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                 temp = new->br_startoff - PREV.br_startoff;
1043                 trace_xfs_bmap_pre_update(ip, idx, 0, _THIS_IP_);
1044                 xfs_bmbt_set_blockcount(ep, temp);
1045                 r[0] = *new;
1046                 r[1].br_state = PREV.br_state;
1047                 r[1].br_startblock = 0;
1048                 r[1].br_startoff = new_endoff;
1049                 temp2 = PREV.br_startoff + PREV.br_blockcount - new_endoff;
1050                 r[1].br_blockcount = temp2;
1051                 xfs_iext_insert(ip, idx + 1, 2, &r[0], state);
1052                 ip->i_df.if_lastex = idx + 1;
1053                 ip->i_d.di_nextents++;
1054                 if (cur == NULL)
1055                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1056                 else {
1057                         rval = XFS_ILOG_CORE;
1058                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1059                                         new->br_startblock, new->br_blockcount,
1060                                         &i)))
1061                                 goto done;
1062                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1063                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
1064                         if ((error = xfs_btree_insert(cur, &i)))
1065                                 goto done;
1066                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1067                 }
1068                 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1069                     ip->i_d.di_nextents > ip->i_df.if_ext_max) {
1070                         error = xfs_bmap_extents_to_btree(ip->i_transp, ip,
1071                                         first, flist, &cur, 1, &tmp_rval,
1072                                         XFS_DATA_FORK);
1073                         rval |= tmp_rval;
1074                         if (error)
1075                                 goto done;
1076                 }
1077                 temp = xfs_bmap_worst_indlen(ip, temp);
1078                 temp2 = xfs_bmap_worst_indlen(ip, temp2);
1079                 diff = (int)(temp + temp2 - startblockval(PREV.br_startblock) -
1080                         (cur ? cur->bc_private.b.allocated : 0));
1081                 if (diff > 0 &&
1082                     xfs_icsb_modify_counters(ip->i_mount, XFS_SBS_FDBLOCKS,
1083                                              -((int64_t)diff), rsvd)) {
1084                         /*
1085                          * Ick gross gag me with a spoon.
1086                          */
1087                         ASSERT(0);      /* want to see if this ever happens! */
1088                         while (diff > 0) {
1089                                 if (temp) {
1090                                         temp--;
1091                                         diff--;
1092                                         if (!diff ||
1093                                             !xfs_icsb_modify_counters(ip->i_mount,
1094                                                     XFS_SBS_FDBLOCKS,
1095                                                     -((int64_t)diff), rsvd))
1096                                                 break;
1097                                 }
1098                                 if (temp2) {
1099                                         temp2--;
1100                                         diff--;
1101                                         if (!diff ||
1102                                             !xfs_icsb_modify_counters(ip->i_mount,
1103                                                     XFS_SBS_FDBLOCKS,
1104                                                     -((int64_t)diff), rsvd))
1105                                                 break;
1106                                 }
1107                         }
1108                 }
1109                 ep = xfs_iext_get_ext(ifp, idx);
1110                 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
1111                 trace_xfs_bmap_post_update(ip, idx, state, _THIS_IP_);
1112                 trace_xfs_bmap_pre_update(ip, idx + 2, state, _THIS_IP_);
1113                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, idx + 2),
1114                         nullstartblock((int)temp2));
1115                 trace_xfs_bmap_post_update(ip, idx + 2, state, _THIS_IP_);
1116                 *dnew = temp + temp2;
1117                 break;
1118
1119         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1120         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1121         case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
1122         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1123         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1124         case BMAP_LEFT_CONTIG:
1125         case BMAP_RIGHT_CONTIG:
1126                 /*
1127                  * These cases are all impossible.
1128                  */
1129                 ASSERT(0);
1130         }
1131         *curp = cur;
1132 done:
1133         *logflagsp = rval;
1134         return error;
1135 #undef  LEFT
1136 #undef  RIGHT
1137 #undef  PREV
1138 }
1139
1140 /*
1141  * Called by xfs_bmap_add_extent to handle cases converting an unwritten
1142  * allocation to a real allocation or vice versa.
1143  */
1144 STATIC int                              /* error */
1145 xfs_bmap_add_extent_unwritten_real(
1146         xfs_inode_t             *ip,    /* incore inode pointer */
1147         xfs_extnum_t            idx,    /* extent number to update/insert */
1148         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
1149         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
1150         int                     *logflagsp) /* inode logging flags */
1151 {
1152         xfs_btree_cur_t         *cur;   /* btree cursor */
1153         xfs_bmbt_rec_host_t     *ep;    /* extent entry for idx */
1154         int                     error;  /* error return value */
1155         int                     i;      /* temp state */
1156         xfs_ifork_t             *ifp;   /* inode fork pointer */
1157         xfs_fileoff_t           new_endoff;     /* end offset of new entry */
1158         xfs_exntst_t            newext; /* new extent state */
1159         xfs_exntst_t            oldext; /* old extent state */
1160         xfs_bmbt_irec_t         r[3];   /* neighbor extent entries */
1161                                         /* left is 0, right is 1, prev is 2 */
1162         int                     rval=0; /* return value (logging flags) */
1163         int                     state = 0;/* state bits, accessed thru macros */
1164
1165 #define LEFT            r[0]
1166 #define RIGHT           r[1]
1167 #define PREV            r[2]
1168         /*
1169          * Set up a bunch of variables to make the tests simpler.
1170          */
1171         error = 0;
1172         cur = *curp;
1173         ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
1174         ep = xfs_iext_get_ext(ifp, idx);
1175         xfs_bmbt_get_all(ep, &PREV);
1176         newext = new->br_state;
1177         oldext = (newext == XFS_EXT_UNWRITTEN) ?
1178                 XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
1179         ASSERT(PREV.br_state == oldext);
1180         new_endoff = new->br_startoff + new->br_blockcount;
1181         ASSERT(PREV.br_startoff <= new->br_startoff);
1182         ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
1183
1184         /*
1185          * Set flags determining what part of the previous oldext allocation
1186          * extent is being replaced by a newext allocation.
1187          */
1188         if (PREV.br_startoff == new->br_startoff)
1189                 state |= BMAP_LEFT_FILLING;
1190         if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
1191                 state |= BMAP_RIGHT_FILLING;
1192
1193         /*
1194          * Check and set flags if this segment has a left neighbor.
1195          * Don't set contiguous if the combined extent would be too large.
1196          */
1197         if (idx > 0) {
1198                 state |= BMAP_LEFT_VALID;
1199                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, idx - 1), &LEFT);
1200
1201                 if (isnullstartblock(LEFT.br_startblock))
1202                         state |= BMAP_LEFT_DELAY;
1203         }
1204
1205         if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
1206             LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
1207             LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
1208             LEFT.br_state == newext &&
1209             LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
1210                 state |= BMAP_LEFT_CONTIG;
1211
1212         /*
1213          * Check and set flags if this segment has a right neighbor.
1214          * Don't set contiguous if the combined extent would be too large.
1215          * Also check for all-three-contiguous being too large.
1216          */
1217         if (idx < ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1) {
1218                 state |= BMAP_RIGHT_VALID;
1219                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, idx + 1), &RIGHT);
1220                 if (isnullstartblock(RIGHT.br_startblock))
1221                         state |= BMAP_RIGHT_DELAY;
1222         }
1223
1224         if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
1225             new_endoff == RIGHT.br_startoff &&
1226             new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
1227             newext == RIGHT.br_state &&
1228             new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
1229             ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1230                        BMAP_RIGHT_FILLING)) !=
1231                       (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1232                        BMAP_RIGHT_FILLING) ||
1233              LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
1234                         <= MAXEXTLEN))
1235                 state |= BMAP_RIGHT_CONTIG;
1236
1237         /*
1238          * Switch out based on the FILLING and CONTIG state bits.
1239          */
1240         switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1241                          BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
1242         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1243              BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1244                 /*
1245                  * Setting all of a previous oldext extent to newext.
1246                  * The left and right neighbors are both contiguous with new.
1247                  */
1248                 trace_xfs_bmap_pre_update(ip, idx - 1, state, _THIS_IP_);
1249                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
1250                         LEFT.br_blockcount + PREV.br_blockcount +
1251                         RIGHT.br_blockcount);
1252                 trace_xfs_bmap_post_update(ip, idx - 1, state, _THIS_IP_);
1253
1254                 xfs_iext_remove(ip, idx, 2, state);
1255                 ip->i_df.if_lastex = idx - 1;
1256                 ip->i_d.di_nextents -= 2;
1257                 if (cur == NULL)
1258                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1259                 else {
1260                         rval = XFS_ILOG_CORE;
1261                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
1262                                         RIGHT.br_startblock,
1263                                         RIGHT.br_blockcount, &i)))
1264                                 goto done;
1265                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1266                         if ((error = xfs_btree_delete(cur, &i)))
1267                                 goto done;
1268                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1269                         if ((error = xfs_btree_decrement(cur, 0, &i)))
1270                                 goto done;
1271                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1272                         if ((error = xfs_btree_delete(cur, &i)))
1273                                 goto done;
1274                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1275                         if ((error = xfs_btree_decrement(cur, 0, &i)))
1276                                 goto done;
1277                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1278                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
1279                                 LEFT.br_startblock,
1280                                 LEFT.br_blockcount + PREV.br_blockcount +
1281                                 RIGHT.br_blockcount, LEFT.br_state)))
1282                                 goto done;
1283                 }
1284                 break;
1285
1286         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1287                 /*
1288                  * Setting all of a previous oldext extent to newext.
1289                  * The left neighbor is contiguous, the right is not.
1290                  */
1291                 trace_xfs_bmap_pre_update(ip, idx - 1, state, _THIS_IP_);
1292                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
1293                         LEFT.br_blockcount + PREV.br_blockcount);
1294                 trace_xfs_bmap_post_update(ip, idx - 1, state, _THIS_IP_);
1295
1296                 ip->i_df.if_lastex = idx - 1;
1297                 xfs_iext_remove(ip, idx, 1, state);
1298                 ip->i_d.di_nextents--;
1299                 if (cur == NULL)
1300                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1301                 else {
1302                         rval = XFS_ILOG_CORE;
1303                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1304                                         PREV.br_startblock, PREV.br_blockcount,
1305                                         &i)))
1306                                 goto done;
1307                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1308                         if ((error = xfs_btree_delete(cur, &i)))
1309                                 goto done;
1310                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1311                         if ((error = xfs_btree_decrement(cur, 0, &i)))
1312                                 goto done;
1313                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1314                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
1315                                 LEFT.br_startblock,
1316                                 LEFT.br_blockcount + PREV.br_blockcount,
1317                                 LEFT.br_state)))
1318                                 goto done;
1319                 }
1320                 break;
1321
1322         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1323                 /*
1324                  * Setting all of a previous oldext extent to newext.
1325                  * The right neighbor is contiguous, the left is not.
1326                  */
1327                 trace_xfs_bmap_pre_update(ip, idx, state, _THIS_IP_);
1328                 xfs_bmbt_set_blockcount(ep,
1329                         PREV.br_blockcount + RIGHT.br_blockcount);
1330                 xfs_bmbt_set_state(ep, newext);
1331                 trace_xfs_bmap_post_update(ip, idx, state, _THIS_IP_);
1332                 ip->i_df.if_lastex = idx;
1333                 xfs_iext_remove(ip, idx + 1, 1, state);
1334                 ip->i_d.di_nextents--;
1335                 if (cur == NULL)
1336                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1337                 else {
1338                         rval = XFS_ILOG_CORE;
1339                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
1340                                         RIGHT.br_startblock,
1341                                         RIGHT.br_blockcount, &i)))
1342                                 goto done;
1343                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1344                         if ((error = xfs_btree_delete(cur, &i)))
1345                                 goto done;
1346                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1347                         if ((error = xfs_btree_decrement(cur, 0, &i)))
1348                                 goto done;
1349                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1350                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
1351                                 new->br_startblock,
1352                                 new->br_blockcount + RIGHT.br_blockcount,
1353                                 newext)))
1354                                 goto done;
1355                 }
1356                 break;
1357
1358         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
1359                 /*
1360                  * Setting all of a previous oldext extent to newext.
1361                  * Neither the left nor right neighbors are contiguous with
1362                  * the new one.
1363                  */
1364                 trace_xfs_bmap_pre_update(ip, idx, state, _THIS_IP_);
1365                 xfs_bmbt_set_state(ep, newext);
1366                 trace_xfs_bmap_post_update(ip, idx, state, _THIS_IP_);
1367
1368                 ip->i_df.if_lastex = idx;
1369                 if (cur == NULL)
1370                         rval = XFS_ILOG_DEXT;
1371                 else {
1372                         rval = 0;
1373                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1374                                         new->br_startblock, new->br_blockcount,
1375                                         &i)))
1376                                 goto done;
1377                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1378                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
1379                                 new->br_startblock, new->br_blockcount,
1380                                 newext)))
1381                                 goto done;
1382                 }
1383                 break;
1384
1385         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
1386                 /*
1387                  * Setting the first part of a previous oldext extent to newext.
1388                  * The left neighbor is contiguous.
1389                  */
1390                 trace_xfs_bmap_pre_update(ip, idx - 1, state, _THIS_IP_);
1391                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
1392                         LEFT.br_blockcount + new->br_blockcount);
1393                 xfs_bmbt_set_startoff(ep,
1394                         PREV.br_startoff + new->br_blockcount);
1395                 trace_xfs_bmap_post_update(ip, idx - 1, state, _THIS_IP_);
1396
1397                 trace_xfs_bmap_pre_update(ip, idx, state, _THIS_IP_);
1398                 xfs_bmbt_set_startblock(ep,
1399                         new->br_startblock + new->br_blockcount);
1400                 xfs_bmbt_set_blockcount(ep,
1401                         PREV.br_blockcount - new->br_blockcount);
1402                 trace_xfs_bmap_post_update(ip, idx, state, _THIS_IP_);
1403
1404                 ip->i_df.if_lastex = idx - 1;
1405                 if (cur == NULL)
1406                         rval = XFS_ILOG_DEXT;
1407                 else {
1408                         rval = 0;
1409                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1410                                         PREV.br_startblock, PREV.br_blockcount,
1411                                         &i)))
1412                                 goto done;
1413                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1414                         if ((error = xfs_bmbt_update(cur,
1415                                 PREV.br_startoff + new->br_blockcount,
1416                                 PREV.br_startblock + new->br_blockcount,
1417                                 PREV.br_blockcount - new->br_blockcount,
1418                                 oldext)))
1419                                 goto done;
1420                         if ((error = xfs_btree_decrement(cur, 0, &i)))
1421                                 goto done;
1422                         if (xfs_bmbt_update(cur, LEFT.br_startoff,
1423                                 LEFT.br_startblock,
1424                                 LEFT.br_blockcount + new->br_blockcount,
1425                                 LEFT.br_state))
1426                                 goto done;
1427                 }
1428                 break;
1429
1430         case BMAP_LEFT_FILLING:
1431                 /*
1432                  * Setting the first part of a previous oldext extent to newext.
1433                  * The left neighbor is not contiguous.
1434                  */
1435                 trace_xfs_bmap_pre_update(ip, idx, state, _THIS_IP_);
1436                 ASSERT(ep && xfs_bmbt_get_state(ep) == oldext);
1437                 xfs_bmbt_set_startoff(ep, new_endoff);
1438                 xfs_bmbt_set_blockcount(ep,
1439                         PREV.br_blockcount - new->br_blockcount);
1440                 xfs_bmbt_set_startblock(ep,
1441                         new->br_startblock + new->br_blockcount);
1442                 trace_xfs_bmap_post_update(ip, idx, state, _THIS_IP_);
1443
1444                 xfs_iext_insert(ip, idx, 1, new, state);
1445                 ip->i_df.if_lastex = idx;
1446                 ip->i_d.di_nextents++;
1447                 if (cur == NULL)
1448                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1449                 else {
1450                         rval = XFS_ILOG_CORE;
1451                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1452                                         PREV.br_startblock, PREV.br_blockcount,
1453                                         &i)))
1454                                 goto done;
1455                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1456                         if ((error = xfs_bmbt_update(cur,
1457                                 PREV.br_startoff + new->br_blockcount,
1458                                 PREV.br_startblock + new->br_blockcount,
1459                                 PREV.br_blockcount - new->br_blockcount,
1460                                 oldext)))
1461                                 goto done;
1462                         cur->bc_rec.b = *new;
1463                         if ((error = xfs_btree_insert(cur, &i)))
1464                                 goto done;
1465                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1466                 }
1467                 break;
1468
1469         case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1470                 /*
1471                  * Setting the last part of a previous oldext extent to newext.
1472                  * The right neighbor is contiguous with the new allocation.
1473                  */
1474                 trace_xfs_bmap_pre_update(ip, idx, state, _THIS_IP_);
1475                 trace_xfs_bmap_pre_update(ip, idx + 1, state, _THIS_IP_);
1476                 xfs_bmbt_set_blockcount(ep,
1477                         PREV.br_blockcount - new->br_blockcount);
1478                 trace_xfs_bmap_post_update(ip, idx, state, _THIS_IP_);
1479                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, idx + 1),
1480                         new->br_startoff, new->br_startblock,
1481                         new->br_blockcount + RIGHT.br_blockcount, newext);
1482                 trace_xfs_bmap_post_update(ip, idx + 1, state, _THIS_IP_);
1483
1484                 ip->i_df.if_lastex = idx + 1;
1485                 if (cur == NULL)
1486                         rval = XFS_ILOG_DEXT;
1487                 else {
1488                         rval = 0;
1489                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1490                                         PREV.br_startblock,
1491                                         PREV.br_blockcount, &i)))
1492                                 goto done;
1493                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1494                         if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
1495                                 PREV.br_startblock,
1496                                 PREV.br_blockcount - new->br_blockcount,
1497                                 oldext)))
1498                                 goto done;
1499                         if ((error = xfs_btree_increment(cur, 0, &i)))
1500                                 goto done;
1501                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
1502                                 new->br_startblock,
1503                                 new->br_blockcount + RIGHT.br_blockcount,
1504                                 newext)))
1505                                 goto done;
1506                 }
1507                 break;
1508
1509         case BMAP_RIGHT_FILLING:
1510                 /*
1511                  * Setting the last part of a previous oldext extent to newext.
1512                  * The right neighbor is not contiguous.
1513                  */
1514                 trace_xfs_bmap_pre_update(ip, idx, state, _THIS_IP_);
1515                 xfs_bmbt_set_blockcount(ep,
1516                         PREV.br_blockcount - new->br_blockcount);
1517                 trace_xfs_bmap_post_update(ip, idx, state, _THIS_IP_);
1518
1519                 xfs_iext_insert(ip, idx + 1, 1, new, state);
1520                 ip->i_df.if_lastex = idx + 1;
1521                 ip->i_d.di_nextents++;
1522                 if (cur == NULL)
1523                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1524                 else {
1525                         rval = XFS_ILOG_CORE;
1526                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1527                                         PREV.br_startblock, PREV.br_blockcount,
1528                                         &i)))
1529                                 goto done;
1530                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1531                         if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
1532                                 PREV.br_startblock,
1533                                 PREV.br_blockcount - new->br_blockcount,
1534                                 oldext)))
1535                                 goto done;
1536                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1537                                         new->br_startblock, new->br_blockcount,
1538                                         &i)))
1539                                 goto done;
1540                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1541                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
1542                         if ((error = xfs_btree_insert(cur, &i)))
1543                                 goto done;
1544                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1545                 }
1546                 break;
1547
1548         case 0:
1549                 /*
1550                  * Setting the middle part of a previous oldext extent to
1551                  * newext.  Contiguity is impossible here.
1552                  * One extent becomes three extents.
1553                  */
1554                 trace_xfs_bmap_pre_update(ip, idx, state, _THIS_IP_);
1555                 xfs_bmbt_set_blockcount(ep,
1556                         new->br_startoff - PREV.br_startoff);
1557                 trace_xfs_bmap_post_update(ip, idx, state, _THIS_IP_);
1558
1559                 r[0] = *new;
1560                 r[1].br_startoff = new_endoff;
1561                 r[1].br_blockcount =
1562                         PREV.br_startoff + PREV.br_blockcount - new_endoff;
1563                 r[1].br_startblock = new->br_startblock + new->br_blockcount;
1564                 r[1].br_state = oldext;
1565                 xfs_iext_insert(ip, idx + 1, 2, &r[0], state);
1566                 ip->i_df.if_lastex = idx + 1;
1567                 ip->i_d.di_nextents += 2;
1568                 if (cur == NULL)
1569                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1570                 else {
1571                         rval = XFS_ILOG_CORE;
1572                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1573                                         PREV.br_startblock, PREV.br_blockcount,
1574                                         &i)))
1575                                 goto done;
1576                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1577                         /* new right extent - oldext */
1578                         if ((error = xfs_bmbt_update(cur, r[1].br_startoff,
1579                                 r[1].br_startblock, r[1].br_blockcount,
1580                                 r[1].br_state)))
1581                                 goto done;
1582                         /* new left extent - oldext */
1583                         cur->bc_rec.b = PREV;
1584                         cur->bc_rec.b.br_blockcount =
1585                                 new->br_startoff - PREV.br_startoff;
1586                         if ((error = xfs_btree_insert(cur, &i)))
1587                                 goto done;
1588                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1589                         /*
1590                          * Reset the cursor to the position of the new extent
1591                          * we are about to insert as we can't trust it after
1592                          * the previous insert.
1593                          */
1594                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1595                                         new->br_startblock, new->br_blockcount,
1596                                         &i)))
1597                                 goto done;
1598                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1599                         /* new middle extent - newext */
1600                         cur->bc_rec.b.br_state = new->br_state;
1601                         if ((error = xfs_btree_insert(cur, &i)))
1602                                 goto done;
1603                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1604                 }
1605                 break;
1606
1607         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1608         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1609         case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
1610         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1611         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1612         case BMAP_LEFT_CONTIG:
1613         case BMAP_RIGHT_CONTIG:
1614                 /*
1615                  * These cases are all impossible.
1616                  */
1617                 ASSERT(0);
1618         }
1619         *curp = cur;
1620 done:
1621         *logflagsp = rval;
1622         return error;
1623 #undef  LEFT
1624 #undef  RIGHT
1625 #undef  PREV
1626 }
1627
1628 /*
1629  * Called by xfs_bmap_add_extent to handle cases converting a hole
1630  * to a delayed allocation.
1631  */
1632 /*ARGSUSED*/
1633 STATIC int                              /* error */
1634 xfs_bmap_add_extent_hole_delay(
1635         xfs_inode_t             *ip,    /* incore inode pointer */
1636         xfs_extnum_t            idx,    /* extent number to update/insert */
1637         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
1638         int                     *logflagsp, /* inode logging flags */
1639         int                     rsvd)           /* OK to allocate reserved blocks */
1640 {
1641         xfs_bmbt_rec_host_t     *ep;    /* extent record for idx */
1642         xfs_ifork_t             *ifp;   /* inode fork pointer */
1643         xfs_bmbt_irec_t         left;   /* left neighbor extent entry */
1644         xfs_filblks_t           newlen=0;       /* new indirect size */
1645         xfs_filblks_t           oldlen=0;       /* old indirect size */
1646         xfs_bmbt_irec_t         right;  /* right neighbor extent entry */
1647         int                     state;  /* state bits, accessed thru macros */
1648         xfs_filblks_t           temp=0; /* temp for indirect calculations */
1649
1650         ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
1651         ep = xfs_iext_get_ext(ifp, idx);
1652         state = 0;
1653         ASSERT(isnullstartblock(new->br_startblock));
1654
1655         /*
1656          * Check and set flags if this segment has a left neighbor
1657          */
1658         if (idx > 0) {
1659                 state |= BMAP_LEFT_VALID;
1660                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, idx - 1), &left);
1661
1662                 if (isnullstartblock(left.br_startblock))
1663                         state |= BMAP_LEFT_DELAY;
1664         }
1665
1666         /*
1667          * Check and set flags if the current (right) segment exists.
1668          * If it doesn't exist, we're converting the hole at end-of-file.
1669          */
1670         if (idx < ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t)) {
1671                 state |= BMAP_RIGHT_VALID;
1672                 xfs_bmbt_get_all(ep, &right);
1673
1674                 if (isnullstartblock(right.br_startblock))
1675                         state |= BMAP_RIGHT_DELAY;
1676         }
1677
1678         /*
1679          * Set contiguity flags on the left and right neighbors.
1680          * Don't let extents get too large, even if the pieces are contiguous.
1681          */
1682         if ((state & BMAP_LEFT_VALID) && (state & BMAP_LEFT_DELAY) &&
1683             left.br_startoff + left.br_blockcount == new->br_startoff &&
1684             left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
1685                 state |= BMAP_LEFT_CONTIG;
1686
1687         if ((state & BMAP_RIGHT_VALID) && (state & BMAP_RIGHT_DELAY) &&
1688             new->br_startoff + new->br_blockcount == right.br_startoff &&
1689             new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
1690             (!(state & BMAP_LEFT_CONTIG) ||
1691              (left.br_blockcount + new->br_blockcount +
1692               right.br_blockcount <= MAXEXTLEN)))
1693                 state |= BMAP_RIGHT_CONTIG;
1694
1695         /*
1696          * Switch out based on the contiguity flags.
1697          */
1698         switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
1699         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1700                 /*
1701                  * New allocation is contiguous with delayed allocations
1702                  * on the left and on the right.
1703                  * Merge all three into a single extent record.
1704                  */
1705                 temp = left.br_blockcount + new->br_blockcount +
1706                         right.br_blockcount;
1707
1708                 trace_xfs_bmap_pre_update(ip, idx - 1, state, _THIS_IP_);
1709                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1), temp);
1710                 oldlen = startblockval(left.br_startblock) +
1711                         startblockval(new->br_startblock) +
1712                         startblockval(right.br_startblock);
1713                 newlen = xfs_bmap_worst_indlen(ip, temp);
1714                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, idx - 1),
1715                         nullstartblock((int)newlen));
1716                 trace_xfs_bmap_post_update(ip, idx - 1, state, _THIS_IP_);
1717
1718                 xfs_iext_remove(ip, idx, 1, state);
1719                 ip->i_df.if_lastex = idx - 1;
1720                 break;
1721
1722         case BMAP_LEFT_CONTIG:
1723                 /*
1724                  * New allocation is contiguous with a delayed allocation
1725                  * on the left.
1726                  * Merge the new allocation with the left neighbor.
1727                  */
1728                 temp = left.br_blockcount + new->br_blockcount;
1729                 trace_xfs_bmap_pre_update(ip, idx - 1, state, _THIS_IP_);
1730                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1), temp);
1731                 oldlen = startblockval(left.br_startblock) +
1732                         startblockval(new->br_startblock);
1733                 newlen = xfs_bmap_worst_indlen(ip, temp);
1734                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, idx - 1),
1735                         nullstartblock((int)newlen));
1736                 trace_xfs_bmap_post_update(ip, idx - 1, state, _THIS_IP_);
1737
1738                 ip->i_df.if_lastex = idx - 1;
1739                 break;
1740
1741         case BMAP_RIGHT_CONTIG:
1742                 /*
1743                  * New allocation is contiguous with a delayed allocation
1744                  * on the right.
1745                  * Merge the new allocation with the right neighbor.
1746                  */
1747                 trace_xfs_bmap_pre_update(ip, idx, state, _THIS_IP_);
1748                 temp = new->br_blockcount + right.br_blockcount;
1749                 oldlen = startblockval(new->br_startblock) +
1750                         startblockval(right.br_startblock);
1751                 newlen = xfs_bmap_worst_indlen(ip, temp);
1752                 xfs_bmbt_set_allf(ep, new->br_startoff,
1753                         nullstartblock((int)newlen), temp, right.br_state);
1754                 trace_xfs_bmap_post_update(ip, idx, state, _THIS_IP_);
1755
1756                 ip->i_df.if_lastex = idx;
1757                 break;
1758
1759         case 0:
1760                 /*
1761                  * New allocation is not contiguous with another
1762                  * delayed allocation.
1763                  * Insert a new entry.
1764                  */
1765                 oldlen = newlen = 0;
1766                 xfs_iext_insert(ip, idx, 1, new, state);
1767                 ip->i_df.if_lastex = idx;
1768                 break;
1769         }
1770         if (oldlen != newlen) {
1771                 ASSERT(oldlen > newlen);
1772                 xfs_icsb_modify_counters(ip->i_mount, XFS_SBS_FDBLOCKS,
1773                         (int64_t)(oldlen - newlen), rsvd);
1774                 /*
1775                  * Nothing to do for disk quota accounting here.
1776                  */
1777         }
1778         *logflagsp = 0;
1779         return 0;
1780 }
1781
1782 /*
1783  * Called by xfs_bmap_add_extent to handle cases converting a hole
1784  * to a real allocation.
1785  */
1786 STATIC int                              /* error */
1787 xfs_bmap_add_extent_hole_real(
1788         xfs_inode_t             *ip,    /* incore inode pointer */
1789         xfs_extnum_t            idx,    /* extent number to update/insert */
1790         xfs_btree_cur_t         *cur,   /* if null, not a btree */
1791         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
1792         int                     *logflagsp, /* inode logging flags */
1793         int                     whichfork) /* data or attr fork */
1794 {
1795         xfs_bmbt_rec_host_t     *ep;    /* pointer to extent entry ins. point */
1796         int                     error;  /* error return value */
1797         int                     i;      /* temp state */
1798         xfs_ifork_t             *ifp;   /* inode fork pointer */
1799         xfs_bmbt_irec_t         left;   /* left neighbor extent entry */
1800         xfs_bmbt_irec_t         right;  /* right neighbor extent entry */
1801         int                     rval=0; /* return value (logging flags) */
1802         int                     state;  /* state bits, accessed thru macros */
1803
1804         ifp = XFS_IFORK_PTR(ip, whichfork);
1805         ASSERT(idx <= ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t));
1806         ep = xfs_iext_get_ext(ifp, idx);
1807         state = 0;
1808
1809         if (whichfork == XFS_ATTR_FORK)
1810                 state |= BMAP_ATTRFORK;
1811
1812         /*
1813          * Check and set flags if this segment has a left neighbor.
1814          */
1815         if (idx > 0) {
1816                 state |= BMAP_LEFT_VALID;
1817                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, idx - 1), &left);
1818                 if (isnullstartblock(left.br_startblock))
1819                         state |= BMAP_LEFT_DELAY;
1820         }
1821
1822         /*
1823          * Check and set flags if this segment has a current value.
1824          * Not true if we're inserting into the "hole" at eof.
1825          */
1826         if (idx < ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)) {
1827                 state |= BMAP_RIGHT_VALID;
1828                 xfs_bmbt_get_all(ep, &right);
1829                 if (isnullstartblock(right.br_startblock))
1830                         state |= BMAP_RIGHT_DELAY;
1831         }
1832
1833         /*
1834          * We're inserting a real allocation between "left" and "right".
1835          * Set the contiguity flags.  Don't let extents get too large.
1836          */
1837         if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
1838             left.br_startoff + left.br_blockcount == new->br_startoff &&
1839             left.br_startblock + left.br_blockcount == new->br_startblock &&
1840             left.br_state == new->br_state &&
1841             left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
1842                 state |= BMAP_LEFT_CONTIG;
1843
1844         if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
1845             new->br_startoff + new->br_blockcount == right.br_startoff &&
1846             new->br_startblock + new->br_blockcount == right.br_startblock &&
1847             new->br_state == right.br_state &&
1848             new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
1849             (!(state & BMAP_LEFT_CONTIG) ||
1850              left.br_blockcount + new->br_blockcount +
1851              right.br_blockcount <= MAXEXTLEN))
1852                 state |= BMAP_RIGHT_CONTIG;
1853
1854         error = 0;
1855         /*
1856          * Select which case we're in here, and implement it.
1857          */
1858         switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
1859         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1860                 /*
1861                  * New allocation is contiguous with real allocations on the
1862                  * left and on the right.
1863                  * Merge all three into a single extent record.
1864                  */
1865                 trace_xfs_bmap_pre_update(ip, idx - 1, state, _THIS_IP_);
1866                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
1867                         left.br_blockcount + new->br_blockcount +
1868                         right.br_blockcount);
1869                 trace_xfs_bmap_post_update(ip, idx - 1, state, _THIS_IP_);
1870
1871                 xfs_iext_remove(ip, idx, 1, state);
1872                 ifp->if_lastex = idx - 1;
1873                 XFS_IFORK_NEXT_SET(ip, whichfork,
1874                         XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
1875                 if (cur == NULL) {
1876                         rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
1877                 } else {
1878                         rval = XFS_ILOG_CORE;
1879                         if ((error = xfs_bmbt_lookup_eq(cur,
1880                                         right.br_startoff,
1881                                         right.br_startblock,
1882                                         right.br_blockcount, &i)))
1883                                 goto done;
1884                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1885                         if ((error = xfs_btree_delete(cur, &i)))
1886                                 goto done;
1887                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1888                         if ((error = xfs_btree_decrement(cur, 0, &i)))
1889                                 goto done;
1890                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1891                         if ((error = xfs_bmbt_update(cur, left.br_startoff,
1892                                         left.br_startblock,
1893                                         left.br_blockcount +
1894                                                 new->br_blockcount +
1895                                                 right.br_blockcount,
1896                                         left.br_state)))
1897                                 goto done;
1898                 }
1899                 break;
1900
1901         case BMAP_LEFT_CONTIG:
1902                 /*
1903                  * New allocation is contiguous with a real allocation
1904                  * on the left.
1905                  * Merge the new allocation with the left neighbor.
1906                  */
1907                 trace_xfs_bmap_pre_update(ip, idx - 1, state, _THIS_IP_);
1908                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, idx - 1),
1909                         left.br_blockcount + new->br_blockcount);
1910                 trace_xfs_bmap_post_update(ip, idx - 1, state, _THIS_IP_);
1911
1912                 ifp->if_lastex = idx - 1;
1913                 if (cur == NULL) {
1914                         rval = xfs_ilog_fext(whichfork);
1915                 } else {
1916                         rval = 0;
1917                         if ((error = xfs_bmbt_lookup_eq(cur,
1918                                         left.br_startoff,
1919                                         left.br_startblock,
1920                                         left.br_blockcount, &i)))
1921                                 goto done;
1922                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1923                         if ((error = xfs_bmbt_update(cur, left.br_startoff,
1924                                         left.br_startblock,
1925                                         left.br_blockcount +
1926                                                 new->br_blockcount,
1927                                         left.br_state)))
1928                                 goto done;
1929                 }
1930                 break;
1931
1932         case BMAP_RIGHT_CONTIG:
1933                 /*
1934                  * New allocation is contiguous with a real allocation
1935                  * on the right.
1936                  * Merge the new allocation with the right neighbor.
1937                  */
1938                 trace_xfs_bmap_pre_update(ip, idx, state, _THIS_IP_);
1939                 xfs_bmbt_set_allf(ep, new->br_startoff, new->br_startblock,
1940                         new->br_blockcount + right.br_blockcount,
1941                         right.br_state);
1942                 trace_xfs_bmap_post_update(ip, idx, state, _THIS_IP_);
1943
1944                 ifp->if_lastex = idx;
1945                 if (cur == NULL) {
1946                         rval = xfs_ilog_fext(whichfork);
1947                 } else {
1948                         rval = 0;
1949                         if ((error = xfs_bmbt_lookup_eq(cur,
1950                                         right.br_startoff,
1951                                         right.br_startblock,
1952                                         right.br_blockcount, &i)))
1953                                 goto done;
1954                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1955                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
1956                                         new->br_startblock,
1957                                         new->br_blockcount +
1958                                                 right.br_blockcount,
1959                                         right.br_state)))
1960                                 goto done;
1961                 }
1962                 break;
1963
1964         case 0:
1965                 /*
1966                  * New allocation is not contiguous with another
1967                  * real allocation.
1968                  * Insert a new entry.
1969                  */
1970                 xfs_iext_insert(ip, idx, 1, new, state);
1971                 ifp->if_lastex = idx;
1972                 XFS_IFORK_NEXT_SET(ip, whichfork,
1973                         XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
1974                 if (cur == NULL) {
1975                         rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
1976                 } else {
1977                         rval = XFS_ILOG_CORE;
1978                         if ((error = xfs_bmbt_lookup_eq(cur,
1979                                         new->br_startoff,
1980                                         new->br_startblock,
1981                                         new->br_blockcount, &i)))
1982                                 goto done;
1983                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1984                         cur->bc_rec.b.br_state = new->br_state;
1985                         if ((error = xfs_btree_insert(cur, &i)))
1986                                 goto done;
1987                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1988                 }
1989                 break;
1990         }
1991 done:
1992         *logflagsp = rval;
1993         return error;
1994 }
1995
1996 /*
1997  * Adjust the size of the new extent based on di_extsize and rt extsize.
1998  */
1999 STATIC int
2000 xfs_bmap_extsize_align(
2001         xfs_mount_t     *mp,
2002         xfs_bmbt_irec_t *gotp,          /* next extent pointer */
2003         xfs_bmbt_irec_t *prevp,         /* previous extent pointer */
2004         xfs_extlen_t    extsz,          /* align to this extent size */
2005         int             rt,             /* is this a realtime inode? */
2006         int             eof,            /* is extent at end-of-file? */
2007         int             delay,          /* creating delalloc extent? */
2008         int             convert,        /* overwriting unwritten extent? */
2009         xfs_fileoff_t   *offp,          /* in/out: aligned offset */
2010         xfs_extlen_t    *lenp)          /* in/out: aligned length */
2011 {
2012         xfs_fileoff_t   orig_off;       /* original offset */
2013         xfs_extlen_t    orig_alen;      /* original length */
2014         xfs_fileoff_t   orig_end;       /* original off+len */
2015         xfs_fileoff_t   nexto;          /* next file offset */
2016         xfs_fileoff_t   prevo;          /* previous file offset */
2017         xfs_fileoff_t   align_off;      /* temp for offset */
2018         xfs_extlen_t    align_alen;     /* temp for length */
2019         xfs_extlen_t    temp;           /* temp for calculations */
2020
2021         if (convert)
2022                 return 0;
2023
2024         orig_off = align_off = *offp;
2025         orig_alen = align_alen = *lenp;
2026         orig_end = orig_off + orig_alen;
2027
2028         /*
2029          * If this request overlaps an existing extent, then don't
2030          * attempt to perform any additional alignment.
2031          */
2032         if (!delay && !eof &&
2033             (orig_off >= gotp->br_startoff) &&
2034             (orig_end <= gotp->br_startoff + gotp->br_blockcount)) {
2035                 return 0;
2036         }
2037
2038         /*
2039          * If the file offset is unaligned vs. the extent size
2040          * we need to align it.  This will be possible unless
2041          * the file was previously written with a kernel that didn't
2042          * perform this alignment, or if a truncate shot us in the
2043          * foot.
2044          */
2045         temp = do_mod(orig_off, extsz);
2046         if (temp) {
2047                 align_alen += temp;
2048                 align_off -= temp;
2049         }
2050         /*
2051          * Same adjustment for the end of the requested area.
2052          */
2053         if ((temp = (align_alen % extsz))) {
2054                 align_alen += extsz - temp;
2055         }
2056         /*
2057          * If the previous block overlaps with this proposed allocation
2058          * then move the start forward without adjusting the length.
2059          */
2060         if (prevp->br_startoff != NULLFILEOFF) {
2061                 if (prevp->br_startblock == HOLESTARTBLOCK)
2062                         prevo = prevp->br_startoff;
2063                 else
2064                         prevo = prevp->br_startoff + prevp->br_blockcount;
2065         } else
2066                 prevo = 0;
2067         if (align_off != orig_off && align_off < prevo)
2068                 align_off = prevo;
2069         /*
2070          * If the next block overlaps with this proposed allocation
2071          * then move the start back without adjusting the length,
2072          * but not before offset 0.
2073          * This may of course make the start overlap previous block,
2074          * and if we hit the offset 0 limit then the next block
2075          * can still overlap too.
2076          */
2077         if (!eof && gotp->br_startoff != NULLFILEOFF) {
2078                 if ((delay && gotp->br_startblock == HOLESTARTBLOCK) ||
2079                     (!delay && gotp->br_startblock == DELAYSTARTBLOCK))
2080                         nexto = gotp->br_startoff + gotp->br_blockcount;
2081                 else
2082                         nexto = gotp->br_startoff;
2083         } else
2084                 nexto = NULLFILEOFF;
2085         if (!eof &&
2086             align_off + align_alen != orig_end &&
2087             align_off + align_alen > nexto)
2088                 align_off = nexto > align_alen ? nexto - align_alen : 0;
2089         /*
2090          * If we're now overlapping the next or previous extent that
2091          * means we can't fit an extsz piece in this hole.  Just move
2092          * the start forward to the first valid spot and set
2093          * the length so we hit the end.
2094          */
2095         if (align_off != orig_off && align_off < prevo)
2096                 align_off = prevo;
2097         if (align_off + align_alen != orig_end &&
2098             align_off + align_alen > nexto &&
2099             nexto != NULLFILEOFF) {
2100                 ASSERT(nexto > prevo);
2101                 align_alen = nexto - align_off;
2102         }
2103
2104         /*
2105          * If realtime, and the result isn't a multiple of the realtime
2106          * extent size we need to remove blocks until it is.
2107          */
2108         if (rt && (temp = (align_alen % mp->m_sb.sb_rextsize))) {
2109                 /*
2110                  * We're not covering the original request, or
2111                  * we won't be able to once we fix the length.
2112                  */
2113                 if (orig_off < align_off ||
2114                     orig_end > align_off + align_alen ||
2115                     align_alen - temp < orig_alen)
2116                         return XFS_ERROR(EINVAL);
2117                 /*
2118                  * Try to fix it by moving the start up.
2119                  */
2120                 if (align_off + temp <= orig_off) {
2121                         align_alen -= temp;
2122                         align_off += temp;
2123                 }
2124                 /*
2125                  * Try to fix it by moving the end in.
2126                  */
2127                 else if (align_off + align_alen - temp >= orig_end)
2128                         align_alen -= temp;
2129                 /*
2130                  * Set the start to the minimum then trim the length.
2131                  */
2132                 else {
2133                         align_alen -= orig_off - align_off;
2134                         align_off = orig_off;
2135                         align_alen -= align_alen % mp->m_sb.sb_rextsize;
2136                 }
2137                 /*
2138                  * Result doesn't cover the request, fail it.
2139                  */
2140                 if (orig_off < align_off || orig_end > align_off + align_alen)
2141                         return XFS_ERROR(EINVAL);
2142         } else {
2143                 ASSERT(orig_off >= align_off);
2144                 ASSERT(orig_end <= align_off + align_alen);
2145         }
2146
2147 #ifdef DEBUG
2148         if (!eof && gotp->br_startoff != NULLFILEOFF)
2149                 ASSERT(align_off + align_alen <= gotp->br_startoff);
2150         if (prevp->br_startoff != NULLFILEOFF)
2151                 ASSERT(align_off >= prevp->br_startoff + prevp->br_blockcount);
2152 #endif
2153
2154         *lenp = align_alen;
2155         *offp = align_off;
2156         return 0;
2157 }
2158
2159 #define XFS_ALLOC_GAP_UNITS     4
2160
2161 STATIC void
2162 xfs_bmap_adjacent(
2163         xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
2164 {
2165         xfs_fsblock_t   adjust;         /* adjustment to block numbers */
2166         xfs_agnumber_t  fb_agno;        /* ag number of ap->firstblock */
2167         xfs_mount_t     *mp;            /* mount point structure */
2168         int             nullfb;         /* true if ap->firstblock isn't set */
2169         int             rt;             /* true if inode is realtime */
2170
2171 #define ISVALID(x,y)    \
2172         (rt ? \
2173                 (x) < mp->m_sb.sb_rblocks : \
2174                 XFS_FSB_TO_AGNO(mp, x) == XFS_FSB_TO_AGNO(mp, y) && \
2175                 XFS_FSB_TO_AGNO(mp, x) < mp->m_sb.sb_agcount && \
2176                 XFS_FSB_TO_AGBNO(mp, x) < mp->m_sb.sb_agblocks)
2177
2178         mp = ap->ip->i_mount;
2179         nullfb = ap->firstblock == NULLFSBLOCK;
2180         rt = XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata;
2181         fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, ap->firstblock);
2182         /*
2183          * If allocating at eof, and there's a previous real block,
2184          * try to use its last block as our starting point.
2185          */
2186         if (ap->eof && ap->prevp->br_startoff != NULLFILEOFF &&
2187             !isnullstartblock(ap->prevp->br_startblock) &&
2188             ISVALID(ap->prevp->br_startblock + ap->prevp->br_blockcount,
2189                     ap->prevp->br_startblock)) {
2190                 ap->rval = ap->prevp->br_startblock + ap->prevp->br_blockcount;
2191                 /*
2192                  * Adjust for the gap between prevp and us.
2193                  */
2194                 adjust = ap->off -
2195                         (ap->prevp->br_startoff + ap->prevp->br_blockcount);
2196                 if (adjust &&
2197                     ISVALID(ap->rval + adjust, ap->prevp->br_startblock))
2198                         ap->rval += adjust;
2199         }
2200         /*
2201          * If not at eof, then compare the two neighbor blocks.
2202          * Figure out whether either one gives us a good starting point,
2203          * and pick the better one.
2204          */
2205         else if (!ap->eof) {
2206                 xfs_fsblock_t   gotbno;         /* right side block number */
2207                 xfs_fsblock_t   gotdiff=0;      /* right side difference */
2208                 xfs_fsblock_t   prevbno;        /* left side block number */
2209                 xfs_fsblock_t   prevdiff=0;     /* left side difference */
2210
2211                 /*
2212                  * If there's a previous (left) block, select a requested
2213                  * start block based on it.
2214                  */
2215                 if (ap->prevp->br_startoff != NULLFILEOFF &&
2216                     !isnullstartblock(ap->prevp->br_startblock) &&
2217                     (prevbno = ap->prevp->br_startblock +
2218                                ap->prevp->br_blockcount) &&
2219                     ISVALID(prevbno, ap->prevp->br_startblock)) {
2220                         /*
2221                          * Calculate gap to end of previous block.
2222                          */
2223                         adjust = prevdiff = ap->off -
2224                                 (ap->prevp->br_startoff +
2225                                  ap->prevp->br_blockcount);
2226                         /*
2227                          * Figure the startblock based on the previous block's
2228                          * end and the gap size.
2229                          * Heuristic!
2230                          * If the gap is large relative to the piece we're
2231                          * allocating, or using it gives us an invalid block
2232                          * number, then just use the end of the previous block.
2233                          */
2234                         if (prevdiff <= XFS_ALLOC_GAP_UNITS * ap->alen &&
2235                             ISVALID(prevbno + prevdiff,
2236                                     ap->prevp->br_startblock))
2237                                 prevbno += adjust;
2238                         else
2239                                 prevdiff += adjust;
2240                         /*
2241                          * If the firstblock forbids it, can't use it,
2242                          * must use default.
2243                          */
2244                         if (!rt && !nullfb &&
2245                             XFS_FSB_TO_AGNO(mp, prevbno) != fb_agno)
2246                                 prevbno = NULLFSBLOCK;
2247                 }
2248                 /*
2249                  * No previous block or can't follow it, just default.
2250                  */
2251                 else
2252                         prevbno = NULLFSBLOCK;
2253                 /*
2254                  * If there's a following (right) block, select a requested
2255                  * start block based on it.
2256                  */
2257                 if (!isnullstartblock(ap->gotp->br_startblock)) {
2258                         /*
2259                          * Calculate gap to start of next block.
2260                          */
2261                         adjust = gotdiff = ap->gotp->br_startoff - ap->off;
2262                         /*
2263                          * Figure the startblock based on the next block's
2264                          * start and the gap size.
2265                          */
2266                         gotbno = ap->gotp->br_startblock;
2267                         /*
2268                          * Heuristic!
2269                          * If the gap is large relative to the piece we're
2270                          * allocating, or using it gives us an invalid block
2271                          * number, then just use the start of the next block
2272                          * offset by our length.
2273                          */
2274                         if (gotdiff <= XFS_ALLOC_GAP_UNITS * ap->alen &&
2275                             ISVALID(gotbno - gotdiff, gotbno))
2276                                 gotbno -= adjust;
2277                         else if (ISVALID(gotbno - ap->alen, gotbno)) {
2278                                 gotbno -= ap->alen;
2279                                 gotdiff += adjust - ap->alen;
2280                         } else
2281                                 gotdiff += adjust;
2282                         /*
2283                          * If the firstblock forbids it, can't use it,
2284                          * must use default.
2285                          */
2286                         if (!rt && !nullfb &&
2287                             XFS_FSB_TO_AGNO(mp, gotbno) != fb_agno)
2288                                 gotbno = NULLFSBLOCK;
2289                 }
2290                 /*
2291                  * No next block, just default.
2292                  */
2293                 else
2294                         gotbno = NULLFSBLOCK;
2295                 /*
2296                  * If both valid, pick the better one, else the only good
2297                  * one, else ap->rval is already set (to 0 or the inode block).
2298                  */
2299                 if (prevbno != NULLFSBLOCK && gotbno != NULLFSBLOCK)
2300                         ap->rval = prevdiff <= gotdiff ? prevbno : gotbno;
2301                 else if (prevbno != NULLFSBLOCK)
2302                         ap->rval = prevbno;
2303                 else if (gotbno != NULLFSBLOCK)
2304                         ap->rval = gotbno;
2305         }
2306 #undef ISVALID
2307 }
2308
2309 STATIC int
2310 xfs_bmap_rtalloc(
2311         xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
2312 {
2313         xfs_alloctype_t atype = 0;      /* type for allocation routines */
2314         int             error;          /* error return value */
2315         xfs_mount_t     *mp;            /* mount point structure */
2316         xfs_extlen_t    prod = 0;       /* product factor for allocators */
2317         xfs_extlen_t    ralen = 0;      /* realtime allocation length */
2318         xfs_extlen_t    align;          /* minimum allocation alignment */
2319         xfs_rtblock_t   rtb;
2320
2321         mp = ap->ip->i_mount;
2322         align = xfs_get_extsz_hint(ap->ip);
2323         prod = align / mp->m_sb.sb_rextsize;
2324         error = xfs_bmap_extsize_align(mp, ap->gotp, ap->prevp,
2325                                         align, 1, ap->eof, 0,
2326                                         ap->conv, &ap->off, &ap->alen);
2327         if (error)
2328                 return error;
2329         ASSERT(ap->alen);
2330         ASSERT(ap->alen % mp->m_sb.sb_rextsize == 0);
2331
2332         /*
2333          * If the offset & length are not perfectly aligned
2334          * then kill prod, it will just get us in trouble.
2335          */
2336         if (do_mod(ap->off, align) || ap->alen % align)
2337                 prod = 1;
2338         /*
2339          * Set ralen to be the actual requested length in rtextents.
2340          */
2341         ralen = ap->alen / mp->m_sb.sb_rextsize;
2342         /*
2343          * If the old value was close enough to MAXEXTLEN that
2344          * we rounded up to it, cut it back so it's valid again.
2345          * Note that if it's a really large request (bigger than
2346          * MAXEXTLEN), we don't hear about that number, and can't
2347          * adjust the starting point to match it.
2348          */
2349         if (ralen * mp->m_sb.sb_rextsize >= MAXEXTLEN)
2350                 ralen = MAXEXTLEN / mp->m_sb.sb_rextsize;
2351         /*
2352          * If it's an allocation to an empty file at offset 0,
2353          * pick an extent that will space things out in the rt area.
2354          */
2355         if (ap->eof && ap->off == 0) {
2356                 xfs_rtblock_t uninitialized_var(rtx); /* realtime extent no */
2357
2358                 error = xfs_rtpick_extent(mp, ap->tp, ralen, &rtx);
2359                 if (error)
2360                         return error;
2361                 ap->rval = rtx * mp->m_sb.sb_rextsize;
2362         } else {
2363                 ap->rval = 0;
2364         }
2365
2366         xfs_bmap_adjacent(ap);
2367
2368         /*
2369          * Realtime allocation, done through xfs_rtallocate_extent.
2370          */
2371         atype = ap->rval == 0 ?  XFS_ALLOCTYPE_ANY_AG : XFS_ALLOCTYPE_NEAR_BNO;
2372         do_div(ap->rval, mp->m_sb.sb_rextsize);
2373         rtb = ap->rval;
2374         ap->alen = ralen;
2375         if ((error = xfs_rtallocate_extent(ap->tp, ap->rval, 1, ap->alen,
2376                                 &ralen, atype, ap->wasdel, prod, &rtb)))
2377                 return error;
2378         if (rtb == NULLFSBLOCK && prod > 1 &&
2379             (error = xfs_rtallocate_extent(ap->tp, ap->rval, 1,
2380                                            ap->alen, &ralen, atype,
2381                                            ap->wasdel, 1, &rtb)))
2382                 return error;
2383         ap->rval = rtb;
2384         if (ap->rval != NULLFSBLOCK) {
2385                 ap->rval *= mp->m_sb.sb_rextsize;
2386                 ralen *= mp->m_sb.sb_rextsize;
2387                 ap->alen = ralen;
2388                 ap->ip->i_d.di_nblocks += ralen;
2389                 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
2390                 if (ap->wasdel)
2391                         ap->ip->i_delayed_blks -= ralen;
2392                 /*
2393                  * Adjust the disk quota also. This was reserved
2394                  * earlier.
2395                  */
2396                 xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
2397                         ap->wasdel ? XFS_TRANS_DQ_DELRTBCOUNT :
2398                                         XFS_TRANS_DQ_RTBCOUNT, (long) ralen);
2399         } else {
2400                 ap->alen = 0;
2401         }
2402         return 0;
2403 }
2404
2405 STATIC int
2406 xfs_bmap_btalloc_nullfb(
2407         struct xfs_bmalloca     *ap,
2408         struct xfs_alloc_arg    *args,
2409         xfs_extlen_t            *blen)
2410 {
2411         struct xfs_mount        *mp = ap->ip->i_mount;
2412         struct xfs_perag        *pag;
2413         xfs_agnumber_t          ag, startag;
2414         int                     notinit = 0;
2415         int                     error;
2416
2417         if (ap->userdata && xfs_inode_is_filestream(ap->ip))
2418                 args->type = XFS_ALLOCTYPE_NEAR_BNO;
2419         else
2420                 args->type = XFS_ALLOCTYPE_START_BNO;
2421         args->total = ap->total;
2422
2423         /*
2424          * Search for an allocation group with a single extent large enough
2425          * for the request.  If one isn't found, then adjust the minimum
2426          * allocation size to the largest space found.
2427          */
2428         startag = ag = XFS_FSB_TO_AGNO(mp, args->fsbno);
2429         if (startag == NULLAGNUMBER)
2430                 startag = ag = 0;
2431
2432         pag = xfs_perag_get(mp, ag);
2433         while (*blen < ap->alen) {
2434                 if (!pag->pagf_init) {
2435                         error = xfs_alloc_pagf_init(mp, args->tp, ag,
2436                                                     XFS_ALLOC_FLAG_TRYLOCK);
2437                         if (error) {
2438                                 xfs_perag_put(pag);
2439                                 return error;
2440                         }
2441                 }
2442
2443                 /*
2444                  * See xfs_alloc_fix_freelist...
2445                  */
2446                 if (pag->pagf_init) {
2447                         xfs_extlen_t    longest;
2448                         longest = xfs_alloc_longest_free_extent(mp, pag);
2449                         if (*blen < longest)
2450                                 *blen = longest;
2451                 } else
2452                         notinit = 1;
2453
2454                 if (xfs_inode_is_filestream(ap->ip)) {
2455                         if (*blen >= ap->alen)
2456                                 break;
2457
2458                         if (ap->userdata) {
2459                                 /*
2460                                  * If startag is an invalid AG, we've
2461                                  * come here once before and
2462                                  * xfs_filestream_new_ag picked the
2463                                  * best currently available.
2464                                  *
2465                                  * Don't continue looping, since we
2466                                  * could loop forever.
2467                                  */
2468                                 if (startag == NULLAGNUMBER)
2469                                         break;
2470
2471                                 error = xfs_filestream_new_ag(ap, &ag);
2472                                 xfs_perag_put(pag);
2473                                 if (error)
2474                                         return error;
2475
2476                                 /* loop again to set 'blen'*/
2477                                 startag = NULLAGNUMBER;
2478                                 pag = xfs_perag_get(mp, ag);
2479                                 continue;
2480                         }
2481                 }
2482                 if (++ag == mp->m_sb.sb_agcount)
2483                         ag = 0;
2484                 if (ag == startag)
2485                         break;
2486                 xfs_perag_put(pag);
2487                 pag = xfs_perag_get(mp, ag);
2488         }
2489         xfs_perag_put(pag);
2490
2491         /*
2492          * Since the above loop did a BUF_TRYLOCK, it is
2493          * possible that there is space for this request.
2494          */
2495         if (notinit || *blen < ap->minlen)
2496                 args->minlen = ap->minlen;
2497         /*
2498          * If the best seen length is less than the request
2499          * length, use the best as the minimum.
2500          */
2501         else if (*blen < ap->alen)
2502                 args->minlen = *blen;
2503         /*
2504          * Otherwise we've seen an extent as big as alen,
2505          * use that as the minimum.
2506          */
2507         else
2508                 args->minlen = ap->alen;
2509
2510         /*
2511          * set the failure fallback case to look in the selected
2512          * AG as the stream may have moved.
2513          */
2514         if (xfs_inode_is_filestream(ap->ip))
2515                 ap->rval = args->fsbno = XFS_AGB_TO_FSB(mp, ag, 0);
2516
2517         return 0;
2518 }
2519
2520 STATIC int
2521 xfs_bmap_btalloc(
2522         xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
2523 {
2524         xfs_mount_t     *mp;            /* mount point structure */
2525         xfs_alloctype_t atype = 0;      /* type for allocation routines */
2526         xfs_extlen_t    align;          /* minimum allocation alignment */
2527         xfs_agnumber_t  fb_agno;        /* ag number of ap->firstblock */
2528         xfs_agnumber_t  ag;
2529         xfs_alloc_arg_t args;
2530         xfs_extlen_t    blen;
2531         xfs_extlen_t    nextminlen = 0;
2532         int             nullfb;         /* true if ap->firstblock isn't set */
2533         int             isaligned;
2534         int             tryagain;
2535         int             error;
2536
2537         mp = ap->ip->i_mount;
2538         align = ap->userdata ? xfs_get_extsz_hint(ap->ip) : 0;
2539         if (unlikely(align)) {
2540                 error = xfs_bmap_extsize_align(mp, ap->gotp, ap->prevp,
2541                                                 align, 0, ap->eof, 0, ap->conv,
2542                                                 &ap->off, &ap->alen);
2543                 ASSERT(!error);
2544                 ASSERT(ap->alen);
2545         }
2546         nullfb = ap->firstblock == NULLFSBLOCK;
2547         fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, ap->firstblock);
2548         if (nullfb) {
2549                 if (ap->userdata && xfs_inode_is_filestream(ap->ip)) {
2550                         ag = xfs_filestream_lookup_ag(ap->ip);
2551                         ag = (ag != NULLAGNUMBER) ? ag : 0;
2552                         ap->rval = XFS_AGB_TO_FSB(mp, ag, 0);
2553                 } else {
2554                         ap->rval = XFS_INO_TO_FSB(mp, ap->ip->i_ino);
2555                 }
2556         } else
2557                 ap->rval = ap->firstblock;
2558
2559         xfs_bmap_adjacent(ap);
2560
2561         /*
2562          * If allowed, use ap->rval; otherwise must use firstblock since
2563          * it's in the right allocation group.
2564          */
2565         if (nullfb || XFS_FSB_TO_AGNO(mp, ap->rval) == fb_agno)
2566                 ;
2567         else
2568                 ap->rval = ap->firstblock;
2569         /*
2570          * Normal allocation, done through xfs_alloc_vextent.
2571          */
2572         tryagain = isaligned = 0;
2573         args.tp = ap->tp;
2574         args.mp = mp;
2575         args.fsbno = ap->rval;
2576         args.maxlen = MIN(ap->alen, mp->m_sb.sb_agblocks);
2577         args.firstblock = ap->firstblock;
2578         blen = 0;
2579         if (nullfb) {
2580                 error = xfs_bmap_btalloc_nullfb(ap, &args, &blen);
2581                 if (error)
2582                         return error;
2583         } else if (ap->low) {
2584                 if (xfs_inode_is_filestream(ap->ip))
2585                         args.type = XFS_ALLOCTYPE_FIRST_AG;
2586                 else
2587                         args.type = XFS_ALLOCTYPE_START_BNO;
2588                 args.total = args.minlen = ap->minlen;
2589         } else {
2590                 args.type = XFS_ALLOCTYPE_NEAR_BNO;
2591                 args.total = ap->total;
2592                 args.minlen = ap->minlen;
2593         }
2594         /* apply extent size hints if obtained earlier */
2595         if (unlikely(align)) {
2596                 args.prod = align;
2597                 if ((args.mod = (xfs_extlen_t)do_mod(ap->off, args.prod)))
2598                         args.mod = (xfs_extlen_t)(args.prod - args.mod);
2599         } else if (mp->m_sb.sb_blocksize >= PAGE_CACHE_SIZE) {
2600                 args.prod = 1;
2601                 args.mod = 0;
2602         } else {
2603                 args.prod = PAGE_CACHE_SIZE >> mp->m_sb.sb_blocklog;
2604                 if ((args.mod = (xfs_extlen_t)(do_mod(ap->off, args.prod))))
2605                         args.mod = (xfs_extlen_t)(args.prod - args.mod);
2606         }
2607         /*
2608          * If we are not low on available data blocks, and the
2609          * underlying logical volume manager is a stripe, and
2610          * the file offset is zero then try to allocate data
2611          * blocks on stripe unit boundary.
2612          * NOTE: ap->aeof is only set if the allocation length
2613          * is >= the stripe unit and the allocation offset is
2614          * at the end of file.
2615          */
2616         if (!ap->low && ap->aeof) {
2617                 if (!ap->off) {
2618                         args.alignment = mp->m_dalign;
2619                         atype = args.type;
2620                         isaligned = 1;
2621                         /*
2622                          * Adjust for alignment
2623                          */
2624                         if (blen > args.alignment && blen <= ap->alen)
2625                                 args.minlen = blen - args.alignment;
2626                         args.minalignslop = 0;
2627                 } else {
2628                         /*
2629                          * First try an exact bno allocation.
2630                          * If it fails then do a near or start bno
2631                          * allocation with alignment turned on.
2632                          */
2633                         atype = args.type;
2634                         tryagain = 1;
2635                         args.type = XFS_ALLOCTYPE_THIS_BNO;
2636                         args.alignment = 1;
2637                         /*
2638                          * Compute the minlen+alignment for the
2639                          * next case.  Set slop so that the value
2640                          * of minlen+alignment+slop doesn't go up
2641                          * between the calls.
2642                          */
2643                         if (blen > mp->m_dalign && blen <= ap->alen)
2644                                 nextminlen = blen - mp->m_dalign;
2645                         else
2646                                 nextminlen = args.minlen;
2647                         if (nextminlen + mp->m_dalign > args.minlen + 1)
2648                                 args.minalignslop =
2649                                         nextminlen + mp->m_dalign -
2650                                         args.minlen - 1;
2651                         else
2652                                 args.minalignslop = 0;
2653                 }
2654         } else {
2655                 args.alignment = 1;
2656                 args.minalignslop = 0;
2657         }
2658         args.minleft = ap->minleft;
2659         args.wasdel = ap->wasdel;
2660         args.isfl = 0;
2661         args.userdata = ap->userdata;
2662         if ((error = xfs_alloc_vextent(&args)))
2663                 return error;
2664         if (tryagain && args.fsbno == NULLFSBLOCK) {
2665                 /*
2666                  * Exact allocation failed. Now try with alignment
2667                  * turned on.
2668                  */
2669                 args.type = atype;
2670                 args.fsbno = ap->rval;
2671                 args.alignment = mp->m_dalign;
2672                 args.minlen = nextminlen;
2673                 args.minalignslop = 0;
2674                 isaligned = 1;
2675                 if ((error = xfs_alloc_vextent(&args)))
2676                         return error;
2677         }
2678         if (isaligned && args.fsbno == NULLFSBLOCK) {
2679                 /*
2680                  * allocation failed, so turn off alignment and
2681                  * try again.
2682                  */
2683                 args.type = atype;
2684                 args.fsbno = ap->rval;
2685                 args.alignment = 0;
2686                 if ((error = xfs_alloc_vextent(&args)))
2687                         return error;
2688         }
2689         if (args.fsbno == NULLFSBLOCK && nullfb &&
2690             args.minlen > ap->minlen) {
2691                 args.minlen = ap->minlen;
2692                 args.type = XFS_ALLOCTYPE_START_BNO;
2693                 args.fsbno = ap->rval;
2694                 if ((error = xfs_alloc_vextent(&args)))
2695                         return error;
2696         }
2697         if (args.fsbno == NULLFSBLOCK && nullfb) {
2698                 args.fsbno = 0;
2699                 args.type = XFS_ALLOCTYPE_FIRST_AG;
2700                 args.total = ap->minlen;
2701                 args.minleft = 0;
2702                 if ((error = xfs_alloc_vextent(&args)))
2703                         return error;
2704                 ap->low = 1;
2705         }
2706         if (args.fsbno != NULLFSBLOCK) {
2707                 ap->firstblock = ap->rval = args.fsbno;
2708                 ASSERT(nullfb || fb_agno == args.agno ||
2709                        (ap->low && fb_agno < args.agno));
2710                 ap->alen = args.len;
2711                 ap->ip->i_d.di_nblocks += args.len;
2712                 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
2713                 if (ap->wasdel)
2714                         ap->ip->i_delayed_blks -= args.len;
2715                 /*
2716                  * Adjust the disk quota also. This was reserved
2717                  * earlier.
2718                  */
2719                 xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
2720                         ap->wasdel ? XFS_TRANS_DQ_DELBCOUNT :
2721                                         XFS_TRANS_DQ_BCOUNT,
2722                         (long) args.len);
2723         } else {
2724                 ap->rval = NULLFSBLOCK;
2725                 ap->alen = 0;
2726         }
2727         return 0;
2728 }
2729
2730 /*
2731  * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
2732  * It figures out where to ask the underlying allocator to put the new extent.
2733  */
2734 STATIC int
2735 xfs_bmap_alloc(
2736         xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
2737 {
2738         if (XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata)
2739                 return xfs_bmap_rtalloc(ap);
2740         return xfs_bmap_btalloc(ap);
2741 }
2742
2743 /*
2744  * Transform a btree format file with only one leaf node, where the
2745  * extents list will fit in the inode, into an extents format file.
2746  * Since the file extents are already in-core, all we have to do is
2747  * give up the space for the btree root and pitch the leaf block.
2748  */
2749 STATIC int                              /* error */
2750 xfs_bmap_btree_to_extents(
2751         xfs_trans_t             *tp,    /* transaction pointer */
2752         xfs_inode_t             *ip,    /* incore inode pointer */
2753         xfs_btree_cur_t         *cur,   /* btree cursor */
2754         int                     *logflagsp, /* inode logging flags */
2755         int                     whichfork)  /* data or attr fork */
2756 {
2757         /* REFERENCED */
2758         struct xfs_btree_block  *cblock;/* child btree block */
2759         xfs_fsblock_t           cbno;   /* child block number */
2760         xfs_buf_t               *cbp;   /* child block's buffer */
2761         int                     error;  /* error return value */
2762         xfs_ifork_t             *ifp;   /* inode fork data */
2763         xfs_mount_t             *mp;    /* mount point structure */
2764         __be64                  *pp;    /* ptr to block address */
2765         struct xfs_btree_block  *rblock;/* root btree block */
2766
2767         mp = ip->i_mount;
2768         ifp = XFS_IFORK_PTR(ip, whichfork);
2769         ASSERT(ifp->if_flags & XFS_IFEXTENTS);
2770         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
2771         rblock = ifp->if_broot;
2772         ASSERT(be16_to_cpu(rblock->bb_level) == 1);
2773         ASSERT(be16_to_cpu(rblock->bb_numrecs) == 1);
2774         ASSERT(xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0) == 1);
2775         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, rblock, 1, ifp->if_broot_bytes);
2776         cbno = be64_to_cpu(*pp);
2777         *logflagsp = 0;
2778 #ifdef DEBUG
2779         if ((error = xfs_btree_check_lptr(cur, cbno, 1)))
2780                 return error;
2781 #endif
2782         if ((error = xfs_btree_read_bufl(mp, tp, cbno, 0, &cbp,
2783                         XFS_BMAP_BTREE_REF)))
2784                 return error;
2785         cblock = XFS_BUF_TO_BLOCK(cbp);
2786         if ((error = xfs_btree_check_block(cur, cblock, 0, cbp)))
2787                 return error;
2788         xfs_bmap_add_free(cbno, 1, cur->bc_private.b.flist, mp);
2789         ip->i_d.di_nblocks--;
2790         xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, -1L);
2791         xfs_trans_binval(tp, cbp);
2792         if (cur->bc_bufs[0] == cbp)
2793                 cur->bc_bufs[0] = NULL;
2794         xfs_iroot_realloc(ip, -1, whichfork);
2795         ASSERT(ifp->if_broot == NULL);
2796         ASSERT((ifp->if_flags & XFS_IFBROOT) == 0);
2797         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
2798         *logflagsp = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
2799         return 0;
2800 }
2801
2802 /*
2803  * Called by xfs_bmapi to update file extent records and the btree
2804  * after removing space (or undoing a delayed allocation).
2805  */
2806 STATIC int                              /* error */
2807 xfs_bmap_del_extent(
2808         xfs_inode_t             *ip,    /* incore inode pointer */
2809         xfs_trans_t             *tp,    /* current transaction pointer */
2810         xfs_extnum_t            idx,    /* extent number to update/delete */
2811         xfs_bmap_free_t         *flist, /* list of extents to be freed */
2812         xfs_btree_cur_t         *cur,   /* if null, not a btree */
2813         xfs_bmbt_irec_t         *del,   /* data to remove from extents */
2814         int                     *logflagsp, /* inode logging flags */
2815         int                     whichfork, /* data or attr fork */
2816         int                     rsvd)   /* OK to allocate reserved blocks */
2817 {
2818         xfs_filblks_t           da_new; /* new delay-alloc indirect blocks */
2819         xfs_filblks_t           da_old; /* old delay-alloc indirect blocks */
2820         xfs_fsblock_t           del_endblock=0; /* first block past del */
2821         xfs_fileoff_t           del_endoff;     /* first offset past del */
2822         int                     delay;  /* current block is delayed allocated */
2823         int                     do_fx;  /* free extent at end of routine */
2824         xfs_bmbt_rec_host_t     *ep;    /* current extent entry pointer */
2825         int                     error;  /* error return value */
2826         int                     flags;  /* inode logging flags */
2827         xfs_bmbt_irec_t         got;    /* current extent entry */
2828         xfs_fileoff_t           got_endoff;     /* first offset past got */
2829         int                     i;      /* temp state */
2830         xfs_ifork_t             *ifp;   /* inode fork pointer */
2831         xfs_mount_t             *mp;    /* mount structure */
2832         xfs_filblks_t           nblks;  /* quota/sb block count */
2833         xfs_bmbt_irec_t         new;    /* new record to be inserted */
2834         /* REFERENCED */
2835         uint                    qfield; /* quota field to update */
2836         xfs_filblks_t           temp;   /* for indirect length calculations */
2837         xfs_filblks_t           temp2;  /* for indirect length calculations */
2838         int                     state = 0;
2839
2840         XFS_STATS_INC(xs_del_exlist);
2841
2842         if (whichfork == XFS_ATTR_FORK)
2843                 state |= BMAP_ATTRFORK;
2844
2845         mp = ip->i_mount;
2846         ifp = XFS_IFORK_PTR(ip, whichfork);
2847         ASSERT((idx >= 0) && (idx < ifp->if_bytes /
2848                 (uint)sizeof(xfs_bmbt_rec_t)));
2849         ASSERT(del->br_blockcount > 0);
2850         ep = xfs_iext_get_ext(ifp, idx);
2851         xfs_bmbt_get_all(ep, &got);
2852         ASSERT(got.br_startoff <= del->br_startoff);
2853         del_endoff = del->br_startoff + del->br_blockcount;
2854         got_endoff = got.br_startoff + got.br_blockcount;
2855         ASSERT(got_endoff >= del_endoff);
2856         delay = isnullstartblock(got.br_startblock);
2857         ASSERT(isnullstartblock(del->br_startblock) == delay);
2858         flags = 0;
2859         qfield = 0;
2860         error = 0;
2861         /*
2862          * If deleting a real allocation, must free up the disk space.
2863          */
2864         if (!delay) {
2865                 flags = XFS_ILOG_CORE;
2866                 /*
2867                  * Realtime allocation.  Free it and record di_nblocks update.
2868                  */
2869                 if (whichfork == XFS_DATA_FORK && XFS_IS_REALTIME_INODE(ip)) {
2870                         xfs_fsblock_t   bno;
2871                         xfs_filblks_t   len;
2872
2873                         ASSERT(do_mod(del->br_blockcount,
2874                                       mp->m_sb.sb_rextsize) == 0);
2875                         ASSERT(do_mod(del->br_startblock,
2876                                       mp->m_sb.sb_rextsize) == 0);
2877                         bno = del->br_startblock;
2878                         len = del->br_blockcount;
2879                         do_div(bno, mp->m_sb.sb_rextsize);
2880                         do_div(len, mp->m_sb.sb_rextsize);
2881                         if ((error = xfs_rtfree_extent(ip->i_transp, bno,
2882                                         (xfs_extlen_t)len)))
2883                                 goto done;
2884                         do_fx = 0;
2885                         nblks = len * mp->m_sb.sb_rextsize;
2886                         qfield = XFS_TRANS_DQ_RTBCOUNT;
2887                 }
2888                 /*
2889                  * Ordinary allocation.
2890                  */
2891                 else {
2892                         do_fx = 1;
2893                         nblks = del->br_blockcount;
2894                         qfield = XFS_TRANS_DQ_BCOUNT;
2895                 }
2896                 /*
2897                  * Set up del_endblock and cur for later.
2898                  */
2899                 del_endblock = del->br_startblock + del->br_blockcount;
2900                 if (cur) {
2901                         if ((error = xfs_bmbt_lookup_eq(cur, got.br_startoff,
2902                                         got.br_startblock, got.br_blockcount,
2903                                         &i)))
2904                                 goto done;
2905                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2906                 }
2907                 da_old = da_new = 0;
2908         } else {
2909                 da_old = startblockval(got.br_startblock);
2910                 da_new = 0;
2911                 nblks = 0;
2912                 do_fx = 0;
2913         }
2914         /*
2915          * Set flag value to use in switch statement.
2916          * Left-contig is 2, right-contig is 1.
2917          */
2918         switch (((got.br_startoff == del->br_startoff) << 1) |
2919                 (got_endoff == del_endoff)) {
2920         case 3:
2921                 /*
2922                  * Matches the whole extent.  Delete the entry.
2923                  */
2924                 xfs_iext_remove(ip, idx, 1,
2925                                 whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0);
2926                 ifp->if_lastex = idx;
2927                 if (delay)
2928                         break;
2929                 XFS_IFORK_NEXT_SET(ip, whichfork,
2930                         XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
2931                 flags |= XFS_ILOG_CORE;
2932                 if (!cur) {
2933                         flags |= xfs_ilog_fext(whichfork);
2934                         break;
2935                 }
2936                 if ((error = xfs_btree_delete(cur, &i)))
2937                         goto done;
2938                 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2939                 break;
2940
2941         case 2:
2942                 /*
2943                  * Deleting the first part of the extent.
2944                  */
2945                 trace_xfs_bmap_pre_update(ip, idx, state, _THIS_IP_);
2946                 xfs_bmbt_set_startoff(ep, del_endoff);
2947                 temp = got.br_blockcount - del->br_blockcount;
2948                 xfs_bmbt_set_blockcount(ep, temp);
2949                 ifp->if_lastex = idx;
2950                 if (delay) {
2951                         temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2952                                 da_old);
2953                         xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
2954                         trace_xfs_bmap_post_update(ip, idx, state, _THIS_IP_);
2955                         da_new = temp;
2956                         break;
2957                 }
2958                 xfs_bmbt_set_startblock(ep, del_endblock);
2959                 trace_xfs_bmap_post_update(ip, idx, state, _THIS_IP_);
2960                 if (!cur) {
2961                         flags |= xfs_ilog_fext(whichfork);
2962                         break;
2963                 }
2964                 if ((error = xfs_bmbt_update(cur, del_endoff, del_endblock,
2965                                 got.br_blockcount - del->br_blockcount,
2966                                 got.br_state)))
2967                         goto done;
2968                 break;
2969
2970         case 1:
2971                 /*
2972                  * Deleting the last part of the extent.
2973                  */
2974                 temp = got.br_blockcount - del->br_blockcount;
2975                 trace_xfs_bmap_pre_update(ip, idx, state, _THIS_IP_);
2976                 xfs_bmbt_set_blockcount(ep, temp);
2977                 ifp->if_lastex = idx;
2978                 if (delay) {
2979                         temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2980                                 da_old);
2981                         xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
2982                         trace_xfs_bmap_post_update(ip, idx, state, _THIS_IP_);
2983                         da_new = temp;
2984                         break;
2985                 }
2986                 trace_xfs_bmap_post_update(ip, idx, state, _THIS_IP_);
2987                 if (!cur) {
2988                         flags |= xfs_ilog_fext(whichfork);
2989                         break;
2990                 }
2991                 if ((error = xfs_bmbt_update(cur, got.br_startoff,
2992                                 got.br_startblock,
2993                                 got.br_blockcount - del->br_blockcount,
2994                                 got.br_state)))
2995                         goto done;
2996                 break;
2997
2998         case 0:
2999                 /*
3000                  * Deleting the middle of the extent.
3001                  */
3002                 temp = del->br_startoff - got.br_startoff;
3003                 trace_xfs_bmap_pre_update(ip, idx, state, _THIS_IP_);
3004                 xfs_bmbt_set_blockcount(ep, temp);
3005                 new.br_startoff = del_endoff;
3006                 temp2 = got_endoff - del_endoff;
3007                 new.br_blockcount = temp2;
3008                 new.br_state = got.br_state;
3009                 if (!delay) {
3010                         new.br_startblock = del_endblock;
3011                         flags |= XFS_ILOG_CORE;
3012                         if (cur) {
3013                                 if ((error = xfs_bmbt_update(cur,
3014                                                 got.br_startoff,
3015                                                 got.br_startblock, temp,
3016                                                 got.br_state)))
3017                                         goto done;
3018                                 if ((error = xfs_btree_increment(cur, 0, &i)))
3019                                         goto done;
3020                                 cur->bc_rec.b = new;
3021                                 error = xfs_btree_insert(cur, &i);
3022                                 if (error && error != ENOSPC)
3023                                         goto done;
3024                                 /*
3025                                  * If get no-space back from btree insert,
3026                                  * it tried a split, and we have a zero
3027                                  * block reservation.
3028                                  * Fix up our state and return the error.
3029                                  */
3030                                 if (error == ENOSPC) {
3031                                         /*
3032                                          * Reset the cursor, don't trust
3033                                          * it after any insert operation.
3034                                          */
3035                                         if ((error = xfs_bmbt_lookup_eq(cur,
3036                                                         got.br_startoff,
3037                                                         got.br_startblock,
3038                                                         temp, &i)))
3039                                                 goto done;
3040                                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3041                                         /*
3042                                          * Update the btree record back
3043                                          * to the original value.
3044                                          */
3045                                         if ((error = xfs_bmbt_update(cur,
3046                                                         got.br_startoff,
3047                                                         got.br_startblock,
3048                                                         got.br_blockcount,
3049                                                         got.br_state)))
3050                                                 goto done;
3051                                         /*
3052                                          * Reset the extent record back
3053                                          * to the original value.
3054                                          */
3055                                         xfs_bmbt_set_blockcount(ep,
3056                                                 got.br_blockcount);
3057                                         flags = 0;
3058                                         error = XFS_ERROR(ENOSPC);
3059                                         goto done;
3060                                 }
3061                                 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3062                         } else
3063                                 flags |= xfs_ilog_fext(whichfork);
3064                         XFS_IFORK_NEXT_SET(ip, whichfork,
3065                                 XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
3066                 } else {
3067                         ASSERT(whichfork == XFS_DATA_FORK);
3068                         temp = xfs_bmap_worst_indlen(ip, temp);
3069                         xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
3070                         temp2 = xfs_bmap_worst_indlen(ip, temp2);
3071                         new.br_startblock = nullstartblock((int)temp2);
3072                         da_new = temp + temp2;
3073                         while (da_new > da_old) {
3074                                 if (temp) {
3075                                         temp--;
3076                                         da_new--;
3077                                         xfs_bmbt_set_startblock(ep,
3078                                                 nullstartblock((int)temp));
3079                                 }
3080                                 if (da_new == da_old)
3081                                         break;
3082                                 if (temp2) {
3083                                         temp2--;
3084                                         da_new--;
3085                                         new.br_startblock =
3086                                                 nullstartblock((int)temp2);
3087                                 }
3088                         }
3089                 }
3090                 trace_xfs_bmap_post_update(ip, idx, state, _THIS_IP_);
3091                 xfs_iext_insert(ip, idx + 1, 1, &new, state);
3092                 ifp->if_lastex = idx + 1;
3093                 break;
3094         }
3095         /*
3096          * If we need to, add to list of extents to delete.
3097          */
3098         if (do_fx)
3099                 xfs_bmap_add_free(del->br_startblock, del->br_blockcount, flist,
3100                         mp);
3101         /*
3102          * Adjust inode # blocks in the file.
3103          */
3104         if (nblks)
3105                 ip->i_d.di_nblocks -= nblks;
3106         /*
3107          * Adjust quota data.
3108          */
3109         if (qfield)
3110                 xfs_trans_mod_dquot_byino(tp, ip, qfield, (long)-nblks);
3111
3112         /*
3113          * Account for change in delayed indirect blocks.
3114          * Nothing to do for disk quota accounting here.
3115          */
3116         ASSERT(da_old >= da_new);
3117         if (da_old > da_new) {
3118                 xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
3119                         (int64_t)(da_old - da_new), rsvd);
3120         }
3121 done:
3122         *logflagsp = flags;
3123         return error;
3124 }
3125
3126 /*
3127  * Remove the entry "free" from the free item list.  Prev points to the
3128  * previous entry, unless "free" is the head of the list.
3129  */
3130 STATIC void
3131 xfs_bmap_del_free(
3132         xfs_bmap_free_t         *flist, /* free item list header */
3133         xfs_bmap_free_item_t    *prev,  /* previous item on list, if any */
3134         xfs_bmap_free_item_t    *free)  /* list item to be freed */
3135 {
3136         if (prev)
3137                 prev->xbfi_next = free->xbfi_next;
3138         else
3139                 flist->xbf_first = free->xbfi_next;
3140         flist->xbf_count--;
3141         kmem_zone_free(xfs_bmap_free_item_zone, free);
3142 }
3143
3144 /*
3145  * Convert an extents-format file into a btree-format file.
3146  * The new file will have a root block (in the inode) and a single child block.
3147  */
3148 STATIC int                                      /* error */
3149 xfs_bmap_extents_to_btree(
3150         xfs_trans_t             *tp,            /* transaction pointer */
3151         xfs_inode_t             *ip,            /* incore inode pointer */
3152         xfs_fsblock_t           *firstblock,    /* first-block-allocated */
3153         xfs_bmap_free_t         *flist,         /* blocks freed in xaction */
3154         xfs_btree_cur_t         **curp,         /* cursor returned to caller */
3155         int                     wasdel,         /* converting a delayed alloc */
3156         int                     *logflagsp,     /* inode logging flags */
3157         int                     whichfork)      /* data or attr fork */
3158 {
3159         struct xfs_btree_block  *ablock;        /* allocated (child) bt block */
3160         xfs_buf_t               *abp;           /* buffer for ablock */
3161         xfs_alloc_arg_t         args;           /* allocation arguments */
3162         xfs_bmbt_rec_t          *arp;           /* child record pointer */
3163         struct xfs_btree_block  *block;         /* btree root block */
3164         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
3165         xfs_bmbt_rec_host_t     *ep;            /* extent record pointer */
3166         int                     error;          /* error return value */
3167         xfs_extnum_t            i, cnt;         /* extent record index */
3168         xfs_ifork_t             *ifp;           /* inode fork pointer */
3169         xfs_bmbt_key_t          *kp;            /* root block key pointer */
3170         xfs_mount_t             *mp;            /* mount structure */
3171         xfs_extnum_t            nextents;       /* number of file extents */
3172         xfs_bmbt_ptr_t          *pp;            /* root block address pointer */
3173
3174         ifp = XFS_IFORK_PTR(ip, whichfork);
3175         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS);
3176         ASSERT(ifp->if_ext_max ==
3177                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
3178         /*
3179          * Make space in the inode incore.
3180          */
3181         xfs_iroot_realloc(ip, 1, whichfork);
3182         ifp->if_flags |= XFS_IFBROOT;
3183
3184         /*
3185          * Fill in the root.
3186          */
3187         block = ifp->if_broot;
3188         block->bb_magic = cpu_to_be32(XFS_BMAP_MAGIC);
3189         block->bb_level = cpu_to_be16(1);
3190         block->bb_numrecs = cpu_to_be16(1);
3191         block->bb_u.l.bb_leftsib = cpu_to_be64(NULLDFSBNO);
3192         block->bb_u.l.bb_rightsib = cpu_to_be64(NULLDFSBNO);
3193
3194         /*
3195          * Need a cursor.  Can't allocate until bb_level is filled in.
3196          */
3197         mp = ip->i_mount;
3198         cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
3199         cur->bc_private.b.firstblock = *firstblock;
3200         cur->bc_private.b.flist = flist;
3201         cur->bc_private.b.flags = wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
3202         /*
3203          * Convert to a btree with two levels, one record in root.
3204          */
3205         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_BTREE);
3206         args.tp = tp;
3207         args.mp = mp;
3208         args.firstblock = *firstblock;
3209         if (*firstblock == NULLFSBLOCK) {
3210                 args.type = XFS_ALLOCTYPE_START_BNO;
3211                 args.fsbno = XFS_INO_TO_FSB(mp, ip->i_ino);
3212         } else if (flist->xbf_low) {
3213                 args.type = XFS_ALLOCTYPE_START_BNO;
3214                 args.fsbno = *firstblock;
3215         } else {
3216                 args.type = XFS_ALLOCTYPE_NEAR_BNO;
3217                 args.fsbno = *firstblock;
3218         }
3219         args.minlen = args.maxlen = args.prod = 1;
3220         args.total = args.minleft = args.alignment = args.mod = args.isfl =
3221                 args.minalignslop = 0;
3222         args.wasdel = wasdel;
3223         *logflagsp = 0;
3224         if ((error = xfs_alloc_vextent(&args))) {
3225                 xfs_iroot_realloc(ip, -1, whichfork);
3226                 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
3227                 return error;
3228         }
3229         /*
3230          * Allocation can't fail, the space was reserved.
3231          */
3232         ASSERT(args.fsbno != NULLFSBLOCK);
3233         ASSERT(*firstblock == NULLFSBLOCK ||
3234                args.agno == XFS_FSB_TO_AGNO(mp, *firstblock) ||
3235                (flist->xbf_low &&
3236                 args.agno > XFS_FSB_TO_AGNO(mp, *firstblock)));
3237         *firstblock = cur->bc_private.b.firstblock = args.fsbno;
3238         cur->bc_private.b.allocated++;
3239         ip->i_d.di_nblocks++;
3240         xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, 1L);
3241         abp = xfs_btree_get_bufl(mp, tp, args.fsbno, 0);
3242         /*
3243          * Fill in the child block.
3244          */
3245         ablock = XFS_BUF_TO_BLOCK(abp);
3246         ablock->bb_magic = cpu_to_be32(XFS_BMAP_MAGIC);
3247         ablock->bb_level = 0;
3248         ablock->bb_u.l.bb_leftsib = cpu_to_be64(NULLDFSBNO);
3249         ablock->bb_u.l.bb_rightsib = cpu_to_be64(NULLDFSBNO);
3250         arp = XFS_BMBT_REC_ADDR(mp, ablock, 1);
3251         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3252         for (cnt = i = 0; i < nextents; i++) {
3253                 ep = xfs_iext_get_ext(ifp, i);
3254                 if (!isnullstartblock(xfs_bmbt_get_startblock(ep))) {
3255                         arp->l0 = cpu_to_be64(ep->l0);
3256                         arp->l1 = cpu_to_be64(ep->l1);
3257                         arp++; cnt++;
3258                 }
3259         }
3260         ASSERT(cnt == XFS_IFORK_NEXTENTS(ip, whichfork));
3261         xfs_btree_set_numrecs(ablock, cnt);
3262
3263         /*
3264          * Fill in the root key and pointer.
3265          */
3266         kp = XFS_BMBT_KEY_ADDR(mp, block, 1);
3267         arp = XFS_BMBT_REC_ADDR(mp, ablock, 1);
3268         kp->br_startoff = cpu_to_be64(xfs_bmbt_disk_get_startoff(arp));
3269         pp = XFS_BMBT_PTR_ADDR(mp, block, 1, xfs_bmbt_get_maxrecs(cur,
3270                                                 be16_to_cpu(block->bb_level)));
3271         *pp = cpu_to_be64(args.fsbno);
3272
3273         /*
3274          * Do all this logging at the end so that
3275          * the root is at the right level.
3276          */
3277         xfs_btree_log_block(cur, abp, XFS_BB_ALL_BITS);
3278         xfs_btree_log_recs(cur, abp, 1, be16_to_cpu(ablock->bb_numrecs));
3279         ASSERT(*curp == NULL);
3280         *curp = cur;
3281         *logflagsp = XFS_ILOG_CORE | xfs_ilog_fbroot(whichfork);
3282         return 0;
3283 }
3284
3285 /*
3286  * Calculate the default attribute fork offset for newly created inodes.
3287  */
3288 uint
3289 xfs_default_attroffset(
3290         struct xfs_inode        *ip)
3291 {
3292         struct xfs_mount        *mp = ip->i_mount;
3293         uint                    offset;
3294
3295         if (mp->m_sb.sb_inodesize == 256) {
3296                 offset = XFS_LITINO(mp) -
3297                                 XFS_BMDR_SPACE_CALC(MINABTPTRS);
3298         } else {
3299                 offset = XFS_BMDR_SPACE_CALC(6 * MINABTPTRS);
3300         }
3301
3302         ASSERT(offset < XFS_LITINO(mp));
3303         return offset;
3304 }
3305
3306 /*
3307  * Helper routine to reset inode di_forkoff field when switching
3308  * attribute fork from local to extent format - we reset it where
3309  * possible to make space available for inline data fork extents.
3310  */
3311 STATIC void
3312 xfs_bmap_forkoff_reset(
3313         xfs_mount_t     *mp,
3314         xfs_inode_t     *ip,
3315         int             whichfork)
3316 {
3317         if (whichfork == XFS_ATTR_FORK &&
3318             ip->i_d.di_format != XFS_DINODE_FMT_DEV &&
3319             ip->i_d.di_format != XFS_DINODE_FMT_UUID &&
3320             ip->i_d.di_format != XFS_DINODE_FMT_BTREE) {
3321                 uint    dfl_forkoff = xfs_default_attroffset(ip) >> 3;
3322
3323                 if (dfl_forkoff > ip->i_d.di_forkoff) {
3324                         ip->i_d.di_forkoff = dfl_forkoff;
3325                         ip->i_df.if_ext_max =
3326                                 XFS_IFORK_DSIZE(ip) / sizeof(xfs_bmbt_rec_t);
3327                         ip->i_afp->if_ext_max =
3328                                 XFS_IFORK_ASIZE(ip) / sizeof(xfs_bmbt_rec_t);
3329                 }
3330         }
3331 }
3332
3333 /*
3334  * Convert a local file to an extents file.
3335  * This code is out of bounds for data forks of regular files,
3336  * since the file data needs to get logged so things will stay consistent.
3337  * (The bmap-level manipulations are ok, though).
3338  */
3339 STATIC int                              /* error */
3340 xfs_bmap_local_to_extents(
3341         xfs_trans_t     *tp,            /* transaction pointer */
3342         xfs_inode_t     *ip,            /* incore inode pointer */
3343         xfs_fsblock_t   *firstblock,    /* first block allocated in xaction */
3344         xfs_extlen_t    total,          /* total blocks needed by transaction */
3345         int             *logflagsp,     /* inode logging flags */
3346         int             whichfork)      /* data or attr fork */
3347 {
3348         int             error;          /* error return value */
3349         int             flags;          /* logging flags returned */
3350         xfs_ifork_t     *ifp;           /* inode fork pointer */
3351
3352         /*
3353          * We don't want to deal with the case of keeping inode data inline yet.
3354          * So sending the data fork of a regular inode is invalid.
3355          */
3356         ASSERT(!((ip->i_d.di_mode & S_IFMT) == S_IFREG &&
3357                  whichfork == XFS_DATA_FORK));
3358         ifp = XFS_IFORK_PTR(ip, whichfork);
3359         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
3360         flags = 0;
3361         error = 0;
3362         if (ifp->if_bytes) {
3363                 xfs_alloc_arg_t args;   /* allocation arguments */
3364                 xfs_buf_t       *bp;    /* buffer for extent block */
3365                 xfs_bmbt_rec_host_t *ep;/* extent record pointer */
3366
3367                 args.tp = tp;
3368                 args.mp = ip->i_mount;
3369                 args.firstblock = *firstblock;
3370                 ASSERT((ifp->if_flags &
3371                         (XFS_IFINLINE|XFS_IFEXTENTS|XFS_IFEXTIREC)) == XFS_IFINLINE);
3372                 /*
3373                  * Allocate a block.  We know we need only one, since the
3374                  * file currently fits in an inode.
3375                  */
3376                 if (*firstblock == NULLFSBLOCK) {
3377                         args.fsbno = XFS_INO_TO_FSB(args.mp, ip->i_ino);
3378                         args.type = XFS_ALLOCTYPE_START_BNO;
3379                 } else {
3380                         args.fsbno = *firstblock;
3381                         args.type = XFS_ALLOCTYPE_NEAR_BNO;
3382                 }
3383                 args.total = total;
3384                 args.mod = args.minleft = args.alignment = args.wasdel =
3385                         args.isfl = args.minalignslop = 0;
3386                 args.minlen = args.maxlen = args.prod = 1;
3387                 if ((error = xfs_alloc_vextent(&args)))
3388                         goto done;
3389                 /*
3390                  * Can't fail, the space was reserved.
3391                  */
3392                 ASSERT(args.fsbno != NULLFSBLOCK);
3393                 ASSERT(args.len == 1);
3394                 *firstblock = args.fsbno;
3395                 bp = xfs_btree_get_bufl(args.mp, tp, args.fsbno, 0);
3396                 memcpy((char *)XFS_BUF_PTR(bp), ifp->if_u1.if_data,
3397                         ifp->if_bytes);
3398                 xfs_trans_log_buf(tp, bp, 0, ifp->if_bytes - 1);
3399                 xfs_bmap_forkoff_reset(args.mp, ip, whichfork);
3400                 xfs_idata_realloc(ip, -ifp->if_bytes, whichfork);
3401                 xfs_iext_add(ifp, 0, 1);
3402                 ep = xfs_iext_get_ext(ifp, 0);
3403                 xfs_bmbt_set_allf(ep, 0, args.fsbno, 1, XFS_EXT_NORM);
3404                 trace_xfs_bmap_post_update(ip, 0,
3405                                 whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0,
3406                                 _THIS_IP_);
3407                 XFS_IFORK_NEXT_SET(ip, whichfork, 1);
3408                 ip->i_d.di_nblocks = 1;
3409                 xfs_trans_mod_dquot_byino(tp, ip,
3410                         XFS_TRANS_DQ_BCOUNT, 1L);
3411                 flags |= xfs_ilog_fext(whichfork);
3412         } else {
3413                 ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) == 0);
3414                 xfs_bmap_forkoff_reset(ip->i_mount, ip, whichfork);
3415         }
3416         ifp->if_flags &= ~XFS_IFINLINE;
3417         ifp->if_flags |= XFS_IFEXTENTS;
3418         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
3419         flags |= XFS_ILOG_CORE;
3420 done:
3421         *logflagsp = flags;
3422         return error;
3423 }
3424
3425 /*
3426  * Search the extent records for the entry containing block bno.
3427  * If bno lies in a hole, point to the next entry.  If bno lies
3428  * past eof, *eofp will be set, and *prevp will contain the last
3429  * entry (null if none).  Else, *lastxp will be set to the index
3430  * of the found entry; *gotp will contain the entry.
3431  */
3432 STATIC xfs_bmbt_rec_host_t *            /* pointer to found extent entry */
3433 xfs_bmap_search_multi_extents(
3434         xfs_ifork_t     *ifp,           /* inode fork pointer */
3435         xfs_fileoff_t   bno,            /* block number searched for */
3436         int             *eofp,          /* out: end of file found */
3437         xfs_extnum_t    *lastxp,        /* out: last extent index */
3438         xfs_bmbt_irec_t *gotp,          /* out: extent entry found */
3439         xfs_bmbt_irec_t *prevp)         /* out: previous extent entry found */
3440 {
3441         xfs_bmbt_rec_host_t *ep;                /* extent record pointer */
3442         xfs_extnum_t    lastx;          /* last extent index */
3443
3444         /*
3445          * Initialize the extent entry structure to catch access to
3446          * uninitialized br_startblock field.
3447          */
3448         gotp->br_startoff = 0xffa5a5a5a5a5a5a5LL;
3449         gotp->br_blockcount = 0xa55a5a5a5a5a5a5aLL;
3450         gotp->br_state = XFS_EXT_INVALID;
3451 #if XFS_BIG_BLKNOS
3452         gotp->br_startblock = 0xffffa5a5a5a5a5a5LL;
3453 #else
3454         gotp->br_startblock = 0xffffa5a5;
3455 #endif
3456         prevp->br_startoff = NULLFILEOFF;
3457
3458         ep = xfs_iext_bno_to_ext(ifp, bno, &lastx);
3459         if (lastx > 0) {
3460                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx - 1), prevp);
3461         }
3462         if (lastx < (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t))) {
3463                 xfs_bmbt_get_all(ep, gotp);
3464                 *eofp = 0;
3465         } else {
3466                 if (lastx > 0) {
3467                         *gotp = *prevp;
3468                 }
3469                 *eofp = 1;
3470                 ep = NULL;
3471         }
3472         *lastxp = lastx;
3473         return ep;
3474 }
3475
3476 /*
3477  * Search the extents list for the inode, for the extent containing bno.
3478  * If bno lies in a hole, point to the next entry.  If bno lies past eof,
3479  * *eofp will be set, and *prevp will contain the last entry (null if none).
3480  * Else, *lastxp will be set to the index of the found
3481  * entry; *gotp will contain the entry.
3482  */
3483 STATIC xfs_bmbt_rec_host_t *                 /* pointer to found extent entry */
3484 xfs_bmap_search_extents(
3485         xfs_inode_t     *ip,            /* incore inode pointer */
3486         xfs_fileoff_t   bno,            /* block number searched for */
3487         int             fork,           /* data or attr fork */
3488         int             *eofp,          /* out: end of file found */
3489         xfs_extnum_t    *lastxp,        /* out: last extent index */
3490         xfs_bmbt_irec_t *gotp,          /* out: extent entry found */
3491         xfs_bmbt_irec_t *prevp)         /* out: previous extent entry found */
3492 {
3493         xfs_ifork_t     *ifp;           /* inode fork pointer */
3494         xfs_bmbt_rec_host_t  *ep;            /* extent record pointer */
3495
3496         XFS_STATS_INC(xs_look_exlist);
3497         ifp = XFS_IFORK_PTR(ip, fork);
3498
3499         ep = xfs_bmap_search_multi_extents(ifp, bno, eofp, lastxp, gotp, prevp);
3500
3501         if (unlikely(!(gotp->br_startblock) && (*lastxp != NULLEXTNUM) &&
3502                      !(XFS_IS_REALTIME_INODE(ip) && fork == XFS_DATA_FORK))) {
3503                 xfs_cmn_err(XFS_PTAG_FSBLOCK_ZERO, CE_ALERT, ip->i_mount,
3504                                 "Access to block zero in inode %llu "
3505                                 "start_block: %llx start_off: %llx "
3506                                 "blkcnt: %llx extent-state: %x lastx: %x\n",
3507                         (unsigned long long)ip->i_ino,
3508                         (unsigned long long)gotp->br_startblock,
3509                         (unsigned long long)gotp->br_startoff,
3510                         (unsigned long long)gotp->br_blockcount,
3511                         gotp->br_state, *lastxp);
3512                 *lastxp = NULLEXTNUM;
3513                 *eofp = 1;
3514                 return NULL;
3515         }
3516         return ep;
3517 }
3518
3519 /*
3520  * Compute the worst-case number of indirect blocks that will be used
3521  * for ip's delayed extent of length "len".
3522  */
3523 STATIC xfs_filblks_t
3524 xfs_bmap_worst_indlen(
3525         xfs_inode_t     *ip,            /* incore inode pointer */
3526         xfs_filblks_t   len)            /* delayed extent length */
3527 {
3528         int             level;          /* btree level number */
3529         int             maxrecs;        /* maximum record count at this level */
3530         xfs_mount_t     *mp;            /* mount structure */
3531         xfs_filblks_t   rval;           /* return value */
3532
3533         mp = ip->i_mount;
3534         maxrecs = mp->m_bmap_dmxr[0];
3535         for (level = 0, rval = 0;
3536              level < XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK);
3537              level++) {
3538                 len += maxrecs - 1;
3539                 do_div(len, maxrecs);
3540                 rval += len;
3541                 if (len == 1)
3542                         return rval + XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK) -
3543                                 level - 1;
3544                 if (level == 0)
3545                         maxrecs = mp->m_bmap_dmxr[1];
3546         }
3547         return rval;
3548 }
3549
3550 /*
3551  * Convert inode from non-attributed to attributed.
3552  * Must not be in a transaction, ip must not be locked.
3553  */
3554 int                                             /* error code */
3555 xfs_bmap_add_attrfork(
3556         xfs_inode_t             *ip,            /* incore inode pointer */
3557         int                     size,           /* space new attribute needs */
3558         int                     rsvd)           /* xact may use reserved blks */
3559 {
3560         xfs_fsblock_t           firstblock;     /* 1st block/ag allocated */
3561         xfs_bmap_free_t         flist;          /* freed extent records */
3562         xfs_mount_t             *mp;            /* mount structure */
3563         xfs_trans_t             *tp;            /* transaction pointer */
3564         int                     blks;           /* space reservation */
3565         int                     version = 1;    /* superblock attr version */
3566         int                     committed;      /* xaction was committed */
3567         int                     logflags;       /* logging flags */
3568         int                     error;          /* error return value */
3569
3570         ASSERT(XFS_IFORK_Q(ip) == 0);
3571         ASSERT(ip->i_df.if_ext_max ==
3572                XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t));
3573
3574         mp = ip->i_mount;
3575         ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
3576         tp = xfs_trans_alloc(mp, XFS_TRANS_ADDAFORK);
3577         blks = XFS_ADDAFORK_SPACE_RES(mp);
3578         if (rsvd)
3579                 tp->t_flags |= XFS_TRANS_RESERVE;
3580         if ((error = xfs_trans_reserve(tp, blks, XFS_ADDAFORK_LOG_RES(mp), 0,
3581                         XFS_TRANS_PERM_LOG_RES, XFS_ADDAFORK_LOG_COUNT)))
3582                 goto error0;
3583         xfs_ilock(ip, XFS_ILOCK_EXCL);
3584         error = xfs_trans_reserve_quota_nblks(tp, ip, blks, 0, rsvd ?
3585                         XFS_QMOPT_RES_REGBLKS | XFS_QMOPT_FORCE_RES :
3586                         XFS_QMOPT_RES_REGBLKS);
3587         if (error) {
3588                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3589                 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES);
3590                 return error;
3591         }
3592         if (XFS_IFORK_Q(ip))
3593                 goto error1;
3594         if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS) {
3595                 /*
3596                  * For inodes coming from pre-6.2 filesystems.
3597                  */
3598                 ASSERT(ip->i_d.di_aformat == 0);
3599                 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
3600         }
3601         ASSERT(ip->i_d.di_anextents == 0);
3602
3603         xfs_trans_ijoin_ref(tp, ip, XFS_ILOCK_EXCL);
3604         xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
3605
3606         switch (ip->i_d.di_format) {
3607         case XFS_DINODE_FMT_DEV:
3608                 ip->i_d.di_forkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
3609                 break;
3610         case XFS_DINODE_FMT_UUID:
3611                 ip->i_d.di_forkoff = roundup(sizeof(uuid_t), 8) >> 3;
3612                 break;
3613         case XFS_DINODE_FMT_LOCAL:
3614         case XFS_DINODE_FMT_EXTENTS:
3615         case XFS_DINODE_FMT_BTREE:
3616                 ip->i_d.di_forkoff = xfs_attr_shortform_bytesfit(ip, size);
3617                 if (!ip->i_d.di_forkoff)
3618                         ip->i_d.di_forkoff = xfs_default_attroffset(ip) >> 3;
3619                 else if (mp->m_flags & XFS_MOUNT_ATTR2)
3620                         version = 2;
3621                 break;
3622         default:
3623                 ASSERT(0);
3624                 error = XFS_ERROR(EINVAL);
3625                 goto error1;
3626         }
3627         ip->i_df.if_ext_max =
3628                 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
3629         ASSERT(ip->i_afp == NULL);
3630         ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP);
3631         ip->i_afp->if_ext_max =
3632                 XFS_IFORK_ASIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
3633         ip->i_afp->if_flags = XFS_IFEXTENTS;
3634         logflags = 0;
3635         xfs_bmap_init(&flist, &firstblock);
3636         switch (ip->i_d.di_format) {
3637         case XFS_DINODE_FMT_LOCAL:
3638                 error = xfs_bmap_add_attrfork_local(tp, ip, &firstblock, &flist,
3639                         &logflags);
3640                 break;
3641         case XFS_DINODE_FMT_EXTENTS:
3642                 error = xfs_bmap_add_attrfork_extents(tp, ip, &firstblock,
3643                         &flist, &logflags);
3644                 break;
3645         case XFS_DINODE_FMT_BTREE:
3646                 error = xfs_bmap_add_attrfork_btree(tp, ip, &firstblock, &flist,
3647                         &logflags);
3648                 break;
3649         default:
3650                 error = 0;
3651                 break;
3652         }
3653         if (logflags)
3654                 xfs_trans_log_inode(tp, ip, logflags);
3655         if (error)
3656                 goto error2;
3657         if (!xfs_sb_version_hasattr(&mp->m_sb) ||
3658            (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2)) {
3659                 __int64_t sbfields = 0;
3660
3661                 spin_lock(&mp->m_sb_lock);
3662                 if (!xfs_sb_version_hasattr(&mp->m_sb)) {
3663                         xfs_sb_version_addattr(&mp->m_sb);
3664                         sbfields |= XFS_SB_VERSIONNUM;
3665                 }
3666                 if (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2) {
3667                         xfs_sb_version_addattr2(&mp->m_sb);
3668                         sbfields |= (XFS_SB_VERSIONNUM | XFS_SB_FEATURES2);
3669                 }
3670                 if (sbfields) {
3671                         spin_unlock(&mp->m_sb_lock);
3672                         xfs_mod_sb(tp, sbfields);
3673                 } else
3674                         spin_unlock(&mp->m_sb_lock);
3675         }
3676         if ((error = xfs_bmap_finish(&tp, &flist, &committed)))
3677                 goto error2;
3678         error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
3679         ASSERT(ip->i_df.if_ext_max ==
3680                XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t));
3681         return error;
3682 error2:
3683         xfs_bmap_cancel(&flist);
3684 error1:
3685         xfs_iunlock(ip, XFS_ILOCK_EXCL);
3686 error0:
3687         xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
3688         ASSERT(ip->i_df.if_ext_max ==
3689                XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t));
3690         return error;
3691 }
3692
3693 /*
3694  * Add the extent to the list of extents to be free at transaction end.
3695  * The list is maintained sorted (by block number).
3696  */
3697 /* ARGSUSED */
3698 void
3699 xfs_bmap_add_free(
3700         xfs_fsblock_t           bno,            /* fs block number of extent */
3701         xfs_filblks_t           len,            /* length of extent */
3702         xfs_bmap_free_t         *flist,         /* list of extents */
3703         xfs_mount_t             *mp)            /* mount point structure */
3704 {
3705         xfs_bmap_free_item_t    *cur;           /* current (next) element */
3706         xfs_bmap_free_item_t    *new;           /* new element */
3707         xfs_bmap_free_item_t    *prev;          /* previous element */
3708 #ifdef DEBUG
3709         xfs_agnumber_t          agno;
3710         xfs_agblock_t           agbno;
3711
3712         ASSERT(bno != NULLFSBLOCK);
3713         ASSERT(len > 0);
3714         ASSERT(len <= MAXEXTLEN);
3715         ASSERT(!isnullstartblock(bno));
3716         agno = XFS_FSB_TO_AGNO(mp, bno);
3717         agbno = XFS_FSB_TO_AGBNO(mp, bno);
3718         ASSERT(agno < mp->m_sb.sb_agcount);
3719         ASSERT(agbno < mp->m_sb.sb_agblocks);
3720         ASSERT(len < mp->m_sb.sb_agblocks);
3721         ASSERT(agbno + len <= mp->m_sb.sb_agblocks);
3722 #endif
3723         ASSERT(xfs_bmap_free_item_zone != NULL);
3724         new = kmem_zone_alloc(xfs_bmap_free_item_zone, KM_SLEEP);
3725         new->xbfi_startblock = bno;
3726         new->xbfi_blockcount = (xfs_extlen_t)len;
3727         for (prev = NULL, cur = flist->xbf_first;
3728              cur != NULL;
3729              prev = cur, cur = cur->xbfi_next) {
3730                 if (cur->xbfi_startblock >= bno)
3731                         break;
3732         }
3733         if (prev)
3734                 prev->xbfi_next = new;
3735         else
3736                 flist->xbf_first = new;
3737         new->xbfi_next = cur;
3738         flist->xbf_count++;
3739 }
3740
3741 /*
3742  * Compute and fill in the value of the maximum depth of a bmap btree
3743  * in this filesystem.  Done once, during mount.
3744  */
3745 void
3746 xfs_bmap_compute_maxlevels(
3747         xfs_mount_t     *mp,            /* file system mount structure */
3748         int             whichfork)      /* data or attr fork */
3749 {
3750         int             level;          /* btree level */
3751         uint            maxblocks;      /* max blocks at this level */
3752         uint            maxleafents;    /* max leaf entries possible */
3753         int             maxrootrecs;    /* max records in root block */
3754         int             minleafrecs;    /* min records in leaf block */
3755         int             minnoderecs;    /* min records in node block */
3756         int             sz;             /* root block size */
3757
3758         /*
3759          * The maximum number of extents in a file, hence the maximum
3760          * number of leaf entries, is controlled by the type of di_nextents
3761          * (a signed 32-bit number, xfs_extnum_t), or by di_anextents
3762          * (a signed 16-bit number, xfs_aextnum_t).
3763          *
3764          * Note that we can no longer assume that if we are in ATTR1 that
3765          * the fork offset of all the inodes will be
3766          * (xfs_default_attroffset(ip) >> 3) because we could have mounted
3767          * with ATTR2 and then mounted back with ATTR1, keeping the
3768          * di_forkoff's fixed but probably at various positions. Therefore,
3769          * for both ATTR1 and ATTR2 we have to assume the worst case scenario
3770          * of a minimum size available.
3771          */
3772         if (whichfork == XFS_DATA_FORK) {
3773                 maxleafents = MAXEXTNUM;
3774                 sz = XFS_BMDR_SPACE_CALC(MINDBTPTRS);
3775         } else {
3776                 maxleafents = MAXAEXTNUM;
3777                 sz = XFS_BMDR_SPACE_CALC(MINABTPTRS);
3778         }
3779         maxrootrecs = xfs_bmdr_maxrecs(mp, sz, 0);
3780         minleafrecs = mp->m_bmap_dmnr[0];
3781         minnoderecs = mp->m_bmap_dmnr[1];
3782         maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs;
3783         for (level = 1; maxblocks > 1; level++) {
3784                 if (maxblocks <= maxrootrecs)
3785                         maxblocks = 1;
3786                 else
3787                         maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs;
3788         }
3789         mp->m_bm_maxlevels[whichfork] = level;
3790 }
3791
3792 /*
3793  * Routine to be called at transaction's end by xfs_bmapi, xfs_bunmapi
3794  * caller.  Frees all the extents that need freeing, which must be done
3795  * last due to locking considerations.  We never free any extents in
3796  * the first transaction.  This is to allow the caller to make the first
3797  * transaction a synchronous one so that the pointers to the data being
3798  * broken in this transaction will be permanent before the data is actually
3799  * freed.  This is necessary to prevent blocks from being reallocated
3800  * and written to before the free and reallocation are actually permanent.
3801  * We do not just make the first transaction synchronous here, because
3802  * there are more efficient ways to gain the same protection in some cases
3803  * (see the file truncation code).
3804  *
3805  * Return 1 if the given transaction was committed and a new one
3806  * started, and 0 otherwise in the committed parameter.
3807  */
3808 /*ARGSUSED*/
3809 int                                             /* error */
3810 xfs_bmap_finish(
3811         xfs_trans_t             **tp,           /* transaction pointer addr */
3812         xfs_bmap_free_t         *flist,         /* i/o: list extents to free */
3813         int                     *committed)     /* xact committed or not */
3814 {
3815         xfs_efd_log_item_t      *efd;           /* extent free data */
3816         xfs_efi_log_item_t      *efi;           /* extent free intention */
3817         int                     error;          /* error return value */
3818         xfs_bmap_free_item_t    *free;          /* free extent item */
3819         unsigned int            logres;         /* new log reservation */
3820         unsigned int            logcount;       /* new log count */
3821         xfs_mount_t             *mp;            /* filesystem mount structure */
3822         xfs_bmap_free_item_t    *next;          /* next item on free list */
3823         xfs_trans_t             *ntp;           /* new transaction pointer */
3824
3825         ASSERT((*tp)->t_flags & XFS_TRANS_PERM_LOG_RES);
3826         if (flist->xbf_count == 0) {
3827                 *committed = 0;
3828                 return 0;
3829         }
3830         ntp = *tp;
3831         efi = xfs_trans_get_efi(ntp, flist->xbf_count);
3832         for (free = flist->xbf_first; free; free = free->xbfi_next)
3833                 xfs_trans_log_efi_extent(ntp, efi, free->xbfi_startblock,
3834                         free->xbfi_blockcount);
3835         logres = ntp->t_log_res;
3836         logcount = ntp->t_log_count;
3837         ntp = xfs_trans_dup(*tp);
3838         error = xfs_trans_commit(*tp, 0);
3839         *tp = ntp;
3840         *committed = 1;
3841         /*
3842          * We have a new transaction, so we should return committed=1,
3843          * even though we're returning an error.
3844          */
3845         if (error)
3846                 return error;
3847
3848         /*
3849          * transaction commit worked ok so we can drop the extra ticket
3850          * reference that we gained in xfs_trans_dup()
3851          */
3852         xfs_log_ticket_put(ntp->t_ticket);
3853
3854         if ((error = xfs_trans_reserve(ntp, 0, logres, 0, XFS_TRANS_PERM_LOG_RES,
3855                         logcount)))
3856                 return error;
3857         efd = xfs_trans_get_efd(ntp, efi, flist->xbf_count);
3858         for (free = flist->xbf_first; free != NULL; free = next) {
3859                 next = free->xbfi_next;
3860                 if ((error = xfs_free_extent(ntp, free->xbfi_startblock,
3861                                 free->xbfi_blockcount))) {
3862                         /*
3863                          * The bmap free list will be cleaned up at a
3864                          * higher level.  The EFI will be canceled when
3865                          * this transaction is aborted.
3866                          * Need to force shutdown here to make sure it
3867                          * happens, since this transaction may not be
3868                          * dirty yet.
3869                          */
3870                         mp = ntp->t_mountp;
3871                         if (!XFS_FORCED_SHUTDOWN(mp))
3872                                 xfs_force_shutdown(mp,
3873                                                    (error == EFSCORRUPTED) ?
3874                                                    SHUTDOWN_CORRUPT_INCORE :
3875                                                    SHUTDOWN_META_IO_ERROR);
3876                         return error;
3877                 }
3878                 xfs_trans_log_efd_extent(ntp, efd, free->xbfi_startblock,
3879                         free->xbfi_blockcount);
3880                 xfs_bmap_del_free(flist, NULL, free);
3881         }
3882         return 0;
3883 }
3884
3885 /*
3886  * Free up any items left in the list.
3887  */
3888 void
3889 xfs_bmap_cancel(
3890         xfs_bmap_free_t         *flist) /* list of bmap_free_items */
3891 {
3892         xfs_bmap_free_item_t    *free;  /* free list item */
3893         xfs_bmap_free_item_t    *next;
3894
3895         if (flist->xbf_count == 0)
3896                 return;
3897         ASSERT(flist->xbf_first != NULL);
3898         for (free = flist->xbf_first; free; free = next) {
3899                 next = free->xbfi_next;
3900                 xfs_bmap_del_free(flist, NULL, free);
3901         }
3902         ASSERT(flist->xbf_count == 0);
3903 }
3904
3905 /*
3906  * Returns the file-relative block number of the first unused block(s)
3907  * in the file with at least "len" logically contiguous blocks free.
3908  * This is the lowest-address hole if the file has holes, else the first block
3909  * past the end of file.
3910  * Return 0 if the file is currently local (in-inode).
3911  */
3912 int                                             /* error */
3913 xfs_bmap_first_unused(
3914         xfs_trans_t     *tp,                    /* transaction pointer */
3915         xfs_inode_t     *ip,                    /* incore inode */
3916         xfs_extlen_t    len,                    /* size of hole to find */
3917         xfs_fileoff_t   *first_unused,          /* unused block */
3918         int             whichfork)              /* data or attr fork */
3919 {
3920         int             error;                  /* error return value */
3921         int             idx;                    /* extent record index */
3922         xfs_ifork_t     *ifp;                   /* inode fork pointer */
3923         xfs_fileoff_t   lastaddr;               /* last block number seen */
3924         xfs_fileoff_t   lowest;                 /* lowest useful block */
3925         xfs_fileoff_t   max;                    /* starting useful block */
3926         xfs_fileoff_t   off;                    /* offset for this block */
3927         xfs_extnum_t    nextents;               /* number of extent entries */
3928
3929         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE ||
3930                XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ||
3931                XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
3932         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
3933                 *first_unused = 0;
3934                 return 0;
3935         }
3936         ifp = XFS_IFORK_PTR(ip, whichfork);
3937         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
3938             (error = xfs_iread_extents(tp, ip, whichfork)))
3939                 return error;
3940         lowest = *first_unused;
3941         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3942         for (idx = 0, lastaddr = 0, max = lowest; idx < nextents; idx++) {
3943                 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, idx);
3944                 off = xfs_bmbt_get_startoff(ep);
3945                 /*
3946                  * See if the hole before this extent will work.
3947                  */
3948                 if (off >= lowest + len && off - max >= len) {
3949                         *first_unused = max;
3950                         return 0;
3951                 }
3952                 lastaddr = off + xfs_bmbt_get_blockcount(ep);
3953                 max = XFS_FILEOFF_MAX(lastaddr, lowest);
3954         }
3955         *first_unused = max;
3956         return 0;
3957 }
3958
3959 /*
3960  * Returns the file-relative block number of the last block + 1 before
3961  * last_block (input value) in the file.
3962  * This is not based on i_size, it is based on the extent records.
3963  * Returns 0 for local files, as they do not have extent records.
3964  */
3965 int                                             /* error */
3966 xfs_bmap_last_before(
3967         xfs_trans_t     *tp,                    /* transaction pointer */
3968         xfs_inode_t     *ip,                    /* incore inode */
3969         xfs_fileoff_t   *last_block,            /* last block */
3970         int             whichfork)              /* data or attr fork */
3971 {
3972         xfs_fileoff_t   bno;                    /* input file offset */
3973         int             eof;                    /* hit end of file */
3974         xfs_bmbt_rec_host_t *ep;                /* pointer to last extent */
3975         int             error;                  /* error return value */
3976         xfs_bmbt_irec_t got;                    /* current extent value */
3977         xfs_ifork_t     *ifp;                   /* inode fork pointer */
3978         xfs_extnum_t    lastx;                  /* last extent used */
3979         xfs_bmbt_irec_t prev;                   /* previous extent value */
3980
3981         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
3982             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
3983             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL)
3984                return XFS_ERROR(EIO);
3985         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
3986                 *last_block = 0;
3987                 return 0;
3988         }
3989         ifp = XFS_IFORK_PTR(ip, whichfork);
3990         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
3991             (error = xfs_iread_extents(tp, ip, whichfork)))
3992                 return error;
3993         bno = *last_block - 1;
3994         ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
3995                 &prev);
3996         if (eof || xfs_bmbt_get_startoff(ep) > bno) {
3997                 if (prev.br_startoff == NULLFILEOFF)
3998                         *last_block = 0;
3999                 else
4000                         *last_block = prev.br_startoff + prev.br_blockcount;
4001         }
4002         /*
4003          * Otherwise *last_block is already the right answer.
4004          */
4005         return 0;
4006 }
4007
4008 /*
4009  * Returns the file-relative block number of the first block past eof in
4010  * the file.  This is not based on i_size, it is based on the extent records.
4011  * Returns 0 for local files, as they do not have extent records.
4012  */
4013 int                                             /* error */
4014 xfs_bmap_last_offset(
4015         xfs_trans_t     *tp,                    /* transaction pointer */
4016         xfs_inode_t     *ip,                    /* incore inode */
4017         xfs_fileoff_t   *last_block,            /* last block */
4018         int             whichfork)              /* data or attr fork */
4019 {
4020         xfs_bmbt_rec_host_t *ep;                /* pointer to last extent */
4021         int             error;                  /* error return value */
4022         xfs_ifork_t     *ifp;                   /* inode fork pointer */
4023         xfs_extnum_t    nextents;               /* number of extent entries */
4024
4025         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
4026             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4027             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL)
4028                return XFS_ERROR(EIO);
4029         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
4030                 *last_block = 0;
4031                 return 0;
4032         }
4033         ifp = XFS_IFORK_PTR(ip, whichfork);
4034         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
4035             (error = xfs_iread_extents(tp, ip, whichfork)))
4036                 return error;
4037         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4038         if (!nextents) {
4039                 *last_block = 0;
4040                 return 0;
4041         }
4042         ep = xfs_iext_get_ext(ifp, nextents - 1);
4043         *last_block = xfs_bmbt_get_startoff(ep) + xfs_bmbt_get_blockcount(ep);
4044         return 0;
4045 }
4046
4047 /*
4048  * Returns whether the selected fork of the inode has exactly one
4049  * block or not.  For the data fork we check this matches di_size,
4050  * implying the file's range is 0..bsize-1.
4051  */
4052 int                                     /* 1=>1 block, 0=>otherwise */
4053 xfs_bmap_one_block(
4054         xfs_inode_t     *ip,            /* incore inode */
4055         int             whichfork)      /* data or attr fork */
4056 {
4057         xfs_bmbt_rec_host_t *ep;        /* ptr to fork's extent */
4058         xfs_ifork_t     *ifp;           /* inode fork pointer */
4059         int             rval;           /* return value */
4060         xfs_bmbt_irec_t s;              /* internal version of extent */
4061
4062 #ifndef DEBUG
4063         if (whichfork == XFS_DATA_FORK) {
4064                 return ((ip->i_d.di_mode & S_IFMT) == S_IFREG) ?
4065                         (ip->i_size == ip->i_mount->m_sb.sb_blocksize) :
4066                         (ip->i_d.di_size == ip->i_mount->m_sb.sb_blocksize);
4067         }
4068 #endif  /* !DEBUG */
4069         if (XFS_IFORK_NEXTENTS(ip, whichfork) != 1)
4070                 return 0;
4071         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
4072                 return 0;
4073         ifp = XFS_IFORK_PTR(ip, whichfork);
4074         ASSERT(ifp->if_flags & XFS_IFEXTENTS);
4075         ep = xfs_iext_get_ext(ifp, 0);
4076         xfs_bmbt_get_all(ep, &s);
4077         rval = s.br_startoff == 0 && s.br_blockcount == 1;
4078         if (rval && whichfork == XFS_DATA_FORK)
4079                 ASSERT(ip->i_size == ip->i_mount->m_sb.sb_blocksize);
4080         return rval;
4081 }
4082
4083 STATIC int
4084 xfs_bmap_sanity_check(
4085         struct xfs_mount        *mp,
4086         struct xfs_buf          *bp,
4087         int                     level)
4088 {
4089         struct xfs_btree_block  *block = XFS_BUF_TO_BLOCK(bp);
4090
4091         if (be32_to_cpu(block->bb_magic) != XFS_BMAP_MAGIC ||
4092             be16_to_cpu(block->bb_level) != level ||
4093             be16_to_cpu(block->bb_numrecs) == 0 ||
4094             be16_to_cpu(block->bb_numrecs) > mp->m_bmap_dmxr[level != 0])
4095                 return 0;
4096         return 1;
4097 }
4098
4099 /*
4100  * Read in the extents to if_extents.
4101  * All inode fields are set up by caller, we just traverse the btree
4102  * and copy the records in. If the file system cannot contain unwritten
4103  * extents, the records are checked for no "state" flags.
4104  */
4105 int                                     /* error */
4106 xfs_bmap_read_extents(
4107         xfs_trans_t             *tp,    /* transaction pointer */
4108         xfs_inode_t             *ip,    /* incore inode */
4109         int                     whichfork) /* data or attr fork */
4110 {
4111         struct xfs_btree_block  *block; /* current btree block */
4112         xfs_fsblock_t           bno;    /* block # of "block" */
4113         xfs_buf_t               *bp;    /* buffer for "block" */
4114         int                     error;  /* error return value */
4115         xfs_exntfmt_t           exntf;  /* XFS_EXTFMT_NOSTATE, if checking */
4116         xfs_extnum_t            i, j;   /* index into the extents list */
4117         xfs_ifork_t             *ifp;   /* fork structure */
4118         int                     level;  /* btree level, for checking */
4119         xfs_mount_t             *mp;    /* file system mount structure */
4120         __be64                  *pp;    /* pointer to block address */
4121         /* REFERENCED */
4122         xfs_extnum_t            room;   /* number of entries there's room for */
4123
4124         bno = NULLFSBLOCK;
4125         mp = ip->i_mount;
4126         ifp = XFS_IFORK_PTR(ip, whichfork);
4127         exntf = (whichfork != XFS_DATA_FORK) ? XFS_EXTFMT_NOSTATE :
4128                                         XFS_EXTFMT_INODE(ip);
4129         block = ifp->if_broot;
4130         /*
4131          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
4132          */
4133         level = be16_to_cpu(block->bb_level);
4134         ASSERT(level > 0);
4135         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
4136         bno = be64_to_cpu(*pp);
4137         ASSERT(bno != NULLDFSBNO);
4138         ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
4139         ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
4140         /*
4141          * Go down the tree until leaf level is reached, following the first
4142          * pointer (leftmost) at each level.
4143          */
4144         while (level-- > 0) {
4145                 if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
4146                                 XFS_BMAP_BTREE_REF)))
4147                         return error;
4148                 block = XFS_BUF_TO_BLOCK(bp);
4149                 XFS_WANT_CORRUPTED_GOTO(
4150                         xfs_bmap_sanity_check(mp, bp, level),
4151                         error0);
4152                 if (level == 0)
4153                         break;
4154                 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
4155                 bno = be64_to_cpu(*pp);
4156                 XFS_WANT_CORRUPTED_GOTO(XFS_FSB_SANITY_CHECK(mp, bno), error0);
4157                 xfs_trans_brelse(tp, bp);
4158         }
4159         /*
4160          * Here with bp and block set to the leftmost leaf node in the tree.
4161          */
4162         room = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4163         i = 0;
4164         /*
4165          * Loop over all leaf nodes.  Copy information to the extent records.
4166          */
4167         for (;;) {
4168                 xfs_bmbt_rec_t  *frp;
4169                 xfs_fsblock_t   nextbno;
4170                 xfs_extnum_t    num_recs;
4171                 xfs_extnum_t    start;
4172
4173
4174                 num_recs = xfs_btree_get_numrecs(block);
4175                 if (unlikely(i + num_recs > room)) {
4176                         ASSERT(i + num_recs <= room);
4177                         xfs_fs_repair_cmn_err(CE_WARN, ip->i_mount,
4178                                 "corrupt dinode %Lu, (btree extents).",
4179                                 (unsigned long long) ip->i_ino);
4180                         XFS_ERROR_REPORT("xfs_bmap_read_extents(1)",
4181                                          XFS_ERRLEVEL_LOW,
4182                                         ip->i_mount);
4183                         goto error0;
4184                 }
4185                 XFS_WANT_CORRUPTED_GOTO(
4186                         xfs_bmap_sanity_check(mp, bp, 0),
4187                         error0);
4188                 /*
4189                  * Read-ahead the next leaf block, if any.
4190                  */
4191                 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
4192                 if (nextbno != NULLFSBLOCK)
4193                         xfs_btree_reada_bufl(mp, nextbno, 1);
4194                 /*
4195                  * Copy records into the extent records.
4196                  */
4197                 frp = XFS_BMBT_REC_ADDR(mp, block, 1);
4198                 start = i;
4199                 for (j = 0; j < num_recs; j++, i++, frp++) {
4200                         xfs_bmbt_rec_host_t *trp = xfs_iext_get_ext(ifp, i);
4201                         trp->l0 = be64_to_cpu(frp->l0);
4202                         trp->l1 = be64_to_cpu(frp->l1);
4203                 }
4204                 if (exntf == XFS_EXTFMT_NOSTATE) {
4205                         /*
4206                          * Check all attribute bmap btree records and
4207                          * any "older" data bmap btree records for a
4208                          * set bit in the "extent flag" position.
4209                          */
4210                         if (unlikely(xfs_check_nostate_extents(ifp,
4211                                         start, num_recs))) {
4212                                 XFS_ERROR_REPORT("xfs_bmap_read_extents(2)",
4213                                                  XFS_ERRLEVEL_LOW,
4214                                                  ip->i_mount);
4215                                 goto error0;
4216                         }
4217                 }
4218                 xfs_trans_brelse(tp, bp);
4219                 bno = nextbno;
4220                 /*
4221                  * If we've reached the end, stop.
4222                  */
4223                 if (bno == NULLFSBLOCK)
4224                         break;
4225                 if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
4226                                 XFS_BMAP_BTREE_REF)))
4227                         return error;
4228                 block = XFS_BUF_TO_BLOCK(bp);
4229         }
4230         ASSERT(i == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)));
4231         ASSERT(i == XFS_IFORK_NEXTENTS(ip, whichfork));
4232         XFS_BMAP_TRACE_EXLIST(ip, i, whichfork);
4233         return 0;
4234 error0:
4235         xfs_trans_brelse(tp, bp);
4236         return XFS_ERROR(EFSCORRUPTED);
4237 }
4238
4239 #ifdef DEBUG
4240 /*
4241  * Add bmap trace insert entries for all the contents of the extent records.
4242  */
4243 void
4244 xfs_bmap_trace_exlist(
4245         xfs_inode_t     *ip,            /* incore inode pointer */
4246         xfs_extnum_t    cnt,            /* count of entries in the list */
4247         int             whichfork,      /* data or attr fork */
4248         unsigned long   caller_ip)
4249 {
4250         xfs_extnum_t    idx;            /* extent record index */
4251         xfs_ifork_t     *ifp;           /* inode fork pointer */
4252         int             state = 0;
4253
4254         if (whichfork == XFS_ATTR_FORK)
4255                 state |= BMAP_ATTRFORK;
4256
4257         ifp = XFS_IFORK_PTR(ip, whichfork);
4258         ASSERT(cnt == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)));
4259         for (idx = 0; idx < cnt; idx++)
4260                 trace_xfs_extlist(ip, idx, whichfork, caller_ip);
4261 }
4262
4263 /*
4264  * Validate that the bmbt_irecs being returned from bmapi are valid
4265  * given the callers original parameters.  Specifically check the
4266  * ranges of the returned irecs to ensure that they only extent beyond
4267  * the given parameters if the XFS_BMAPI_ENTIRE flag was set.
4268  */
4269 STATIC void
4270 xfs_bmap_validate_ret(
4271         xfs_fileoff_t           bno,
4272         xfs_filblks_t           len,
4273         int                     flags,
4274         xfs_bmbt_irec_t         *mval,
4275         int                     nmap,
4276         int                     ret_nmap)
4277 {
4278         int                     i;              /* index to map values */
4279
4280         ASSERT(ret_nmap <= nmap);
4281
4282         for (i = 0; i < ret_nmap; i++) {
4283                 ASSERT(mval[i].br_blockcount > 0);
4284                 if (!(flags & XFS_BMAPI_ENTIRE)) {
4285                         ASSERT(mval[i].br_startoff >= bno);
4286                         ASSERT(mval[i].br_blockcount <= len);
4287                         ASSERT(mval[i].br_startoff + mval[i].br_blockcount <=
4288                                bno + len);
4289                 } else {
4290                         ASSERT(mval[i].br_startoff < bno + len);
4291                         ASSERT(mval[i].br_startoff + mval[i].br_blockcount >
4292                                bno);
4293                 }
4294                 ASSERT(i == 0 ||
4295                        mval[i - 1].br_startoff + mval[i - 1].br_blockcount ==
4296                        mval[i].br_startoff);
4297                 if ((flags & XFS_BMAPI_WRITE) && !(flags & XFS_BMAPI_DELAY))
4298                         ASSERT(mval[i].br_startblock != DELAYSTARTBLOCK &&
4299                                mval[i].br_startblock != HOLESTARTBLOCK);
4300                 ASSERT(mval[i].br_state == XFS_EXT_NORM ||
4301                        mval[i].br_state == XFS_EXT_UNWRITTEN);
4302         }
4303 }
4304 #endif /* DEBUG */
4305
4306
4307 /*
4308  * Map file blocks to filesystem blocks.
4309  * File range is given by the bno/len pair.
4310  * Adds blocks to file if a write ("flags & XFS_BMAPI_WRITE" set)
4311  * into a hole or past eof.
4312  * Only allocates blocks from a single allocation group,
4313  * to avoid locking problems.
4314  * The returned value in "firstblock" from the first call in a transaction
4315  * must be remembered and presented to subsequent calls in "firstblock".
4316  * An upper bound for the number of blocks to be allocated is supplied to
4317  * the first call in "total"; if no allocation group has that many free
4318  * blocks then the call will fail (return NULLFSBLOCK in "firstblock").
4319  */
4320 int                                     /* error */
4321 xfs_bmapi(
4322         xfs_trans_t     *tp,            /* transaction pointer */
4323         xfs_inode_t     *ip,            /* incore inode */
4324         xfs_fileoff_t   bno,            /* starting file offs. mapped */
4325         xfs_filblks_t   len,            /* length to map in file */
4326         int             flags,          /* XFS_BMAPI_... */
4327         xfs_fsblock_t   *firstblock,    /* first allocated block
4328                                            controls a.g. for allocs */
4329         xfs_extlen_t    total,          /* total blocks needed */
4330         xfs_bmbt_irec_t *mval,          /* output: map values */
4331         int             *nmap,          /* i/o: mval size/count */
4332         xfs_bmap_free_t *flist)         /* i/o: list extents to free */
4333 {
4334         xfs_fsblock_t   abno;           /* allocated block number */
4335         xfs_extlen_t    alen;           /* allocated extent length */
4336         xfs_fileoff_t   aoff;           /* allocated file offset */
4337         xfs_bmalloca_t  bma = { 0 };    /* args for xfs_bmap_alloc */
4338         xfs_btree_cur_t *cur;           /* bmap btree cursor */
4339         xfs_fileoff_t   end;            /* end of mapped file region */
4340         int             eof;            /* we've hit the end of extents */
4341         xfs_bmbt_rec_host_t *ep;        /* extent record pointer */
4342         int             error;          /* error return */
4343         xfs_bmbt_irec_t got;            /* current file extent record */
4344         xfs_ifork_t     *ifp;           /* inode fork pointer */
4345         xfs_extlen_t    indlen;         /* indirect blocks length */
4346         xfs_extnum_t    lastx;          /* last useful extent number */
4347         int             logflags;       /* flags for transaction logging */
4348         xfs_extlen_t    minleft;        /* min blocks left after allocation */
4349         xfs_extlen_t    minlen;         /* min allocation size */
4350         xfs_mount_t     *mp;            /* xfs mount structure */
4351         int             n;              /* current extent index */
4352         int             nallocs;        /* number of extents alloc'd */
4353         xfs_extnum_t    nextents;       /* number of extents in file */
4354         xfs_fileoff_t   obno;           /* old block number (offset) */
4355         xfs_bmbt_irec_t prev;           /* previous file extent record */
4356         int             tmp_logflags;   /* temp flags holder */
4357         int             whichfork;      /* data or attr fork */
4358         char            inhole;         /* current location is hole in file */
4359         char            wasdelay;       /* old extent was delayed */
4360         char            wr;             /* this is a write request */
4361         char            rt;             /* this is a realtime file */
4362 #ifdef DEBUG
4363         xfs_fileoff_t   orig_bno;       /* original block number value */
4364         int             orig_flags;     /* original flags arg value */
4365         xfs_filblks_t   orig_len;       /* original value of len arg */
4366         xfs_bmbt_irec_t *orig_mval;     /* original value of mval */
4367         int             orig_nmap;      /* original value of *nmap */
4368
4369         orig_bno = bno;
4370         orig_len = len;
4371         orig_flags = flags;
4372         orig_mval = mval;
4373         orig_nmap = *nmap;
4374 #endif
4375         ASSERT(*nmap >= 1);
4376         ASSERT(*nmap <= XFS_BMAP_MAX_NMAP || !(flags & XFS_BMAPI_WRITE));
4377         whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4378                 XFS_ATTR_FORK : XFS_DATA_FORK;
4379         mp = ip->i_mount;
4380         if (unlikely(XFS_TEST_ERROR(
4381             (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4382              XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
4383              XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL),
4384              mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4385                 XFS_ERROR_REPORT("xfs_bmapi", XFS_ERRLEVEL_LOW, mp);
4386                 return XFS_ERROR(EFSCORRUPTED);
4387         }
4388         if (XFS_FORCED_SHUTDOWN(mp))
4389                 return XFS_ERROR(EIO);
4390         rt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
4391         ifp = XFS_IFORK_PTR(ip, whichfork);
4392         ASSERT(ifp->if_ext_max ==
4393                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
4394         if ((wr = (flags & XFS_BMAPI_WRITE)) != 0)
4395                 XFS_STATS_INC(xs_blk_mapw);
4396         else
4397                 XFS_STATS_INC(xs_blk_mapr);
4398         /*
4399          * IGSTATE flag is used to combine extents which
4400          * differ only due to the state of the extents.
4401          * This technique is used from xfs_getbmap()
4402          * when the caller does not wish to see the
4403          * separation (which is the default).
4404          *
4405          * This technique is also used when writing a
4406          * buffer which has been partially written,
4407          * (usually by being flushed during a chunkread),
4408          * to ensure one write takes place. This also
4409          * prevents a change in the xfs inode extents at
4410          * this time, intentionally. This change occurs
4411          * on completion of the write operation, in
4412          * xfs_strat_comp(), where the xfs_bmapi() call
4413          * is transactioned, and the extents combined.
4414          */
4415         if ((flags & XFS_BMAPI_IGSTATE) && wr)  /* if writing unwritten space */
4416                 wr = 0;                         /* no allocations are allowed */
4417         ASSERT(wr || !(flags & XFS_BMAPI_DELAY));
4418         logflags = 0;
4419         nallocs = 0;
4420         cur = NULL;
4421         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
4422                 ASSERT(wr && tp);
4423                 if ((error = xfs_bmap_local_to_extents(tp, ip,
4424                                 firstblock, total, &logflags, whichfork)))
4425                         goto error0;
4426         }
4427         if (wr && *firstblock == NULLFSBLOCK) {
4428                 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE)
4429                         minleft = be16_to_cpu(ifp->if_broot->bb_level) + 1;
4430                 else
4431                         minleft = 1;
4432         } else
4433                 minleft = 0;
4434         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
4435             (error = xfs_iread_extents(tp, ip, whichfork)))
4436                 goto error0;
4437         ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
4438                 &prev);
4439         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4440         n = 0;
4441         end = bno + len;
4442         obno = bno;
4443         bma.ip = NULL;
4444
4445         while (bno < end && n < *nmap) {
4446                 /*
4447                  * Reading past eof, act as though there's a hole
4448                  * up to end.
4449                  */
4450                 if (eof && !wr)
4451                         got.br_startoff = end;
4452                 inhole = eof || got.br_startoff > bno;
4453                 wasdelay = wr && !inhole && !(flags & XFS_BMAPI_DELAY) &&
4454                         isnullstartblock(got.br_startblock);
4455                 /*
4456                  * First, deal with the hole before the allocated space
4457                  * that we found, if any.
4458                  */
4459                 if (wr && (inhole || wasdelay)) {
4460                         /*
4461                          * For the wasdelay case, we could also just
4462                          * allocate the stuff asked for in this bmap call
4463                          * but that wouldn't be as good.
4464                          */
4465                         if (wasdelay) {
4466                                 alen = (xfs_extlen_t)got.br_blockcount;
4467                                 aoff = got.br_startoff;
4468                                 if (lastx != NULLEXTNUM && lastx) {
4469                                         ep = xfs_iext_get_ext(ifp, lastx - 1);
4470                                         xfs_bmbt_get_all(ep, &prev);
4471                                 }
4472                         } else {
4473                                 alen = (xfs_extlen_t)
4474                                         XFS_FILBLKS_MIN(len, MAXEXTLEN);
4475                                 if (!eof)
4476                                         alen = (xfs_extlen_t)
4477                                                 XFS_FILBLKS_MIN(alen,
4478                                                         got.br_startoff - bno);
4479                                 aoff = bno;
4480                         }
4481                         minlen = (flags & XFS_BMAPI_CONTIG) ? alen : 1;
4482                         if (flags & XFS_BMAPI_DELAY) {
4483                                 xfs_extlen_t    extsz;
4484
4485                                 /* Figure out the extent size, adjust alen */
4486                                 extsz = xfs_get_extsz_hint(ip);
4487                                 if (extsz) {
4488                                         error = xfs_bmap_extsize_align(mp,
4489                                                         &got, &prev, extsz,
4490                                                         rt, eof,
4491                                                         flags&XFS_BMAPI_DELAY,
4492                                                         flags&XFS_BMAPI_CONVERT,
4493                                                         &aoff, &alen);
4494                                         ASSERT(!error);
4495                                 }
4496
4497                                 if (rt)
4498                                         extsz = alen / mp->m_sb.sb_rextsize;
4499
4500                                 /*
4501                                  * Make a transaction-less quota reservation for
4502                                  * delayed allocation blocks. This number gets
4503                                  * adjusted later.  We return if we haven't
4504                                  * allocated blocks already inside this loop.
4505                                  */
4506                                 error = xfs_trans_reserve_quota_nblks(
4507                                                 NULL, ip, (long)alen, 0,
4508                                                 rt ? XFS_QMOPT_RES_RTBLKS :
4509                                                      XFS_QMOPT_RES_REGBLKS);
4510                                 if (error) {
4511                                         if (n == 0) {
4512                                                 *nmap = 0;
4513                                                 ASSERT(cur == NULL);
4514                                                 return error;
4515                                         }
4516                                         break;
4517                                 }
4518
4519                                 /*
4520                                  * Split changing sb for alen and indlen since
4521                                  * they could be coming from different places.
4522                                  */
4523                                 indlen = (xfs_extlen_t)
4524                                         xfs_bmap_worst_indlen(ip, alen);
4525                                 ASSERT(indlen > 0);
4526
4527                                 if (rt) {
4528                                         error = xfs_mod_incore_sb(mp,
4529                                                         XFS_SBS_FREXTENTS,
4530                                                         -((int64_t)extsz), (flags &
4531                                                         XFS_BMAPI_RSVBLOCKS));
4532                                 } else {
4533                                         error = xfs_icsb_modify_counters(mp,
4534                                                         XFS_SBS_FDBLOCKS,
4535                                                         -((int64_t)alen), (flags &
4536                                                         XFS_BMAPI_RSVBLOCKS));
4537                                 }
4538                                 if (!error) {
4539                                         error = xfs_icsb_modify_counters(mp,
4540                                                         XFS_SBS_FDBLOCKS,
4541                                                         -((int64_t)indlen), (flags &
4542                                                         XFS_BMAPI_RSVBLOCKS));
4543                                         if (error && rt)
4544                                                 xfs_mod_incore_sb(mp,
4545                                                         XFS_SBS_FREXTENTS,
4546                                                         (int64_t)extsz, (flags &
4547                                                         XFS_BMAPI_RSVBLOCKS));
4548                                         else if (error)
4549                                                 xfs_icsb_modify_counters(mp,
4550                                                         XFS_SBS_FDBLOCKS,
4551                                                         (int64_t)alen, (flags &
4552                                                         XFS_BMAPI_RSVBLOCKS));
4553                                 }
4554
4555                                 if (error) {
4556                                         if (XFS_IS_QUOTA_ON(mp))
4557                                                 /* unreserve the blocks now */
4558                                                 (void)
4559                                                 xfs_trans_unreserve_quota_nblks(
4560                                                         NULL, ip,
4561                                                         (long)alen, 0, rt ?
4562                                                         XFS_QMOPT_RES_RTBLKS :
4563                                                         XFS_QMOPT_RES_REGBLKS);
4564                                         break;
4565                                 }
4566
4567                                 ip->i_delayed_blks += alen;
4568                                 abno = nullstartblock(indlen);
4569                         } else {
4570                                 /*
4571                                  * If first time, allocate and fill in
4572                                  * once-only bma fields.
4573                                  */
4574                                 if (bma.ip == NULL) {
4575                                         bma.tp = tp;
4576                                         bma.ip = ip;
4577                                         bma.prevp = &prev;
4578                                         bma.gotp = &got;
4579                                         bma.total = total;
4580                                         bma.userdata = 0;
4581                                 }
4582                                 /* Indicate if this is the first user data
4583                                  * in the file, or just any user data.
4584                                  */
4585                                 if (!(flags & XFS_BMAPI_METADATA)) {
4586                                         bma.userdata = (aoff == 0) ?
4587                                                 XFS_ALLOC_INITIAL_USER_DATA :
4588                                                 XFS_ALLOC_USERDATA;
4589                                 }
4590                                 /*
4591                                  * Fill in changeable bma fields.
4592                                  */
4593                                 bma.eof = eof;
4594                                 bma.firstblock = *firstblock;
4595                                 bma.alen = alen;
4596                                 bma.off = aoff;
4597                                 bma.conv = !!(flags & XFS_BMAPI_CONVERT);
4598                                 bma.wasdel = wasdelay;
4599                                 bma.minlen = minlen;
4600                                 bma.low = flist->xbf_low;
4601                                 bma.minleft = minleft;
4602                                 /*
4603                                  * Only want to do the alignment at the
4604                                  * eof if it is userdata and allocation length
4605                                  * is larger than a stripe unit.
4606                                  */
4607                                 if (mp->m_dalign && alen >= mp->m_dalign &&
4608                                     (!(flags & XFS_BMAPI_METADATA)) &&
4609                                     (whichfork == XFS_DATA_FORK)) {
4610                                         if ((error = xfs_bmap_isaeof(ip, aoff,
4611                                                         whichfork, &bma.aeof)))
4612                                                 goto error0;
4613                                 } else
4614                                         bma.aeof = 0;
4615                                 /*
4616                                  * Call allocator.
4617                                  */
4618                                 if ((error = xfs_bmap_alloc(&bma)))
4619                                         goto error0;
4620                                 /*
4621                                  * Copy out result fields.
4622                                  */
4623                                 abno = bma.rval;
4624                                 if ((flist->xbf_low = bma.low))
4625                                         minleft = 0;
4626                                 alen = bma.alen;
4627                                 aoff = bma.off;
4628                                 ASSERT(*firstblock == NULLFSBLOCK ||
4629                                        XFS_FSB_TO_AGNO(mp, *firstblock) ==
4630                                        XFS_FSB_TO_AGNO(mp, bma.firstblock) ||
4631                                        (flist->xbf_low &&
4632                                         XFS_FSB_TO_AGNO(mp, *firstblock) <
4633                                         XFS_FSB_TO_AGNO(mp, bma.firstblock)));
4634                                 *firstblock = bma.firstblock;
4635                                 if (cur)
4636                                         cur->bc_private.b.firstblock =
4637                                                 *firstblock;
4638                                 if (abno == NULLFSBLOCK)
4639                                         break;
4640                                 if ((ifp->if_flags & XFS_IFBROOT) && !cur) {
4641                                         cur = xfs_bmbt_init_cursor(mp, tp,
4642                                                 ip, whichfork);
4643                                         cur->bc_private.b.firstblock =
4644                                                 *firstblock;
4645                                         cur->bc_private.b.flist = flist;
4646                                 }
4647                                 /*
4648                                  * Bump the number of extents we've allocated
4649                                  * in this call.
4650                                  */
4651                                 nallocs++;
4652                         }
4653                         if (cur)
4654                                 cur->bc_private.b.flags =
4655                                         wasdelay ? XFS_BTCUR_BPRV_WASDEL : 0;
4656                         got.br_startoff = aoff;
4657                         got.br_startblock = abno;
4658                         got.br_blockcount = alen;
4659                         got.br_state = XFS_EXT_NORM;    /* assume normal */
4660                         /*
4661                          * Determine state of extent, and the filesystem.
4662                          * A wasdelay extent has been initialized, so
4663                          * shouldn't be flagged as unwritten.
4664                          */
4665                         if (wr && xfs_sb_version_hasextflgbit(&mp->m_sb)) {
4666                                 if (!wasdelay && (flags & XFS_BMAPI_PREALLOC))
4667                                         got.br_state = XFS_EXT_UNWRITTEN;
4668                         }
4669                         error = xfs_bmap_add_extent(ip, lastx, &cur, &got,
4670                                 firstblock, flist, &tmp_logflags,
4671                                 whichfork, (flags & XFS_BMAPI_RSVBLOCKS));
4672                         logflags |= tmp_logflags;
4673                         if (error)
4674                                 goto error0;
4675                         lastx = ifp->if_lastex;
4676                         ep = xfs_iext_get_ext(ifp, lastx);
4677                         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4678                         xfs_bmbt_get_all(ep, &got);
4679                         ASSERT(got.br_startoff <= aoff);
4680                         ASSERT(got.br_startoff + got.br_blockcount >=
4681                                 aoff + alen);
4682 #ifdef DEBUG
4683                         if (flags & XFS_BMAPI_DELAY) {
4684                                 ASSERT(isnullstartblock(got.br_startblock));
4685                                 ASSERT(startblockval(got.br_startblock) > 0);
4686                         }
4687                         ASSERT(got.br_state == XFS_EXT_NORM ||
4688                                got.br_state == XFS_EXT_UNWRITTEN);
4689 #endif
4690                         /*
4691                          * Fall down into the found allocated space case.
4692                          */
4693                 } else if (inhole) {
4694                         /*
4695                          * Reading in a hole.
4696                          */
4697                         mval->br_startoff = bno;
4698                         mval->br_startblock = HOLESTARTBLOCK;
4699                         mval->br_blockcount =
4700                                 XFS_FILBLKS_MIN(len, got.br_startoff - bno);
4701                         mval->br_state = XFS_EXT_NORM;
4702                         bno += mval->br_blockcount;
4703                         len -= mval->br_blockcount;
4704                         mval++;
4705                         n++;
4706                         continue;
4707                 }
4708                 /*
4709                  * Then deal with the allocated space we found.
4710                  */
4711                 ASSERT(ep != NULL);
4712                 if (!(flags & XFS_BMAPI_ENTIRE) &&
4713                     (got.br_startoff + got.br_blockcount > obno)) {
4714                         if (obno > bno)
4715                                 bno = obno;
4716                         ASSERT((bno >= obno) || (n == 0));
4717                         ASSERT(bno < end);
4718                         mval->br_startoff = bno;
4719                         if (isnullstartblock(got.br_startblock)) {
4720                                 ASSERT(!wr || (flags & XFS_BMAPI_DELAY));
4721                                 mval->br_startblock = DELAYSTARTBLOCK;
4722                         } else
4723                                 mval->br_startblock =
4724                                         got.br_startblock +
4725                                         (bno - got.br_startoff);
4726                         /*
4727                          * Return the minimum of what we got and what we
4728                          * asked for for the length.  We can use the len
4729                          * variable here because it is modified below
4730                          * and we could have been there before coming
4731                          * here if the first part of the allocation
4732                          * didn't overlap what was asked for.
4733                          */
4734                         mval->br_blockcount =
4735                                 XFS_FILBLKS_MIN(end - bno, got.br_blockcount -
4736                                         (bno - got.br_startoff));
4737                         mval->br_state = got.br_state;
4738                         ASSERT(mval->br_blockcount <= len);
4739                 } else {
4740                         *mval = got;
4741                         if (isnullstartblock(mval->br_startblock)) {
4742                                 ASSERT(!wr || (flags & XFS_BMAPI_DELAY));
4743                                 mval->br_startblock = DELAYSTARTBLOCK;
4744                         }
4745                 }
4746
4747                 /*
4748                  * Check if writing previously allocated but
4749                  * unwritten extents.
4750                  */
4751                 if (wr &&
4752                     ((mval->br_state == XFS_EXT_UNWRITTEN &&
4753                       ((flags & (XFS_BMAPI_PREALLOC|XFS_BMAPI_DELAY)) == 0)) ||
4754                      (mval->br_state == XFS_EXT_NORM &&
4755                       ((flags & (XFS_BMAPI_PREALLOC|XFS_BMAPI_CONVERT)) ==
4756                                 (XFS_BMAPI_PREALLOC|XFS_BMAPI_CONVERT))))) {
4757                         /*
4758                          * Modify (by adding) the state flag, if writing.
4759                          */
4760                         ASSERT(mval->br_blockcount <= len);
4761                         if ((ifp->if_flags & XFS_IFBROOT) && !cur) {
4762                                 cur = xfs_bmbt_init_cursor(mp,
4763                                         tp, ip, whichfork);
4764                                 cur->bc_private.b.firstblock =
4765                                         *firstblock;
4766                                 cur->bc_private.b.flist = flist;
4767                         }
4768                         mval->br_state = (mval->br_state == XFS_EXT_UNWRITTEN)
4769                                                 ? XFS_EXT_NORM
4770                                                 : XFS_EXT_UNWRITTEN;
4771                         error = xfs_bmap_add_extent(ip, lastx, &cur, mval,
4772                                 firstblock, flist, &tmp_logflags,
4773                                 whichfork, (flags & XFS_BMAPI_RSVBLOCKS));
4774                         logflags |= tmp_logflags;
4775                         if (error)
4776                                 goto error0;
4777                         lastx = ifp->if_lastex;
4778                         ep = xfs_iext_get_ext(ifp, lastx);
4779                         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4780                         xfs_bmbt_get_all(ep, &got);
4781                         /*
4782                          * We may have combined previously unwritten
4783                          * space with written space, so generate
4784                          * another request.
4785                          */
4786                         if (mval->br_blockcount < len)
4787                                 continue;
4788                 }
4789
4790                 ASSERT((flags & XFS_BMAPI_ENTIRE) ||
4791                        ((mval->br_startoff + mval->br_blockcount) <= end));
4792                 ASSERT((flags & XFS_BMAPI_ENTIRE) ||
4793                        (mval->br_blockcount <= len) ||
4794                        (mval->br_startoff < obno));
4795                 bno = mval->br_startoff + mval->br_blockcount;
4796                 len = end - bno;
4797                 if (n > 0 && mval->br_startoff == mval[-1].br_startoff) {
4798                         ASSERT(mval->br_startblock == mval[-1].br_startblock);
4799                         ASSERT(mval->br_blockcount > mval[-1].br_blockcount);
4800                         ASSERT(mval->br_state == mval[-1].br_state);
4801                         mval[-1].br_blockcount = mval->br_blockcount;
4802                         mval[-1].br_state = mval->br_state;
4803                 } else if (n > 0 && mval->br_startblock != DELAYSTARTBLOCK &&
4804                            mval[-1].br_startblock != DELAYSTARTBLOCK &&
4805                            mval[-1].br_startblock != HOLESTARTBLOCK &&
4806                            mval->br_startblock ==
4807                            mval[-1].br_startblock + mval[-1].br_blockcount &&
4808                            ((flags & XFS_BMAPI_IGSTATE) ||
4809                                 mval[-1].br_state == mval->br_state)) {
4810                         ASSERT(mval->br_startoff ==
4811                                mval[-1].br_startoff + mval[-1].br_blockcount);
4812                         mval[-1].br_blockcount += mval->br_blockcount;
4813                 } else if (n > 0 &&
4814                            mval->br_startblock == DELAYSTARTBLOCK &&
4815                            mval[-1].br_startblock == DELAYSTARTBLOCK &&
4816                            mval->br_startoff ==
4817                            mval[-1].br_startoff + mval[-1].br_blockcount) {
4818                         mval[-1].br_blockcount += mval->br_blockcount;
4819                         mval[-1].br_state = mval->br_state;
4820                 } else if (!((n == 0) &&
4821                              ((mval->br_startoff + mval->br_blockcount) <=
4822                               obno))) {
4823                         mval++;
4824                         n++;
4825                 }
4826                 /*
4827                  * If we're done, stop now.  Stop when we've allocated
4828                  * XFS_BMAP_MAX_NMAP extents no matter what.  Otherwise
4829                  * the transaction may get too big.
4830                  */
4831                 if (bno >= end || n >= *nmap || nallocs >= *nmap)
4832                         break;
4833                 /*
4834                  * Else go on to the next record.
4835                  */
4836                 ep = xfs_iext_get_ext(ifp, ++lastx);
4837                 prev = got;
4838                 if (lastx >= nextents)
4839                         eof = 1;
4840                 else
4841                         xfs_bmbt_get_all(ep, &got);
4842         }
4843         ifp->if_lastex = lastx;
4844         *nmap = n;
4845         /*
4846          * Transform from btree to extents, give it cur.
4847          */
4848         if (tp && XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE &&
4849             XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max) {
4850                 ASSERT(wr && cur);
4851                 error = xfs_bmap_btree_to_extents(tp, ip, cur,
4852                         &tmp_logflags, whichfork);
4853                 logflags |= tmp_logflags;
4854                 if (error)
4855                         goto error0;
4856         }
4857         ASSERT(ifp->if_ext_max ==
4858                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
4859         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE ||
4860                XFS_IFORK_NEXTENTS(ip, whichfork) > ifp->if_ext_max);
4861         error = 0;
4862 error0:
4863         /*
4864          * Log everything.  Do this after conversion, there's no point in
4865          * logging the extent records if we've converted to btree format.
4866          */
4867         if ((logflags & xfs_ilog_fext(whichfork)) &&
4868             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
4869                 logflags &= ~xfs_ilog_fext(whichfork);
4870         else if ((logflags & xfs_ilog_fbroot(whichfork)) &&
4871                  XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
4872                 logflags &= ~xfs_ilog_fbroot(whichfork);
4873         /*
4874          * Log whatever the flags say, even if error.  Otherwise we might miss
4875          * detecting a case where the data is changed, there's an error,
4876          * and it's not logged so we don't shutdown when we should.
4877          */
4878         if (logflags) {
4879                 ASSERT(tp && wr);
4880                 xfs_trans_log_inode(tp, ip, logflags);
4881         }
4882         if (cur) {
4883                 if (!error) {
4884                         ASSERT(*firstblock == NULLFSBLOCK ||
4885                                XFS_FSB_TO_AGNO(mp, *firstblock) ==
4886                                XFS_FSB_TO_AGNO(mp,
4887                                        cur->bc_private.b.firstblock) ||
4888                                (flist->xbf_low &&
4889                                 XFS_FSB_TO_AGNO(mp, *firstblock) <
4890                                 XFS_FSB_TO_AGNO(mp,
4891                                         cur->bc_private.b.firstblock)));
4892                         *firstblock = cur->bc_private.b.firstblock;
4893                 }
4894                 xfs_btree_del_cursor(cur,
4895                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
4896         }
4897         if (!error)
4898                 xfs_bmap_validate_ret(orig_bno, orig_len, orig_flags, orig_mval,
4899                         orig_nmap, *nmap);
4900         return error;
4901 }
4902
4903 /*
4904  * Map file blocks to filesystem blocks, simple version.
4905  * One block (extent) only, read-only.
4906  * For flags, only the XFS_BMAPI_ATTRFORK flag is examined.
4907  * For the other flag values, the effect is as if XFS_BMAPI_METADATA
4908  * was set and all the others were clear.
4909  */
4910 int                                             /* error */
4911 xfs_bmapi_single(
4912         xfs_trans_t     *tp,            /* transaction pointer */
4913         xfs_inode_t     *ip,            /* incore inode */
4914         int             whichfork,      /* data or attr fork */
4915         xfs_fsblock_t   *fsb,           /* output: mapped block */
4916         xfs_fileoff_t   bno)            /* starting file offs. mapped */
4917 {
4918         int             eof;            /* we've hit the end of extents */
4919         int             error;          /* error return */
4920         xfs_bmbt_irec_t got;            /* current file extent record */
4921         xfs_ifork_t     *ifp;           /* inode fork pointer */
4922         xfs_extnum_t    lastx;          /* last useful extent number */
4923         xfs_bmbt_irec_t prev;           /* previous file extent record */
4924
4925         ifp = XFS_IFORK_PTR(ip, whichfork);
4926         if (unlikely(
4927             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
4928             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)) {
4929                XFS_ERROR_REPORT("xfs_bmapi_single", XFS_ERRLEVEL_LOW,
4930                                 ip->i_mount);
4931                return XFS_ERROR(EFSCORRUPTED);
4932         }
4933         if (XFS_FORCED_SHUTDOWN(ip->i_mount))
4934                 return XFS_ERROR(EIO);
4935         XFS_STATS_INC(xs_blk_mapr);
4936         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
4937             (error = xfs_iread_extents(tp, ip, whichfork)))
4938                 return error;
4939         (void)xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
4940                 &prev);
4941         /*
4942          * Reading past eof, act as though there's a hole
4943          * up to end.
4944          */
4945         if (eof || got.br_startoff > bno) {
4946                 *fsb = NULLFSBLOCK;
4947                 return 0;
4948         }
4949         ASSERT(!isnullstartblock(got.br_startblock));
4950         ASSERT(bno < got.br_startoff + got.br_blockcount);
4951         *fsb = got.br_startblock + (bno - got.br_startoff);
4952         ifp->if_lastex = lastx;
4953         return 0;
4954 }
4955
4956 /*
4957  * Unmap (remove) blocks from a file.
4958  * If nexts is nonzero then the number of extents to remove is limited to
4959  * that value.  If not all extents in the block range can be removed then
4960  * *done is set.
4961  */
4962 int                                             /* error */
4963 xfs_bunmapi(
4964         xfs_trans_t             *tp,            /* transaction pointer */
4965         struct xfs_inode        *ip,            /* incore inode */
4966         xfs_fileoff_t           bno,            /* starting offset to unmap */
4967         xfs_filblks_t           len,            /* length to unmap in file */
4968         int                     flags,          /* misc flags */
4969         xfs_extnum_t            nexts,          /* number of extents max */
4970         xfs_fsblock_t           *firstblock,    /* first allocated block
4971                                                    controls a.g. for allocs */
4972         xfs_bmap_free_t         *flist,         /* i/o: list extents to free */
4973         int                     *done)          /* set if not done yet */
4974 {
4975         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
4976         xfs_bmbt_irec_t         del;            /* extent being deleted */
4977         int                     eof;            /* is deleting at eof */
4978         xfs_bmbt_rec_host_t     *ep;            /* extent record pointer */
4979         int                     error;          /* error return value */
4980         xfs_extnum_t            extno;          /* extent number in list */
4981         xfs_bmbt_irec_t         got;            /* current extent record */
4982         xfs_ifork_t             *ifp;           /* inode fork pointer */
4983         int                     isrt;           /* freeing in rt area */
4984         xfs_extnum_t            lastx;          /* last extent index used */
4985         int                     logflags;       /* transaction logging flags */
4986         xfs_extlen_t            mod;            /* rt extent offset */
4987         xfs_mount_t             *mp;            /* mount structure */
4988         xfs_extnum_t            nextents;       /* number of file extents */
4989         xfs_bmbt_irec_t         prev;           /* previous extent record */
4990         xfs_fileoff_t           start;          /* first file offset deleted */
4991         int                     tmp_logflags;   /* partial logging flags */
4992         int                     wasdel;         /* was a delayed alloc extent */
4993         int                     whichfork;      /* data or attribute fork */
4994         int                     rsvd;           /* OK to allocate reserved blocks */
4995         xfs_fsblock_t           sum;
4996
4997         trace_xfs_bunmap(ip, bno, len, flags, _RET_IP_);
4998
4999         whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
5000                 XFS_ATTR_FORK : XFS_DATA_FORK;
5001         ifp = XFS_IFORK_PTR(ip, whichfork);
5002         if (unlikely(
5003             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
5004             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
5005                 XFS_ERROR_REPORT("xfs_bunmapi", XFS_ERRLEVEL_LOW,
5006                                  ip->i_mount);
5007                 return XFS_ERROR(EFSCORRUPTED);
5008         }
5009         mp = ip->i_mount;
5010         if (XFS_FORCED_SHUTDOWN(mp))
5011                 return XFS_ERROR(EIO);
5012         rsvd = (flags & XFS_BMAPI_RSVBLOCKS) != 0;
5013         ASSERT(len > 0);
5014         ASSERT(nexts >= 0);
5015         ASSERT(ifp->if_ext_max ==
5016                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5017         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5018             (error = xfs_iread_extents(tp, ip, whichfork)))
5019                 return error;
5020         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5021         if (nextents == 0) {
5022                 *done = 1;
5023                 return 0;
5024         }
5025         XFS_STATS_INC(xs_blk_unmap);
5026         isrt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
5027         start = bno;
5028         bno = start + len - 1;
5029         ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
5030                 &prev);
5031
5032         /*
5033          * Check to see if the given block number is past the end of the
5034          * file, back up to the last block if so...
5035          */
5036         if (eof) {
5037                 ep = xfs_iext_get_ext(ifp, --lastx);
5038                 xfs_bmbt_get_all(ep, &got);
5039                 bno = got.br_startoff + got.br_blockcount - 1;
5040         }
5041         logflags = 0;
5042         if (ifp->if_flags & XFS_IFBROOT) {
5043                 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
5044                 cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5045                 cur->bc_private.b.firstblock = *firstblock;
5046                 cur->bc_private.b.flist = flist;
5047                 cur->bc_private.b.flags = 0;
5048         } else
5049                 cur = NULL;
5050         extno = 0;
5051         while (bno != (xfs_fileoff_t)-1 && bno >= start && lastx >= 0 &&
5052                (nexts == 0 || extno < nexts)) {
5053                 /*
5054                  * Is the found extent after a hole in which bno lives?
5055                  * Just back up to the previous extent, if so.
5056                  */
5057                 if (got.br_startoff > bno) {
5058                         if (--lastx < 0)
5059                                 break;
5060                         ep = xfs_iext_get_ext(ifp, lastx);
5061                         xfs_bmbt_get_all(ep, &got);
5062                 }
5063                 /*
5064                  * Is the last block of this extent before the range
5065                  * we're supposed to delete?  If so, we're done.
5066                  */
5067                 bno = XFS_FILEOFF_MIN(bno,
5068                         got.br_startoff + got.br_blockcount - 1);
5069                 if (bno < start)
5070                         break;
5071                 /*
5072                  * Then deal with the (possibly delayed) allocated space
5073                  * we found.
5074                  */
5075                 ASSERT(ep != NULL);
5076                 del = got;
5077                 wasdel = isnullstartblock(del.br_startblock);
5078                 if (got.br_startoff < start) {
5079                         del.br_startoff = start;
5080                         del.br_blockcount -= start - got.br_startoff;
5081                         if (!wasdel)
5082                                 del.br_startblock += start - got.br_startoff;
5083                 }
5084                 if (del.br_startoff + del.br_blockcount > bno + 1)
5085                         del.br_blockcount = bno + 1 - del.br_startoff;
5086                 sum = del.br_startblock + del.br_blockcount;
5087                 if (isrt &&
5088                     (mod = do_mod(sum, mp->m_sb.sb_rextsize))) {
5089                         /*
5090                          * Realtime extent not lined up at the end.
5091                          * The extent could have been split into written
5092                          * and unwritten pieces, or we could just be
5093                          * unmapping part of it.  But we can't really
5094                          * get rid of part of a realtime extent.
5095                          */
5096                         if (del.br_state == XFS_EXT_UNWRITTEN ||
5097                             !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5098                                 /*
5099                                  * This piece is unwritten, or we're not
5100                                  * using unwritten extents.  Skip over it.
5101                                  */
5102                                 ASSERT(bno >= mod);
5103                                 bno -= mod > del.br_blockcount ?
5104                                         del.br_blockcount : mod;
5105                                 if (bno < got.br_startoff) {
5106                                         if (--lastx >= 0)
5107                                                 xfs_bmbt_get_all(xfs_iext_get_ext(
5108                                                         ifp, lastx), &got);
5109                                 }
5110                                 continue;
5111                         }
5112                         /*
5113                          * It's written, turn it unwritten.
5114                          * This is better than zeroing it.
5115                          */
5116                         ASSERT(del.br_state == XFS_EXT_NORM);
5117                         ASSERT(xfs_trans_get_block_res(tp) > 0);
5118                         /*
5119                          * If this spans a realtime extent boundary,
5120                          * chop it back to the start of the one we end at.
5121                          */
5122                         if (del.br_blockcount > mod) {
5123                                 del.br_startoff += del.br_blockcount - mod;
5124                                 del.br_startblock += del.br_blockcount - mod;
5125                                 del.br_blockcount = mod;
5126                         }
5127                         del.br_state = XFS_EXT_UNWRITTEN;
5128                         error = xfs_bmap_add_extent(ip, lastx, &cur, &del,
5129                                 firstblock, flist, &logflags,
5130                                 XFS_DATA_FORK, 0);
5131                         if (error)
5132                                 goto error0;
5133                         goto nodelete;
5134                 }
5135                 if (isrt && (mod = do_mod(del.br_startblock, mp->m_sb.sb_rextsize))) {
5136                         /*
5137                          * Realtime extent is lined up at the end but not
5138                          * at the front.  We'll get rid of full extents if
5139                          * we can.
5140                          */
5141                         mod = mp->m_sb.sb_rextsize - mod;
5142                         if (del.br_blockcount > mod) {
5143                                 del.br_blockcount -= mod;
5144                                 del.br_startoff += mod;
5145                                 del.br_startblock += mod;
5146                         } else if ((del.br_startoff == start &&
5147                                     (del.br_state == XFS_EXT_UNWRITTEN ||
5148                                      xfs_trans_get_block_res(tp) == 0)) ||
5149                                    !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5150                                 /*
5151                                  * Can't make it unwritten.  There isn't
5152                                  * a full extent here so just skip it.
5153                                  */
5154                                 ASSERT(bno >= del.br_blockcount);
5155                                 bno -= del.br_blockcount;
5156                                 if (bno < got.br_startoff) {
5157                                         if (--lastx >= 0)
5158                                                 xfs_bmbt_get_all(--ep, &got);
5159                                 }
5160                                 continue;
5161                         } else if (del.br_state == XFS_EXT_UNWRITTEN) {
5162                                 /*
5163                                  * This one is already unwritten.
5164                                  * It must have a written left neighbor.
5165                                  * Unwrite the killed part of that one and
5166                                  * try again.
5167                                  */
5168                                 ASSERT(lastx > 0);
5169                                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp,
5170                                                 lastx - 1), &prev);
5171                                 ASSERT(prev.br_state == XFS_EXT_NORM);
5172                                 ASSERT(!isnullstartblock(prev.br_startblock));
5173                                 ASSERT(del.br_startblock ==
5174                                        prev.br_startblock + prev.br_blockcount);
5175                                 if (prev.br_startoff < start) {
5176                                         mod = start - prev.br_startoff;
5177                                         prev.br_blockcount -= mod;
5178                                         prev.br_startblock += mod;
5179                                         prev.br_startoff = start;
5180                                 }
5181                                 prev.br_state = XFS_EXT_UNWRITTEN;
5182                                 error = xfs_bmap_add_extent(ip, lastx - 1, &cur,
5183                                         &prev, firstblock, flist, &logflags,
5184                                         XFS_DATA_FORK, 0);
5185                                 if (error)
5186                                         goto error0;
5187                                 goto nodelete;
5188                         } else {
5189                                 ASSERT(del.br_state == XFS_EXT_NORM);
5190                                 del.br_state = XFS_EXT_UNWRITTEN;
5191                                 error = xfs_bmap_add_extent(ip, lastx, &cur,
5192                                         &del, firstblock, flist, &logflags,
5193                                         XFS_DATA_FORK, 0);
5194                                 if (error)
5195                                         goto error0;
5196                                 goto nodelete;
5197                         }
5198                 }
5199                 if (wasdel) {
5200                         ASSERT(startblockval(del.br_startblock) > 0);
5201                         /* Update realtime/data freespace, unreserve quota */
5202                         if (isrt) {
5203                                 xfs_filblks_t rtexts;
5204
5205                                 rtexts = XFS_FSB_TO_B(mp, del.br_blockcount);
5206                                 do_div(rtexts, mp->m_sb.sb_rextsize);
5207                                 xfs_mod_incore_sb(mp, XFS_SBS_FREXTENTS,
5208                                                 (int64_t)rtexts, rsvd);
5209                                 (void)xfs_trans_reserve_quota_nblks(NULL,
5210                                         ip, -((long)del.br_blockcount), 0,
5211                                         XFS_QMOPT_RES_RTBLKS);
5212                         } else {
5213                                 xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
5214                                                 (int64_t)del.br_blockcount, rsvd);
5215                                 (void)xfs_trans_reserve_quota_nblks(NULL,
5216                                         ip, -((long)del.br_blockcount), 0,
5217                                         XFS_QMOPT_RES_REGBLKS);
5218                         }
5219                         ip->i_delayed_blks -= del.br_blockcount;
5220                         if (cur)
5221                                 cur->bc_private.b.flags |=
5222                                         XFS_BTCUR_BPRV_WASDEL;
5223                 } else if (cur)
5224                         cur->bc_private.b.flags &= ~XFS_BTCUR_BPRV_WASDEL;
5225                 /*
5226                  * If it's the case where the directory code is running
5227                  * with no block reservation, and the deleted block is in
5228                  * the middle of its extent, and the resulting insert
5229                  * of an extent would cause transformation to btree format,
5230                  * then reject it.  The calling code will then swap
5231                  * blocks around instead.
5232                  * We have to do this now, rather than waiting for the
5233                  * conversion to btree format, since the transaction
5234                  * will be dirty.
5235                  */
5236                 if (!wasdel && xfs_trans_get_block_res(tp) == 0 &&
5237                     XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
5238                     XFS_IFORK_NEXTENTS(ip, whichfork) >= ifp->if_ext_max &&
5239                     del.br_startoff > got.br_startoff &&
5240                     del.br_startoff + del.br_blockcount <
5241                     got.br_startoff + got.br_blockcount) {
5242                         error = XFS_ERROR(ENOSPC);
5243                         goto error0;
5244                 }
5245                 error = xfs_bmap_del_extent(ip, tp, lastx, flist, cur, &del,
5246                                 &tmp_logflags, whichfork, rsvd);
5247                 logflags |= tmp_logflags;
5248                 if (error)
5249                         goto error0;
5250                 bno = del.br_startoff - 1;
5251 nodelete:
5252                 lastx = ifp->if_lastex;
5253                 /*
5254                  * If not done go on to the next (previous) record.
5255                  * Reset ep in case the extents array was re-alloced.
5256                  */
5257                 ep = xfs_iext_get_ext(ifp, lastx);
5258                 if (bno != (xfs_fileoff_t)-1 && bno >= start) {
5259                         if (lastx >= XFS_IFORK_NEXTENTS(ip, whichfork) ||
5260                             xfs_bmbt_get_startoff(ep) > bno) {
5261                                 if (--lastx >= 0)
5262                                         ep = xfs_iext_get_ext(ifp, lastx);
5263                         }
5264                         if (lastx >= 0)
5265                                 xfs_bmbt_get_all(ep, &got);
5266                         extno++;
5267                 }
5268         }
5269         ifp->if_lastex = lastx;
5270         *done = bno == (xfs_fileoff_t)-1 || bno < start || lastx < 0;
5271         ASSERT(ifp->if_ext_max ==
5272                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5273         /*
5274          * Convert to a btree if necessary.
5275          */
5276         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
5277             XFS_IFORK_NEXTENTS(ip, whichfork) > ifp->if_ext_max) {
5278                 ASSERT(cur == NULL);
5279                 error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist,
5280                         &cur, 0, &tmp_logflags, whichfork);
5281                 logflags |= tmp_logflags;
5282                 if (error)
5283                         goto error0;
5284         }
5285         /*
5286          * transform from btree to extents, give it cur
5287          */
5288         else if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE &&
5289                  XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max) {
5290                 ASSERT(cur != NULL);
5291                 error = xfs_bmap_btree_to_extents(tp, ip, cur, &tmp_logflags,
5292                         whichfork);
5293                 logflags |= tmp_logflags;
5294                 if (error)
5295                         goto error0;
5296         }
5297         /*
5298          * transform from extents to local?
5299          */
5300         ASSERT(ifp->if_ext_max ==
5301                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5302         error = 0;
5303 error0:
5304         /*
5305          * Log everything.  Do this after conversion, there's no point in
5306          * logging the extent records if we've converted to btree format.
5307          */
5308         if ((logflags & xfs_ilog_fext(whichfork)) &&
5309             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
5310                 logflags &= ~xfs_ilog_fext(whichfork);
5311         else if ((logflags & xfs_ilog_fbroot(whichfork)) &&
5312                  XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
5313                 logflags &= ~xfs_ilog_fbroot(whichfork);
5314         /*
5315          * Log inode even in the error case, if the transaction
5316          * is dirty we'll need to shut down the filesystem.
5317          */
5318         if (logflags)
5319                 xfs_trans_log_inode(tp, ip, logflags);
5320         if (cur) {
5321                 if (!error) {
5322                         *firstblock = cur->bc_private.b.firstblock;
5323                         cur->bc_private.b.allocated = 0;
5324                 }
5325                 xfs_btree_del_cursor(cur,
5326                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5327         }
5328         return error;
5329 }
5330
5331 /*
5332  * returns 1 for success, 0 if we failed to map the extent.
5333  */
5334 STATIC int
5335 xfs_getbmapx_fix_eof_hole(
5336         xfs_inode_t             *ip,            /* xfs incore inode pointer */
5337         struct getbmapx         *out,           /* output structure */
5338         int                     prealloced,     /* this is a file with
5339                                                  * preallocated data space */
5340         __int64_t               end,            /* last block requested */
5341         xfs_fsblock_t           startblock)
5342 {
5343         __int64_t               fixlen;
5344         xfs_mount_t             *mp;            /* file system mount point */
5345         xfs_ifork_t             *ifp;           /* inode fork pointer */
5346         xfs_extnum_t            lastx;          /* last extent pointer */
5347         xfs_fileoff_t           fileblock;
5348
5349         if (startblock == HOLESTARTBLOCK) {
5350                 mp = ip->i_mount;
5351                 out->bmv_block = -1;
5352                 fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, ip->i_size));
5353                 fixlen -= out->bmv_offset;
5354                 if (prealloced && out->bmv_offset + out->bmv_length == end) {
5355                         /* Came to hole at EOF. Trim it. */
5356                         if (fixlen <= 0)
5357                                 return 0;
5358                         out->bmv_length = fixlen;
5359                 }
5360         } else {
5361                 if (startblock == DELAYSTARTBLOCK)
5362                         out->bmv_block = -2;
5363                 else
5364                         out->bmv_block = xfs_fsb_to_db(ip, startblock);
5365                 fileblock = XFS_BB_TO_FSB(ip->i_mount, out->bmv_offset);
5366                 ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
5367                 if (xfs_iext_bno_to_ext(ifp, fileblock, &lastx) &&
5368                    (lastx == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t))-1))
5369                         out->bmv_oflags |= BMV_OF_LAST;
5370         }
5371
5372         return 1;
5373 }
5374
5375 /*
5376  * Get inode's extents as described in bmv, and format for output.
5377  * Calls formatter to fill the user's buffer until all extents
5378  * are mapped, until the passed-in bmv->bmv_count slots have
5379  * been filled, or until the formatter short-circuits the loop,
5380  * if it is tracking filled-in extents on its own.
5381  */
5382 int                                             /* error code */
5383 xfs_getbmap(
5384         xfs_inode_t             *ip,
5385         struct getbmapx         *bmv,           /* user bmap structure */
5386         xfs_bmap_format_t       formatter,      /* format to user */
5387         void                    *arg)           /* formatter arg */
5388 {
5389         __int64_t               bmvend;         /* last block requested */
5390         int                     error = 0;      /* return value */
5391         __int64_t               fixlen;         /* length for -1 case */
5392         int                     i;              /* extent number */
5393         int                     lock;           /* lock state */
5394         xfs_bmbt_irec_t         *map;           /* buffer for user's data */
5395         xfs_mount_t             *mp;            /* file system mount point */
5396         int                     nex;            /* # of user extents can do */
5397         int                     nexleft;        /* # of user extents left */
5398         int                     subnex;         /* # of bmapi's can do */
5399         int                     nmap;           /* number of map entries */
5400         struct getbmapx         *out;           /* output structure */
5401         int                     whichfork;      /* data or attr fork */
5402         int                     prealloced;     /* this is a file with
5403                                                  * preallocated data space */
5404         int                     iflags;         /* interface flags */
5405         int                     bmapi_flags;    /* flags for xfs_bmapi */
5406         int                     cur_ext = 0;
5407
5408         mp = ip->i_mount;
5409         iflags = bmv->bmv_iflags;
5410         whichfork = iflags & BMV_IF_ATTRFORK ? XFS_ATTR_FORK : XFS_DATA_FORK;
5411
5412         if (whichfork == XFS_ATTR_FORK) {
5413                 if (XFS_IFORK_Q(ip)) {
5414                         if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS &&
5415                             ip->i_d.di_aformat != XFS_DINODE_FMT_BTREE &&
5416                             ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)
5417                                 return XFS_ERROR(EINVAL);
5418                 } else if (unlikely(
5419                            ip->i_d.di_aformat != 0 &&
5420                            ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS)) {
5421                         XFS_ERROR_REPORT("xfs_getbmap", XFS_ERRLEVEL_LOW,
5422                                          ip->i_mount);
5423                         return XFS_ERROR(EFSCORRUPTED);
5424                 }
5425
5426                 prealloced = 0;
5427                 fixlen = 1LL << 32;
5428         } else {
5429                 if (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS &&
5430                     ip->i_d.di_format != XFS_DINODE_FMT_BTREE &&
5431                     ip->i_d.di_format != XFS_DINODE_FMT_LOCAL)
5432                         return XFS_ERROR(EINVAL);
5433
5434                 if (xfs_get_extsz_hint(ip) ||
5435                     ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC|XFS_DIFLAG_APPEND)){
5436                         prealloced = 1;
5437                         fixlen = XFS_MAXIOFFSET(mp);
5438                 } else {
5439                         prealloced = 0;
5440                         fixlen = ip->i_size;
5441                 }
5442         }
5443
5444         if (bmv->bmv_length == -1) {
5445                 fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, fixlen));
5446                 bmv->bmv_length =
5447                         max_t(__int64_t, fixlen - bmv->bmv_offset, 0);
5448         } else if (bmv->bmv_length == 0) {
5449                 bmv->bmv_entries = 0;
5450                 return 0;
5451         } else if (bmv->bmv_length < 0) {
5452                 return XFS_ERROR(EINVAL);
5453         }
5454
5455         nex = bmv->bmv_count - 1;
5456         if (nex <= 0)
5457                 return XFS_ERROR(EINVAL);
5458         bmvend = bmv->bmv_offset + bmv->bmv_length;
5459
5460
5461         if (bmv->bmv_count > ULONG_MAX / sizeof(struct getbmapx))
5462                 return XFS_ERROR(ENOMEM);
5463         out = kmem_zalloc(bmv->bmv_count * sizeof(struct getbmapx), KM_MAYFAIL);
5464         if (!out)
5465                 return XFS_ERROR(ENOMEM);
5466
5467         xfs_ilock(ip, XFS_IOLOCK_SHARED);
5468         if (whichfork == XFS_DATA_FORK && !(iflags & BMV_IF_DELALLOC)) {
5469                 if (ip->i_delayed_blks || ip->i_size > ip->i_d.di_size) {
5470                         error = xfs_flush_pages(ip, 0, -1, 0, FI_REMAPF);
5471                         if (error)
5472                                 goto out_unlock_iolock;
5473                 }
5474                 /*
5475                  * even after flushing the inode, there can still be delalloc
5476                  * blocks on the inode beyond EOF due to speculative
5477                  * preallocation. These are not removed until the release
5478                  * function is called or the inode is inactivated. Hence we
5479                  * cannot assert here that ip->i_delayed_blks == 0.
5480                  */
5481         }
5482
5483         lock = xfs_ilock_map_shared(ip);
5484
5485         /*
5486          * Don't let nex be bigger than the number of extents
5487          * we can have assuming alternating holes and real extents.
5488          */
5489         if (nex > XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1)
5490                 nex = XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1;
5491
5492         bmapi_flags = xfs_bmapi_aflag(whichfork);
5493         if (!(iflags & BMV_IF_PREALLOC))
5494                 bmapi_flags |= XFS_BMAPI_IGSTATE;
5495
5496         /*
5497          * Allocate enough space to handle "subnex" maps at a time.
5498          */
5499         error = ENOMEM;
5500         subnex = 16;
5501         map = kmem_alloc(subnex * sizeof(*map), KM_MAYFAIL | KM_NOFS);
5502         if (!map)
5503                 goto out_unlock_ilock;
5504
5505         bmv->bmv_entries = 0;
5506
5507         if (XFS_IFORK_NEXTENTS(ip, whichfork) == 0 &&
5508             (whichfork == XFS_ATTR_FORK || !(iflags & BMV_IF_DELALLOC))) {
5509                 error = 0;
5510                 goto out_free_map;
5511         }
5512
5513         nexleft = nex;
5514
5515         do {
5516                 nmap = (nexleft > subnex) ? subnex : nexleft;
5517                 error = xfs_bmapi(NULL, ip, XFS_BB_TO_FSBT(mp, bmv->bmv_offset),
5518                                   XFS_BB_TO_FSB(mp, bmv->bmv_length),
5519                                   bmapi_flags, NULL, 0, map, &nmap,
5520                                   NULL);
5521                 if (error)
5522                         goto out_free_map;
5523                 ASSERT(nmap <= subnex);
5524
5525                 for (i = 0; i < nmap && nexleft && bmv->bmv_length; i++) {
5526                         out[cur_ext].bmv_oflags = 0;
5527                         if (map[i].br_state == XFS_EXT_UNWRITTEN)
5528                                 out[cur_ext].bmv_oflags |= BMV_OF_PREALLOC;
5529                         else if (map[i].br_startblock == DELAYSTARTBLOCK)
5530                                 out[cur_ext].bmv_oflags |= BMV_OF_DELALLOC;
5531                         out[cur_ext].bmv_offset =
5532                                 XFS_FSB_TO_BB(mp, map[i].br_startoff);
5533                         out[cur_ext].bmv_length =
5534                                 XFS_FSB_TO_BB(mp, map[i].br_blockcount);
5535                         out[cur_ext].bmv_unused1 = 0;
5536                         out[cur_ext].bmv_unused2 = 0;
5537                         ASSERT(((iflags & BMV_IF_DELALLOC) != 0) ||
5538                               (map[i].br_startblock != DELAYSTARTBLOCK));
5539                         if (map[i].br_startblock == HOLESTARTBLOCK &&
5540                             whichfork == XFS_ATTR_FORK) {
5541                                 /* came to the end of attribute fork */
5542                                 out[cur_ext].bmv_oflags |= BMV_OF_LAST;
5543                                 goto out_free_map;
5544                         }
5545
5546                         if (!xfs_getbmapx_fix_eof_hole(ip, &out[cur_ext],
5547                                         prealloced, bmvend,
5548                                         map[i].br_startblock))
5549                                 goto out_free_map;
5550
5551                         bmv->bmv_offset =
5552                                 out[cur_ext].bmv_offset +
5553                                 out[cur_ext].bmv_length;
5554                         bmv->bmv_length =
5555                                 max_t(__int64_t, 0, bmvend - bmv->bmv_offset);
5556
5557                         /*
5558                          * In case we don't want to return the hole,
5559                          * don't increase cur_ext so that we can reuse
5560                          * it in the next loop.
5561                          */
5562                         if ((iflags & BMV_IF_NO_HOLES) &&
5563                             map[i].br_startblock == HOLESTARTBLOCK) {
5564                                 memset(&out[cur_ext], 0, sizeof(out[cur_ext]));
5565                                 continue;
5566                         }
5567
5568                         nexleft--;
5569                         bmv->bmv_entries++;
5570                         cur_ext++;
5571                 }
5572         } while (nmap && nexleft && bmv->bmv_length);
5573
5574  out_free_map:
5575         kmem_free(map);
5576  out_unlock_ilock:
5577         xfs_iunlock_map_shared(ip, lock);
5578  out_unlock_iolock:
5579         xfs_iunlock(ip, XFS_IOLOCK_SHARED);
5580
5581         for (i = 0; i < cur_ext; i++) {
5582                 int full = 0;   /* user array is full */
5583
5584                 /* format results & advance arg */
5585                 error = formatter(&arg, &out[i], &full);
5586                 if (error || full)
5587                         break;
5588         }
5589
5590         kmem_free(out);
5591         return error;
5592 }
5593
5594 /*
5595  * Check the last inode extent to determine whether this allocation will result
5596  * in blocks being allocated at the end of the file. When we allocate new data
5597  * blocks at the end of the file which do not start at the previous data block,
5598  * we will try to align the new blocks at stripe unit boundaries.
5599  */
5600 STATIC int                              /* error */
5601 xfs_bmap_isaeof(
5602         xfs_inode_t     *ip,            /* incore inode pointer */
5603         xfs_fileoff_t   off,            /* file offset in fsblocks */
5604         int             whichfork,      /* data or attribute fork */
5605         char            *aeof)          /* return value */
5606 {
5607         int             error;          /* error return value */
5608         xfs_ifork_t     *ifp;           /* inode fork pointer */
5609         xfs_bmbt_rec_host_t *lastrec;   /* extent record pointer */
5610         xfs_extnum_t    nextents;       /* number of file extents */
5611         xfs_bmbt_irec_t s;              /* expanded extent record */
5612
5613         ASSERT(whichfork == XFS_DATA_FORK);
5614         ifp = XFS_IFORK_PTR(ip, whichfork);
5615         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5616             (error = xfs_iread_extents(NULL, ip, whichfork)))
5617                 return error;
5618         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5619         if (nextents == 0) {
5620                 *aeof = 1;
5621                 return 0;
5622         }
5623         /*
5624          * Go to the last extent
5625          */
5626         lastrec = xfs_iext_get_ext(ifp, nextents - 1);
5627         xfs_bmbt_get_all(lastrec, &s);
5628         /*
5629          * Check we are allocating in the last extent (for delayed allocations)
5630          * or past the last extent for non-delayed allocations.
5631          */
5632         *aeof = (off >= s.br_startoff &&
5633                  off < s.br_startoff + s.br_blockcount &&
5634                  isnullstartblock(s.br_startblock)) ||
5635                 off >= s.br_startoff + s.br_blockcount;
5636         return 0;
5637 }
5638
5639 /*
5640  * Check if the endoff is outside the last extent. If so the caller will grow
5641  * the allocation to a stripe unit boundary.
5642  */
5643 int                                     /* error */
5644 xfs_bmap_eof(
5645         xfs_inode_t     *ip,            /* incore inode pointer */
5646         xfs_fileoff_t   endoff,         /* file offset in fsblocks */
5647         int             whichfork,      /* data or attribute fork */
5648         int             *eof)           /* result value */
5649 {
5650         xfs_fsblock_t   blockcount;     /* extent block count */
5651         int             error;          /* error return value */
5652         xfs_ifork_t     *ifp;           /* inode fork pointer */
5653         xfs_bmbt_rec_host_t *lastrec;   /* extent record pointer */
5654         xfs_extnum_t    nextents;       /* number of file extents */
5655         xfs_fileoff_t   startoff;       /* extent starting file offset */
5656
5657         ASSERT(whichfork == XFS_DATA_FORK);
5658         ifp = XFS_IFORK_PTR(ip, whichfork);
5659         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5660             (error = xfs_iread_extents(NULL, ip, whichfork)))
5661                 return error;
5662         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5663         if (nextents == 0) {
5664                 *eof = 1;
5665                 return 0;
5666         }
5667         /*
5668          * Go to the last extent
5669          */
5670         lastrec = xfs_iext_get_ext(ifp, nextents - 1);
5671         startoff = xfs_bmbt_get_startoff(lastrec);
5672         blockcount = xfs_bmbt_get_blockcount(lastrec);
5673         *eof = endoff >= startoff + blockcount;
5674         return 0;
5675 }
5676
5677 #ifdef DEBUG
5678 STATIC struct xfs_buf *
5679 xfs_bmap_get_bp(
5680         struct xfs_btree_cur    *cur,
5681         xfs_fsblock_t           bno)
5682 {
5683         struct xfs_log_item_desc *lidp;
5684         int                     i;
5685
5686         if (!cur)
5687                 return NULL;
5688
5689         for (i = 0; i < XFS_BTREE_MAXLEVELS; i++) {
5690                 if (!cur->bc_bufs[i])
5691                         break;
5692                 if (XFS_BUF_ADDR(cur->bc_bufs[i]) == bno)
5693                         return cur->bc_bufs[i];
5694         }
5695
5696         /* Chase down all the log items to see if the bp is there */
5697         list_for_each_entry(lidp, &cur->bc_tp->t_items, lid_trans) {
5698                 struct xfs_buf_log_item *bip;
5699                 bip = (struct xfs_buf_log_item *)lidp->lid_item;
5700                 if (bip->bli_item.li_type == XFS_LI_BUF &&
5701                     XFS_BUF_ADDR(bip->bli_buf) == bno)
5702                         return bip->bli_buf;
5703         }
5704
5705         return NULL;
5706 }
5707
5708 STATIC void
5709 xfs_check_block(
5710         struct xfs_btree_block  *block,
5711         xfs_mount_t             *mp,
5712         int                     root,
5713         short                   sz)
5714 {
5715         int                     i, j, dmxr;
5716         __be64                  *pp, *thispa;   /* pointer to block address */
5717         xfs_bmbt_key_t          *prevp, *keyp;
5718
5719         ASSERT(be16_to_cpu(block->bb_level) > 0);
5720
5721         prevp = NULL;
5722         for( i = 1; i <= xfs_btree_get_numrecs(block); i++) {
5723                 dmxr = mp->m_bmap_dmxr[0];
5724                 keyp = XFS_BMBT_KEY_ADDR(mp, block, i);
5725
5726                 if (prevp) {
5727                         ASSERT(be64_to_cpu(prevp->br_startoff) <
5728                                be64_to_cpu(keyp->br_startoff));
5729                 }
5730                 prevp = keyp;
5731
5732                 /*
5733                  * Compare the block numbers to see if there are dups.
5734                  */
5735                 if (root)
5736                         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, i, sz);
5737                 else
5738                         pp = XFS_BMBT_PTR_ADDR(mp, block, i, dmxr);
5739
5740                 for (j = i+1; j <= be16_to_cpu(block->bb_numrecs); j++) {
5741                         if (root)
5742                                 thispa = XFS_BMAP_BROOT_PTR_ADDR(mp, block, j, sz);
5743                         else
5744                                 thispa = XFS_BMBT_PTR_ADDR(mp, block, j, dmxr);
5745                         if (*thispa == *pp) {
5746                                 cmn_err(CE_WARN, "%s: thispa(%d) == pp(%d) %Ld",
5747                                         __func__, j, i,
5748                                         (unsigned long long)be64_to_cpu(*thispa));
5749                                 panic("%s: ptrs are equal in node\n",
5750                                         __func__);
5751                         }
5752                 }
5753         }
5754 }
5755
5756 /*
5757  * Check that the extents for the inode ip are in the right order in all
5758  * btree leaves.
5759  */
5760
5761 STATIC void
5762 xfs_bmap_check_leaf_extents(
5763         xfs_btree_cur_t         *cur,   /* btree cursor or null */
5764         xfs_inode_t             *ip,            /* incore inode pointer */
5765         int                     whichfork)      /* data or attr fork */
5766 {
5767         struct xfs_btree_block  *block; /* current btree block */
5768         xfs_fsblock_t           bno;    /* block # of "block" */
5769         xfs_buf_t               *bp;    /* buffer for "block" */
5770         int                     error;  /* error return value */
5771         xfs_extnum_t            i=0, j; /* index into the extents list */
5772         xfs_ifork_t             *ifp;   /* fork structure */
5773         int                     level;  /* btree level, for checking */
5774         xfs_mount_t             *mp;    /* file system mount structure */
5775         __be64                  *pp;    /* pointer to block address */
5776         xfs_bmbt_rec_t          *ep;    /* pointer to current extent */
5777         xfs_bmbt_rec_t          last = {0, 0}; /* last extent in prev block */
5778         xfs_bmbt_rec_t          *nextp; /* pointer to next extent */
5779         int                     bp_release = 0;
5780
5781         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE) {
5782                 return;
5783         }
5784
5785         bno = NULLFSBLOCK;
5786         mp = ip->i_mount;
5787         ifp = XFS_IFORK_PTR(ip, whichfork);
5788         block = ifp->if_broot;
5789         /*
5790          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
5791          */
5792         level = be16_to_cpu(block->bb_level);
5793         ASSERT(level > 0);
5794         xfs_check_block(block, mp, 1, ifp->if_broot_bytes);
5795         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
5796         bno = be64_to_cpu(*pp);
5797
5798         ASSERT(bno != NULLDFSBNO);
5799         ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
5800         ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
5801
5802         /*
5803          * Go down the tree until leaf level is reached, following the first
5804          * pointer (leftmost) at each level.
5805          */
5806         while (level-- > 0) {
5807                 /* See if buf is in cur first */
5808                 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
5809                 if (bp) {
5810                         bp_release = 0;
5811                 } else {
5812                         bp_release = 1;
5813                 }
5814                 if (!bp && (error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
5815                                 XFS_BMAP_BTREE_REF)))
5816                         goto error_norelse;
5817                 block = XFS_BUF_TO_BLOCK(bp);
5818                 XFS_WANT_CORRUPTED_GOTO(
5819                         xfs_bmap_sanity_check(mp, bp, level),
5820                         error0);
5821                 if (level == 0)
5822                         break;
5823
5824                 /*
5825                  * Check this block for basic sanity (increasing keys and
5826                  * no duplicate blocks).
5827                  */
5828
5829                 xfs_check_block(block, mp, 0, 0);
5830                 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
5831                 bno = be64_to_cpu(*pp);
5832                 XFS_WANT_CORRUPTED_GOTO(XFS_FSB_SANITY_CHECK(mp, bno), error0);
5833                 if (bp_release) {
5834                         bp_release = 0;
5835                         xfs_trans_brelse(NULL, bp);
5836                 }
5837         }
5838
5839         /*
5840          * Here with bp and block set to the leftmost leaf node in the tree.
5841          */
5842         i = 0;
5843
5844         /*
5845          * Loop over all leaf nodes checking that all extents are in the right order.
5846          */
5847         for (;;) {
5848                 xfs_fsblock_t   nextbno;
5849                 xfs_extnum_t    num_recs;
5850
5851
5852                 num_recs = xfs_btree_get_numrecs(block);
5853
5854                 /*
5855                  * Read-ahead the next leaf block, if any.
5856                  */
5857
5858                 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
5859
5860                 /*
5861                  * Check all the extents to make sure they are OK.
5862                  * If we had a previous block, the last entry should
5863                  * conform with the first entry in this one.
5864                  */
5865
5866                 ep = XFS_BMBT_REC_ADDR(mp, block, 1);
5867                 if (i) {
5868                         ASSERT(xfs_bmbt_disk_get_startoff(&last) +
5869                                xfs_bmbt_disk_get_blockcount(&last) <=
5870                                xfs_bmbt_disk_get_startoff(ep));
5871                 }
5872                 for (j = 1; j < num_recs; j++) {
5873                         nextp = XFS_BMBT_REC_ADDR(mp, block, j + 1);
5874                         ASSERT(xfs_bmbt_disk_get_startoff(ep) +
5875                                xfs_bmbt_disk_get_blockcount(ep) <=
5876                                xfs_bmbt_disk_get_startoff(nextp));
5877                         ep = nextp;
5878                 }
5879
5880                 last = *ep;
5881                 i += num_recs;
5882                 if (bp_release) {
5883                         bp_release = 0;
5884                         xfs_trans_brelse(NULL, bp);
5885                 }
5886                 bno = nextbno;
5887                 /*
5888                  * If we've reached the end, stop.
5889                  */
5890                 if (bno == NULLFSBLOCK)
5891                         break;
5892
5893                 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
5894                 if (bp) {
5895                         bp_release = 0;
5896                 } else {
5897                         bp_release = 1;
5898                 }
5899                 if (!bp && (error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
5900                                 XFS_BMAP_BTREE_REF)))
5901                         goto error_norelse;
5902                 block = XFS_BUF_TO_BLOCK(bp);
5903         }
5904         if (bp_release) {
5905                 bp_release = 0;
5906                 xfs_trans_brelse(NULL, bp);
5907         }
5908         return;
5909
5910 error0:
5911         cmn_err(CE_WARN, "%s: at error0", __func__);
5912         if (bp_release)
5913                 xfs_trans_brelse(NULL, bp);
5914 error_norelse:
5915         cmn_err(CE_WARN, "%s: BAD after btree leaves for %d extents",
5916                 __func__, i);
5917         panic("%s: CORRUPTED BTREE OR SOMETHING", __func__);
5918         return;
5919 }
5920 #endif
5921
5922 /*
5923  * Count fsblocks of the given fork.
5924  */
5925 int                                             /* error */
5926 xfs_bmap_count_blocks(
5927         xfs_trans_t             *tp,            /* transaction pointer */
5928         xfs_inode_t             *ip,            /* incore inode */
5929         int                     whichfork,      /* data or attr fork */
5930         int                     *count)         /* out: count of blocks */
5931 {
5932         struct xfs_btree_block  *block; /* current btree block */
5933         xfs_fsblock_t           bno;    /* block # of "block" */
5934         xfs_ifork_t             *ifp;   /* fork structure */
5935         int                     level;  /* btree level, for checking */
5936         xfs_mount_t             *mp;    /* file system mount structure */
5937         __be64                  *pp;    /* pointer to block address */
5938
5939         bno = NULLFSBLOCK;
5940         mp = ip->i_mount;
5941         ifp = XFS_IFORK_PTR(ip, whichfork);
5942         if ( XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ) {
5943                 xfs_bmap_count_leaves(ifp, 0,
5944                         ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t),
5945                         count);
5946                 return 0;
5947         }
5948
5949         /*
5950          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
5951          */
5952         block = ifp->if_broot;
5953         level = be16_to_cpu(block->bb_level);
5954         ASSERT(level > 0);
5955         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
5956         bno = be64_to_cpu(*pp);
5957         ASSERT(bno != NULLDFSBNO);
5958         ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
5959         ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
5960
5961         if (unlikely(xfs_bmap_count_tree(mp, tp, ifp, bno, level, count) < 0)) {
5962                 XFS_ERROR_REPORT("xfs_bmap_count_blocks(2)", XFS_ERRLEVEL_LOW,
5963                                  mp);
5964                 return XFS_ERROR(EFSCORRUPTED);
5965         }
5966
5967         return 0;
5968 }
5969
5970 /*
5971  * Recursively walks each level of a btree
5972  * to count total fsblocks is use.
5973  */
5974 STATIC int                                     /* error */
5975 xfs_bmap_count_tree(
5976         xfs_mount_t     *mp,            /* file system mount point */
5977         xfs_trans_t     *tp,            /* transaction pointer */
5978         xfs_ifork_t     *ifp,           /* inode fork pointer */
5979         xfs_fsblock_t   blockno,        /* file system block number */
5980         int             levelin,        /* level in btree */
5981         int             *count)         /* Count of blocks */
5982 {
5983         int                     error;
5984         xfs_buf_t               *bp, *nbp;
5985         int                     level = levelin;
5986         __be64                  *pp;
5987         xfs_fsblock_t           bno = blockno;
5988         xfs_fsblock_t           nextbno;
5989         struct xfs_btree_block  *block, *nextblock;
5990         int                     numrecs;
5991
5992         if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp, XFS_BMAP_BTREE_REF)))
5993                 return error;
5994         *count += 1;
5995         block = XFS_BUF_TO_BLOCK(bp);
5996
5997         if (--level) {
5998                 /* Not at node above leaves, count this level of nodes */
5999                 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
6000                 while (nextbno != NULLFSBLOCK) {
6001                         if ((error = xfs_btree_read_bufl(mp, tp, nextbno,
6002                                 0, &nbp, XFS_BMAP_BTREE_REF)))
6003                                 return error;
6004                         *count += 1;
6005                         nextblock = XFS_BUF_TO_BLOCK(nbp);
6006                         nextbno = be64_to_cpu(nextblock->bb_u.l.bb_rightsib);
6007                         xfs_trans_brelse(tp, nbp);
6008                 }
6009
6010                 /* Dive to the next level */
6011                 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
6012                 bno = be64_to_cpu(*pp);
6013                 if (unlikely((error =
6014                      xfs_bmap_count_tree(mp, tp, ifp, bno, level, count)) < 0)) {
6015                         xfs_trans_brelse(tp, bp);
6016                         XFS_ERROR_REPORT("xfs_bmap_count_tree(1)",
6017                                          XFS_ERRLEVEL_LOW, mp);
6018                         return XFS_ERROR(EFSCORRUPTED);
6019                 }
6020                 xfs_trans_brelse(tp, bp);
6021         } else {
6022                 /* count all level 1 nodes and their leaves */
6023                 for (;;) {
6024                         nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
6025                         numrecs = be16_to_cpu(block->bb_numrecs);
6026                         xfs_bmap_disk_count_leaves(mp, block, numrecs, count);
6027                         xfs_trans_brelse(tp, bp);
6028                         if (nextbno == NULLFSBLOCK)
6029                                 break;
6030                         bno = nextbno;
6031                         if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
6032                                 XFS_BMAP_BTREE_REF)))
6033                                 return error;
6034                         *count += 1;
6035                         block = XFS_BUF_TO_BLOCK(bp);
6036                 }
6037         }
6038         return 0;
6039 }
6040
6041 /*
6042  * Count leaf blocks given a range of extent records.
6043  */
6044 STATIC void
6045 xfs_bmap_count_leaves(
6046         xfs_ifork_t             *ifp,
6047         xfs_extnum_t            idx,
6048         int                     numrecs,
6049         int                     *count)
6050 {
6051         int             b;
6052
6053         for (b = 0; b < numrecs; b++) {
6054                 xfs_bmbt_rec_host_t *frp = xfs_iext_get_ext(ifp, idx + b);
6055                 *count += xfs_bmbt_get_blockcount(frp);
6056         }
6057 }
6058
6059 /*
6060  * Count leaf blocks given a range of extent records originally
6061  * in btree format.
6062  */
6063 STATIC void
6064 xfs_bmap_disk_count_leaves(
6065         struct xfs_mount        *mp,
6066         struct xfs_btree_block  *block,
6067         int                     numrecs,
6068         int                     *count)
6069 {
6070         int             b;
6071         xfs_bmbt_rec_t  *frp;
6072
6073         for (b = 1; b <= numrecs; b++) {
6074                 frp = XFS_BMBT_REC_ADDR(mp, block, b);
6075                 *count += xfs_bmbt_disk_get_blockcount(frp);
6076         }
6077 }
6078
6079 /*
6080  * dead simple method of punching delalyed allocation blocks from a range in
6081  * the inode. Walks a block at a time so will be slow, but is only executed in
6082  * rare error cases so the overhead is not critical. This will alays punch out
6083  * both the start and end blocks, even if the ranges only partially overlap
6084  * them, so it is up to the caller to ensure that partial blocks are not
6085  * passed in.
6086  */
6087 int
6088 xfs_bmap_punch_delalloc_range(
6089         struct xfs_inode        *ip,
6090         xfs_fileoff_t           start_fsb,
6091         xfs_fileoff_t           length)
6092 {
6093         xfs_fileoff_t           remaining = length;
6094         int                     error = 0;
6095
6096         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
6097
6098         do {
6099                 int             done;
6100                 xfs_bmbt_irec_t imap;
6101                 int             nimaps = 1;
6102                 xfs_fsblock_t   firstblock;
6103                 xfs_bmap_free_t flist;
6104
6105                 /*
6106                  * Map the range first and check that it is a delalloc extent
6107                  * before trying to unmap the range. Otherwise we will be
6108                  * trying to remove a real extent (which requires a
6109                  * transaction) or a hole, which is probably a bad idea...
6110                  */
6111                 error = xfs_bmapi(NULL, ip, start_fsb, 1,
6112                                 XFS_BMAPI_ENTIRE,  NULL, 0, &imap,
6113                                 &nimaps, NULL);
6114
6115                 if (error) {
6116                         /* something screwed, just bail */
6117                         if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
6118                                 xfs_fs_cmn_err(CE_ALERT, ip->i_mount,
6119                         "Failed delalloc mapping lookup ino %lld fsb %lld.",
6120                                                 ip->i_ino, start_fsb);
6121                         }
6122                         break;
6123                 }
6124                 if (!nimaps) {
6125                         /* nothing there */
6126                         goto next_block;
6127                 }
6128                 if (imap.br_startblock != DELAYSTARTBLOCK) {
6129                         /* been converted, ignore */
6130                         goto next_block;
6131                 }
6132                 WARN_ON(imap.br_blockcount == 0);
6133
6134                 /*
6135                  * Note: while we initialise the firstblock/flist pair, they
6136                  * should never be used because blocks should never be
6137                  * allocated or freed for a delalloc extent and hence we need
6138                  * don't cancel or finish them after the xfs_bunmapi() call.
6139                  */
6140                 xfs_bmap_init(&flist, &firstblock);
6141                 error = xfs_bunmapi(NULL, ip, start_fsb, 1, 0, 1, &firstblock,
6142                                         &flist, &done);
6143                 if (error)
6144                         break;
6145
6146                 ASSERT(!flist.xbf_count && !flist.xbf_first);
6147 next_block:
6148                 start_fsb++;
6149                 remaining--;
6150         } while(remaining > 0);
6151
6152         return error;
6153 }