO(
4211                         xfs_bmap_sanity_check(mp, bp, 0),
4212                         error0);
4213                 /*
4214                  * Read-ahead the next leaf block, if any.
4215                  */
4216                 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
4217                 if (nextbno != NULLFSBLOCK)
4218                         xfs_btree_reada_bufl(mp, nextbno, 1);
4219                 /*
4220                  * Copy records into the extent records.
4221                  */
4222                 frp = XFS_BMBT_REC_ADDR(mp, block, 1);
4223                 start = i;
4224                 for (j = 0; j < num_recs; j++, i++, frp++) {
4225                         xfs_bmbt_rec_host_t *trp = xfs_iext_get_ext(ifp, i);
4226                         trp->l0 = be64_to_cpu(frp->l0);
4227                         trp->l1 = be64_to_cpu(frp->l1);
4228                 }
4229                 if (exntf == XFS_EXTFMT_NOSTATE) {
4230                         /*
4231                          * Check all attribute bmap btree records and
4232                          * any "older" data bmap btree records for a
4233                          * set bit in the "extent flag" position.
4234                          */
4235                         if (unlikely(xfs_check_nostate_extents(ifp,
4236                                         start, num_recs))) {
4237                                 XFS_ERROR_REPORT("xfs_bmap_read_extents(2)",
4238                                                  XFS_ERRLEVEL_LOW,
4239                                                  ip->i_mount);
4240                                 goto error0;
4241                         }
4242                 }
4243                 xfs_trans_brelse(tp, bp);
4244                 bno = nextbno;
4245                 /*
4246                  * If we've reached the end, stop.
4247                  */
4248                 if (bno == NULLFSBLOCK)
4249                         break;
4250                 if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
4251                                 XFS_BMAP_BTREE_REF)))
4252                         return error;
4253                 block = XFS_BUF_TO_BLOCK(bp);
4254         }
4255         ASSERT(i == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)));
4256         ASSERT(i == XFS_IFORK_NEXTENTS(ip, whichfork));
4257         XFS_BMAP_TRACE_EXLIST(ip, i, whichfork);
4258         return 0;
4259 error0:
4260         xfs_trans_brelse(tp, bp);
4261         return XFS_ERROR(EFSCORRUPTED);
4262 }
4263
4264 #ifdef DEBUG
4265 /*
4266  * Add bmap trace insert entries for all the contents of the extent records.
4267  */
4268 void
4269 xfs_bmap_trace_exlist(
4270         xfs_inode_t     *ip,            /* incore inode pointer */
4271         xfs_extnum_t    cnt,            /* count of entries in the list */
4272         int             whichfork,      /* data or attr fork */
4273         unsigned long   caller_ip)
4274 {
4275         xfs_extnum_t    idx;            /* extent record index */
4276         xfs_ifork_t     *ifp;           /* inode fork pointer */
4277         int             state = 0;
4278
4279         if (whichfork == XFS_ATTR_FORK)
4280                 state |= BMAP_ATTRFORK;
4281
4282         ifp = XFS_IFORK_PTR(ip, whichfork);
4283         ASSERT(cnt == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)));
4284         for (idx = 0; idx < cnt; idx++)
4285                 trace_xfs_extlist(ip, idx, whichfork, caller_ip);
4286 }
4287
4288 /*
4289  * Validate that the bmbt_irecs being returned from bmapi are valid
4290  * given the callers original parameters.  Specifically check the
4291  * ranges of the returned irecs to ensure that they only extent beyond
4292  * the given parameters if the XFS_BMAPI_ENTIRE flag was set.
4293  */
4294 STATIC void
4295 xfs_bmap_validate_ret(
4296         xfs_fileoff_t           bno,
4297         xfs_filblks_t           len,
4298         int                     flags,
4299         xfs_bmbt_irec_t         *mval,
4300         int                     nmap,
4301         int                     ret_nmap)
4302 {
4303         int                     i;              /* index to map values */
4304
4305         ASSERT(ret_nmap <= nmap);
4306
4307         for (i = 0; i < ret_nmap; i++) {
4308                 ASSERT(mval[i].br_blockcount > 0);
4309                 if (!(flags & XFS_BMAPI_ENTIRE)) {
4310                         ASSERT(mval[i].br_startoff >= bno);
4311                         ASSERT(mval[i].br_blockcount <= len);
4312                         ASSERT(mval[i].br_startoff + mval[i].br_blockcount <=
4313                                bno + len);
4314                 } else {
4315                         ASSERT(mval[i].br_startoff < bno + len);
4316                         ASSERT(mval[i].br_startoff + mval[i].br_blockcount >
4317                                bno);
4318                 }
4319                 ASSERT(i == 0 ||
4320                        mval[i - 1].br_startoff + mval[i - 1].br_blockcount ==
4321                        mval[i].br_startoff);
4322                 if ((flags & XFS_BMAPI_WRITE) && !(flags & XFS_BMAPI_DELAY))
4323                         ASSERT(mval[i].br_startblock != DELAYSTARTBLOCK &&
4324                                mval[i].br_startblock != HOLESTARTBLOCK);
4325                 ASSERT(mval[i].br_state == XFS_EXT_NORM ||
4326                        mval[i].br_state == XFS_EXT_UNWRITTEN);
4327         }
4328 }
4329 #endif /* DEBUG */
4330
4331
4332 /*
4333  * Map file blocks to filesystem blocks.
4334  * File range is given by the bno/len pair.
4335  * Adds blocks to file if a write ("flags & XFS_BMAPI_WRITE" set)
4336  * into a hole or past eof.
4337  * Only allocates blocks from a single allocation group,
4338  * to avoid locking problems.
4339  * The returned value in "firstblock" from the first call in a transaction
4340  * must be remembered and presented to subsequent calls in "firstblock".
4341  * An upper bound for the number of blocks to be allocated is supplied to
4342  * the first call in "total"; if no allocation group has that many free
4343  * blocks then the call will fail (return NULLFSBLOCK in "firstblock").
4344  */
4345 int                                     /* error */
4346 xfs_bmapi(
4347         xfs_trans_t     *tp,            /* transaction pointer */
4348         xfs_inode_t     *ip,            /* incore inode */
4349         xfs_fileoff_t   bno,            /* starting file offs. mapped */
4350         xfs_filblks_t   len,            /* length to map in file */
4351         int             flags,          /* XFS_BMAPI_... */
4352         xfs_fsblock_t   *firstblock,    /* first allocated block
4353                                            controls a.g. for allocs */
4354         xfs_extlen_t    total,          /* total blocks needed */
4355         xfs_bmbt_irec_t *mval,          /* output: map values */
4356         int             *nmap,          /* i/o: mval size/count */
4357         xfs_bmap_free_t *flist)         /* i/o: list extents to free */
4358 {
4359         xfs_fsblock_t   abno;           /* allocated block number */
4360         xfs_extlen_t    alen;           /* allocated extent length */
4361         xfs_fileoff_t   aoff;           /* allocated file offset */
4362         xfs_bmalloca_t  bma = { 0 };    /* args for xfs_bmap_alloc */
4363         xfs_btree_cur_t *cur;           /* bmap btree cursor */
4364         xfs_fileoff_t   end;            /* end of mapped file region */
4365         int             eof;            /* we've hit the end of extents */
4366         xfs_bmbt_rec_host_t *ep;        /* extent record pointer */
4367         int             error;          /* error return */
4368         xfs_bmbt_irec_t got;            /* current file extent record */
4369         xfs_ifork_t     *ifp;           /* inode fork pointer */
4370         xfs_extlen_t    indlen;         /* indirect blocks length */
4371         xfs_extnum_t    lastx;          /* last useful extent number */
4372         int             logflags;       /* flags for transaction logging */
4373         xfs_extlen_t    minleft;        /* min blocks left after allocation */
4374         xfs_extlen_t    minlen;         /* min allocation size */
4375         xfs_mount_t     *mp;            /* xfs mount structure */
4376         int             n;              /* current extent index */
4377         int             nallocs;        /* number of extents alloc'd */
4378         xfs_extnum_t    nextents;       /* number of extents in file */
4379         xfs_fileoff_t   obno;           /* old block number (offset) */
4380         xfs_bmbt_irec_t prev;           /* previous file extent record */
4381         int             tmp_logflags;   /* temp flags holder */
4382         int             whichfork;      /* data or attr fork */
4383         char            inhole;         /* current location is hole in file */
4384         char            wasdelay;       /* old extent was delayed */
4385         char            wr;             /* this is a write request */
4386         char            rt;             /* this is a realtime file */
4387 #ifdef DEBUG
4388         xfs_fileoff_t   orig_bno;       /* original block number value */
4389         int             orig_flags;     /* original flags arg value */
4390         xfs_filblks_t   orig_len;       /* original value of len arg */
4391         xfs_bmbt_irec_t *orig_mval;     /* original value of mval */
4392         int             orig_nmap;      /* original value of *nmap */
4393
4394         orig_bno = bno;
4395         orig_len = len;
4396         orig_flags = flags;
4397         orig_mval = mval;
4398         orig_nmap = *nmap;
4399 #endif
4400         ASSERT(*nmap >= 1);
4401         ASSERT(*nmap <= XFS_BMAP_MAX_NMAP || !(flags & XFS_BMAPI_WRITE));
4402         whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4403                 XFS_ATTR_FORK : XFS_DATA_FORK;
4404         mp = ip->i_mount;
4405         if (unlikely(XFS_TEST_ERROR(
4406             (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4407              XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
4408              XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL),
4409              mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4410                 XFS_ERROR_REPORT("xfs_bmapi", XFS_ERRLEVEL_LOW, mp);
4411                 return XFS_ERROR(EFSCORRUPTED);
4412         }
4413         if (XFS_FORCED_SHUTDOWN(mp))
4414                 return XFS_ERROR(EIO);
4415         rt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
4416         ifp = XFS_IFORK_PTR(ip, whichfork);
4417         ASSERT(ifp->if_ext_max ==
4418                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
4419         if ((wr = (flags & XFS_BMAPI_WRITE)) != 0)
4420                 XFS_STATS_INC(xs_blk_mapw);
4421         else
4422                 XFS_STATS_INC(xs_blk_mapr);
4423         /*
4424          * IGSTATE flag is used to combine extents which
4425          * differ only due to the state of the extents.
4426          * This technique is used from xfs_getbmap()
4427          * when the caller does not wish to see the
4428          * separation (which is the default).
4429          *
4430          * This technique is also used when writing a
4431          * buffer which has been partially written,
4432          * (usually by being flushed during a chunkread),
4433          * to ensure one write takes place. This also
4434          * prevents a change in the xfs inode extents at
4435          * this time, intentionally. This change occurs
4436          * on completion of the write operation, in
4437          * xfs_strat_comp(), where the xfs_bmapi() call
4438          * is transactioned, and the extents combined.
4439          */
4440         if ((flags & XFS_BMAPI_IGSTATE) && wr)  /* if writing unwritten space */
4441                 wr = 0;                         /* no allocations are allowed */
4442         ASSERT(wr || !(flags & XFS_BMAPI_DELAY));
4443         logflags = 0;
4444         nallocs = 0;
4445         cur = NULL;
4446         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
4447                 ASSERT(wr && tp);
4448                 if ((error = xfs_bmap_local_to_extents(tp, ip,
4449                                 firstblock, total, &logflags, whichfork)))
4450                         goto error0;
4451         }
4452         if (wr && *firstblock == NULLFSBLOCK) {
4453                 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE)
4454                         minleft = be16_to_cpu(ifp->if_broot->bb_level) + 1;
4455                 else
4456                         minleft = 1;
4457         } else
4458                 minleft = 0;
4459         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
4460             (error = xfs_iread_extents(tp, ip, whichfork)))
4461                 goto error0;
4462         ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
4463                 &prev);
4464         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4465         n = 0;
4466         end = bno + len;
4467         obno = bno;
4468         bma.ip = NULL;
4469
4470         while (bno < end && n < *nmap) {
4471                 /*
4472                  * Reading past eof, act as though there's a hole
4473                  * up to end.
4474                  */
4475                 if (eof && !wr)
4476                         got.br_startoff = end;
4477                 inhole = eof || got.br_startoff > bno;
4478                 wasdelay = wr && !inhole && !(flags & XFS_BMAPI_DELAY) &&
4479                         isnullstartblock(got.br_startblock);
4480                 /*
4481                  * First, deal with the hole before the allocated space
4482                  * that we found, if any.
4483                  */
4484                 if (wr && (inhole || wasdelay)) {
4485                         /*
4486                          * For the wasdelay case, we could also just
4487                          * allocate the stuff asked for in this bmap call
4488                          * but that wouldn't be as good.
4489                          */
4490                         if (wasdelay) {
4491                                 alen = (xfs_extlen_t)got.br_blockcount;
4492                                 aoff = got.br_startoff;
4493                                 if (lastx != NULLEXTNUM && lastx) {
4494                                         ep = xfs_iext_get_ext(ifp, lastx - 1);
4495                                         xfs_bmbt_get_all(ep, &prev);
4496                                 }
4497                         } else {
4498                                 alen = (xfs_extlen_t)
4499                                         XFS_FILBLKS_MIN(len, MAXEXTLEN);
4500                                 if (!eof)
4501                                         alen = (xfs_extlen_t)
4502                                                 XFS_FILBLKS_MIN(alen,
4503                                                         got.br_startoff - bno);
4504                                 aoff = bno;
4505                         }
4506                         minlen = (flags & XFS_BMAPI_CONTIG) ? alen : 1;
4507                         if (flags & XFS_BMAPI_DELAY) {
4508                                 xfs_extlen_t    extsz;
4509
4510                                 /* Figure out the extent size, adjust alen */
4511                                 extsz = xfs_get_extsz_hint(ip);
4512                                 if (extsz) {
4513                                         /*
4514                                          * make sure we don't exceed a single
4515                                          * extent length when we align the
4516                                          * extent by reducing length we are
4517                                          * going to allocate by the maximum
4518                                          * amount extent size aligment may
4519                                          * require.
4520                                          */
4521                                         alen = XFS_FILBLKS_MIN(len,
4522                                                    MAXEXTLEN - (2 * extsz - 1));
4523                                         error = xfs_bmap_extsize_align(mp,
4524                                                         &got, &prev, extsz,
4525                                                         rt, eof,
4526                                                         flags&XFS_BMAPI_DELAY,
4527                                                         flags&XFS_BMAPI_CONVERT,
4528                                                         &aoff, &alen);
4529                                         ASSERT(!error);
4530                                 }
4531
4532                                 if (rt)
4533                                         extsz = alen / mp->m_sb.sb_rextsize;
4534
4535                                 /*
4536                                  * Make a transaction-less quota reservation for
4537                                  * delayed allocation blocks. This number gets
4538                                  * adjusted later.  We return if we haven't
4539                                  * allocated blocks already inside this loop.
4540                                  */
4541                                 error = xfs_trans_reserve_quota_nblks(
4542                                                 NULL, ip, (long)alen, 0,
4543                                                 rt ? XFS_QMOPT_RES_RTBLKS :
4544                                                      XFS_QMOPT_RES_REGBLKS);
4545                                 if (error) {
4546                                         if (n == 0) {
4547                                                 *nmap = 0;
4548                                                 ASSERT(cur == NULL);
4549                                                 return error;
4550                                         }
4551                                         break;
4552                                 }
4553
4554                                 /*
4555                                  * Split changing sb for alen and indlen since
4556                                  * they could be coming from different places.
4557                                  */
4558                                 indlen = (xfs_extlen_t)
4559                                         xfs_bmap_worst_indlen(ip, alen);
4560                                 ASSERT(indlen > 0);
4561
4562                                 if (rt) {
4563                                         error = xfs_mod_incore_sb(mp,
4564                                                         XFS_SBS_FREXTENTS,
4565                                                         -((int64_t)extsz), (flags &
4566                                                         XFS_BMAPI_RSVBLOCKS));
4567                                 } else {
4568                                         error = xfs_icsb_modify_counters(mp,
4569                                                         XFS_SBS_FDBLOCKS,
4570                                                         -((int64_t)alen), (flags &
4571                                                         XFS_BMAPI_RSVBLOCKS));
4572                                 }
4573                                 if (!error) {
4574                                         error = xfs_icsb_modify_counters(mp,
4575                                                         XFS_SBS_FDBLOCKS,
4576                                                         -((int64_t)indlen), (flags &
4577                                                         XFS_BMAPI_RSVBLOCKS));
4578                                         if (error && rt)
4579                                                 xfs_mod_incore_sb(mp,
4580                                                         XFS_SBS_FREXTENTS,
4581                                                         (int64_t)extsz, (flags &
4582                                                         XFS_BMAPI_RSVBLOCKS));
4583                                         else if (error)
4584                                                 xfs_icsb_modify_counters(mp,
4585                                                         XFS_SBS_FDBLOCKS,
4586                                                         (int64_t)alen, (flags &
4587                                                         XFS_BMAPI_RSVBLOCKS));
4588                                 }
4589
4590                                 if (error) {
4591                                         if (XFS_IS_QUOTA_ON(mp))
4592                                                 /* unreserve the blocks now */
4593                                                 (void)
4594                                                 xfs_trans_unreserve_quota_nblks(
4595                                                         NULL, ip,
4596                                                         (long)alen, 0, rt ?
4597                                                         XFS_QMOPT_RES_RTBLKS :
4598                                                         XFS_QMOPT_RES_REGBLKS);
4599                                         break;
4600                                 }
4601
4602                                 ip->i_delayed_blks += alen;
4603                                 abno = nullstartblock(indlen);
4604                         } else {
4605                                 /*
4606                                  * If first time, allocate and fill in
4607                                  * once-only bma fields.
4608                                  */
4609                                 if (bma.ip == NULL) {
4610                                         bma.tp = tp;
4611                                         bma.ip = ip;
4612                                         bma.prevp = &prev;
4613                                         bma.gotp = &got;
4614                                         bma.total = total;
4615                                         bma.userdata = 0;
4616                                 }
4617                                 /* Indicate if this is the first user data
4618                                  * in the file, or just any user data.
4619                                  */
4620                                 if (!(flags & XFS_BMAPI_METADATA)) {
4621                                         bma.userdata = (aoff == 0) ?
4622                                                 XFS_ALLOC_INITIAL_USER_DATA :
4623                                                 XFS_ALLOC_USERDATA;
4624                                 }
4625                                 /*
4626                                  * Fill in changeable bma fields.
4627                                  */
4628                                 bma.eof = eof;
4629                                 bma.firstblock = *firstblock;
4630                                 bma.alen = alen;
4631                                 bma.off = aoff;
4632                                 bma.conv = !!(flags & XFS_BMAPI_CONVERT);
4633                                 bma.wasdel = wasdelay;
4634                                 bma.minlen = minlen;
4635                                 bma.low = flist->xbf_low;
4636                                 bma.minleft = minleft;
4637                                 /*
4638                                  * Only want to do the alignment at the
4639                                  * eof if it is userdata and allocation length
4640                                  * is larger than a stripe unit.
4641                                  */
4642                                 if (mp->m_dalign && alen >= mp->m_dalign &&
4643                                     (!(flags & XFS_BMAPI_METADATA)) &&
4644                                     (whichfork == XFS_DATA_FORK)) {
4645                                         if ((error = xfs_bmap_isaeof(ip, aoff,
4646                                                         whichfork, &bma.aeof)))
4647                                                 goto error0;
4648                                 } else
4649                                         bma.aeof = 0;
4650                                 /*
4651                                  * Call allocator.
4652                                  */
4653                                 if ((error = xfs_bmap_alloc(&bma)))
4654                                         goto error0;
4655                                 /*
4656                                  * Copy out result fields.
4657                                  */
4658                                 abno = bma.rval;
4659                                 if ((flist->xbf_low = bma.low))
4660                                         minleft = 0;
4661                                 alen = bma.alen;
4662                                 aoff = bma.off;
4663                                 ASSERT(*firstblock == NULLFSBLOCK ||
4664                                        XFS_FSB_TO_AGNO(mp, *firstblock) ==
4665                                        XFS_FSB_TO_AGNO(mp, bma.firstblock) ||
4666                                        (flist->xbf_low &&
4667                                         XFS_FSB_TO_AGNO(mp, *firstblock) <
4668                                         XFS_FSB_TO_AGNO(mp, bma.firstblock)));
4669                                 *firstblock = bma.firstblock;
4670                                 if (cur)
4671                                         cur->bc_private.b.firstblock =
4672                                                 *firstblock;
4673                                 if (abno == NULLFSBLOCK)
4674                                         break;
4675                                 if ((ifp->if_flags & XFS_IFBROOT) && !cur) {
4676                                         cur = xfs_bmbt_init_cursor(mp, tp,
4677                                                 ip, whichfork);
4678                                         cur->bc_private.b.firstblock =
4679                                                 *firstblock;
4680                                         cur->bc_private.b.flist = flist;
4681                                 }
4682                                 /*
4683                                  * Bump the number of extents we've allocated
4684                                  * in this call.
4685                                  */
4686                                 nallocs++;
4687                         }
4688                         if (cur)
4689                                 cur->bc_private.b.flags =
4690                                         wasdelay ? XFS_BTCUR_BPRV_WASDEL : 0;
4691                         got.br_startoff = aoff;
4692                         got.br_startblock = abno;
4693                         got.br_blockcount = alen;
4694                         got.br_state = XFS_EXT_NORM;    /* assume normal */
4695                         /*
4696                          * Determine state of extent, and the filesystem.
4697                          * A wasdelay extent has been initialized, so
4698                          * shouldn't be flagged as unwritten.
4699                          */
4700                         if (wr && xfs_sb_version_hasextflgbit(&mp->m_sb)) {
4701                                 if (!wasdelay && (flags & XFS_BMAPI_PREALLOC))
4702                                         got.br_state = XFS_EXT_UNWRITTEN;
4703                         }
4704                         error = xfs_bmap_add_extent(ip, lastx, &cur, &got,
4705                                 firstblock, flist, &tmp_logflags,
4706                                 whichfork, (flags & XFS_BMAPI_RSVBLOCKS));
4707                         logflags |= tmp_logflags;
4708                         if (error)
4709                                 goto error0;
4710                         lastx = ifp->if_lastex;
4711                         ep = xfs_iext_get_ext(ifp, lastx);
4712                         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4713                         xfs_bmbt_get_all(ep, &got);
4714                         ASSERT(got.br_startoff <= aoff);
4715                         ASSERT(got.br_startoff + got.br_blockcount >=
4716                                 aoff + alen);
4717 #ifdef DEBUG
4718                         if (flags & XFS_BMAPI_DELAY) {
4719                                 ASSERT(isnullstartblock(got.br_startblock));
4720                                 ASSERT(startblockval(got.br_startblock) > 0);
4721                         }
4722                         ASSERT(got.br_state == XFS_EXT_NORM ||
4723                                got.br_state == XFS_EXT_UNWRITTEN);
4724 #endif
4725                         /*
4726                          * Fall down into the found allocated space case.
4727                          */
4728                 } else if (inhole) {
4729                         /*
4730                          * Reading in a hole.
4731                          */
4732                         mval->br_startoff = bno;
4733                         mval->br_startblock = HOLESTARTBLOCK;
4734                         mval->br_blockcount =
4735                                 XFS_FILBLKS_MIN(len, got.br_startoff - bno);
4736                         mval->br_state = XFS_EXT_NORM;
4737                         bno += mval->br_blockcount;
4738                         len -= mval->br_blockcount;
4739                         mval++;
4740                         n++;
4741                         continue;
4742                 }
4743                 /*
4744                  * Then deal with the allocated space we found.
4745                  */
4746                 ASSERT(ep != NULL);
4747                 if (!(flags & XFS_BMAPI_ENTIRE) &&
4748                     (got.br_startoff + got.br_blockcount > obno)) {
4749                         if (obno > bno)
4750                                 bno = obno;
4751                         ASSERT((bno >= obno) || (n == 0));
4752                         ASSERT(bno < end);
4753                         mval->br_startoff = bno;
4754                         if (isnullstartblock(got.br_startblock)) {
4755                                 ASSERT(!wr || (flags & XFS_BMAPI_DELAY));
4756                                 mval->br_startblock = DELAYSTARTBLOCK;
4757                         } else
4758                                 mval->br_startblock =
4759                                         got.br_startblock +
4760                                         (bno - got.br_startoff);
4761                         /*
4762                          * Return the minimum of what we got and what we
4763                          * asked for for the length.  We can use the len
4764                          * variable here because it is modified below
4765                          * and we could have been there before coming
4766                          * here if the first part of the allocation
4767                          * didn't overlap what was asked for.
4768                          */
4769                         mval->br_blockcount =
4770                                 XFS_FILBLKS_MIN(end - bno, got.br_blockcount -
4771                                         (bno - got.br_startoff));
4772                         mval->br_state = got.br_state;
4773                         ASSERT(mval->br_blockcount <= len);
4774                 } else {
4775                         *mval = got;
4776                         if (isnullstartblock(mval->br_startblock)) {
4777                                 ASSERT(!wr || (flags & XFS_BMAPI_DELAY));
4778                                 mval->br_startblock = DELAYSTARTBLOCK;
4779                         }
4780                 }
4781
4782                 /*
4783                  * Check if writing previously allocated but
4784                  * unwritten extents.
4785                  */
4786                 if (wr &&
4787                     ((mval->br_state == XFS_EXT_UNWRITTEN &&
4788                       ((flags & (XFS_BMAPI_PREALLOC|XFS_BMAPI_DELAY)) == 0)) ||
4789                      (mval->br_state == XFS_EXT_NORM &&
4790                       ((flags & (XFS_BMAPI_PREALLOC|XFS_BMAPI_CONVERT)) ==
4791                                 (XFS_BMAPI_PREALLOC|XFS_BMAPI_CONVERT))))) {
4792                         /*
4793                          * Modify (by adding) the state flag, if writing.
4794                          */
4795                         ASSERT(mval->br_blockcount <= len);
4796                         if ((ifp->if_flags & XFS_IFBROOT) && !cur) {
4797                                 cur = xfs_bmbt_init_cursor(mp,
4798                                         tp, ip, whichfork);
4799                                 cur->bc_private.b.firstblock =
4800                                         *firstblock;
4801                                 cur->bc_private.b.flist = flist;
4802                         }
4803                         mval->br_state = (mval->br_state == XFS_EXT_UNWRITTEN)
4804                                                 ? XFS_EXT_NORM
4805                                                 : XFS_EXT_UNWRITTEN;
4806                         error = xfs_bmap_add_extent(ip, lastx, &cur, mval,
4807                                 firstblock, flist, &tmp_logflags,
4808                                 whichfork, (flags & XFS_BMAPI_RSVBLOCKS));
4809                         logflags |= tmp_logflags;
4810                         if (error)
4811                                 goto error0;
4812                         lastx = ifp->if_lastex;
4813                         ep = xfs_iext_get_ext(ifp, lastx);
4814                         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4815                         xfs_bmbt_get_all(ep, &got);
4816                         /*
4817                          * We may have combined previously unwritten
4818                          * space with written space, so generate
4819                          * another request.
4820                          */
4821                         if (mval->br_blockcount < len)
4822                                 continue;
4823                 }
4824
4825                 ASSERT((flags & XFS_BMAPI_ENTIRE) ||
4826                        ((mval->br_startoff + mval->br_blockcount) <= end));
4827                 ASSERT((flags & XFS_BMAPI_ENTIRE) ||
4828                        (mval->br_blockcount <= len) ||
4829                        (mval->br_startoff < obno));
4830                 bno = mval->br_startoff + mval->br_blockcount;
4831                 len = end - bno;
4832                 if (n > 0 && mval->br_startoff == mval[-1].br_startoff) {
4833                         ASSERT(mval->br_startblock == mval[-1].br_startblock);
4834                         ASSERT(mval->br_blockcount > mval[-1].br_blockcount);
4835                         ASSERT(mval->br_state == mval[-1].br_state);
4836                         mval[-1].br_blockcount = mval->br_blockcount;
4837                         mval[-1].br_state = mval->br_state;
4838                 } else if (n > 0 && mval->br_startblock != DELAYSTARTBLOCK &&
4839                            mval[-1].br_startblock != DELAYSTARTBLOCK &&
4840                            mval[-1].br_startblock != HOLESTARTBLOCK &&
4841                            mval->br_startblock ==
4842                            mval[-1].br_startblock + mval[-1].br_blockcount &&
4843                            ((flags & XFS_BMAPI_IGSTATE) ||
4844                                 mval[-1].br_state == mval->br_state)) {
4845                         ASSERT(mval->br_startoff ==
4846                                mval[-1].br_startoff + mval[-1].br_blockcount);
4847                         mval[-1].br_blockcount += mval->br_blockcount;
4848                 } else if (n > 0 &&
4849                            mval->br_startblock == DELAYSTARTBLOCK &&
4850                            mval[-1].br_startblock == DELAYSTARTBLOCK &&
4851                            mval->br_startoff ==
4852                            mval[-1].br_startoff + mval[-1].br_blockcount) {
4853                         mval[-1].br_blockcount += mval->br_blockcount;
4854                         mval[-1].br_state = mval->br_state;
4855                 } else if (!((n == 0) &&
4856                              ((mval->br_startoff + mval->br_blockcount) <=
4857                               obno))) {
4858                         mval++;
4859                         n++;
4860                 }
4861                 /*
4862                  * If we're done, stop now.  Stop when we've allocated
4863                  * XFS_BMAP_MAX_NMAP extents no matter what.  Otherwise
4864                  * the transaction may get too big.
4865                  */
4866                 if (bno >= end || n >= *nmap || nallocs >= *nmap)
4867                         break;
4868                 /*
4869                  * Else go on to the next record.
4870                  */
4871                 ep = xfs_iext_get_ext(ifp, ++lastx);
4872                 prev = got;
4873                 if (lastx >= nextents)
4874                         eof = 1;
4875                 else
4876                         xfs_bmbt_get_all(ep, &got);
4877         }
4878         ifp->if_lastex = lastx;
4879         *nmap = n;
4880         /*
4881          * Transform from btree to extents, give it cur.
4882          */
4883         if (tp && XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE &&
4884             XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max) {
4885                 ASSERT(wr && cur);
4886                 error = xfs_bmap_btree_to_extents(tp, ip, cur,
4887                         &tmp_logflags, whichfork);
4888                 logflags |= tmp_logflags;
4889                 if (error)
4890                         goto error0;
4891         }
4892         ASSERT(ifp->if_ext_max ==
4893                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
4894         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE ||
4895                XFS_IFORK_NEXTENTS(ip, whichfork) > ifp->if_ext_max);
4896         error = 0;
4897 error0:
4898         /*
4899          * Log everything.  Do this after conversion, there's no point in
4900          * logging the extent records if we've converted to btree format.
4901          */
4902         if ((logflags & xfs_ilog_fext(whichfork)) &&
4903             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
4904                 logflags &= ~xfs_ilog_fext(whichfork);
4905         else if ((logflags & xfs_ilog_fbroot(whichfork)) &&
4906                  XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
4907                 logflags &= ~xfs_ilog_fbroot(whichfork);
4908         /*
4909          * Log whatever the flags say, even if error.  Otherwise we might miss
4910          * detecting a case where the data is changed, there's an error,
4911          * and it's not logged so we don't shutdown when we should.
4912          */
4913         if (logflags) {
4914                 ASSERT(tp && wr);
4915                 xfs_trans_log_inode(tp, ip, logflags);
4916         }
4917         if (cur) {
4918                 if (!error) {
4919                         ASSERT(*firstblock == NULLFSBLOCK ||
4920                                XFS_FSB_TO_AGNO(mp, *firstblock) ==
4921                                XFS_FSB_TO_AGNO(mp,
4922                                        cur->bc_private.b.firstblock) ||
4923                                (flist->xbf_low &&
4924                                 XFS_FSB_TO_AGNO(mp, *firstblock) <
4925                                 XFS_FSB_TO_AGNO(mp,
4926                                         cur->bc_private.b.firstblock)));
4927                         *firstblock = cur->bc_private.b.firstblock;
4928                 }
4929                 xfs_btree_del_cursor(cur,
4930                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
4931         }
4932         if (!error)
4933                 xfs_bmap_validate_ret(orig_bno, orig_len, orig_flags, orig_mval,
4934                         orig_nmap, *nmap);
4935         return error;
4936 }
4937
4938 /*
4939  * Map file blocks to filesystem blocks, simple version.
4940  * One block (extent) only, read-only.
4941  * For flags, only the XFS_BMAPI_ATTRFORK flag is examined.
4942  * For the other flag values, the effect is as if XFS_BMAPI_METADATA
4943  * was set and all the others were clear.
4944  */
4945 int                                             /* error */
4946 xfs_bmapi_single(
4947         xfs_trans_t     *tp,            /* transaction pointer */
4948         xfs_inode_t     *ip,            /* incore inode */
4949         int             whichfork,      /* data or attr fork */
4950         xfs_fsblock_t   *fsb,           /* output: mapped block */
4951         xfs_fileoff_t   bno)            /* starting file offs. mapped */
4952 {
4953         int             eof;            /* we've hit the end of extents */
4954         int             error;          /* error return */
4955         xfs_bmbt_irec_t got;            /* current file extent record */
4956         xfs_ifork_t     *ifp;           /* inode fork pointer */
4957         xfs_extnum_t    lastx;          /* last useful extent number */
4958         xfs_bmbt_irec_t prev;           /* previous file extent record */
4959
4960         ifp = XFS_IFORK_PTR(ip, whichfork);
4961         if (unlikely(
4962             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
4963             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)) {
4964                XFS_ERROR_REPORT("xfs_bmapi_single", XFS_ERRLEVEL_LOW,
4965                                 ip->i_mount);
4966                return XFS_ERROR(EFSCORRUPTED);
4967         }
4968         if (XFS_FORCED_SHUTDOWN(ip->i_mount))
4969                 return XFS_ERROR(EIO);
4970         XFS_STATS_INC(xs_blk_mapr);
4971         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
4972             (error = xfs_iread_extents(tp, ip, whichfork)))
4973                 return error;
4974         (void)xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
4975                 &prev);
4976         /*
4977          * Reading past eof, act as though there's a hole
4978          * up to end.
4979          */
4980         if (eof || got.br_startoff > bno) {
4981                 *fsb = NULLFSBLOCK;
4982                 return 0;
4983         }
4984         ASSERT(!isnullstartblock(got.br_startblock));
4985         ASSERT(bno < got.br_startoff + got.br_blockcount);
4986         *fsb = got.br_startblock + (bno - got.br_startoff);
4987         ifp->if_lastex = lastx;
4988         return 0;
4989 }
4990
4991 /*
4992  * Unmap (remove) blocks from a file.
4993  * If nexts is nonzero then the number of extents to remove is limited to
4994  * that value.  If not all extents in the block range can be removed then
4995  * *done is set.
4996  */
4997 int                                             /* error */
4998 xfs_bunmapi(
4999         xfs_trans_t             *tp,            /* transaction pointer */
5000         struct xfs_inode        *ip,            /* incore inode */
5001         xfs_fileoff_t           bno,            /* starting offset to unmap */
5002         xfs_filblks_t           len,            /* length to unmap in file */
5003         int                     flags,          /* misc flags */
5004         xfs_extnum_t            nexts,          /* number of extents max */
5005         xfs_fsblock_t           *firstblock,    /* first allocated block
5006                                                    controls a.g. for allocs */
5007         xfs_bmap_free_t         *flist,         /* i/o: list extents to free */
5008         int                     *done)          /* set if not done yet */
5009 {
5010         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
5011         xfs_bmbt_irec_t         del;            /* extent being deleted */
5012         int                     eof;            /* is deleting at eof */
5013         xfs_bmbt_rec_host_t     *ep;            /* extent record pointer */
5014         int                     error;          /* error return value */
5015         xfs_extnum_t            extno;          /* extent number in list */
5016         xfs_bmbt_irec_t         got;            /* current extent record */
5017         xfs_ifork_t             *ifp;           /* inode fork pointer */
5018         int                     isrt;           /* freeing in rt area */
5019         xfs_extnum_t            lastx;          /* last extent index used */
5020         int                     logflags;       /* transaction logging flags */
5021         xfs_extlen_t            mod;            /* rt extent offset */
5022         xfs_mount_t             *mp;            /* mount structure */
5023         xfs_extnum_t            nextents;       /* number of file extents */
5024         xfs_bmbt_irec_t         prev;           /* previous extent record */
5025         xfs_fileoff_t           start;          /* first file offset deleted */
5026         int                     tmp_logflags;   /* partial logging flags */
5027         int                     wasdel;         /* was a delayed alloc extent */
5028         int                     whichfork;      /* data or attribute fork */
5029         int                     rsvd;           /* OK to allocate reserved blocks */
5030         xfs_fsblock_t           sum;
5031
5032         trace_xfs_bunmap(ip, bno, len, flags, _RET_IP_);
5033
5034         whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
5035                 XFS_ATTR_FORK : XFS_DATA_FORK;
5036         ifp = XFS_IFORK_PTR(ip, whichfork);
5037         if (unlikely(
5038             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
5039             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
5040                 XFS_ERROR_REPORT("xfs_bunmapi", XFS_ERRLEVEL_LOW,
5041                                  ip->i_mount);
5042                 return XFS_ERROR(EFSCORRUPTED);
5043         }
5044         mp = ip->i_mount;
5045         if (XFS_FORCED_SHUTDOWN(mp))
5046                 return XFS_ERROR(EIO);
5047         rsvd = (flags & XFS_BMAPI_RSVBLOCKS) != 0;
5048         ASSERT(len > 0);
5049         ASSERT(nexts >= 0);
5050         ASSERT(ifp->if_ext_max ==
5051                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5052         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5053             (error = xfs_iread_extents(tp, ip, whichfork)))
5054                 return error;
5055         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5056         if (nextents == 0) {
5057                 *done = 1;
5058                 return 0;
5059         }
5060         XFS_STATS_INC(xs_blk_unmap);
5061         isrt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
5062         start = bno;
5063         bno = start + len - 1;
5064         ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
5065                 &prev);
5066
5067         /*
5068          * Check to see if the given block number is past the end of the
5069          * file, back up to the last block if so...
5070          */
5071         if (eof) {
5072                 ep = xfs_iext_get_ext(ifp, --lastx);
5073                 xfs_bmbt_get_all(ep, &got);
5074                 bno = got.br_startoff + got.br_blockcount - 1;
5075         }
5076         logflags = 0;
5077         if (ifp->if_flags & XFS_IFBROOT) {
5078                 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
5079                 cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5080                 cur->bc_private.b.firstblock = *firstblock;
5081                 cur->bc_private.b.flist = flist;
5082                 cur->bc_private.b.flags = 0;
5083         } else
5084                 cur = NULL;
5085         extno = 0;
5086         while (bno != (xfs_fileoff_t)-1 && bno >= start && lastx >= 0 &&
5087                (nexts == 0 || extno < nexts)) {
5088                 /*
5089                  * Is the found extent after a hole in which bno lives?
5090                  * Just back up to the previous extent, if so.
5091                  */
5092                 if (got.br_startoff > bno) {
5093                         if (--lastx < 0)
5094                                 break;
5095                         ep = xfs_iext_get_ext(ifp, lastx);
5096                         xfs_bmbt_get_all(ep, &got);
5097                 }
5098                 /*
5099                  * Is the last block of this extent before the range
5100                  * we're supposed to delete?  If so, we're done.
5101                  */
5102                 bno = XFS_FILEOFF_MIN(bno,
5103                         got.br_startoff + got.br_blockcount - 1);
5104                 if (bno < start)
5105                         break;
5106                 /*
5107                  * Then deal with the (possibly delayed) allocated space
5108                  * we found.
5109                  */
5110                 ASSERT(ep != NULL);
5111                 del = got;
5112                 wasdel = isnullstartblock(del.br_startblock);
5113                 if (got.br_startoff < start) {
5114                         del.br_startoff = start;
5115                         del.br_blockcount -= start - got.br_startoff;
5116                         if (!wasdel)
5117                                 del.br_startblock += start - got.br_startoff;
5118                 }
5119                 if (del.br_startoff + del.br_blockcount > bno + 1)
5120                         del.br_blockcount = bno + 1 - del.br_startoff;
5121                 sum = del.br_startblock + del.br_blockcount;
5122                 if (isrt &&
5123                     (mod = do_mod(sum, mp->m_sb.sb_rextsize))) {
5124                         /*
5125                          * Realtime extent not lined up at the end.
5126                          * The extent could have been split into written
5127                          * and unwritten pieces, or we could just be
5128                          * unmapping part of it.  But we can't really
5129                          * get rid of part of a realtime extent.
5130                          */
5131                         if (del.br_state == XFS_EXT_UNWRITTEN ||
5132                             !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5133                                 /*
5134                                  * This piece is unwritten, or we're not
5135                                  * using unwritten extents.  Skip over it.
5136                                  */
5137                                 ASSERT(bno >= mod);
5138                                 bno -= mod > del.br_blockcount ?
5139                                         del.br_blockcount : mod;
5140                                 if (bno < got.br_startoff) {
5141                                         if (--lastx >= 0)
5142                                                 xfs_bmbt_get_all(xfs_iext_get_ext(
5143                                                         ifp, lastx), &got);
5144                                 }
5145                                 continue;
5146                         }
5147                         /*
5148                          * It's written, turn it unwritten.
5149                          * This is better than zeroing it.
5150                          */
5151                         ASSERT(del.br_state == XFS_EXT_NORM);
5152                         ASSERT(xfs_trans_get_block_res(tp) > 0);
5153                         /*
5154                          * If this spans a realtime extent boundary,
5155                          * chop it back to the start of the one we end at.
5156                          */
5157                         if (del.br_blockcount > mod) {
5158                                 del.br_startoff += del.br_blockcount - mod;
5159                                 del.br_startblock += del.br_blockcount - mod;
5160                                 del.br_blockcount = mod;
5161                         }
5162                         del.br_state = XFS_EXT_UNWRITTEN;
5163                         error = xfs_bmap_add_extent(ip, lastx, &cur, &del,
5164                                 firstblock, flist, &logflags,
5165                                 XFS_DATA_FORK, 0);
5166                         if (error)
5167                                 goto error0;
5168                         goto nodelete;
5169                 }
5170                 if (isrt && (mod = do_mod(del.br_startblock, mp->m_sb.sb_rextsize))) {
5171                         /*
5172                          * Realtime extent is lined up at the end but not
5173                          * at the front.  We'll get rid of full extents if
5174                          * we can.
5175                          */
5176                         mod = mp->m_sb.sb_rextsize - mod;
5177                         if (del.br_blockcount > mod) {
5178                                 del.br_blockcount -= mod;
5179                                 del.br_startoff += mod;
5180                                 del.br_startblock += mod;
5181                         } else if ((del.br_startoff == start &&
5182                                     (del.br_state == XFS_EXT_UNWRITTEN ||
5183                                      xfs_trans_get_block_res(tp) == 0)) ||
5184                                    !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5185                                 /*
5186                                  * Can't make it unwritten.  There isn't
5187                                  * a full extent here so just skip it.
5188                                  */
5189                                 ASSERT(bno >= del.br_blockcount);
5190                                 bno -= del.br_blockcount;
5191                                 if (bno < got.br_startoff) {
5192                                         if (--lastx >= 0)
5193                                                 xfs_bmbt_get_all(--ep, &got);
5194                                 }
5195                                 continue;
5196                         } else if (del.br_state == XFS_EXT_UNWRITTEN) {
5197                                 /*
5198                                  * This one is already unwritten.
5199                                  * It must have a written left neighbor.
5200                                  * Unwrite the killed part of that one and
5201                                  * try again.
5202                                  */
5203                                 ASSERT(lastx > 0);
5204                                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp,
5205                                                 lastx - 1), &prev);
5206                                 ASSERT(prev.br_state == XFS_EXT_NORM);
5207                                 ASSERT(!isnullstartblock(prev.br_startblock));
5208                                 ASSERT(del.br_startblock ==
5209                                        prev.br_startblock + prev.br_blockcount);
5210                                 if (prev.br_startoff < start) {
5211                                         mod = start - prev.br_startoff;
5212                                         prev.br_blockcount -= mod;
5213                                         prev.br_startblock += mod;
5214                                         prev.br_startoff = start;
5215                                 }
5216                                 prev.br_state = XFS_EXT_UNWRITTEN;
5217                                 error = xfs_bmap_add_extent(ip, lastx - 1, &cur,
5218                                         &prev, firstblock, flist, &logflags,
5219                                         XFS_DATA_FORK, 0);
5220                                 if (error)
5221                                         goto error0;
5222                                 goto nodelete;
5223                         } else {
5224                                 ASSERT(del.br_state == XFS_EXT_NORM);
5225                                 del.br_state = XFS_EXT_UNWRITTEN;
5226                                 error = xfs_bmap_add_extent(ip, lastx, &cur,
5227                                         &del, firstblock, flist, &logflags,
5228                                         XFS_DATA_FORK, 0);
5229                                 if (error)
5230                                         goto error0;
5231                                 goto nodelete;
5232                         }
5233                 }
5234                 if (wasdel) {
5235                         ASSERT(startblockval(del.br_startblock) > 0);
5236                         /* Update realtime/data freespace, unreserve quota */
5237                         if (isrt) {
5238                                 xfs_filblks_t rtexts;
5239
5240                                 rtexts = XFS_FSB_TO_B(mp, del.br_blockcount);
5241                                 do_div(rtexts, mp->m_sb.sb_rextsize);
5242                                 xfs_mod_incore_sb(mp, XFS_SBS_FREXTENTS,
5243                                                 (int64_t)rtexts, rsvd);
5244                                 (void)xfs_trans_reserve_quota_nblks(NULL,
5245                                         ip, -((long)del.br_blockcount), 0,
5246                                         XFS_QMOPT_RES_RTBLKS);
5247                         } else {
5248                                 xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
5249                                                 (int64_t)del.br_blockcount, rsvd);
5250                                 (void)xfs_trans_reserve_quota_nblks(NULL,
5251                                         ip, -((long)del.br_blockcount), 0,
5252                                         XFS_QMOPT_RES_REGBLKS);
5253                         }
5254                         ip->i_delayed_blks -= del.br_blockcount;
5255                         if (cur)
5256                                 cur->bc_private.b.flags |=
5257                                         XFS_BTCUR_BPRV_WASDEL;
5258                 } else if (cur)
5259                         cur->bc_private.b.flags &= ~XFS_BTCUR_BPRV_WASDEL;
5260                 /*
5261                  * If it's the case where the directory code is running
5262                  * with no block reservation, and the deleted block is in
5263                  * the middle of its extent, and the resulting insert
5264                  * of an extent would cause transformation to btree format,
5265                  * then reject it.  The calling code will then swap
5266                  * blocks around instead.
5267                  * We have to do this now, rather than waiting for the
5268                  * conversion to btree format, since the transaction
5269                  * will be dirty.
5270                  */
5271                 if (!wasdel && xfs_trans_get_block_res(tp) == 0 &&
5272                     XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
5273                     XFS_IFORK_NEXTENTS(ip, whichfork) >= ifp->if_ext_max &&
5274                     del.br_startoff > got.br_startoff &&
5275                     del.br_startoff + del.br_blockcount <
5276                     got.br_startoff + got.br_blockcount) {
5277                         error = XFS_ERROR(ENOSPC);
5278                         goto error0;
5279                 }
5280                 error = xfs_bmap_del_extent(ip, tp, lastx, flist, cur, &del,
5281                                 &tmp_logflags, whichfork, rsvd);
5282                 logflags |= tmp_logflags;
5283                 if (error)
5284                         goto error0;
5285                 bno = del.br_startoff - 1;
5286 nodelete:
5287                 lastx = ifp->if_lastex;
5288                 /*
5289                  * If not done go on to the next (previous) record.
5290                  * Reset ep in case the extents array was re-alloced.
5291                  */
5292                 ep = xfs_iext_get_ext(ifp, lastx);
5293                 if (bno != (xfs_fileoff_t)-1 && bno >= start) {
5294                         if (lastx >= XFS_IFORK_NEXTENTS(ip, whichfork) ||
5295                             xfs_bmbt_get_startoff(ep) > bno) {
5296                                 if (--lastx >= 0)
5297                                         ep = xfs_iext_get_ext(ifp, lastx);
5298                         }
5299                         if (lastx >= 0)
5300                                 xfs_bmbt_get_all(ep, &got);
5301                         extno++;
5302                 }
5303         }
5304         ifp->if_lastex = lastx;
5305         *done = bno == (xfs_fileoff_t)-1 || bno < start || lastx < 0;
5306         ASSERT(ifp->if_ext_max ==
5307                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5308         /*
5309          * Convert to a btree if necessary.
5310          */
5311         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
5312             XFS_IFORK_NEXTENTS(ip, whichfork) > ifp->if_ext_max) {
5313                 ASSERT(cur == NULL);
5314                 error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist,
5315                         &cur, 0, &tmp_logflags, whichfork);
5316                 logflags |= tmp_logflags;
5317                 if (error)
5318                         goto error0;
5319         }
5320         /*
5321          * transform from btree to extents, give it cur
5322          */
5323         else if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE &&
5324                  XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max) {
5325                 ASSERT(cur != NULL);
5326                 error = xfs_bmap_btree_to_extents(tp, ip, cur, &tmp_logflags,
5327                         whichfork);
5328                 logflags |= tmp_logflags;
5329                 if (error)
5330                         goto error0;
5331         }
5332         /*
5333          * transform from extents to local?
5334          */
5335         ASSERT(ifp->if_ext_max ==
5336                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5337         error = 0;
5338 error0:
5339         /*
5340          * Log everything.  Do this after conversion, there's no point in
5341          * logging the extent records if we've converted to btree format.
5342          */
5343         if ((logflags & xfs_ilog_fext(whichfork)) &&
5344             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
5345                 logflags &= ~xfs_ilog_fext(whichfork);
5346         else if ((logflags & xfs_ilog_fbroot(whichfork)) &&
5347                  XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
5348                 logflags &= ~xfs_ilog_fbroot(whichfork);
5349         /*
5350          * Log inode even in the error case, if the transaction
5351          * is dirty we'll need to shut down the filesystem.
5352          */
5353         if (logflags)
5354                 xfs_trans_log_inode(tp, ip, logflags);
5355         if (cur) {
5356                 if (!error) {
5357                         *firstblock = cur->bc_private.b.firstblock;
5358                         cur->bc_private.b.allocated = 0;
5359                 }
5360                 xfs_btree_del_cursor(cur,
5361                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5362         }
5363         return error;
5364 }
5365
5366 /*
5367  * returns 1 for success, 0 if we failed to map the extent.
5368  */
5369 STATIC int
5370 xfs_getbmapx_fix_eof_hole(
5371         xfs_inode_t             *ip,            /* xfs incore inode pointer */
5372         struct getbmapx         *out,           /* output structure */
5373         int                     prealloced,     /* this is a file with
5374                                                  * preallocated data space */
5375         __int64_t               end,            /* last block requested */
5376         xfs_fsblock_t           startblock)
5377 {
5378         __int64_t               fixlen;
5379         xfs_mount_t             *mp;            /* file system mount point */
5380         xfs_ifork_t             *ifp;           /* inode fork pointer */
5381         xfs_extnum_t            lastx;          /* last extent pointer */
5382         xfs_fileoff_t           fileblock;
5383
5384         if (startblock == HOLESTARTBLOCK) {
5385                 mp = ip->i_mount;
5386                 out->bmv_block = -1;
5387                 fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, ip->i_size));
5388                 fixlen -= out->bmv_offset;
5389                 if (prealloced && out->bmv_offset + out->bmv_length == end) {
5390                         /* Came to hole at EOF. Trim it. */
5391                         if (fixlen <= 0)
5392                                 return 0;
5393                         out->bmv_length = fixlen;
5394                 }
5395         } else {
5396                 if (startblock == DELAYSTARTBLOCK)
5397                         out->bmv_block = -2;
5398                 else
5399                         out->bmv_block = xfs_fsb_to_db(ip, startblock);
5400                 fileblock = XFS_BB_TO_FSB(ip->i_mount, out->bmv_offset);
5401                 ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
5402                 if (xfs_iext_bno_to_ext(ifp, fileblock, &lastx) &&
5403                    (lastx == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t))-1))
5404                         out->bmv_oflags |= BMV_OF_LAST;
5405         }
5406
5407         return 1;
5408 }
5409
5410 /*
5411  * Get inode's extents as described in bmv, and format for output.
5412  * Calls formatter to fill the user's buffer until all extents
5413  * are mapped, until the passed-in bmv->bmv_count slots have
5414  * been filled, or until the formatter short-circuits the loop,
5415  * if it is tracking filled-in extents on its own.
5416  */
5417 int                                             /* error code */
5418 xfs_getbmap(
5419         xfs_inode_t             *ip,
5420         struct getbmapx         *bmv,           /* user bmap structure */
5421         xfs_bmap_format_t       formatter,      /* format to user */
5422         void                    *arg)           /* formatter arg */
5423 {
5424         __int64_t               bmvend;         /* last block requested */
5425         int                     error = 0;      /* return value */
5426         __int64_t               fixlen;         /* length for -1 case */
5427         int                     i;              /* extent number */
5428         int                     lock;           /* lock state */
5429         xfs_bmbt_irec_t         *map;           /* buffer for user's data */
5430         xfs_mount_t             *mp;            /* file system mount point */
5431         int                     nex;            /* # of user extents can do */
5432         int                     nexleft;        /* # of user extents left */
5433         int                     subnex;         /* # of bmapi's can do */
5434         int                     nmap;           /* number of map entries */
5435         struct getbmapx         *out;           /* output structure */
5436         int                     whichfork;      /* data or attr fork */
5437         int                     prealloced;     /* this is a file with
5438                                                  * preallocated data space */
5439         int                     iflags;         /* interface flags */
5440         int                     bmapi_flags;    /* flags for xfs_bmapi */
5441         int                     cur_ext = 0;
5442
5443         mp = ip->i_mount;
5444         iflags = bmv->bmv_iflags;
5445         whichfork = iflags & BMV_IF_ATTRFORK ? XFS_ATTR_FORK : XFS_DATA_FORK;
5446
5447         if (whichfork == XFS_ATTR_FORK) {
5448                 if (XFS_IFORK_Q(ip)) {
5449                         if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS &&
5450                             ip->i_d.di_aformat != XFS_DINODE_FMT_BTREE &&
5451                             ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)
5452                                 return XFS_ERROR(EINVAL);
5453                 } else if (unlikely(
5454                            ip->i_d.di_aformat != 0 &&
5455                            ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS)) {
5456                         XFS_ERROR_REPORT("xfs_getbmap", XFS_ERRLEVEL_LOW,
5457                                          ip->i_mount);
5458                         return XFS_ERROR(EFSCORRUPTED);
5459                 }
5460
5461                 prealloced = 0;
5462                 fixlen = 1LL << 32;
5463         } else {
5464                 if (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS &&
5465                     ip->i_d.di_format != XFS_DINODE_FMT_BTREE &&
5466                     ip->i_d.di_format != XFS_DINODE_FMT_LOCAL)
5467                         return XFS_ERROR(EINVAL);
5468
5469                 if (xfs_get_extsz_hint(ip) ||
5470                     ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC|XFS_DIFLAG_APPEND)){
5471                         prealloced = 1;
5472                         fixlen = XFS_MAXIOFFSET(mp);
5473                 } else {
5474                         prealloced = 0;
5475                         fixlen = ip->i_size;
5476                 }
5477         }
5478
5479         if (bmv->bmv_length == -1) {
5480                 fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, fixlen));
5481                 bmv->bmv_length =
5482                         max_t(__int64_t, fixlen - bmv->bmv_offset, 0);
5483         } else if (bmv->bmv_length == 0) {
5484                 bmv->bmv_entries = 0;
5485                 return 0;
5486         } else if (bmv->bmv_length < 0) {
5487                 return XFS_ERROR(EINVAL);
5488         }
5489
5490         nex = bmv->bmv_count - 1;
5491         if (nex <= 0)
5492                 return XFS_ERROR(EINVAL);
5493         bmvend = bmv->bmv_offset + bmv->bmv_length;
5494
5495
5496         if (bmv->bmv_count > ULONG_MAX / sizeof(struct getbmapx))
5497                 return XFS_ERROR(ENOMEM);
5498         out = kmem_zalloc(bmv->bmv_count * sizeof(struct getbmapx), KM_MAYFAIL);
5499         if (!out)
5500                 return XFS_ERROR(ENOMEM);
5501
5502         xfs_ilock(ip, XFS_IOLOCK_SHARED);
5503         if (whichfork == XFS_DATA_FORK && !(iflags & BMV_IF_DELALLOC)) {
5504                 if (ip->i_delayed_blks || ip->i_size > ip->i_d.di_size) {
5505                         error = xfs_flush_pages(ip, 0, -1, 0, FI_REMAPF);
5506                         if (error)
5507                                 goto out_unlock_iolock;
5508                 }
5509                 /*
5510                  * even after flushing the inode, there can still be delalloc
5511                  * blocks on the inode beyond EOF due to speculative
5512                  * preallocation. These are not removed until the release
5513                  * function is called or the inode is inactivated. Hence we
5514                  * cannot assert here that ip->i_delayed_blks == 0.
5515                  */
5516         }
5517
5518         lock = xfs_ilock_map_shared(ip);
5519
5520         /*
5521          * Don't let nex be bigger than the number of extents
5522          * we can have assuming alternating holes and real extents.
5523          */
5524         if (nex > XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1)
5525                 nex = XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1;
5526
5527         bmapi_flags = xfs_bmapi_aflag(whichfork);
5528         if (!(iflags & BMV_IF_PREALLOC))
5529                 bmapi_flags |= XFS_BMAPI_IGSTATE;
5530
5531         /*
5532          * Allocate enough space to handle "subnex" maps at a time.
5533          */
5534         error = ENOMEM;
5535         subnex = 16;
5536         map = kmem_alloc(subnex * sizeof(*map), KM_MAYFAIL | KM_NOFS);
5537         if (!map)
5538                 goto out_unlock_ilock;
5539
5540         bmv->bmv_entries = 0;
5541
5542         if (XFS_IFORK_NEXTENTS(ip, whichfork) == 0 &&
5543             (whichfork == XFS_ATTR_FORK || !(iflags & BMV_IF_DELALLOC))) {
5544                 error = 0;
5545                 goto out_free_map;
5546         }
5547
5548         nexleft = nex;
5549
5550         do {
5551                 nmap = (nexleft > subnex) ? subnex : nexleft;
5552                 error = xfs_bmapi(NULL, ip, XFS_BB_TO_FSBT(mp, bmv->bmv_offset),
5553                                   XFS_BB_TO_FSB(mp, bmv->bmv_length),
5554                                   bmapi_flags, NULL, 0, map, &nmap,
5555                                   NULL);
5556                 if (error)
5557                         goto out_free_map;
5558                 ASSERT(nmap <= subnex);
5559
5560                 for (i = 0; i < nmap && nexleft && bmv->bmv_length; i++) {
5561                         out[cur_ext].bmv_oflags = 0;
5562                         if (map[i].br_state == XFS_EXT_UNWRITTEN)
5563                                 out[cur_ext].bmv_oflags |= BMV_OF_PREALLOC;
5564                         else if (map[i].br_startblock == DELAYSTARTBLOCK)
5565                                 out[cur_ext].bmv_oflags |= BMV_OF_DELALLOC;
5566                         out[cur_ext].bmv_offset =
5567                                 XFS_FSB_TO_BB(mp, map[i].br_startoff);
5568                         out[cur_ext].bmv_length =
5569                                 XFS_FSB_TO_BB(mp, map[i].br_blockcount);
5570                         out[cur_ext].bmv_unused1 = 0;
5571                         out[cur_ext].bmv_unused2 = 0;
5572                         ASSERT(((iflags & BMV_IF_DELALLOC) != 0) ||
5573                               (map[i].br_startblock != DELAYSTARTBLOCK));
5574                         if (map[i].br_startblock == HOLESTARTBLOCK &&
5575                             whichfork == XFS_ATTR_FORK) {
5576                                 /* came to the end of attribute fork */
5577                                 out[cur_ext].bmv_oflags |= BMV_OF_LAST;
5578                                 goto out_free_map;
5579                         }
5580
5581                         if (!xfs_getbmapx_fix_eof_hole(ip, &out[cur_ext],
5582                                         prealloced, bmvend,
5583                                         map[i].br_startblock))
5584                                 goto out_free_map;
5585
5586                         bmv->bmv_offset =
5587                                 out[cur_ext].bmv_offset +
5588                                 out[cur_ext].bmv_length;
5589                         bmv->bmv_length =
5590                                 max_t(__int64_t, 0, bmvend - bmv->bmv_offset);
5591
5592                         /*
5593                          * In case we don't want to return the hole,
5594                          * don't increase cur_ext so that we can reuse
5595                          * it in the next loop.
5596                          */
5597                         if ((iflags & BMV_IF_NO_HOLES) &&
5598                             map[i].br_startblock == HOLESTARTBLOCK) {
5599                                 memset(&out[cur_ext], 0, sizeof(out[cur_ext]));
5600                                 continue;
5601                         }
5602
5603                         nexleft--;
5604                         bmv->bmv_entries++;
5605                         cur_ext++;
5606                 }
5607         } while (nmap && nexleft && bmv->bmv_length);
5608
5609  out_free_map:
5610         kmem_free(map);
5611  out_unlock_ilock:
5612         xfs_iunlock_map_shared(ip, lock);
5613  out_unlock_iolock:
5614         xfs_iunlock(ip, XFS_IOLOCK_SHARED);
5615
5616         for (i = 0; i < cur_ext; i++) {
5617                 int full = 0;   /* user array is full */
5618
5619                 /* format results & advance arg */
5620                 error = formatter(&arg, &out[i], &full);
5621                 if (error || full)
5622                         break;
5623         }
5624
5625         kmem_free(out);
5626         return error;
5627 }
5628
5629 /*
5630  * Check the last inode extent to determine whether this allocation will result
5631  * in blocks being allocated at the end of the file. When we allocate new data
5632  * blocks at the end of the file which do not start at the previous data block,
5633  * we will try to align the new blocks at stripe unit boundaries.
5634  */
5635 STATIC int                              /* error */
5636 xfs_bmap_isaeof(
5637         xfs_inode_t     *ip,            /* incore inode pointer */
5638         xfs_fileoff_t   off,            /* file offset in fsblocks */
5639         int             whichfork,      /* data or attribute fork */
5640         char            *aeof)          /* return value */
5641 {
5642         int             error;          /* error return value */
5643         xfs_ifork_t     *ifp;           /* inode fork pointer */
5644         xfs_bmbt_rec_host_t *lastrec;   /* extent record pointer */
5645         xfs_extnum_t    nextents;       /* number of file extents */
5646         xfs_bmbt_irec_t s;              /* expanded extent record */
5647
5648         ASSERT(whichfork == XFS_DATA_FORK);
5649         ifp = XFS_IFORK_PTR(ip, whichfork);
5650         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5651             (error = xfs_iread_extents(NULL, ip, whichfork)))
5652                 return error;
5653         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5654         if (nextents == 0) {
5655                 *aeof = 1;
5656                 return 0;
5657         }
5658         /*
5659          * Go to the last extent
5660          */
5661         lastrec = xfs_iext_get_ext(ifp, nextents - 1);
5662         xfs_bmbt_get_all(lastrec, &s);
5663         /*
5664          * Check we are allocating in the last extent (for delayed allocations)
5665          * or past the last extent for non-delayed allocations.
5666          */
5667         *aeof = (off >= s.br_startoff &&
5668                  off < s.br_startoff + s.br_blockcount &&
5669                  isnullstartblock(s.br_startblock)) ||
5670                 off >= s.br_startoff + s.br_blockcount;
5671         return 0;
5672 }
5673
5674 /*
5675  * Check if the endoff is outside the last extent. If so the caller will grow
5676  * the allocation to a stripe unit boundary.
5677  */
5678 int                                     /* error */
5679 xfs_bmap_eof(
5680         xfs_inode_t     *ip,            /* incore inode pointer */
5681         xfs_fileoff_t   endoff,         /* file offset in fsblocks */
5682         int             whichfork,      /* data or attribute fork */
5683         int             *eof)           /* result value */
5684 {
5685         xfs_fsblock_t   blockcount;     /* extent block count */
5686         int             error;          /* error return value */
5687         xfs_ifork_t     *ifp;           /* inode fork pointer */
5688         xfs_bmbt_rec_host_t *lastrec;   /* extent record pointer */
5689         xfs_extnum_t    nextents;       /* number of file extents */
5690         xfs_fileoff_t   startoff;       /* extent starting file offset */
5691
5692         ASSERT(whichfork == XFS_DATA_FORK);
5693         ifp = XFS_IFORK_PTR(ip, whichfork);
5694         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5695             (error = xfs_iread_extents(NULL, ip, whichfork)))
5696                 return error;
5697         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5698         if (nextents == 0) {
5699                 *eof = 1;
5700                 return 0;
5701         }
5702         /*
5703          * Go to the last extent
5704          */
5705         lastrec = xfs_iext_get_ext(ifp, nextents - 1);
5706         startoff = xfs_bmbt_get_startoff(lastrec);
5707         blockcount = xfs_bmbt_get_blockcount(lastrec);
5708         *eof = endoff >= startoff + blockcount;
5709         return 0;
5710 }
5711
5712 #ifdef DEBUG
5713 STATIC struct xfs_buf *
5714 xfs_bmap_get_bp(
5715         struct xfs_btree_cur    *cur,
5716         xfs_fsblock_t           bno)
5717 {
5718         struct xfs_log_item_desc *lidp;
5719         int                     i;
5720
5721         if (!cur)
5722                 return NULL;
5723
5724         for (i = 0; i < XFS_BTREE_MAXLEVELS; i++) {
5725                 if (!cur->bc_bufs[i])
5726                         break;
5727                 if (XFS_BUF_ADDR(cur->bc_bufs[i]) == bno)
5728                         return cur->bc_bufs[i];
5729         }
5730
5731         /* Chase down all the log items to see if the bp is there */
5732         list_for_each_entry(lidp, &cur->bc_tp->t_items, lid_trans) {
5733                 struct xfs_buf_log_item *bip;
5734                 bip = (struct xfs_buf_log_item *)lidp->lid_item;
5735                 if (bip->bli_item.li_type == XFS_LI_BUF &&
5736                     XFS_BUF_ADDR(bip->bli_buf) == bno)
5737                         return bip->bli_buf;
5738         }
5739
5740         return NULL;
5741 }
5742
5743 STATIC void
5744 xfs_check_block(
5745         struct xfs_btree_block  *block,
5746         xfs_mount_t             *mp,
5747         int                     root,
5748         short                   sz)
5749 {
5750         int                     i, j, dmxr;
5751         __be64                  *pp, *thispa;   /* pointer to block address */
5752         xfs_bmbt_key_t          *prevp, *keyp;
5753
5754         ASSERT(be16_to_cpu(block->bb_level) > 0);
5755
5756         prevp = NULL;
5757         for( i = 1; i <= xfs_btree_get_numrecs(block); i++) {
5758                 dmxr = mp->m_bmap_dmxr[0];
5759                 keyp = XFS_BMBT_KEY_ADDR(mp, block, i);
5760
5761                 if (prevp) {
5762                         ASSERT(be64_to_cpu(prevp->br_startoff) <
5763                                be64_to_cpu(keyp->br_startoff));
5764                 }
5765                 prevp = keyp;
5766
5767                 /*
5768                  * Compare the block numbers to see if there are dups.
5769                  */
5770                 if (root)
5771                         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, i, sz);
5772                 else
5773                         pp = XFS_BMBT_PTR_ADDR(mp, block, i, dmxr);
5774
5775                 for (j = i+1; j <= be16_to_cpu(block->bb_numrecs); j++) {
5776                         if (root)
5777                                 thispa = XFS_BMAP_BROOT_PTR_ADDR(mp, block, j, sz);
5778                         else
5779                                 thispa = XFS_BMBT_PTR_ADDR(mp, block, j, dmxr);
5780                         if (*thispa == *pp) {
5781                                 xfs_warn(mp, "%s: thispa(%d) == pp(%d) %Ld",
5782                                         __func__, j, i,
5783                                         (unsigned long long)be64_to_cpu(*thispa));
5784                                 panic("%s: ptrs are equal in node\n",
5785                                         __func__);
5786                         }
5787                 }
5788         }
5789 }
5790
5791 /*
5792  * Check that the extents for the inode ip are in the right order in all
5793  * btree leaves.
5794  */
5795
5796 STATIC void
5797 xfs_bmap_check_leaf_extents(
5798         xfs_btree_cur_t         *cur,   /* btree cursor or null */
5799         xfs_inode_t             *ip,            /* incore inode pointer */
5800         int                     whichfork)      /* data or attr fork */
5801 {
5802         struct xfs_btree_block  *block; /* current btree block */
5803         xfs_fsblock_t           bno;    /* block # of "block" */
5804         xfs_buf_t               *bp;    /* buffer for "block" */
5805         int                     error;  /* error return value */
5806         xfs_extnum_t            i=0, j; /* index into the extents list */
5807         xfs_ifork_t             *ifp;   /* fork structure */
5808         int                     level;  /* btree level, for checking */
5809         xfs_mount_t             *mp;    /* file system mount structure */
5810         __be64                  *pp;    /* pointer to block address */
5811         xfs_bmbt_rec_t          *ep;    /* pointer to current extent */
5812         xfs_bmbt_rec_t          last = {0, 0}; /* last extent in prev block */
5813         xfs_bmbt_rec_t          *nextp; /* pointer to next extent */
5814         int                     bp_release = 0;
5815
5816         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE) {
5817                 return;
5818         }
5819
5820         bno = NULLFSBLOCK;
5821         mp = ip->i_mount;
5822         ifp = XFS_IFORK_PTR(ip, whichfork);
5823         block = ifp->if_broot;
5824         /*
5825          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
5826          */
5827         level = be16_to_cpu(block->bb_level);
5828         ASSERT(level > 0);
5829         xfs_check_block(block, mp, 1, ifp->if_broot_bytes);
5830         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
5831         bno = be64_to_cpu(*pp);
5832
5833         ASSERT(bno != NULLDFSBNO);
5834         ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
5835         ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
5836
5837         /*
5838          * Go down the tree until leaf level is reached, following the first
5839          * pointer (leftmost) at each level.
5840          */
5841         while (level-- > 0) {
5842                 /* See if buf is in cur first */
5843                 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
5844                 if (bp) {
5845                         bp_release = 0;
5846                 } else {
5847                         bp_release = 1;
5848                 }
5849                 if (!bp && (error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
5850                                 XFS_BMAP_BTREE_REF)))
5851                         goto error_norelse;
5852                 block = XFS_BUF_TO_BLOCK(bp);
5853                 XFS_WANT_CORRUPTED_GOTO(
5854                         xfs_bmap_sanity_check(mp, bp, level),
5855                         error0);
5856                 if (level == 0)
5857                         break;
5858
5859                 /*
5860                  * Check this block for basic sanity (increasing keys and
5861                  * no duplicate blocks).
5862                  */
5863
5864                 xfs_check_block(block, mp, 0, 0);
5865                 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
5866                 bno = be64_to_cpu(*pp);
5867                 XFS_WANT_CORRUPTED_GOTO(XFS_FSB_SANITY_CHECK(mp, bno), error0);
5868                 if (bp_release) {
5869                         bp_release = 0;
5870                         xfs_trans_brelse(NULL, bp);
5871                 }
5872         }
5873
5874         /*
5875          * Here with bp and block set to the leftmost leaf node in the tree.
5876          */
5877         i = 0;
5878
5879         /*
5880          * Loop over all leaf nodes checking that all extents are in the right order.
5881          */
5882         for (;;) {
5883                 xfs_fsblock_t   nextbno;
5884                 xfs_extnum_t    num_recs;
5885
5886
5887                 num_recs = xfs_btree_get_numrecs(block);
5888
5889                 /*
5890                  * Read-ahead the next leaf block, if any.
5891                  */
5892
5893                 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
5894
5895                 /*
5896                  * Check all the extents to make sure they are OK.
5897                  * If we had a previous block, the last entry should
5898                  * conform with the first entry in this one.
5899                  */
5900
5901                 ep = XFS_BMBT_REC_ADDR(mp, block, 1);
5902                 if (i) {
5903                         ASSERT(xfs_bmbt_disk_get_startoff(&last) +
5904                                xfs_bmbt_disk_get_blockcount(&last) <=
5905                                xfs_bmbt_disk_get_startoff(ep));
5906                 }
5907                 for (j = 1; j < num_recs; j++) {
5908                         nextp = XFS_BMBT_REC_ADDR(mp, block, j + 1);
5909                         ASSERT(xfs_bmbt_disk_get_startoff(ep) +
5910                                xfs_bmbt_disk_get_blockcount(ep) <=
5911                                xfs_bmbt_disk_get_startoff(nextp));
5912                         ep = nextp;
5913                 }
5914
5915                 last = *ep;
5916                 i += num_recs;
5917                 if (bp_release) {
5918                         bp_release = 0;
5919                         xfs_trans_brelse(NULL, bp);
5920                 }
5921                 bno = nextbno;
5922                 /*
5923                  * If we've reached the end, stop.
5924                  */
5925                 if (bno == NULLFSBLOCK)
5926                         break;
5927
5928                 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
5929                 if (bp) {
5930                         bp_release = 0;
5931                 } else {
5932                         bp_release = 1;
5933                 }
5934                 if (!bp && (error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
5935                                 XFS_BMAP_BTREE_REF)))
5936                         goto error_norelse;
5937                 block = XFS_BUF_TO_BLOCK(bp);
5938         }
5939         if (bp_release) {
5940                 bp_release = 0;
5941                 xfs_trans_brelse(NULL, bp);
5942         }
5943         return;
5944
5945 error0:
5946         xfs_warn(mp, "%s: at error0", __func__);
5947         if (bp_release)
5948                 xfs_trans_brelse(NULL, bp);
5949 error_norelse:
5950         xfs_warn(mp, "%s: BAD after btree leaves for %d extents",
5951                 __func__, i);
5952         panic("%s: CORRUPTED BTREE OR SOMETHING", __func__);
5953         return;
5954 }
5955 #endif
5956
5957 /*
5958  * Count fsblocks of the given fork.
5959  */
5960 int                                             /* error */
5961 xfs_bmap_count_blocks(
5962         xfs_trans_t             *tp,            /* transaction pointer */
5963         xfs_inode_t             *ip,            /* incore inode */
5964         int                     whichfork,      /* data or attr fork */
5965         int                     *count)         /* out: count of blocks */
5966 {
5967         struct xfs_btree_block  *block; /* current btree block */
5968         xfs_fsblock_t           bno;    /* block # of "block" */
5969         xfs_ifork_t             *ifp;   /* fork structure */
5970         int                     level;  /* btree level, for checking */
5971         xfs_mount_t             *mp;    /* file system mount structure */
5972         __be64                  *pp;    /* pointer to block address */
5973
5974         bno = NULLFSBLOCK;
5975         mp = ip->i_mount;
5976         ifp = XFS_IFORK_PTR(ip, whichfork);
5977         if ( XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ) {
5978                 xfs_bmap_count_leaves(ifp, 0,
5979                         ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t),
5980                         count);
5981                 return 0;
5982         }
5983
5984         /*
5985          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
5986          */
5987         block = ifp->if_broot;
5988         level = be16_to_cpu(block->bb_level);
5989         ASSERT(level > 0);
5990         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
5991         bno = be64_to_cpu(*pp);
5992         ASSERT(bno != NULLDFSBNO);
5993         ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
5994         ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
5995
5996         if (unlikely(xfs_bmap_count_tree(mp, tp, ifp, bno, level, count) < 0)) {
5997                 XFS_ERROR_REPORT("xfs_bmap_count_blocks(2)", XFS_ERRLEVEL_LOW,
5998                                  mp);
5999                 return XFS_ERROR(EFSCORRUPTED);
6000         }
6001
6002         return 0;
6003 }
6004
6005 /*
6006  * Recursively walks each level of a btree
6007  * to count total fsblocks is use.
6008  */
6009 STATIC int                                     /* error */
6010 xfs_bmap_count_tree(
6011         xfs_mount_t     *mp,            /* file system mount point */
6012         xfs_trans_t     *tp,            /* transaction pointer */
6013         xfs_ifork_t     *ifp,           /* inode fork pointer */
6014         xfs_fsblock_t   blockno,        /* file system block number */
6015         int             levelin,        /* level in btree */
6016         int             *count)         /* Count of blocks */
6017 {
6018         int                     error;
6019         xfs_buf_t               *bp, *nbp;
6020         int                     level = levelin;
6021         __be64                  *pp;
6022         xfs_fsblock_t           bno = blockno;
6023         xfs_fsblock_t           nextbno;
6024         struct xfs_btree_block  *block, *nextblock;
6025         int                     numrecs;
6026
6027         if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp, XFS_BMAP_BTREE_REF)))
6028                 return error;
6029         *count += 1;
6030         block = XFS_BUF_TO_BLOCK(bp);
6031
6032         if (--level) {
6033                 /* Not at node above leaves, count this level of nodes */
6034                 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
6035                 while (nextbno != NULLFSBLOCK) {
6036                         if ((error = xfs_btree_read_bufl(mp, tp, nextbno,
6037                                 0, &nbp, XFS_BMAP_BTREE_REF)))
6038                                 return error;
6039                         *count += 1;
6040                         nextblock = XFS_BUF_TO_BLOCK(nbp);
6041                         nextbno = be64_to_cpu(nextblock->bb_u.l.bb_rightsib);
6042                         xfs_trans_brelse(tp, nbp);
6043                 }
6044
6045                 /* Dive to the next level */
6046                 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
6047                 bno = be64_to_cpu(*pp);
6048                 if (unlikely((error =
6049                      xfs_bmap_count_tree(mp, tp, ifp, bno, level, count)) < 0)) {
6050                         xfs_trans_brelse(tp, bp);
6051                         XFS_ERROR_REPORT("xfs_bmap_count_tree(1)",
6052                                          XFS_ERRLEVEL_LOW, mp);
6053                         return XFS_ERROR(EFSCORRUPTED);
6054                 }
6055                 xfs_trans_brelse(tp, bp);
6056         } else {
6057                 /* count all level 1 nodes and their leaves */
6058                 for (;;) {
6059                         nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
6060                         numrecs = be16_to_cpu(block->bb_numrecs);
6061                         xfs_bmap_disk_count_leaves(mp, block, numrecs, count);
6062                         xfs_trans_brelse(tp, bp);
6063                         if (nextbno == NULLFSBLOCK)
6064                                 break;
6065                         bno = nextbno;
6066                         if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
6067                                 XFS_BMAP_BTREE_REF)))
6068                                 return error;
6069                         *count += 1;
6070                         block = XFS_BUF_TO_BLOCK(bp);
6071                 }
6072         }
6073         return 0;
6074 }
6075
6076 /*
6077  * Count leaf blocks given a range of extent records.
6078  */
6079 STATIC void
6080 xfs_bmap_count_leaves(
6081         xfs_ifork_t             *ifp,
6082         xfs_extnum_t            idx,
6083         int                     numrecs,
6084         int                     *count)
6085 {
6086         int             b;
6087
6088         for (b = 0; b < numrecs; b++) {
6089                 xfs_bmbt_rec_host_t *frp = xfs_iext_get_ext(ifp, idx + b);
6090                 *count += xfs_bmbt_get_blockcount(frp);
6091         }
6092 }
6093
6094 /*
6095  * Count leaf blocks given a range of extent records originally
6096  * in btree format.
6097  */
6098 STATIC void
6099 xfs_bmap_disk_count_leaves(
6100         struct xfs_mount        *mp,
6101         struct xfs_btree_block  *block,
6102         int                     numrecs,
6103         int                     *count)
6104 {
6105         int             b;
6106         xfs_bmbt_rec_t  *frp;
6107
6108         for (b = 1; b <= numrecs; b++) {
6109                 frp = XFS_BMBT_REC_ADDR(mp, block, b);
6110                 *count += xfs_bmbt_disk_get_blockcount(frp);
6111         }
6112 }
6113
6114 /*
6115  * dead simple method of punching delalyed allocation blocks from a range in
6116  * the inode. Walks a block at a time so will be slow, but is only executed in
6117  * rare error cases so the overhead is not critical. This will alays punch out
6118  * both the start and end blocks, even if the ranges only partially overlap
6119  * them, so it is up to the caller to ensure that partial blocks are not
6120  * passed in.
6121  */
6122 int
6123 xfs_bmap_punch_delalloc_range(
6124         struct xfs_inode        *ip,
6125         xfs_fileoff_t           start_fsb,
6126         xfs_fileoff_t           length)
6127 {
6128         xfs_fileoff_t           remaining = length;
6129         int                     error = 0;
6130
6131         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
6132
6133         do {
6134                 int             done;
6135                 xfs_bmbt_irec_t imap;
6136                 int             nimaps = 1;
6137                 xfs_fsblock_t   firstblock;
6138                 xfs_bmap_free_t flist;
6139
6140                 /*
6141                  * Map the range first and check that it is a delalloc extent
6142                  * before trying to unmap the range. Otherwise we will be
6143                  * trying to remove a real extent (which requires a
6144                  * transaction) or a hole, which is probably a bad idea...
6145                  */
6146                 error = xfs_bmapi(NULL, ip, start_fsb, 1,
6147                                 XFS_BMAPI_ENTIRE,  NULL, 0, &imap,
6148                                 &nimaps, NULL);
6149
6150                 if (error) {
6151                         /* something screwed, just bail */
6152                         if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
6153                                 xfs_alert(ip->i_mount,
6154                         "Failed delalloc mapping lookup ino %lld fsb %lld.",
6155                                                 ip->i_ino, start_fsb);
6156                         }
6157                         break;
6158                 }
6159                 if (!nimaps) {
6160                         /* nothing there */
6161                         goto next_block;
6162                 }
6163                 if (imap.br_startblock != DELAYSTARTBLOCK) {
6164                         /* been converted, ignore */
6165                         goto next_block;
6166                 }
6167                 WARN_ON(imap.br_blockcount == 0);
6168
6169                 /*
6170                  * Note: while we initialise the firstblock/flist pair, they
6171                  * should never be used because blocks should never be
6172                  * allocated or freed for a delalloc extent and hence we need
6173                  * don't cancel or finish them after the xfs_bunmapi() call.
6174                  */
6175                 xfs_bmap_init(&flist, &firstblock);
6176                 error = xfs_bunmapi(NULL, ip, start_fsb, 1, 0, 1, &firstblock,
6177                                         &flist, &done);
6178                 if (error)
6179                         break;
6180
6181                 ASSERT(!flist.xbf_count && !flist.xbf_first);
6182 next_block:
6183                 start_fsb++;
6184                 remaining--;
6185         } while(remaining > 0);
6186
6187         return error;
6188 }