fs/xfs/xfs_bmap.c

   1 /*
   2  * Copyright (c) 2000-2006 Silicon Graphics, Inc.
   3  * All Rights Reserved.
   4  *
   5  * This program is free software; you can redistribute it and/or
   6  * modify it under the terms of the GNU General Public License as
   7  * published by the Free Software Foundation.
   8  *
   9  * This program is distributed in the hope that it would be useful,
  10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12  * GNU General Public License for more details.
  13  *
  14  * You should have received a copy of the GNU General Public License
  15  * along with this program; if not, write the Free Software Foundation,
  16  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  17  */
  18 #include "xfs.h"
  19 #include "xfs_fs.h"
  20 #include "xfs_types.h"
  21 #include "xfs_bit.h"
  22 #include "xfs_log.h"
  23 #include "xfs_inum.h"
  24 #include "xfs_trans.h"
  25 #include "xfs_sb.h"
  26 #include "xfs_ag.h"
  27 #include "xfs_dir2.h"
  28 #include "xfs_da_btree.h"
  29 #include "xfs_bmap_btree.h"
  30 #include "xfs_alloc_btree.h"
  31 #include "xfs_ialloc_btree.h"
  32 #include "xfs_dinode.h"
  33 #include "xfs_inode.h"
  34 #include "xfs_btree.h"
  35 #include "xfs_mount.h"
  36 #include "xfs_itable.h"
  37 #include "xfs_inode_item.h"
  38 #include "xfs_extfree_item.h"
  39 #include "xfs_alloc.h"
  40 #include "xfs_bmap.h"
  41 #include "xfs_rtalloc.h"
  42 #include "xfs_error.h"
  43 #include "xfs_attr_leaf.h"
  44 #include "xfs_rw.h"
  45 #include "xfs_quota.h"
  46 #include "xfs_trans_space.h"
  47 #include "xfs_buf_item.h"
  48 #include "xfs_filestream.h"
  49 #include "xfs_vnodeops.h"
  50 #include "xfs_trace.h"
  51
  52
  53 #ifdef DEBUG
  54 STATIC void
  55 xfs_bmap_check_leaf_extents(xfs_btree_cur_t *cur, xfs_inode_t *ip, int whichfork);
  56 #endif
  57
  58 kmem_zone_t             *xfs_bmap_free_item_zone;
  59
  60 /*
  61  * Prototypes for internal bmap routines.
  62  */
  63
  64
  65 /*
  66  * Called from xfs_bmap_add_attrfork to handle extents format files.
  67  */
  68 STATIC int                                      /* error */
  69 xfs_bmap_add_attrfork_extents(
  70         xfs_trans_t             *tp,            /* transaction pointer */
  71         xfs_inode_t             *ip,            /* incore inode pointer */
  72         xfs_fsblock_t           *firstblock,    /* first block allocated */
  73         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
  74         int                     *flags);        /* inode logging flags */
  75
  76 /*
  77  * Called from xfs_bmap_add_attrfork to handle local format files.
  78  */
  79 STATIC int                                      /* error */
  80 xfs_bmap_add_attrfork_local(
  81         xfs_trans_t             *tp,            /* transaction pointer */
  82         xfs_inode_t             *ip,            /* incore inode pointer */
  83         xfs_fsblock_t           *firstblock,    /* first block allocated */
  84         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
  85         int                     *flags);        /* inode logging flags */
  86
  87 /*
  88  * Called by xfs_bmap_add_extent to handle cases converting a delayed
  89  * allocation to a real allocation.
  90  */
  91 STATIC int                              /* error */
  92 xfs_bmap_add_extent_delay_real(
  93         struct xfs_trans        *tp,    /* transaction pointer */
  94         xfs_inode_t             *ip,    /* incore inode pointer */
  95         xfs_extnum_t            *idx,   /* extent number to update/insert */
  96         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
  97         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
  98         xfs_filblks_t           *dnew,  /* new delayed-alloc indirect blocks */
  99         xfs_fsblock_t           *first, /* pointer to firstblock variable */
 100         xfs_bmap_free_t         *flist, /* list of extents to be freed */
 101         int                     *logflagsp); /* inode logging flags */
 102
 103 /*
 104  * Called by xfs_bmap_add_extent to handle cases converting a hole
 105  * to a real allocation.
 106  */
 107 STATIC int                              /* error */
 108 xfs_bmap_add_extent_hole_real(
 109         xfs_inode_t             *ip,    /* incore inode pointer */
 110         xfs_extnum_t            *idx,   /* extent number to update/insert */
 111         xfs_btree_cur_t         *cur,   /* if null, not a btree */
 112         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
 113         int                     *logflagsp, /* inode logging flags */
 114         int                     whichfork); /* data or attr fork */
 115
 116 /*
 117  * Called by xfs_bmap_add_extent to handle cases converting an unwritten
 118  * allocation to a real allocation or vice versa.
 119  */
 120 STATIC int                              /* error */
 121 xfs_bmap_add_extent_unwritten_real(
 122         xfs_inode_t             *ip,    /* incore inode pointer */
 123         xfs_extnum_t            *idx,   /* extent number to update/insert */
 124         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
 125         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
 126         int                     *logflagsp); /* inode logging flags */
 127
 128 /*
 129  * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
 130  * It figures out where to ask the underlying allocator to put the new extent.
 131  */
 132 STATIC int                              /* error */
 133 xfs_bmap_alloc(
 134         xfs_bmalloca_t          *ap);   /* bmap alloc argument struct */
 135
 136 /*
 137  * Transform a btree format file with only one leaf node, where the
 138  * extents list will fit in the inode, into an extents format file.
 139  * Since the file extents are already in-core, all we have to do is
 140  * give up the space for the btree root and pitch the leaf block.
 141  */
 142 STATIC int                              /* error */
 143 xfs_bmap_btree_to_extents(
 144         xfs_trans_t             *tp,    /* transaction pointer */
 145         xfs_inode_t             *ip,    /* incore inode pointer */
 146         xfs_btree_cur_t         *cur,   /* btree cursor */
 147         int                     *logflagsp, /* inode logging flags */
 148         int                     whichfork); /* data or attr fork */
 149
 150 /*
 151  * Remove the entry "free" from the free item list.  Prev points to the
 152  * previous entry, unless "free" is the head of the list.
 153  */
 154 STATIC void
 155 xfs_bmap_del_free(
 156         xfs_bmap_free_t         *flist, /* free item list header */
 157         xfs_bmap_free_item_t    *prev,  /* previous item on list, if any */
 158         xfs_bmap_free_item_t    *free); /* list item to be freed */
 159
 160 /*
 161  * Convert an extents-format file into a btree-format file.
 162  * The new file will have a root block (in the inode) and a single child block.
 163  */
 164 STATIC int                                      /* error */
 165 xfs_bmap_extents_to_btree(
 166         xfs_trans_t             *tp,            /* transaction pointer */
 167         xfs_inode_t             *ip,            /* incore inode pointer */
 168         xfs_fsblock_t           *firstblock,    /* first-block-allocated */
 169         xfs_bmap_free_t         *flist,         /* blocks freed in xaction */
 170         xfs_btree_cur_t         **curp,         /* cursor returned to caller */
 171         int                     wasdel,         /* converting a delayed alloc */
 172         int                     *logflagsp,     /* inode logging flags */
 173         int                     whichfork);     /* data or attr fork */
 174
 175 /*
 176  * Convert a local file to an extents file.
 177  * This code is sort of bogus, since the file data needs to get
 178  * logged so it won't be lost.  The bmap-level manipulations are ok, though.
 179  */
 180 STATIC int                              /* error */
 181 xfs_bmap_local_to_extents(
 182         xfs_trans_t     *tp,            /* transaction pointer */
 183         xfs_inode_t     *ip,            /* incore inode pointer */
 184         xfs_fsblock_t   *firstblock,    /* first block allocated in xaction */
 185         xfs_extlen_t    total,          /* total blocks needed by transaction */
 186         int             *logflagsp,     /* inode logging flags */
 187         int             whichfork);     /* data or attr fork */
 188
 189 /*
 190  * Search the extents list for the inode, for the extent containing bno.
 191  * If bno lies in a hole, point to the next entry.  If bno lies past eof,
 192  * *eofp will be set, and *prevp will contain the last entry (null if none).
 193  * Else, *lastxp will be set to the index of the found
 194  * entry; *gotp will contain the entry.
 195  */
 196 STATIC xfs_bmbt_rec_host_t *            /* pointer to found extent entry */
 197 xfs_bmap_search_extents(
 198         xfs_inode_t     *ip,            /* incore inode pointer */
 199         xfs_fileoff_t   bno,            /* block number searched for */
 200         int             whichfork,      /* data or attr fork */
 201         int             *eofp,          /* out: end of file found */
 202         xfs_extnum_t    *lastxp,        /* out: last extent index */
 203         xfs_bmbt_irec_t *gotp,          /* out: extent entry found */
 204         xfs_bmbt_irec_t *prevp);        /* out: previous extent entry found */
 205
 206 /*
 207  * Compute the worst-case number of indirect blocks that will be used
 208  * for ip's delayed extent of length "len".
 209  */
 210 STATIC xfs_filblks_t
 211 xfs_bmap_worst_indlen(
 212         xfs_inode_t             *ip,    /* incore inode pointer */
 213         xfs_filblks_t           len);   /* delayed extent length */
 214
 215 #ifdef DEBUG
 216 /*
 217  * Perform various validation checks on the values being returned
 218  * from xfs_bmapi().
 219  */
 220 STATIC void
 221 xfs_bmap_validate_ret(
 222         xfs_fileoff_t           bno,
 223         xfs_filblks_t           len,
 224         int                     flags,
 225         xfs_bmbt_irec_t         *mval,
 226         int                     nmap,
 227         int                     ret_nmap);
 228 #else
 229 #define xfs_bmap_validate_ret(bno,len,flags,mval,onmap,nmap)
 230 #endif /* DEBUG */
 231
 232 STATIC int
 233 xfs_bmap_count_tree(
 234         xfs_mount_t     *mp,
 235         xfs_trans_t     *tp,
 236         xfs_ifork_t     *ifp,
 237         xfs_fsblock_t   blockno,
 238         int             levelin,
 239         int             *count);
 240
 241 STATIC void
 242 xfs_bmap_count_leaves(
 243         xfs_ifork_t             *ifp,
 244         xfs_extnum_t            idx,
 245         int                     numrecs,
 246         int                     *count);
 247
 248 STATIC void
 249 xfs_bmap_disk_count_leaves(
 250         struct xfs_mount        *mp,
 251         struct xfs_btree_block  *block,
 252         int                     numrecs,
 253         int                     *count);
 254
 255 /*
 256  * Bmap internal routines.
 257  */
 258
 259 STATIC int                              /* error */
 260 xfs_bmbt_lookup_eq(
 261         struct xfs_btree_cur    *cur,
 262         xfs_fileoff_t           off,
 263         xfs_fsblock_t           bno,
 264         xfs_filblks_t           len,
 265         int                     *stat)  /* success/failure */
 266 {
 267         cur->bc_rec.b.br_startoff = off;
 268         cur->bc_rec.b.br_startblock = bno;
 269         cur->bc_rec.b.br_blockcount = len;
 270         return xfs_btree_lookup(cur, XFS_LOOKUP_EQ, stat);
 271 }
 272
 273 STATIC int                              /* error */
 274 xfs_bmbt_lookup_ge(
 275         struct xfs_btree_cur    *cur,
 276         xfs_fileoff_t           off,
 277         xfs_fsblock_t           bno,
 278         xfs_filblks_t           len,
 279         int                     *stat)  /* success/failure */
 280 {
 281         cur->bc_rec.b.br_startoff = off;
 282         cur->bc_rec.b.br_startblock = bno;
 283         cur->bc_rec.b.br_blockcount = len;
 284         return xfs_btree_lookup(cur, XFS_LOOKUP_GE, stat);
 285 }
 286
 287 /*
 288 * Update the record referred to by cur to the value given
 289  * by [off, bno, len, state].
 290  * This either works (return 0) or gets an EFSCORRUPTED error.
 291  */
 292 STATIC int
 293 xfs_bmbt_update(
 294         struct xfs_btree_cur    *cur,
 295         xfs_fileoff_t           off,
 296         xfs_fsblock_t           bno,
 297         xfs_filblks_t           len,
 298         xfs_exntst_t            state)
 299 {
 300         union xfs_btree_rec     rec;
 301
 302         xfs_bmbt_disk_set_allf(&rec.bmbt, off, bno, len, state);
 303         return xfs_btree_update(cur, &rec);
 304 }
 305
 306 /*
 307  * Called from xfs_bmap_add_attrfork to handle btree format files.
 308  */
 309 STATIC int                                      /* error */
 310 xfs_bmap_add_attrfork_btree(
 311         xfs_trans_t             *tp,            /* transaction pointer */
 312         xfs_inode_t             *ip,            /* incore inode pointer */
 313         xfs_fsblock_t           *firstblock,    /* first block allocated */
 314         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
 315         int                     *flags)         /* inode logging flags */
 316 {
 317         xfs_btree_cur_t         *cur;           /* btree cursor */
 318         int                     error;          /* error return value */
 319         xfs_mount_t             *mp;            /* file system mount struct */
 320         int                     stat;           /* newroot status */
 321
 322         mp = ip->i_mount;
 323         if (ip->i_df.if_broot_bytes <= XFS_IFORK_DSIZE(ip))
 324                 *flags |= XFS_ILOG_DBROOT;
 325         else {
 326                 cur = xfs_bmbt_init_cursor(mp, tp, ip, XFS_DATA_FORK);
 327                 cur->bc_private.b.flist = flist;
 328                 cur->bc_private.b.firstblock = *firstblock;
 329                 if ((error = xfs_bmbt_lookup_ge(cur, 0, 0, 0, &stat)))
 330                         goto error0;
 331                 /* must be at least one entry */
 332                 XFS_WANT_CORRUPTED_GOTO(stat == 1, error0);
 333                 if ((error = xfs_btree_new_iroot(cur, flags, &stat)))
 334                         goto error0;
 335                 if (stat == 0) {
 336                         xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
 337                         return XFS_ERROR(ENOSPC);
 338                 }
 339                 *firstblock = cur->bc_private.b.firstblock;
 340                 cur->bc_private.b.allocated = 0;
 341                 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
 342         }
 343         return 0;
 344 error0:
 345         xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
 346         return error;
 347 }
 348
 349 /*
 350  * Called from xfs_bmap_add_attrfork to handle extents format files.
 351  */
 352 STATIC int                                      /* error */
 353 xfs_bmap_add_attrfork_extents(
 354         xfs_trans_t             *tp,            /* transaction pointer */
 355         xfs_inode_t             *ip,            /* incore inode pointer */
 356         xfs_fsblock_t           *firstblock,    /* first block allocated */
 357         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
 358         int                     *flags)         /* inode logging flags */
 359 {
 360         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
 361         int                     error;          /* error return value */
 362
 363         if (ip->i_d.di_nextents * sizeof(xfs_bmbt_rec_t) <= XFS_IFORK_DSIZE(ip))
 364                 return 0;
 365         cur = NULL;
 366         error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist, &cur, 0,
 367                 flags, XFS_DATA_FORK);
 368         if (cur) {
 369                 cur->bc_private.b.allocated = 0;
 370                 xfs_btree_del_cursor(cur,
 371                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
 372         }
 373         return error;
 374 }
 375
 376 /*
 377  * Called from xfs_bmap_add_attrfork to handle local format files.
 378  */
 379 STATIC int                                      /* error */
 380 xfs_bmap_add_attrfork_local(
 381         xfs_trans_t             *tp,            /* transaction pointer */
 382         xfs_inode_t             *ip,            /* incore inode pointer */
 383         xfs_fsblock_t           *firstblock,    /* first block allocated */
 384         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
 385         int                     *flags)         /* inode logging flags */
 386 {
 387         xfs_da_args_t           dargs;          /* args for dir/attr code */
 388         int                     error;          /* error return value */
 389         xfs_mount_t             *mp;            /* mount structure pointer */
 390
 391         if (ip->i_df.if_bytes <= XFS_IFORK_DSIZE(ip))
 392                 return 0;
 393         if (S_ISDIR(ip->i_d.di_mode)) {
 394                 mp = ip->i_mount;
 395                 memset(&dargs, 0, sizeof(dargs));
 396                 dargs.dp = ip;
 397                 dargs.firstblock = firstblock;
 398                 dargs.flist = flist;
 399                 dargs.total = mp->m_dirblkfsbs;
 400                 dargs.whichfork = XFS_DATA_FORK;
 401                 dargs.trans = tp;
 402                 error = xfs_dir2_sf_to_block(&dargs);
 403         } else
 404                 error = xfs_bmap_local_to_extents(tp, ip, firstblock, 1, flags,
 405                         XFS_DATA_FORK);
 406         return error;
 407 }
 408
 409 /*
 410  * Update file extent records and the btree after allocating space.
 411  */
 412 STATIC int                              /* error */
 413 xfs_bmap_add_extent(
 414         struct xfs_trans        *tp,    /* transaction pointer */
 415         xfs_inode_t             *ip,    /* incore inode pointer */
 416         xfs_extnum_t            *idx,   /* extent number to update/insert */
 417         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
 418         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
 419         xfs_fsblock_t           *first, /* pointer to firstblock variable */
 420         xfs_bmap_free_t         *flist, /* list of extents to be freed */
 421         int                     *logflagsp, /* inode logging flags */
 422         int                     whichfork) /* data or attr fork */
 423 {
 424         xfs_btree_cur_t         *cur;   /* btree cursor or null */
 425         xfs_filblks_t           da_new; /* new count del alloc blocks used */
 426         xfs_filblks_t           da_old; /* old count del alloc blocks used */
 427         int                     error;  /* error return value */
 428         xfs_ifork_t             *ifp;   /* inode fork ptr */
 429         int                     logflags; /* returned value */
 430         xfs_extnum_t            nextents; /* number of extents in file now */
 431
 432         XFS_STATS_INC(xs_add_exlist);
 433
 434         cur = *curp;
 435         ifp = XFS_IFORK_PTR(ip, whichfork);
 436         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
 437         da_old = da_new = 0;
 438         error = 0;
 439
 440         ASSERT(*idx >= 0);
 441         ASSERT(*idx <= nextents);
 442         ASSERT(!isnullstartblock(new->br_startblock));
 443
 444         /*
 445          * Real allocation off the end of the file.
 446          */
 447         if (*idx == nextents) {
 448                 if (cur)
 449                         ASSERT((cur->bc_private.b.flags &
 450                                 XFS_BTCUR_BPRV_WASDEL) == 0);
 451                 error = xfs_bmap_add_extent_hole_real(ip, idx, cur, new,
 452                                 &logflags, whichfork);
 453         } else {
 454                 xfs_bmbt_irec_t prev;   /* old extent at offset idx */
 455
 456                 /*
 457                  * Get the record referred to by idx.
 458                  */
 459                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx), &prev);
 460                 /*
 461                  * If it's a real allocation record, and the new allocation ends
 462                  * after the start of the referred to record, then we're filling
 463                  * in a delayed or unwritten allocation with a real one, or
 464                  * converting real back to unwritten.
 465                  */
 466                 if (!isnullstartblock(new->br_startblock) &&
 467                     new->br_startoff + new->br_blockcount > prev.br_startoff) {
 468                         if (prev.br_state != XFS_EXT_UNWRITTEN &&
 469                             isnullstartblock(prev.br_startblock)) {
 470                                 da_old = startblockval(prev.br_startblock);
 471                                 if (cur)
 472                                         ASSERT(cur->bc_private.b.flags &
 473                                                 XFS_BTCUR_BPRV_WASDEL);
 474                                 error = xfs_bmap_add_extent_delay_real(tp, ip,
 475                                                 idx, &cur, new, &da_new,
 476                                                 first, flist, &logflags);
 477                         } else {
 478                                 ASSERT(new->br_state == XFS_EXT_NORM ||
 479                                        new->br_state == XFS_EXT_UNWRITTEN);
 480
 481                                 error = xfs_bmap_add_extent_unwritten_real(ip,
 482                                                 idx, &cur, new, &logflags);
 483                                 if (error)
 484                                         goto done;
 485                         }
 486                 }
 487                 /*
 488                  * Otherwise we're filling in a hole with an allocation.
 489                  */
 490                 else {
 491                         if (cur)
 492                                 ASSERT((cur->bc_private.b.flags &
 493                                         XFS_BTCUR_BPRV_WASDEL) == 0);
 494                         error = xfs_bmap_add_extent_hole_real(ip, idx, cur,
 495                                         new, &logflags, whichfork);
 496                 }
 497         }
 498
 499         if (error)
 500                 goto done;
 501         ASSERT(*curp == cur || *curp == NULL);
 502
 503         /*
 504          * Convert to a btree if necessary.
 505          */
 506         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
 507             XFS_IFORK_NEXTENTS(ip, whichfork) > ifp->if_ext_max) {
 508                 int     tmp_logflags;   /* partial log flag return val */
 509
 510                 ASSERT(cur == NULL);
 511                 error = xfs_bmap_extents_to_btree(tp, ip, first,
 512                         flist, &cur, da_old > 0, &tmp_logflags, whichfork);
 513                 logflags |= tmp_logflags;
 514                 if (error)
 515                         goto done;
 516         }
 517         /*
 518          * Adjust for changes in reserved delayed indirect blocks.
 519          * Nothing to do for disk quotas here.
 520          */
 521         if (da_old || da_new) {
 522                 xfs_filblks_t   nblks;
 523
 524                 nblks = da_new;
 525                 if (cur)
 526                         nblks += cur->bc_private.b.allocated;
 527                 ASSERT(nblks <= da_old);
 528                 if (nblks < da_old)
 529                         xfs_icsb_modify_counters(ip->i_mount, XFS_SBS_FDBLOCKS,
 530                                 (int64_t)(da_old - nblks), 0);
 531         }
 532         /*
 533          * Clear out the allocated field, done with it now in any case.
 534          */
 535         if (cur) {
 536                 cur->bc_private.b.allocated = 0;
 537                 *curp = cur;
 538         }
 539 done:
 540 #ifdef DEBUG
 541         if (!error)
 542                 xfs_bmap_check_leaf_extents(*curp, ip, whichfork);
 543 #endif
 544         *logflagsp = logflags;
 545         return error;
 546 }
 547
 548 /*
 549  * Called by xfs_bmap_add_extent to handle cases converting a delayed
 550  * allocation to a real allocation.
 551  */
 552 STATIC int                              /* error */
 553 xfs_bmap_add_extent_delay_real(
 554         struct xfs_trans        *tp,    /* transaction pointer */
 555         xfs_inode_t             *ip,    /* incore inode pointer */
 556         xfs_extnum_t            *idx,   /* extent number to update/insert */
 557         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
 558         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
 559         xfs_filblks_t           *dnew,  /* new delayed-alloc indirect blocks */
 560         xfs_fsblock_t           *first, /* pointer to firstblock variable */
 561         xfs_bmap_free_t         *flist, /* list of extents to be freed */
 562         int                     *logflagsp) /* inode logging flags */
 563 {
 564         xfs_btree_cur_t         *cur;   /* btree cursor */
 565         int                     diff;   /* temp value */
 566         xfs_bmbt_rec_host_t     *ep;    /* extent entry for idx */
 567         int                     error;  /* error return value */
 568         int                     i;      /* temp state */
 569         xfs_ifork_t             *ifp;   /* inode fork pointer */
 570         xfs_fileoff_t           new_endoff;     /* end offset of new entry */
 571         xfs_bmbt_irec_t         r[3];   /* neighbor extent entries */
 572                                         /* left is 0, right is 1, prev is 2 */
 573         int                     rval=0; /* return value (logging flags) */
 574         int                     state = 0;/* state bits, accessed thru macros */
 575         xfs_filblks_t           temp=0; /* value for dnew calculations */
 576         xfs_filblks_t           temp2=0;/* value for dnew calculations */
 577         int                     tmp_rval;       /* partial logging flags */
 578
 579 #define LEFT            r[0]
 580 #define RIGHT           r[1]
 581 #define PREV            r[2]
 582
 583         /*
 584          * Set up a bunch of variables to make the tests simpler.
 585          */
 586         cur = *curp;
 587         ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
 588         ep = xfs_iext_get_ext(ifp, *idx);
 589         xfs_bmbt_get_all(ep, &PREV);
 590         new_endoff = new->br_startoff + new->br_blockcount;
 591         ASSERT(PREV.br_startoff <= new->br_startoff);
 592         ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
 593
 594         /*
 595          * Set flags determining what part of the previous delayed allocation
 596          * extent is being replaced by a real allocation.
 597          */
 598         if (PREV.br_startoff == new->br_startoff)
 599                 state |= BMAP_LEFT_FILLING;
 600         if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
 601                 state |= BMAP_RIGHT_FILLING;
 602
 603         /*
 604          * Check and set flags if this segment has a left neighbor.
 605          * Don't set contiguous if the combined extent would be too large.
 606          */
 607         if (*idx > 0) {
 608                 state |= BMAP_LEFT_VALID;
 609                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &LEFT);
 610
 611                 if (isnullstartblock(LEFT.br_startblock))
 612                         state |= BMAP_LEFT_DELAY;
 613         }
 614
 615         if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
 616             LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
 617             LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
 618             LEFT.br_state == new->br_state &&
 619             LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
 620                 state |= BMAP_LEFT_CONTIG;
 621
 622         /*
 623          * Check and set flags if this segment has a right neighbor.
 624          * Don't set contiguous if the combined extent would be too large.
 625          * Also check for all-three-contiguous being too large.
 626          */
 627         if (*idx < ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1) {
 628                 state |= BMAP_RIGHT_VALID;
 629                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx + 1), &RIGHT);
 630
 631                 if (isnullstartblock(RIGHT.br_startblock))
 632                         state |= BMAP_RIGHT_DELAY;
 633         }
 634
 635         if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
 636             new_endoff == RIGHT.br_startoff &&
 637             new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
 638             new->br_state == RIGHT.br_state &&
 639             new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
 640             ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
 641                        BMAP_RIGHT_FILLING)) !=
 642                       (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
 643                        BMAP_RIGHT_FILLING) ||
 644              LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
 645                         <= MAXEXTLEN))
 646                 state |= BMAP_RIGHT_CONTIG;
 647
 648         error = 0;
 649         /*
 650          * Switch out based on the FILLING and CONTIG state bits.
 651          */
 652         switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
 653                          BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
 654         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
 655              BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
 656                 /*
 657                  * Filling in all of a previously delayed allocation extent.
 658                  * The left and right neighbors are both contiguous with new.
 659                  */
 660                 --*idx;
 661                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
 662                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
 663                         LEFT.br_blockcount + PREV.br_blockcount +
 664                         RIGHT.br_blockcount);
 665                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
 666
 667                 xfs_iext_remove(ip, *idx + 1, 2, state);
 668                 ip->i_d.di_nextents--;
 669                 if (cur == NULL)
 670                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
 671                 else {
 672                         rval = XFS_ILOG_CORE;
 673                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
 674                                         RIGHT.br_startblock,
 675                                         RIGHT.br_blockcount, &i)))
 676                                 goto done;
 677                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 678                         if ((error = xfs_btree_delete(cur, &i)))
 679                                 goto done;
 680                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 681                         if ((error = xfs_btree_decrement(cur, 0, &i)))
 682                                 goto done;
 683                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 684                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
 685                                         LEFT.br_startblock,
 686                                         LEFT.br_blockcount +
 687                                         PREV.br_blockcount +
 688                                         RIGHT.br_blockcount, LEFT.br_state)))
 689                                 goto done;
 690                 }
 691                 *dnew = 0;
 692                 break;
 693
 694         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
 695                 /*
 696                  * Filling in all of a previously delayed allocation extent.
 697                  * The left neighbor is contiguous, the right is not.
 698                  */
 699                 --*idx;
 700
 701                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
 702                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
 703                         LEFT.br_blockcount + PREV.br_blockcount);
 704                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
 705
 706                 xfs_iext_remove(ip, *idx + 1, 1, state);
 707                 if (cur == NULL)
 708                         rval = XFS_ILOG_DEXT;
 709                 else {
 710                         rval = 0;
 711                         if ((error = xfs_bmbt_lookup_eq(cur, LEFT.br_startoff,
 712                                         LEFT.br_startblock, LEFT.br_blockcount,
 713                                         &i)))
 714                                 goto done;
 715                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 716                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
 717                                         LEFT.br_startblock,
 718                                         LEFT.br_blockcount +
 719                                         PREV.br_blockcount, LEFT.br_state)))
 720                                 goto done;
 721                 }
 722                 *dnew = 0;
 723                 break;
 724
 725         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
 726                 /*
 727                  * Filling in all of a previously delayed allocation extent.
 728                  * The right neighbor is contiguous, the left is not.
 729                  */
 730                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
 731                 xfs_bmbt_set_startblock(ep, new->br_startblock);
 732                 xfs_bmbt_set_blockcount(ep,
 733                         PREV.br_blockcount + RIGHT.br_blockcount);
 734                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
 735
 736                 xfs_iext_remove(ip, *idx + 1, 1, state);
 737                 if (cur == NULL)
 738                         rval = XFS_ILOG_DEXT;
 739                 else {
 740                         rval = 0;
 741                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
 742                                         RIGHT.br_startblock,
 743                                         RIGHT.br_blockcount, &i)))
 744                                 goto done;
 745                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 746                         if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
 747                                         new->br_startblock,
 748                                         PREV.br_blockcount +
 749                                         RIGHT.br_blockcount, PREV.br_state)))
 750                                 goto done;
 751                 }
 752
 753                 *dnew = 0;
 754                 break;
 755
 756         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
 757                 /*
 758                  * Filling in all of a previously delayed allocation extent.
 759                  * Neither the left nor right neighbors are contiguous with
 760                  * the new one.
 761                  */
 762                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
 763                 xfs_bmbt_set_startblock(ep, new->br_startblock);
 764                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
 765
 766                 ip->i_d.di_nextents++;
 767                 if (cur == NULL)
 768                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
 769                 else {
 770                         rval = XFS_ILOG_CORE;
 771                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
 772                                         new->br_startblock, new->br_blockcount,
 773                                         &i)))
 774                                 goto done;
 775                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
 776                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
 777                         if ((error = xfs_btree_insert(cur, &i)))
 778                                 goto done;
 779                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 780                 }
 781
 782                 *dnew = 0;
 783                 break;
 784
 785         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
 786                 /*
 787                  * Filling in the first part of a previous delayed allocation.
 788                  * The left neighbor is contiguous.
 789                  */
 790                 trace_xfs_bmap_pre_update(ip, *idx - 1, state, _THIS_IP_);
 791                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx - 1),
 792                         LEFT.br_blockcount + new->br_blockcount);
 793                 xfs_bmbt_set_startoff(ep,
 794                         PREV.br_startoff + new->br_blockcount);
 795                 trace_xfs_bmap_post_update(ip, *idx - 1, state, _THIS_IP_);
 796
 797                 temp = PREV.br_blockcount - new->br_blockcount;
 798                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
 799                 xfs_bmbt_set_blockcount(ep, temp);
 800                 if (cur == NULL)
 801                         rval = XFS_ILOG_DEXT;
 802                 else {
 803                         rval = 0;
 804                         if ((error = xfs_bmbt_lookup_eq(cur, LEFT.br_startoff,
 805                                         LEFT.br_startblock, LEFT.br_blockcount,
 806                                         &i)))
 807                                 goto done;
 808                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 809                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
 810                                         LEFT.br_startblock,
 811                                         LEFT.br_blockcount +
 812                                         new->br_blockcount,
 813                                         LEFT.br_state)))
 814                                 goto done;
 815                 }
 816                 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
 817                         startblockval(PREV.br_startblock));
 818                 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
 819                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
 820
 821                 --*idx;
 822                 *dnew = temp;
 823                 break;
 824
 825         case BMAP_LEFT_FILLING:
 826                 /*
 827                  * Filling in the first part of a previous delayed allocation.
 828                  * The left neighbor is not contiguous.
 829                  */
 830                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
 831                 xfs_bmbt_set_startoff(ep, new_endoff);
 832                 temp = PREV.br_blockcount - new->br_blockcount;
 833                 xfs_bmbt_set_blockcount(ep, temp);
 834                 xfs_iext_insert(ip, *idx, 1, new, state);
 835                 ip->i_d.di_nextents++;
 836                 if (cur == NULL)
 837                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
 838                 else {
 839                         rval = XFS_ILOG_CORE;
 840                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
 841                                         new->br_startblock, new->br_blockcount,
 842                                         &i)))
 843                                 goto done;
 844                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
 845                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
 846                         if ((error = xfs_btree_insert(cur, &i)))
 847                                 goto done;
 848                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 849                 }
 850                 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
 851                     ip->i_d.di_nextents > ip->i_df.if_ext_max) {
 852                         error = xfs_bmap_extents_to_btree(tp, ip,
 853                                         first, flist, &cur, 1, &tmp_rval,
 854                                         XFS_DATA_FORK);
 855                         rval |= tmp_rval;
 856                         if (error)
 857                                 goto done;
 858                 }
 859                 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
 860                         startblockval(PREV.br_startblock) -
 861                         (cur ? cur->bc_private.b.allocated : 0));
 862                 ep = xfs_iext_get_ext(ifp, *idx + 1);
 863                 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
 864                 trace_xfs_bmap_post_update(ip, *idx + 1, state, _THIS_IP_);
 865
 866                 *dnew = temp;
 867                 break;
 868
 869         case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
 870                 /*
 871                  * Filling in the last part of a previous delayed allocation.
 872                  * The right neighbor is contiguous with the new allocation.
 873                  */
 874                 temp = PREV.br_blockcount - new->br_blockcount;
 875                 trace_xfs_bmap_pre_update(ip, *idx + 1, state, _THIS_IP_);
 876                 xfs_bmbt_set_blockcount(ep, temp);
 877                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx + 1),
 878                         new->br_startoff, new->br_startblock,
 879                         new->br_blockcount + RIGHT.br_blockcount,
 880                         RIGHT.br_state);
 881                 trace_xfs_bmap_post_update(ip, *idx + 1, state, _THIS_IP_);
 882                 if (cur == NULL)
 883                         rval = XFS_ILOG_DEXT;
 884                 else {
 885                         rval = 0;
 886                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
 887                                         RIGHT.br_startblock,
 888                                         RIGHT.br_blockcount, &i)))
 889                                 goto done;
 890                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 891                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
 892                                         new->br_startblock,
 893                                         new->br_blockcount +
 894                                         RIGHT.br_blockcount,
 895                                         RIGHT.br_state)))
 896                                 goto done;
 897                 }
 898
 899                 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
 900                         startblockval(PREV.br_startblock));
 901                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
 902                 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
 903                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
 904
 905                 ++*idx;
 906                 *dnew = temp;
 907                 break;
 908
 909         case BMAP_RIGHT_FILLING:
 910                 /*
 911                  * Filling in the last part of a previous delayed allocation.
 912                  * The right neighbor is not contiguous.
 913                  */
 914                 temp = PREV.br_blockcount - new->br_blockcount;
 915                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
 916                 xfs_bmbt_set_blockcount(ep, temp);
 917                 xfs_iext_insert(ip, *idx + 1, 1, new, state);
 918                 ip->i_d.di_nextents++;
 919                 if (cur == NULL)
 920                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
 921                 else {
 922                         rval = XFS_ILOG_CORE;
 923                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
 924                                         new->br_startblock, new->br_blockcount,
 925                                         &i)))
 926                                 goto done;
 927                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
 928                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
 929                         if ((error = xfs_btree_insert(cur, &i)))
 930                                 goto done;
 931                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 932                 }
 933                 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
 934                     ip->i_d.di_nextents > ip->i_df.if_ext_max) {
 935                         error = xfs_bmap_extents_to_btree(tp, ip,
 936                                 first, flist, &cur, 1, &tmp_rval,
 937                                 XFS_DATA_FORK);
 938                         rval |= tmp_rval;
 939                         if (error)
 940                                 goto done;
 941                 }
 942                 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
 943                         startblockval(PREV.br_startblock) -
 944                         (cur ? cur->bc_private.b.allocated : 0));
 945                 ep = xfs_iext_get_ext(ifp, *idx);
 946                 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
 947                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
 948
 949                 ++*idx;
 950                 *dnew = temp;
 951                 break;
 952
 953         case 0:
 954                 /*
 955                  * Filling in the middle part of a previous delayed allocation.
 956                  * Contiguity is impossible here.
 957                  * This case is avoided almost all the time.
 958                  *
 959                  * We start with a delayed allocation:
 960                  *
 961                  * +ddddddddddddddddddddddddddddddddddddddddddddddddddddddd+
 962                  *  PREV @ idx
 963                  *
 964                  * and we are allocating:
 965                  *                     +rrrrrrrrrrrrrrrrr+
 966                  *                            new
 967                  *
 968                  * and we set it up for insertion as:
 969                  * +ddddddddddddddddddd+rrrrrrrrrrrrrrrrr+ddddddddddddddddd+
 970                  *                            new
 971                  *  PREV @ idx          LEFT              RIGHT
 972                  *                      inserted at idx + 1
 973                  */
 974                 temp = new->br_startoff - PREV.br_startoff;
 975                 temp2 = PREV.br_startoff + PREV.br_blockcount - new_endoff;
 976                 trace_xfs_bmap_pre_update(ip, *idx, 0, _THIS_IP_);
 977                 xfs_bmbt_set_blockcount(ep, temp);      /* truncate PREV */
 978                 LEFT = *new;
 979                 RIGHT.br_state = PREV.br_state;
 980                 RIGHT.br_startblock = nullstartblock(
 981                                 (int)xfs_bmap_worst_indlen(ip, temp2));
 982                 RIGHT.br_startoff = new_endoff;
 983                 RIGHT.br_blockcount = temp2;
 984                 /* insert LEFT (r[0]) and RIGHT (r[1]) at the same time */
 985                 xfs_iext_insert(ip, *idx + 1, 2, &LEFT, state);
 986                 ip->i_d.di_nextents++;
 987                 if (cur == NULL)
 988                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
 989                 else {
 990                         rval = XFS_ILOG_CORE;
 991                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
 992                                         new->br_startblock, new->br_blockcount,
 993                                         &i)))
 994                                 goto done;
 995                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
 996                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
 997                         if ((error = xfs_btree_insert(cur, &i)))
 998                                 goto done;
 999                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1000                 }
1001                 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1002                     ip->i_d.di_nextents > ip->i_df.if_ext_max) {
1003                         error = xfs_bmap_extents_to_btree(tp, ip,
1004                                         first, flist, &cur, 1, &tmp_rval,
1005                                         XFS_DATA_FORK);
1006                         rval |= tmp_rval;
1007                         if (error)
1008                                 goto done;
1009                 }
1010                 temp = xfs_bmap_worst_indlen(ip, temp);
1011                 temp2 = xfs_bmap_worst_indlen(ip, temp2);
1012                 diff = (int)(temp + temp2 - startblockval(PREV.br_startblock) -
1013                         (cur ? cur->bc_private.b.allocated : 0));
1014                 if (diff > 0) {
1015                         error = xfs_icsb_modify_counters(ip->i_mount,
1016                                         XFS_SBS_FDBLOCKS,
1017                                         -((int64_t)diff), 0);
1018                         ASSERT(!error);
1019                         if (error)
1020                                 goto done;
1021                 }
1022
1023                 ep = xfs_iext_get_ext(ifp, *idx);
1024                 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
1025                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1026                 trace_xfs_bmap_pre_update(ip, *idx + 2, state, _THIS_IP_);
1027                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, *idx + 2),
1028                         nullstartblock((int)temp2));
1029                 trace_xfs_bmap_post_update(ip, *idx + 2, state, _THIS_IP_);
1030
1031                 ++*idx;
1032                 *dnew = temp + temp2;
1033                 break;
1034
1035         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1036         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1037         case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
1038         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1039         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1040         case BMAP_LEFT_CONTIG:
1041         case BMAP_RIGHT_CONTIG:
1042                 /*
1043                  * These cases are all impossible.
1044                  */
1045                 ASSERT(0);
1046         }
1047         *curp = cur;
1048 done:
1049         *logflagsp = rval;
1050         return error;
1051 #undef  LEFT
1052 #undef  RIGHT
1053 #undef  PREV
1054 }
1055
1056 /*
1057  * Called by xfs_bmap_add_extent to handle cases converting an unwritten
1058  * allocation to a real allocation or vice versa.
1059  */
1060 STATIC int                              /* error */
1061 xfs_bmap_add_extent_unwritten_real(
1062         xfs_inode_t             *ip,    /* incore inode pointer */
1063         xfs_extnum_t            *idx,   /* extent number to update/insert */
1064         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
1065         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
1066         int                     *logflagsp) /* inode logging flags */
1067 {
1068         xfs_btree_cur_t         *cur;   /* btree cursor */
1069         xfs_bmbt_rec_host_t     *ep;    /* extent entry for idx */
1070         int                     error;  /* error return value */
1071         int                     i;      /* temp state */
1072         xfs_ifork_t             *ifp;   /* inode fork pointer */
1073         xfs_fileoff_t           new_endoff;     /* end offset of new entry */
1074         xfs_exntst_t            newext; /* new extent state */
1075         xfs_exntst_t            oldext; /* old extent state */
1076         xfs_bmbt_irec_t         r[3];   /* neighbor extent entries */
1077                                         /* left is 0, right is 1, prev is 2 */
1078         int                     rval=0; /* return value (logging flags) */
1079         int                     state = 0;/* state bits, accessed thru macros */
1080
1081 #define LEFT            r[0]
1082 #define RIGHT           r[1]
1083 #define PREV            r[2]
1084         /*
1085          * Set up a bunch of variables to make the tests simpler.
1086          */
1087         error = 0;
1088         cur = *curp;
1089         ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
1090         ep = xfs_iext_get_ext(ifp, *idx);
1091         xfs_bmbt_get_all(ep, &PREV);
1092         newext = new->br_state;
1093         oldext = (newext == XFS_EXT_UNWRITTEN) ?
1094                 XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
1095         ASSERT(PREV.br_state == oldext);
1096         new_endoff = new->br_startoff + new->br_blockcount;
1097         ASSERT(PREV.br_startoff <= new->br_startoff);
1098         ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
1099
1100         /*
1101          * Set flags determining what part of the previous oldext allocation
1102          * extent is being replaced by a newext allocation.
1103          */
1104         if (PREV.br_startoff == new->br_startoff)
1105                 state |= BMAP_LEFT_FILLING;
1106         if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
1107                 state |= BMAP_RIGHT_FILLING;
1108
1109         /*
1110          * Check and set flags if this segment has a left neighbor.
1111          * Don't set contiguous if the combined extent would be too large.
1112          */
1113         if (*idx > 0) {
1114                 state |= BMAP_LEFT_VALID;
1115                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &LEFT);
1116
1117                 if (isnullstartblock(LEFT.br_startblock))
1118                         state |= BMAP_LEFT_DELAY;
1119         }
1120
1121         if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
1122             LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
1123             LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
1124             LEFT.br_state == newext &&
1125             LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
1126                 state |= BMAP_LEFT_CONTIG;
1127
1128         /*
1129          * Check and set flags if this segment has a right neighbor.
1130          * Don't set contiguous if the combined extent would be too large.
1131          * Also check for all-three-contiguous being too large.
1132          */
1133         if (*idx < ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1) {
1134                 state |= BMAP_RIGHT_VALID;
1135                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx + 1), &RIGHT);
1136                 if (isnullstartblock(RIGHT.br_startblock))
1137                         state |= BMAP_RIGHT_DELAY;
1138         }
1139
1140         if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
1141             new_endoff == RIGHT.br_startoff &&
1142             new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
1143             newext == RIGHT.br_state &&
1144             new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
1145             ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1146                        BMAP_RIGHT_FILLING)) !=
1147                       (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1148                        BMAP_RIGHT_FILLING) ||
1149              LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
1150                         <= MAXEXTLEN))
1151                 state |= BMAP_RIGHT_CONTIG;
1152
1153         /*
1154          * Switch out based on the FILLING and CONTIG state bits.
1155          */
1156         switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1157                          BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
1158         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1159              BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1160                 /*
1161                  * Setting all of a previous oldext extent to newext.
1162                  * The left and right neighbors are both contiguous with new.
1163                  */
1164                 --*idx;
1165
1166                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1167                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
1168                         LEFT.br_blockcount + PREV.br_blockcount +
1169                         RIGHT.br_blockcount);
1170                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1171
1172                 xfs_iext_remove(ip, *idx + 1, 2, state);
1173                 ip->i_d.di_nextents -= 2;
1174                 if (cur == NULL)
1175                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1176                 else {
1177                         rval = XFS_ILOG_CORE;
1178                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
1179                                         RIGHT.br_startblock,
1180                                         RIGHT.br_blockcount, &i)))
1181                                 goto done;
1182                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1183                         if ((error = xfs_btree_delete(cur, &i)))
1184                                 goto done;
1185                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1186                         if ((error = xfs_btree_decrement(cur, 0, &i)))
1187                                 goto done;
1188                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1189                         if ((error = xfs_btree_delete(cur, &i)))
1190                                 goto done;
1191                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1192                         if ((error = xfs_btree_decrement(cur, 0, &i)))
1193                                 goto done;
1194                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1195                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
1196                                 LEFT.br_startblock,
1197                                 LEFT.br_blockcount + PREV.br_blockcount +
1198                                 RIGHT.br_blockcount, LEFT.br_state)))
1199                                 goto done;
1200                 }
1201                 break;
1202
1203         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1204                 /*
1205                  * Setting all of a previous oldext extent to newext.
1206                  * The left neighbor is contiguous, the right is not.
1207                  */
1208                 --*idx;
1209
1210                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1211                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
1212                         LEFT.br_blockcount + PREV.br_blockcount);
1213                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1214
1215                 xfs_iext_remove(ip, *idx + 1, 1, state);
1216                 ip->i_d.di_nextents--;
1217                 if (cur == NULL)
1218                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1219                 else {
1220                         rval = XFS_ILOG_CORE;
1221                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1222                                         PREV.br_startblock, PREV.br_blockcount,
1223                                         &i)))
1224                                 goto done;
1225                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1226                         if ((error = xfs_btree_delete(cur, &i)))
1227                                 goto done;
1228                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1229                         if ((error = xfs_btree_decrement(cur, 0, &i)))
1230                                 goto done;
1231                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1232                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
1233                                 LEFT.br_startblock,
1234                                 LEFT.br_blockcount + PREV.br_blockcount,
1235                                 LEFT.br_state)))
1236                                 goto done;
1237                 }
1238                 break;
1239
1240         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1241                 /*
1242                  * Setting all of a previous oldext extent to newext.
1243                  * The right neighbor is contiguous, the left is not.
1244                  */
1245                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1246                 xfs_bmbt_set_blockcount(ep,
1247                         PREV.br_blockcount + RIGHT.br_blockcount);
1248                 xfs_bmbt_set_state(ep, newext);
1249                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1250                 xfs_iext_remove(ip, *idx + 1, 1, state);
1251                 ip->i_d.di_nextents--;
1252                 if (cur == NULL)
1253                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1254                 else {
1255                         rval = XFS_ILOG_CORE;
1256                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
1257                                         RIGHT.br_startblock,
1258                                         RIGHT.br_blockcount, &i)))
1259                                 goto done;
1260                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1261                         if ((error = xfs_btree_delete(cur, &i)))
1262                                 goto done;
1263                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1264                         if ((error = xfs_btree_decrement(cur, 0, &i)))
1265                                 goto done;
1266                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1267                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
1268                                 new->br_startblock,
1269                                 new->br_blockcount + RIGHT.br_blockcount,
1270                                 newext)))
1271                                 goto done;
1272                 }
1273                 break;
1274
1275         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
1276                 /*
1277                  * Setting all of a previous oldext extent to newext.
1278                  * Neither the left nor right neighbors are contiguous with
1279                  * the new one.
1280                  */
1281                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1282                 xfs_bmbt_set_state(ep, newext);
1283                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1284
1285                 if (cur == NULL)
1286                         rval = XFS_ILOG_DEXT;
1287                 else {
1288                         rval = 0;
1289                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1290                                         new->br_startblock, new->br_blockcount,
1291                                         &i)))
1292                                 goto done;
1293                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1294                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
1295                                 new->br_startblock, new->br_blockcount,
1296                                 newext)))
1297                                 goto done;
1298                 }
1299                 break;
1300
1301         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
1302                 /*
1303                  * Setting the first part of a previous oldext extent to newext.
1304                  * The left neighbor is contiguous.
1305                  */
1306                 trace_xfs_bmap_pre_update(ip, *idx - 1, state, _THIS_IP_);
1307                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx - 1),
1308                         LEFT.br_blockcount + new->br_blockcount);
1309                 xfs_bmbt_set_startoff(ep,
1310                         PREV.br_startoff + new->br_blockcount);
1311                 trace_xfs_bmap_post_update(ip, *idx - 1, state, _THIS_IP_);
1312
1313                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1314                 xfs_bmbt_set_startblock(ep,
1315                         new->br_startblock + new->br_blockcount);
1316                 xfs_bmbt_set_blockcount(ep,
1317                         PREV.br_blockcount - new->br_blockcount);
1318                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1319
1320                 --*idx;
1321
1322                 if (cur == NULL)
1323                         rval = XFS_ILOG_DEXT;
1324                 else {
1325                         rval = 0;
1326                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1327                                         PREV.br_startblock, PREV.br_blockcount,
1328                                         &i)))
1329                                 goto done;
1330                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1331                         if ((error = xfs_bmbt_update(cur,
1332                                 PREV.br_startoff + new->br_blockcount,
1333                                 PREV.br_startblock + new->br_blockcount,
1334                                 PREV.br_blockcount - new->br_blockcount,
1335                                 oldext)))
1336                                 goto done;
1337                         if ((error = xfs_btree_decrement(cur, 0, &i)))
1338                                 goto done;
1339                         if (xfs_bmbt_update(cur, LEFT.br_startoff,
1340                                 LEFT.br_startblock,
1341                                 LEFT.br_blockcount + new->br_blockcount,
1342                                 LEFT.br_state))
1343                                 goto done;
1344                 }
1345                 break;
1346
1347         case BMAP_LEFT_FILLING:
1348                 /*
1349                  * Setting the first part of a previous oldext extent to newext.
1350                  * The left neighbor is not contiguous.
1351                  */
1352                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1353                 ASSERT(ep && xfs_bmbt_get_state(ep) == oldext);
1354                 xfs_bmbt_set_startoff(ep, new_endoff);
1355                 xfs_bmbt_set_blockcount(ep,
1356                         PREV.br_blockcount - new->br_blockcount);
1357                 xfs_bmbt_set_startblock(ep,
1358                         new->br_startblock + new->br_blockcount);
1359                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1360
1361                 xfs_iext_insert(ip, *idx, 1, new, state);
1362                 ip->i_d.di_nextents++;
1363                 if (cur == NULL)
1364                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1365                 else {
1366                         rval = XFS_ILOG_CORE;
1367                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1368                                         PREV.br_startblock, PREV.br_blockcount,
1369                                         &i)))
1370                                 goto done;
1371                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1372                         if ((error = xfs_bmbt_update(cur,
1373                                 PREV.br_startoff + new->br_blockcount,
1374                                 PREV.br_startblock + new->br_blockcount,
1375                                 PREV.br_blockcount - new->br_blockcount,
1376                                 oldext)))
1377                                 goto done;
1378                         cur->bc_rec.b = *new;
1379                         if ((error = xfs_btree_insert(cur, &i)))
1380                                 goto done;
1381                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1382                 }
1383                 break;
1384
1385         case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1386                 /*
1387                  * Setting the last part of a previous oldext extent to newext.
1388                  * The right neighbor is contiguous with the new allocation.
1389                  */
1390                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1391                 xfs_bmbt_set_blockcount(ep,
1392                         PREV.br_blockcount - new->br_blockcount);
1393                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1394
1395                 ++*idx;
1396
1397                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1398                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx),
1399                         new->br_startoff, new->br_startblock,
1400                         new->br_blockcount + RIGHT.br_blockcount, newext);
1401                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1402
1403                 if (cur == NULL)
1404                         rval = XFS_ILOG_DEXT;
1405                 else {
1406                         rval = 0;
1407                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1408                                         PREV.br_startblock,
1409                                         PREV.br_blockcount, &i)))
1410                                 goto done;
1411                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1412                         if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
1413                                 PREV.br_startblock,
1414                                 PREV.br_blockcount - new->br_blockcount,
1415                                 oldext)))
1416                                 goto done;
1417                         if ((error = xfs_btree_increment(cur, 0, &i)))
1418                                 goto done;
1419                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
1420                                 new->br_startblock,
1421                                 new->br_blockcount + RIGHT.br_blockcount,
1422                                 newext)))
1423                                 goto done;
1424                 }
1425                 break;
1426
1427         case BMAP_RIGHT_FILLING:
1428                 /*
1429                  * Setting the last part of a previous oldext extent to newext.
1430                  * The right neighbor is not contiguous.
1431                  */
1432                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1433                 xfs_bmbt_set_blockcount(ep,
1434                         PREV.br_blockcount - new->br_blockcount);
1435                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1436
1437                 ++*idx;
1438                 xfs_iext_insert(ip, *idx, 1, new, state);
1439
1440                 ip->i_d.di_nextents++;
1441                 if (cur == NULL)
1442                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1443                 else {
1444                         rval = XFS_ILOG_CORE;
1445                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1446                                         PREV.br_startblock, PREV.br_blockcount,
1447                                         &i)))
1448                                 goto done;
1449                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1450                         if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
1451                                 PREV.br_startblock,
1452                                 PREV.br_blockcount - new->br_blockcount,
1453                                 oldext)))
1454                                 goto done;
1455                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1456                                         new->br_startblock, new->br_blockcount,
1457                                         &i)))
1458                                 goto done;
1459                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1460                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
1461                         if ((error = xfs_btree_insert(cur, &i)))
1462                                 goto done;
1463                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1464                 }
1465                 break;
1466
1467         case 0:
1468                 /*
1469                  * Setting the middle part of a previous oldext extent to
1470                  * newext.  Contiguity is impossible here.
1471                  * One extent becomes three extents.
1472                  */
1473                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1474                 xfs_bmbt_set_blockcount(ep,
1475                         new->br_startoff - PREV.br_startoff);
1476                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1477
1478                 r[0] = *new;
1479                 r[1].br_startoff = new_endoff;
1480                 r[1].br_blockcount =
1481                         PREV.br_startoff + PREV.br_blockcount - new_endoff;
1482                 r[1].br_startblock = new->br_startblock + new->br_blockcount;
1483                 r[1].br_state = oldext;
1484
1485                 ++*idx;
1486                 xfs_iext_insert(ip, *idx, 2, &r[0], state);
1487
1488                 ip->i_d.di_nextents += 2;
1489                 if (cur == NULL)
1490                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1491                 else {
1492                         rval = XFS_ILOG_CORE;
1493                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1494                                         PREV.br_startblock, PREV.br_blockcount,
1495                                         &i)))
1496                                 goto done;
1497                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1498                         /* new right extent - oldext */
1499                         if ((error = xfs_bmbt_update(cur, r[1].br_startoff,
1500                                 r[1].br_startblock, r[1].br_blockcount,
1501                                 r[1].br_state)))
1502                                 goto done;
1503                         /* new left extent - oldext */
1504                         cur->bc_rec.b = PREV;
1505                         cur->bc_rec.b.br_blockcount =
1506                                 new->br_startoff - PREV.br_startoff;
1507                         if ((error = xfs_btree_insert(cur, &i)))
1508                                 goto done;
1509                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1510                         /*
1511                          * Reset the cursor to the position of the new extent
1512                          * we are about to insert as we can't trust it after
1513                          * the previous insert.
1514                          */
1515                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1516                                         new->br_startblock, new->br_blockcount,
1517                                         &i)))
1518                                 goto done;
1519                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1520                         /* new middle extent - newext */
1521                         cur->bc_rec.b.br_state = new->br_state;
1522                         if ((error = xfs_btree_insert(cur, &i)))
1523                                 goto done;
1524                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1525                 }
1526                 break;
1527
1528         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1529         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1530         case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
1531         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1532         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1533         case BMAP_LEFT_CONTIG:
1534         case BMAP_RIGHT_CONTIG:
1535                 /*
1536                  * These cases are all impossible.
1537                  */
1538                 ASSERT(0);
1539         }
1540         *curp = cur;
1541 done:
1542         *logflagsp = rval;
1543         return error;
1544 #undef  LEFT
1545 #undef  RIGHT
1546 #undef  PREV
1547 }
1548
1549 /*
1550  * Convert a hole to a delayed allocation.
1551  */
1552 STATIC void
1553 xfs_bmap_add_extent_hole_delay(
1554         xfs_inode_t             *ip,    /* incore inode pointer */
1555         xfs_extnum_t            *idx,   /* extent number to update/insert */
1556         xfs_bmbt_irec_t         *new)   /* new data to add to file extents */
1557 {
1558         xfs_ifork_t             *ifp;   /* inode fork pointer */
1559         xfs_bmbt_irec_t         left;   /* left neighbor extent entry */
1560         xfs_filblks_t           newlen=0;       /* new indirect size */
1561         xfs_filblks_t           oldlen=0;       /* old indirect size */
1562         xfs_bmbt_irec_t         right;  /* right neighbor extent entry */
1563         int                     state;  /* state bits, accessed thru macros */
1564         xfs_filblks_t           temp=0; /* temp for indirect calculations */
1565
1566         ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
1567         state = 0;
1568         ASSERT(isnullstartblock(new->br_startblock));
1569
1570         /*
1571          * Check and set flags if this segment has a left neighbor
1572          */
1573         if (*idx > 0) {
1574                 state |= BMAP_LEFT_VALID;
1575                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &left);
1576
1577                 if (isnullstartblock(left.br_startblock))
1578                         state |= BMAP_LEFT_DELAY;
1579         }
1580
1581         /*
1582          * Check and set flags if the current (right) segment exists.
1583          * If it doesn't exist, we're converting the hole at end-of-file.
1584          */
1585         if (*idx < ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t)) {
1586                 state |= BMAP_RIGHT_VALID;
1587                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx), &right);
1588
1589                 if (isnullstartblock(right.br_startblock))
1590                         state |= BMAP_RIGHT_DELAY;
1591         }
1592
1593         /*
1594          * Set contiguity flags on the left and right neighbors.
1595          * Don't let extents get too large, even if the pieces are contiguous.
1596          */
1597         if ((state & BMAP_LEFT_VALID) && (state & BMAP_LEFT_DELAY) &&
1598             left.br_startoff + left.br_blockcount == new->br_startoff &&
1599             left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
1600                 state |= BMAP_LEFT_CONTIG;
1601
1602         if ((state & BMAP_RIGHT_VALID) && (state & BMAP_RIGHT_DELAY) &&
1603             new->br_startoff + new->br_blockcount == right.br_startoff &&
1604             new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
1605             (!(state & BMAP_LEFT_CONTIG) ||
1606              (left.br_blockcount + new->br_blockcount +
1607               right.br_blockcount <= MAXEXTLEN)))
1608                 state |= BMAP_RIGHT_CONTIG;
1609
1610         /*
1611          * Switch out based on the contiguity flags.
1612          */
1613         switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
1614         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1615                 /*
1616                  * New allocation is contiguous with delayed allocations
1617                  * on the left and on the right.
1618                  * Merge all three into a single extent record.
1619                  */
1620                 --*idx;
1621                 temp = left.br_blockcount + new->br_blockcount +
1622                         right.br_blockcount;
1623
1624                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1625                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx), temp);
1626                 oldlen = startblockval(left.br_startblock) +
1627                         startblockval(new->br_startblock) +
1628                         startblockval(right.br_startblock);
1629                 newlen = xfs_bmap_worst_indlen(ip, temp);
1630                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, *idx),
1631                         nullstartblock((int)newlen));
1632                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1633
1634                 xfs_iext_remove(ip, *idx + 1, 1, state);
1635                 break;
1636
1637         case BMAP_LEFT_CONTIG:
1638                 /*
1639                  * New allocation is contiguous with a delayed allocation
1640                  * on the left.
1641                  * Merge the new allocation with the left neighbor.
1642                  */
1643                 --*idx;
1644                 temp = left.br_blockcount + new->br_blockcount;
1645
1646                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1647                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx), temp);
1648                 oldlen = startblockval(left.br_startblock) +
1649                         startblockval(new->br_startblock);
1650                 newlen = xfs_bmap_worst_indlen(ip, temp);
1651                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, *idx),
1652                         nullstartblock((int)newlen));
1653                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1654                 break;
1655
1656         case BMAP_RIGHT_CONTIG:
1657                 /*
1658                  * New allocation is contiguous with a delayed allocation
1659                  * on the right.
1660                  * Merge the new allocation with the right neighbor.
1661                  */
1662                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1663                 temp = new->br_blockcount + right.br_blockcount;
1664                 oldlen = startblockval(new->br_startblock) +
1665                         startblockval(right.br_startblock);
1666                 newlen = xfs_bmap_worst_indlen(ip, temp);
1667                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx),
1668                         new->br_startoff,
1669                         nullstartblock((int)newlen), temp, right.br_state);
1670                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1671                 break;
1672
1673         case 0:
1674                 /*
1675                  * New allocation is not contiguous with another
1676                  * delayed allocation.
1677                  * Insert a new entry.
1678                  */
1679                 oldlen = newlen = 0;
1680                 xfs_iext_insert(ip, *idx, 1, new, state);
1681                 break;
1682         }
1683         if (oldlen != newlen) {
1684                 ASSERT(oldlen > newlen);
1685                 xfs_icsb_modify_counters(ip->i_mount, XFS_SBS_FDBLOCKS,
1686                         (int64_t)(oldlen - newlen), 0);
1687                 /*
1688                  * Nothing to do for disk quota accounting here.
1689                  */
1690         }
1691 }
1692
1693 /*
1694  * Called by xfs_bmap_add_extent to handle cases converting a hole
1695  * to a real allocation.
1696  */
1697 STATIC int                              /* error */
1698 xfs_bmap_add_extent_hole_real(
1699         xfs_inode_t             *ip,    /* incore inode pointer */
1700         xfs_extnum_t            *idx,   /* extent number to update/insert */
1701         xfs_btree_cur_t         *cur,   /* if null, not a btree */
1702         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
1703         int                     *logflagsp, /* inode logging flags */
1704         int                     whichfork) /* data or attr fork */
1705 {
1706         int                     error;  /* error return value */
1707         int                     i;      /* temp state */
1708         xfs_ifork_t             *ifp;   /* inode fork pointer */
1709         xfs_bmbt_irec_t         left;   /* left neighbor extent entry */
1710         xfs_bmbt_irec_t         right;  /* right neighbor extent entry */
1711         int                     rval=0; /* return value (logging flags) */
1712         int                     state;  /* state bits, accessed thru macros */
1713
1714         ifp = XFS_IFORK_PTR(ip, whichfork);
1715         ASSERT(*idx <= ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t));
1716         state = 0;
1717
1718         if (whichfork == XFS_ATTR_FORK)
1719                 state |= BMAP_ATTRFORK;
1720
1721         /*
1722          * Check and set flags if this segment has a left neighbor.
1723          */
1724         if (*idx > 0) {
1725                 state |= BMAP_LEFT_VALID;
1726                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &left);
1727                 if (isnullstartblock(left.br_startblock))
1728                         state |= BMAP_LEFT_DELAY;
1729         }
1730
1731         /*
1732          * Check and set flags if this segment has a current value.
1733          * Not true if we're inserting into the "hole" at eof.
1734          */
1735         if (*idx < ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)) {
1736                 state |= BMAP_RIGHT_VALID;
1737                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx), &right);
1738                 if (isnullstartblock(right.br_startblock))
1739                         state |= BMAP_RIGHT_DELAY;
1740         }
1741
1742         /*
1743          * We're inserting a real allocation between "left" and "right".
1744          * Set the contiguity flags.  Don't let extents get too large.
1745          */
1746         if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
1747             left.br_startoff + left.br_blockcount == new->br_startoff &&
1748             left.br_startblock + left.br_blockcount == new->br_startblock &&
1749             left.br_state == new->br_state &&
1750             left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
1751                 state |= BMAP_LEFT_CONTIG;
1752
1753         if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
1754             new->br_startoff + new->br_blockcount == right.br_startoff &&
1755             new->br_startblock + new->br_blockcount == right.br_startblock &&
1756             new->br_state == right.br_state &&
1757             new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
1758             (!(state & BMAP_LEFT_CONTIG) ||
1759              left.br_blockcount + new->br_blockcount +
1760              right.br_blockcount <= MAXEXTLEN))
1761                 state |= BMAP_RIGHT_CONTIG;
1762
1763         error = 0;
1764         /*
1765          * Select which case we're in here, and implement it.
1766          */
1767         switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
1768         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1769                 /*
1770                  * New allocation is contiguous with real allocations on the
1771                  * left and on the right.
1772                  * Merge all three into a single extent record.
1773                  */
1774                 --*idx;
1775                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1776                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
1777                         left.br_blockcount + new->br_blockcount +
1778                         right.br_blockcount);
1779                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1780
1781                 xfs_iext_remove(ip, *idx + 1, 1, state);
1782
1783                 XFS_IFORK_NEXT_SET(ip, whichfork,
1784                         XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
1785                 if (cur == NULL) {
1786                         rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
1787                 } else {
1788                         rval = XFS_ILOG_CORE;
1789                         if ((error = xfs_bmbt_lookup_eq(cur,
1790                                         right.br_startoff,
1791                                         right.br_startblock,
1792                                         right.br_blockcount, &i)))
1793                                 goto done;
1794                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1795                         if ((error = xfs_btree_delete(cur, &i)))
1796                                 goto done;
1797                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1798                         if ((error = xfs_btree_decrement(cur, 0, &i)))
1799                                 goto done;
1800                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1801                         if ((error = xfs_bmbt_update(cur, left.br_startoff,
1802                                         left.br_startblock,
1803                                         left.br_blockcount +
1804                                                 new->br_blockcount +
1805                                                 right.br_blockcount,
1806                                         left.br_state)))
1807                                 goto done;
1808                 }
1809                 break;
1810
1811         case BMAP_LEFT_CONTIG:
1812                 /*
1813                  * New allocation is contiguous with a real allocation
1814                  * on the left.
1815                  * Merge the new allocation with the left neighbor.
1816                  */
1817                 --*idx;
1818                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1819                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
1820                         left.br_blockcount + new->br_blockcount);
1821                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1822
1823                 if (cur == NULL) {
1824                         rval = xfs_ilog_fext(whichfork);
1825                 } else {
1826                         rval = 0;
1827                         if ((error = xfs_bmbt_lookup_eq(cur,
1828                                         left.br_startoff,
1829                                         left.br_startblock,
1830                                         left.br_blockcount, &i)))
1831                                 goto done;
1832                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1833                         if ((error = xfs_bmbt_update(cur, left.br_startoff,
1834                                         left.br_startblock,
1835                                         left.br_blockcount +
1836                                                 new->br_blockcount,
1837                                         left.br_state)))
1838                                 goto done;
1839                 }
1840                 break;
1841
1842         case BMAP_RIGHT_CONTIG:
1843                 /*
1844                  * New allocation is contiguous with a real allocation
1845                  * on the right.
1846                  * Merge the new allocation with the right neighbor.
1847                  */
1848                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1849                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx),
1850                         new->br_startoff, new->br_startblock,
1851                         new->br_blockcount + right.br_blockcount,
1852                         right.br_state);
1853                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1854
1855                 if (cur == NULL) {
1856                         rval = xfs_ilog_fext(whichfork);
1857                 } else {
1858                         rval = 0;
1859                         if ((error = xfs_bmbt_lookup_eq(cur,
1860                                         right.br_startoff,
1861                                         right.br_startblock,
1862                                         right.br_blockcount, &i)))
1863                                 goto done;
1864                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1865                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
1866                                         new->br_startblock,
1867                                         new->br_blockcount +
1868                                                 right.br_blockcount,
1869                                         right.br_state)))
1870                                 goto done;
1871                 }
1872                 break;
1873
1874         case 0:
1875                 /*
1876                  * New allocation is not contiguous with another
1877                  * real allocation.
1878                  * Insert a new entry.
1879                  */
1880                 xfs_iext_insert(ip, *idx, 1, new, state);
1881                 XFS_IFORK_NEXT_SET(ip, whichfork,
1882                         XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
1883                 if (cur == NULL) {
1884                         rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
1885                 } else {
1886                         rval = XFS_ILOG_CORE;
1887                         if ((error = xfs_bmbt_lookup_eq(cur,
1888                                         new->br_startoff,
1889                                         new->br_startblock,
1890                                         new->br_blockcount, &i)))
1891                                 goto done;
1892                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1893                         cur->bc_rec.b.br_state = new->br_state;
1894                         if ((error = xfs_btree_insert(cur, &i)))
1895                                 goto done;
1896                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1897                 }
1898                 break;
1899         }
1900 done:
1901         *logflagsp = rval;
1902         return error;
1903 }
1904
1905 /*
1906  * Adjust the size of the new extent based on di_extsize and rt extsize.
1907  */
1908 STATIC int
1909 xfs_bmap_extsize_align(
1910         xfs_mount_t     *mp,
1911         xfs_bmbt_irec_t *gotp,          /* next extent pointer */
1912         xfs_bmbt_irec_t *prevp,         /* previous extent pointer */
1913         xfs_extlen_t    extsz,          /* align to this extent size */
1914         int             rt,             /* is this a realtime inode? */
1915         int             eof,            /* is extent at end-of-file? */
1916         int             delay,          /* creating delalloc extent? */
1917         int             convert,        /* overwriting unwritten extent? */
1918         xfs_fileoff_t   *offp,          /* in/out: aligned offset */
1919         xfs_extlen_t    *lenp)          /* in/out: aligned length */
1920 {
1921         xfs_fileoff_t   orig_off;       /* original offset */
1922         xfs_extlen_t    orig_alen;      /* original length */
1923         xfs_fileoff_t   orig_end;       /* original off+len */
1924         xfs_fileoff_t   nexto;          /* next file offset */
1925         xfs_fileoff_t   prevo;          /* previous file offset */
1926         xfs_fileoff_t   align_off;      /* temp for offset */
1927         xfs_extlen_t    align_alen;     /* temp for length */
1928         xfs_extlen_t    temp;           /* temp for calculations */
1929
1930         if (convert)
1931                 return 0;
1932
1933         orig_off = align_off = *offp;
1934         orig_alen = align_alen = *lenp;
1935         orig_end = orig_off + orig_alen;
1936
1937         /*
1938          * If this request overlaps an existing extent, then don't
1939          * attempt to perform any additional alignment.
1940          */
1941         if (!delay && !eof &&
1942             (orig_off >= gotp->br_startoff) &&
1943             (orig_end <= gotp->br_startoff + gotp->br_blockcount)) {
1944                 return 0;
1945         }
1946
1947         /*
1948          * If the file offset is unaligned vs. the extent size
1949          * we need to align it.  This will be possible unless
1950          * the file was previously written with a kernel that didn't
1951          * perform this alignment, or if a truncate shot us in the
1952          * foot.
1953          */
1954         temp = do_mod(orig_off, extsz);
1955         if (temp) {
1956                 align_alen += temp;
1957                 align_off -= temp;
1958         }
1959         /*
1960          * Same adjustment for the end of the requested area.
1961          */
1962         if ((temp = (align_alen % extsz))) {
1963                 align_alen += extsz - temp;
1964         }
1965         /*
1966          * If the previous block overlaps with this proposed allocation
1967          * then move the start forward without adjusting the length.
1968          */
1969         if (prevp->br_startoff != NULLFILEOFF) {
1970                 if (prevp->br_startblock == HOLESTARTBLOCK)
1971                         prevo = prevp->br_startoff;
1972                 else
1973                         prevo = prevp->br_startoff + prevp->br_blockcount;
1974         } else
1975                 prevo = 0;
1976         if (align_off != orig_off && align_off < prevo)
1977                 align_off = prevo;
1978         /*
1979          * If the next block overlaps with this proposed allocation
1980          * then move the start back without adjusting the length,
1981          * but not before offset 0.
1982          * This may of course make the start overlap previous block,
1983          * and if we hit the offset 0 limit then the next block
1984          * can still overlap too.
1985          */
1986         if (!eof && gotp->br_startoff != NULLFILEOFF) {
1987                 if ((delay && gotp->br_startblock == HOLESTARTBLOCK) ||
1988                     (!delay && gotp->br_startblock == DELAYSTARTBLOCK))
1989                         nexto = gotp->br_startoff + gotp->br_blockcount;
1990                 else
1991                         nexto = gotp->br_startoff;
1992         } else
1993                 nexto = NULLFILEOFF;
1994         if (!eof &&
1995             align_off + align_alen != orig_end &&
1996             align_off + align_alen > nexto)
1997                 align_off = nexto > align_alen ? nexto - align_alen : 0;
1998         /*
1999          * If we're now overlapping the next or previous extent that
2000          * means we can't fit an extsz piece in this hole.  Just move
2001          * the start forward to the first valid spot and set
2002          * the length so we hit the end.
2003          */
2004         if (align_off != orig_off && align_off < prevo)
2005                 align_off = prevo;
2006         if (align_off + align_alen != orig_end &&
2007             align_off + align_alen > nexto &&
2008             nexto != NULLFILEOFF) {
2009                 ASSERT(nexto > prevo);
2010                 align_alen = nexto - align_off;
2011         }
2012
2013         /*
2014          * If realtime, and the result isn't a multiple of the realtime
2015          * extent size we need to remove blocks until it is.
2016          */
2017         if (rt && (temp = (align_alen % mp->m_sb.sb_rextsize))) {
2018                 /*
2019                  * We're not covering the original request, or
2020                  * we won't be able to once we fix the length.
2021                  */
2022                 if (orig_off < align_off ||
2023                     orig_end > align_off + align_alen ||
2024                     align_alen - temp < orig_alen)
2025                         return XFS_ERROR(EINVAL);
2026                 /*
2027                  * Try to fix it by moving the start up.
2028                  */
2029                 if (align_off + temp <= orig_off) {
2030                         align_alen -= temp;
2031                         align_off += temp;
2032                 }
2033                 /*
2034                  * Try to fix it by moving the end in.
2035                  */
2036                 else if (align_off + align_alen - temp >= orig_end)
2037                         align_alen -= temp;
2038                 /*
2039                  * Set the start to the minimum then trim the length.
2040                  */
2041                 else {
2042                         align_alen -= orig_off - align_off;
2043                         align_off = orig_off;
2044                         align_alen -= align_alen % mp->m_sb.sb_rextsize;
2045                 }
2046                 /*
2047                  * Result doesn't cover the request, fail it.
2048                  */
2049                 if (orig_off < align_off || orig_end > align_off + align_alen)
2050                         return XFS_ERROR(EINVAL);
2051         } else {
2052                 ASSERT(orig_off >= align_off);
2053                 ASSERT(orig_end <= align_off + align_alen);
2054         }
2055
2056 #ifdef DEBUG
2057         if (!eof && gotp->br_startoff != NULLFILEOFF)
2058                 ASSERT(align_off + align_alen <= gotp->br_startoff);
2059         if (prevp->br_startoff != NULLFILEOFF)
2060                 ASSERT(align_off >= prevp->br_startoff + prevp->br_blockcount);
2061 #endif
2062
2063         *lenp = align_alen;
2064         *offp = align_off;
2065         return 0;
2066 }
2067
2068 #define XFS_ALLOC_GAP_UNITS     4
2069
2070 STATIC void
2071 xfs_bmap_adjacent(
2072         xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
2073 {
2074         xfs_fsblock_t   adjust;         /* adjustment to block numbers */
2075         xfs_agnumber_t  fb_agno;        /* ag number of ap->firstblock */
2076         xfs_mount_t     *mp;            /* mount point structure */
2077         int             nullfb;         /* true if ap->firstblock isn't set */
2078         int             rt;             /* true if inode is realtime */
2079
2080 #define ISVALID(x,y)    \
2081         (rt ? \
2082                 (x) < mp->m_sb.sb_rblocks : \
2083                 XFS_FSB_TO_AGNO(mp, x) == XFS_FSB_TO_AGNO(mp, y) && \
2084                 XFS_FSB_TO_AGNO(mp, x) < mp->m_sb.sb_agcount && \
2085                 XFS_FSB_TO_AGBNO(mp, x) < mp->m_sb.sb_agblocks)
2086
2087         mp = ap->ip->i_mount;
2088         nullfb = ap->firstblock == NULLFSBLOCK;
2089         rt = XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata;
2090         fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, ap->firstblock);
2091         /*
2092          * If allocating at eof, and there's a previous real block,
2093          * try to use its last block as our starting point.
2094          */
2095         if (ap->eof && ap->prevp->br_startoff != NULLFILEOFF &&
2096             !isnullstartblock(ap->prevp->br_startblock) &&
2097             ISVALID(ap->prevp->br_startblock + ap->prevp->br_blockcount,
2098                     ap->prevp->br_startblock)) {
2099                 ap->rval = ap->prevp->br_startblock + ap->prevp->br_blockcount;
2100                 /*
2101                  * Adjust for the gap between prevp and us.
2102                  */
2103                 adjust = ap->off -
2104                         (ap->prevp->br_startoff + ap->prevp->br_blockcount);
2105                 if (adjust &&
2106                     ISVALID(ap->rval + adjust, ap->prevp->br_startblock))
2107                         ap->rval += adjust;
2108         }
2109         /*
2110          * If not at eof, then compare the two neighbor blocks.
2111          * Figure out whether either one gives us a good starting point,
2112          * and pick the better one.
2113          */
2114         else if (!ap->eof) {
2115                 xfs_fsblock_t   gotbno;         /* right side block number */
2116                 xfs_fsblock_t   gotdiff=0;      /* right side difference */
2117                 xfs_fsblock_t   prevbno;        /* left side block number */
2118                 xfs_fsblock_t   prevdiff=0;     /* left side difference */
2119
2120                 /*
2121                  * If there's a previous (left) block, select a requested
2122                  * start block based on it.
2123                  */
2124                 if (ap->prevp->br_startoff != NULLFILEOFF &&
2125                     !isnullstartblock(ap->prevp->br_startblock) &&
2126                     (prevbno = ap->prevp->br_startblock +
2127                                ap->prevp->br_blockcount) &&
2128                     ISVALID(prevbno, ap->prevp->br_startblock)) {
2129                         /*
2130                          * Calculate gap to end of previous block.
2131                          */
2132                         adjust = prevdiff = ap->off -
2133                                 (ap->prevp->br_startoff +
2134                                  ap->prevp->br_blockcount);
2135                         /*
2136                          * Figure the startblock based on the previous block's
2137                          * end and the gap size.
2138                          * Heuristic!
2139                          * If the gap is large relative to the piece we're
2140                          * allocating, or using it gives us an invalid block
2141                          * number, then just use the end of the previous block.
2142                          */
2143                         if (prevdiff <= XFS_ALLOC_GAP_UNITS * ap->alen &&
2144                             ISVALID(prevbno + prevdiff,
2145                                     ap->prevp->br_startblock))
2146                                 prevbno += adjust;
2147                         else
2148                                 prevdiff += adjust;
2149                         /*
2150                          * If the firstblock forbids it, can't use it,
2151                          * must use default.
2152                          */
2153                         if (!rt && !nullfb &&
2154                             XFS_FSB_TO_AGNO(mp, prevbno) != fb_agno)
2155                                 prevbno = NULLFSBLOCK;
2156                 }
2157                 /*
2158                  * No previous block or can't follow it, just default.
2159                  */
2160                 else
2161                         prevbno = NULLFSBLOCK;
2162                 /*
2163                  * If there's a following (right) block, select a requested
2164                  * start block based on it.
2165                  */
2166                 if (!isnullstartblock(ap->gotp->br_startblock)) {
2167                         /*
2168                          * Calculate gap to start of next block.
2169                          */
2170                         adjust = gotdiff = ap->gotp->br_startoff - ap->off;
2171                         /*
2172                          * Figure the startblock based on the next block's
2173                          * start and the gap size.
2174                          */
2175                         gotbno = ap->gotp->br_startblock;
2176                         /*
2177                          * Heuristic!
2178                          * If the gap is large relative to the piece we're
2179                          * allocating, or using it gives us an invalid block
2180                          * number, then just use the start of the next block
2181                          * offset by our length.
2182                          */
2183                         if (gotdiff <= XFS_ALLOC_GAP_UNITS * ap->alen &&
2184                             ISVALID(gotbno - gotdiff, gotbno))
2185                                 gotbno -= adjust;
2186                         else if (ISVALID(gotbno - ap->alen, gotbno)) {
2187                                 gotbno -= ap->alen;
2188                                 gotdiff += adjust - ap->alen;
2189                         } else
2190                                 gotdiff += adjust;
2191                         /*
2192                          * If the firstblock forbids it, can't use it,
2193                          * must use default.
2194                          */
2195                         if (!rt && !nullfb &&
2196                             XFS_FSB_TO_AGNO(mp, gotbno) != fb_agno)
2197                                 gotbno = NULLFSBLOCK;
2198                 }
2199                 /*
2200                  * No next block, just default.
2201                  */
2202                 else
2203                         gotbno = NULLFSBLOCK;
2204                 /*
2205                  * If both valid, pick the better one, else the only good
2206                  * one, else ap->rval is already set (to 0 or the inode block).
2207                  */
2208                 if (prevbno != NULLFSBLOCK && gotbno != NULLFSBLOCK)
2209                         ap->rval = prevdiff <= gotdiff ? prevbno : gotbno;
2210                 else if (prevbno != NULLFSBLOCK)
2211                         ap->rval = prevbno;
2212                 else if (gotbno != NULLFSBLOCK)
2213                         ap->rval = gotbno;
2214         }
2215 #undef ISVALID
2216 }
2217
2218 STATIC int
2219 xfs_bmap_rtalloc(
2220         xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
2221 {
2222         xfs_alloctype_t atype = 0;      /* type for allocation routines */
2223         int             error;          /* error return value */
2224         xfs_mount_t     *mp;            /* mount point structure */
2225         xfs_extlen_t    prod = 0;       /* product factor for allocators */
2226         xfs_extlen_t    ralen = 0;      /* realtime allocation length */
2227         xfs_extlen_t    align;          /* minimum allocation alignment */
2228         xfs_rtblock_t   rtb;
2229
2230         mp = ap->ip->i_mount;
2231         align = xfs_get_extsz_hint(ap->ip);
2232         prod = align / mp->m_sb.sb_rextsize;
2233         error = xfs_bmap_extsize_align(mp, ap->gotp, ap->prevp,
2234                                         align, 1, ap->eof, 0,
2235                                         ap->conv, &ap->off, &ap->alen);
2236         if (error)
2237                 return error;
2238         ASSERT(ap->alen);
2239         ASSERT(ap->alen % mp->m_sb.sb_rextsize == 0);
2240
2241         /*
2242          * If the offset & length are not perfectly aligned
2243          * then kill prod, it will just get us in trouble.
2244          */
2245         if (do_mod(ap->off, align) || ap->alen % align)
2246                 prod = 1;
2247         /*
2248          * Set ralen to be the actual requested length in rtextents.
2249          */
2250         ralen = ap->alen / mp->m_sb.sb_rextsize;
2251         /*
2252          * If the old value was close enough to MAXEXTLEN that
2253          * we rounded up to it, cut it back so it's valid again.
2254          * Note that if it's a really large request (bigger than
2255          * MAXEXTLEN), we don't hear about that number, and can't
2256          * adjust the starting point to match it.
2257          */
2258         if (ralen * mp->m_sb.sb_rextsize >= MAXEXTLEN)
2259                 ralen = MAXEXTLEN / mp->m_sb.sb_rextsize;
2260
2261         /*
2262          * Lock out other modifications to the RT bitmap inode.
2263          */
2264         xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL);
2265         xfs_trans_ijoin_ref(ap->tp, mp->m_rbmip, XFS_ILOCK_EXCL);
2266
2267         /*
2268          * If it's an allocation to an empty file at offset 0,
2269          * pick an extent that will space things out in the rt area.
2270          */
2271         if (ap->eof && ap->off == 0) {
2272                 xfs_rtblock_t uninitialized_var(rtx); /* realtime extent no */
2273
2274                 error = xfs_rtpick_extent(mp, ap->tp, ralen, &rtx);
2275                 if (error)
2276                         return error;
2277                 ap->rval = rtx * mp->m_sb.sb_rextsize;
2278         } else {
2279                 ap->rval = 0;
2280         }
2281
2282         xfs_bmap_adjacent(ap);
2283
2284         /*
2285          * Realtime allocation, done through xfs_rtallocate_extent.
2286          */
2287         atype = ap->rval == 0 ?  XFS_ALLOCTYPE_ANY_AG : XFS_ALLOCTYPE_NEAR_BNO;
2288         do_div(ap->rval, mp->m_sb.sb_rextsize);
2289         rtb = ap->rval;
2290         ap->alen = ralen;
2291         if ((error = xfs_rtallocate_extent(ap->tp, ap->rval, 1, ap->alen,
2292                                 &ralen, atype, ap->wasdel, prod, &rtb)))
2293                 return error;
2294         if (rtb == NULLFSBLOCK && prod > 1 &&
2295             (error = xfs_rtallocate_extent(ap->tp, ap->rval, 1,
2296                                            ap->alen, &ralen, atype,
2297                                            ap->wasdel, 1, &rtb)))
2298                 return error;
2299         ap->rval = rtb;
2300         if (ap->rval != NULLFSBLOCK) {
2301                 ap->rval *= mp->m_sb.sb_rextsize;
2302                 ralen *= mp->m_sb.sb_rextsize;
2303                 ap->alen = ralen;
2304                 ap->ip->i_d.di_nblocks += ralen;
2305                 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
2306                 if (ap->wasdel)
2307                         ap->ip->i_delayed_blks -= ralen;
2308                 /*
2309                  * Adjust the disk quota also. This was reserved
2310                  * earlier.
2311                  */
2312                 xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
2313                         ap->wasdel ? XFS_TRANS_DQ_DELRTBCOUNT :
2314                                         XFS_TRANS_DQ_RTBCOUNT, (long) ralen);
2315         } else {
2316                 ap->alen = 0;
2317         }
2318         return 0;
2319 }
2320
2321 STATIC int
2322 xfs_bmap_btalloc_nullfb(
2323         struct xfs_bmalloca     *ap,
2324         struct xfs_alloc_arg    *args,
2325         xfs_extlen_t            *blen)
2326 {
2327         struct xfs_mount        *mp = ap->ip->i_mount;
2328         struct xfs_perag        *pag;
2329         xfs_agnumber_t          ag, startag;
2330         int                     notinit = 0;
2331         int                     error;
2332
2333         if (ap->userdata && xfs_inode_is_filestream(ap->ip))
2334                 args->type = XFS_ALLOCTYPE_NEAR_BNO;
2335         else
2336                 args->type = XFS_ALLOCTYPE_START_BNO;
2337         args->total = ap->total;
2338
2339         /*
2340          * Search for an allocation group with a single extent large enough
2341          * for the request.  If one isn't found, then adjust the minimum
2342          * allocation size to the largest space found.
2343          */
2344         startag = ag = XFS_FSB_TO_AGNO(mp, args->fsbno);
2345         if (startag == NULLAGNUMBER)
2346                 startag = ag = 0;
2347
2348         pag = xfs_perag_get(mp, ag);
2349         while (*blen < args->maxlen) {
2350                 if (!pag->pagf_init) {
2351                         error = xfs_alloc_pagf_init(mp, args->tp, ag,
2352                                                     XFS_ALLOC_FLAG_TRYLOCK);
2353                         if (error) {
2354                                 xfs_perag_put(pag);
2355                                 return error;
2356                         }
2357                 }
2358
2359                 /*
2360                  * See xfs_alloc_fix_freelist...
2361                  */
2362                 if (pag->pagf_init) {
2363                         xfs_extlen_t    longest;
2364                         longest = xfs_alloc_longest_free_extent(mp, pag);
2365                         if (*blen < longest)
2366                                 *blen = longest;
2367                 } else
2368                         notinit = 1;
2369
2370                 if (xfs_inode_is_filestream(ap->ip)) {
2371                         if (*blen >= args->maxlen)
2372                                 break;
2373
2374                         if (ap->userdata) {
2375                                 /*
2376                                  * If startag is an invalid AG, we've
2377                                  * come here once before and
2378                                  * xfs_filestream_new_ag picked the
2379                                  * best currently available.
2380                                  *
2381                                  * Don't continue looping, since we
2382                                  * could loop forever.
2383                                  */
2384                                 if (startag == NULLAGNUMBER)
2385                                         break;
2386
2387                                 error = xfs_filestream_new_ag(ap, &ag);
2388                                 xfs_perag_put(pag);
2389                                 if (error)
2390                                         return error;
2391
2392                                 /* loop again to set 'blen'*/
2393                                 startag = NULLAGNUMBER;
2394                                 pag = xfs_perag_get(mp, ag);
2395                                 continue;
2396                         }
2397                 }
2398                 if (++ag == mp->m_sb.sb_agcount)
2399                         ag = 0;
2400                 if (ag == startag)
2401                         break;
2402                 xfs_perag_put(pag);
2403                 pag = xfs_perag_get(mp, ag);
2404         }
2405         xfs_perag_put(pag);
2406
2407         /*
2408          * Since the above loop did a BUF_TRYLOCK, it is
2409          * possible that there is space for this request.
2410          */
2411         if (notinit || *blen < ap->minlen)
2412                 args->minlen = ap->minlen;
2413         /*
2414          * If the best seen length is less than the request
2415          * length, use the best as the minimum.
2416          */
2417         else if (*blen < args->maxlen)
2418                 args->minlen = *blen;
2419         /*
2420          * Otherwise we've seen an extent as big as maxlen,
2421          * use that as the minimum.
2422          */
2423         else
2424                 args->minlen = args->maxlen;
2425
2426         /*
2427          * set the failure fallback case to look in the selected
2428          * AG as the stream may have moved.
2429          */
2430         if (xfs_inode_is_filestream(ap->ip))
2431                 ap->rval = args->fsbno = XFS_AGB_TO_FSB(mp, ag, 0);
2432
2433         return 0;
2434 }
2435
2436 STATIC int
2437 xfs_bmap_btalloc(
2438         xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
2439 {
2440         xfs_mount_t     *mp;            /* mount point structure */
2441         xfs_alloctype_t atype = 0;      /* type for allocation routines */
2442         xfs_extlen_t    align;          /* minimum allocation alignment */
2443         xfs_agnumber_t  fb_agno;        /* ag number of ap->firstblock */
2444         xfs_agnumber_t  ag;
2445         xfs_alloc_arg_t args;
2446         xfs_extlen_t    blen;
2447         xfs_extlen_t    nextminlen = 0;
2448         int             nullfb;         /* true if ap->firstblock isn't set */
2449         int             isaligned;
2450         int             tryagain;
2451         int             error;
2452
2453         mp = ap->ip->i_mount;
2454         align = ap->userdata ? xfs_get_extsz_hint(ap->ip) : 0;
2455         if (unlikely(align)) {
2456                 error = xfs_bmap_extsize_align(mp, ap->gotp, ap->prevp,
2457                                                 align, 0, ap->eof, 0, ap->conv,
2458                                                 &ap->off, &ap->alen);
2459                 ASSERT(!error);
2460                 ASSERT(ap->alen);
2461         }
2462         nullfb = ap->firstblock == NULLFSBLOCK;
2463         fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, ap->firstblock);
2464         if (nullfb) {
2465                 if (ap->userdata && xfs_inode_is_filestream(ap->ip)) {
2466                         ag = xfs_filestream_lookup_ag(ap->ip);
2467                         ag = (ag != NULLAGNUMBER) ? ag : 0;
2468                         ap->rval = XFS_AGB_TO_FSB(mp, ag, 0);
2469                 } else {
2470                         ap->rval = XFS_INO_TO_FSB(mp, ap->ip->i_ino);
2471                 }
2472         } else
2473                 ap->rval = ap->firstblock;
2474
2475         xfs_bmap_adjacent(ap);
2476
2477         /*
2478          * If allowed, use ap->rval; otherwise must use firstblock since
2479          * it's in the right allocation group.
2480          */
2481         if (nullfb || XFS_FSB_TO_AGNO(mp, ap->rval) == fb_agno)
2482                 ;
2483         else
2484                 ap->rval = ap->firstblock;
2485         /*
2486          * Normal allocation, done through xfs_alloc_vextent.
2487          */
2488         tryagain = isaligned = 0;
2489         args.tp = ap->tp;
2490         args.mp = mp;
2491         args.fsbno = ap->rval;
2492
2493         /* Trim the allocation back to the maximum an AG can fit. */
2494         args.maxlen = MIN(ap->alen, XFS_ALLOC_AG_MAX_USABLE(mp));
2495         args.firstblock = ap->firstblock;
2496         blen = 0;
2497         if (nullfb) {
2498                 error = xfs_bmap_btalloc_nullfb(ap, &args, &blen);
2499                 if (error)
2500                         return error;
2501         } else if (ap->low) {
2502                 if (xfs_inode_is_filestream(ap->ip))
2503                         args.type = XFS_ALLOCTYPE_FIRST_AG;
2504                 else
2505                         args.type = XFS_ALLOCTYPE_START_BNO;
2506                 args.total = args.minlen = ap->minlen;
2507         } else {
2508                 args.type = XFS_ALLOCTYPE_NEAR_BNO;
2509                 args.total = ap->total;
2510                 args.minlen = ap->minlen;
2511         }
2512         /* apply extent size hints if obtained earlier */
2513         if (unlikely(align)) {
2514                 args.prod = align;
2515                 if ((args.mod = (xfs_extlen_t)do_mod(ap->off, args.prod)))
2516                         args.mod = (xfs_extlen_t)(args.prod - args.mod);
2517         } else if (mp->m_sb.sb_blocksize >= PAGE_CACHE_SIZE) {
2518                 args.prod = 1;
2519                 args.mod = 0;
2520         } else {
2521                 args.prod = PAGE_CACHE_SIZE >> mp->m_sb.sb_blocklog;
2522                 if ((args.mod = (xfs_extlen_t)(do_mod(ap->off, args.prod))))
2523                         args.mod = (xfs_extlen_t)(args.prod - args.mod);
2524         }
2525         /*
2526          * If we are not low on available data blocks, and the
2527          * underlying logical volume manager is a stripe, and
2528          * the file offset is zero then try to allocate data
2529          * blocks on stripe unit boundary.
2530          * NOTE: ap->aeof is only set if the allocation length
2531          * is >= the stripe unit and the allocation offset is
2532          * at the end of file.
2533          */
2534         if (!ap->low && ap->aeof) {
2535                 if (!ap->off) {
2536                         args.alignment = mp->m_dalign;
2537                         atype = args.type;
2538                         isaligned = 1;
2539                         /*
2540                          * Adjust for alignment
2541                          */
2542                         if (blen > args.alignment && blen <= args.maxlen)
2543                                 args.minlen = blen - args.alignment;
2544                         args.minalignslop = 0;
2545                 } else {
2546                         /*
2547                          * First try an exact bno allocation.
2548                          * If it fails then do a near or start bno
2549                          * allocation with alignment turned on.
2550                          */
2551                         atype = args.type;
2552                         tryagain = 1;
2553                         args.type = XFS_ALLOCTYPE_THIS_BNO;
2554                         args.alignment = 1;
2555                         /*
2556                          * Compute the minlen+alignment for the
2557                          * next case.  Set slop so that the value
2558                          * of minlen+alignment+slop doesn't go up
2559                          * between the calls.
2560                          */
2561                         if (blen > mp->m_dalign && blen <= args.maxlen)
2562                                 nextminlen = blen - mp->m_dalign;
2563                         else
2564                                 nextminlen = args.minlen;
2565                         if (nextminlen + mp->m_dalign > args.minlen + 1)
2566                                 args.minalignslop =
2567                                         nextminlen + mp->m_dalign -
2568                                         args.minlen - 1;
2569                         else
2570                                 args.minalignslop = 0;
2571                 }
2572         } else {
2573                 args.alignment = 1;
2574                 args.minalignslop = 0;
2575         }
2576         args.minleft = ap->minleft;
2577         args.wasdel = ap->wasdel;
2578         args.isfl = 0;
2579         args.userdata = ap->userdata;
2580         if ((error = xfs_alloc_vextent(&args)))
2581                 return error;
2582         if (tryagain && args.fsbno == NULLFSBLOCK) {
2583                 /*
2584                  * Exact allocation failed. Now try with alignment
2585                  * turned on.
2586                  */
2587                 args.type = atype;
2588                 args.fsbno = ap->rval;
2589                 args.alignment = mp->m_dalign;
2590                 args.minlen = nextminlen;
2591                 args.minalignslop = 0;
2592                 isaligned = 1;
2593                 if ((error = xfs_alloc_vextent(&args)))
2594                         return error;
2595         }
2596         if (isaligned && args.fsbno == NULLFSBLOCK) {
2597                 /*
2598                  * allocation failed, so turn off alignment and
2599                  * try again.
2600                  */
2601                 args.type = atype;
2602                 args.fsbno = ap->rval;
2603                 args.alignment = 0;
2604                 if ((error = xfs_alloc_vextent(&args)))
2605                         return error;
2606         }
2607         if (args.fsbno == NULLFSBLOCK && nullfb &&
2608             args.minlen > ap->minlen) {
2609                 args.minlen = ap->minlen;
2610                 args.type = XFS_ALLOCTYPE_START_BNO;
2611                 args.fsbno = ap->rval;
2612                 if ((error = xfs_alloc_vextent(&args)))
2613                         return error;
2614         }
2615         if (args.fsbno == NULLFSBLOCK && nullfb) {
2616                 args.fsbno = 0;
2617                 args.type = XFS_ALLOCTYPE_FIRST_AG;
2618                 args.total = ap->minlen;
2619                 args.minleft = 0;
2620                 if ((error = xfs_alloc_vextent(&args)))
2621                         return error;
2622                 ap->low = 1;
2623         }
2624         if (args.fsbno != NULLFSBLOCK) {
2625                 ap->firstblock = ap->rval = args.fsbno;
2626                 ASSERT(nullfb || fb_agno == args.agno ||
2627                        (ap->low && fb_agno < args.agno));
2628                 ap->alen = args.len;
2629                 ap->ip->i_d.di_nblocks += args.len;
2630                 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
2631                 if (ap->wasdel)
2632                         ap->ip->i_delayed_blks -= args.len;
2633                 /*
2634                  * Adjust the disk quota also. This was reserved
2635                  * earlier.
2636                  */
2637                 xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
2638                         ap->wasdel ? XFS_TRANS_DQ_DELBCOUNT :
2639                                         XFS_TRANS_DQ_BCOUNT,
2640                         (long) args.len);
2641         } else {
2642                 ap->rval = NULLFSBLOCK;
2643                 ap->alen = 0;
2644         }
2645         return 0;
2646 }
2647
2648 /*
2649  * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
2650  * It figures out where to ask the underlying allocator to put the new extent.
2651  */
2652 STATIC int
2653 xfs_bmap_alloc(
2654         xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
2655 {
2656         if (XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata)
2657                 return xfs_bmap_rtalloc(ap);
2658         return xfs_bmap_btalloc(ap);
2659 }
2660
2661 /*
2662  * Transform a btree format file with only one leaf node, where the
2663  * extents list will fit in the inode, into an extents format file.
2664  * Since the file extents are already in-core, all we have to do is
2665  * give up the space for the btree root and pitch the leaf block.
2666  */
2667 STATIC int                              /* error */
2668 xfs_bmap_btree_to_extents(
2669         xfs_trans_t             *tp,    /* transaction pointer */
2670         xfs_inode_t             *ip,    /* incore inode pointer */
2671         xfs_btree_cur_t         *cur,   /* btree cursor */
2672         int                     *logflagsp, /* inode logging flags */
2673         int                     whichfork)  /* data or attr fork */
2674 {
2675         /* REFERENCED */
2676         struct xfs_btree_block  *cblock;/* child btree block */
2677         xfs_fsblock_t           cbno;   /* child block number */
2678         xfs_buf_t               *cbp;   /* child block's buffer */
2679         int                     error;  /* error return value */
2680         xfs_ifork_t             *ifp;   /* inode fork data */
2681         xfs_mount_t             *mp;    /* mount point structure */
2682         __be64                  *pp;    /* ptr to block address */
2683         struct xfs_btree_block  *rblock;/* root btree block */
2684
2685         mp = ip->i_mount;
2686         ifp = XFS_IFORK_PTR(ip, whichfork);
2687         ASSERT(ifp->if_flags & XFS_IFEXTENTS);
2688         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
2689         rblock = ifp->if_broot;
2690         ASSERT(be16_to_cpu(rblock->bb_level) == 1);
2691         ASSERT(be16_to_cpu(rblock->bb_numrecs) == 1);
2692         ASSERT(xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0) == 1);
2693         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, rblock, 1, ifp->if_broot_bytes);
2694         cbno = be64_to_cpu(*pp);
2695         *logflagsp = 0;
2696 #ifdef DEBUG
2697         if ((error = xfs_btree_check_lptr(cur, cbno, 1)))
2698                 return error;
2699 #endif
2700         if ((error = xfs_btree_read_bufl(mp, tp, cbno, 0, &cbp,
2701                         XFS_BMAP_BTREE_REF)))
2702                 return error;
2703         cblock = XFS_BUF_TO_BLOCK(cbp);
2704         if ((error = xfs_btree_check_block(cur, cblock, 0, cbp)))
2705                 return error;
2706         xfs_bmap_add_free(cbno, 1, cur->bc_private.b.flist, mp);
2707         ip->i_d.di_nblocks--;
2708         xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, -1L);
2709         xfs_trans_binval(tp, cbp);
2710         if (cur->bc_bufs[0] == cbp)
2711                 cur->bc_bufs[0] = NULL;
2712         xfs_iroot_realloc(ip, -1, whichfork);
2713         ASSERT(ifp->if_broot == NULL);
2714         ASSERT((ifp->if_flags & XFS_IFBROOT) == 0);
2715         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
2716         *logflagsp = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
2717         return 0;
2718 }
2719
2720 /*
2721  * Called by xfs_bmapi to update file extent records and the btree
2722  * after removing space (or undoing a delayed allocation).
2723  */
2724 STATIC int                              /* error */
2725 xfs_bmap_del_extent(
2726         xfs_inode_t             *ip,    /* incore inode pointer */
2727         xfs_trans_t             *tp,    /* current transaction pointer */
2728         xfs_extnum_t            *idx,   /* extent number to update/delete */
2729         xfs_bmap_free_t         *flist, /* list of extents to be freed */
2730         xfs_btree_cur_t         *cur,   /* if null, not a btree */
2731         xfs_bmbt_irec_t         *del,   /* data to remove from extents */
2732         int                     *logflagsp, /* inode logging flags */
2733         int                     whichfork) /* data or attr fork */
2734 {
2735         xfs_filblks_t           da_new; /* new delay-alloc indirect blocks */
2736         xfs_filblks_t           da_old; /* old delay-alloc indirect blocks */
2737         xfs_fsblock_t           del_endblock=0; /* first block past del */
2738         xfs_fileoff_t           del_endoff;     /* first offset past del */
2739         int                     delay;  /* current block is delayed allocated */
2740         int                     do_fx;  /* free extent at end of routine */
2741         xfs_bmbt_rec_host_t     *ep;    /* current extent entry pointer */
2742         int                     error;  /* error return value */
2743         int                     flags;  /* inode logging flags */
2744         xfs_bmbt_irec_t         got;    /* current extent entry */
2745         xfs_fileoff_t           got_endoff;     /* first offset past got */
2746         int                     i;      /* temp state */
2747         xfs_ifork_t             *ifp;   /* inode fork pointer */
2748         xfs_mount_t             *mp;    /* mount structure */
2749         xfs_filblks_t           nblks;  /* quota/sb block count */
2750         xfs_bmbt_irec_t         new;    /* new record to be inserted */
2751         /* REFERENCED */
2752         uint                    qfield; /* quota field to update */
2753         xfs_filblks_t           temp;   /* for indirect length calculations */
2754         xfs_filblks_t           temp2;  /* for indirect length calculations */
2755         int                     state = 0;
2756
2757         XFS_STATS_INC(xs_del_exlist);
2758
2759         if (whichfork == XFS_ATTR_FORK)
2760                 state |= BMAP_ATTRFORK;
2761
2762         mp = ip->i_mount;
2763         ifp = XFS_IFORK_PTR(ip, whichfork);
2764         ASSERT((*idx >= 0) && (*idx < ifp->if_bytes /
2765                 (uint)sizeof(xfs_bmbt_rec_t)));
2766         ASSERT(del->br_blockcount > 0);
2767         ep = xfs_iext_get_ext(ifp, *idx);
2768         xfs_bmbt_get_all(ep, &got);
2769         ASSERT(got.br_startoff <= del->br_startoff);
2770         del_endoff = del->br_startoff + del->br_blockcount;
2771         got_endoff = got.br_startoff + got.br_blockcount;
2772         ASSERT(got_endoff >= del_endoff);
2773         delay = isnullstartblock(got.br_startblock);
2774         ASSERT(isnullstartblock(del->br_startblock) == delay);
2775         flags = 0;
2776         qfield = 0;
2777         error = 0;
2778         /*
2779          * If deleting a real allocation, must free up the disk space.
2780          */
2781         if (!delay) {
2782                 flags = XFS_ILOG_CORE;
2783                 /*
2784                  * Realtime allocation.  Free it and record di_nblocks update.
2785                  */
2786                 if (whichfork == XFS_DATA_FORK && XFS_IS_REALTIME_INODE(ip)) {
2787                         xfs_fsblock_t   bno;
2788                         xfs_filblks_t   len;
2789
2790                         ASSERT(do_mod(del->br_blockcount,
2791                                       mp->m_sb.sb_rextsize) == 0);
2792                         ASSERT(do_mod(del->br_startblock,
2793                                       mp->m_sb.sb_rextsize) == 0);
2794                         bno = del->br_startblock;
2795                         len = del->br_blockcount;
2796                         do_div(bno, mp->m_sb.sb_rextsize);
2797                         do_div(len, mp->m_sb.sb_rextsize);
2798                         error = xfs_rtfree_extent(tp, bno, (xfs_extlen_t)len);
2799                         if (error)
2800                                 goto done;
2801                         do_fx = 0;
2802                         nblks = len * mp->m_sb.sb_rextsize;
2803                         qfield = XFS_TRANS_DQ_RTBCOUNT;
2804                 }
2805                 /*
2806                  * Ordinary allocation.
2807                  */
2808                 else {
2809                         do_fx = 1;
2810                         nblks = del->br_blockcount;
2811                         qfield = XFS_TRANS_DQ_BCOUNT;
2812                 }
2813                 /*
2814                  * Set up del_endblock and cur for later.
2815                  */
2816                 del_endblock = del->br_startblock + del->br_blockcount;
2817                 if (cur) {
2818                         if ((error = xfs_bmbt_lookup_eq(cur, got.br_startoff,
2819                                         got.br_startblock, got.br_blockcount,
2820                                         &i)))
2821                                 goto done;
2822                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2823                 }
2824                 da_old = da_new = 0;
2825         } else {
2826                 da_old = startblockval(got.br_startblock);
2827                 da_new = 0;
2828                 nblks = 0;
2829                 do_fx = 0;
2830         }
2831         /*
2832          * Set flag value to use in switch statement.
2833          * Left-contig is 2, right-contig is 1.
2834          */
2835         switch (((got.br_startoff == del->br_startoff) << 1) |
2836                 (got_endoff == del_endoff)) {
2837         case 3:
2838                 /*
2839                  * Matches the whole extent.  Delete the entry.
2840                  */
2841                 xfs_iext_remove(ip, *idx, 1,
2842                                 whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0);
2843                 --*idx;
2844                 if (delay)
2845                         break;
2846
2847                 XFS_IFORK_NEXT_SET(ip, whichfork,
2848                         XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
2849                 flags |= XFS_ILOG_CORE;
2850                 if (!cur) {
2851                         flags |= xfs_ilog_fext(whichfork);
2852                         break;
2853                 }
2854                 if ((error = xfs_btree_delete(cur, &i)))
2855                         goto done;
2856                 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2857                 break;
2858
2859         case 2:
2860                 /*
2861                  * Deleting the first part of the extent.
2862                  */
2863                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2864                 xfs_bmbt_set_startoff(ep, del_endoff);
2865                 temp = got.br_blockcount - del->br_blockcount;
2866                 xfs_bmbt_set_blockcount(ep, temp);
2867                 if (delay) {
2868                         temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2869                                 da_old);
2870                         xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
2871                         trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2872                         da_new = temp;
2873                         break;
2874                 }
2875                 xfs_bmbt_set_startblock(ep, del_endblock);
2876                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2877                 if (!cur) {
2878                         flags |= xfs_ilog_fext(whichfork);
2879                         break;
2880                 }
2881                 if ((error = xfs_bmbt_update(cur, del_endoff, del_endblock,
2882                                 got.br_blockcount - del->br_blockcount,
2883                                 got.br_state)))
2884                         goto done;
2885                 break;
2886
2887         case 1:
2888                 /*
2889                  * Deleting the last part of the extent.
2890                  */
2891                 temp = got.br_blockcount - del->br_blockcount;
2892                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2893                 xfs_bmbt_set_blockcount(ep, temp);
2894                 if (delay) {
2895                         temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2896                                 da_old);
2897                         xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
2898                         trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2899                         da_new = temp;
2900                         break;
2901                 }
2902                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2903                 if (!cur) {
2904                         flags |= xfs_ilog_fext(whichfork);
2905                         break;
2906                 }
2907                 if ((error = xfs_bmbt_update(cur, got.br_startoff,
2908                                 got.br_startblock,
2909                                 got.br_blockcount - del->br_blockcount,
2910                                 got.br_state)))
2911                         goto done;
2912                 break;
2913
2914         case 0:
2915                 /*
2916                  * Deleting the middle of the extent.
2917                  */
2918                 temp = del->br_startoff - got.br_startoff;
2919                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2920                 xfs_bmbt_set_blockcount(ep, temp);
2921                 new.br_startoff = del_endoff;
2922                 temp2 = got_endoff - del_endoff;
2923                 new.br_blockcount = temp2;
2924                 new.br_state = got.br_state;
2925                 if (!delay) {
2926                         new.br_startblock = del_endblock;
2927                         flags |= XFS_ILOG_CORE;
2928                         if (cur) {
2929                                 if ((error = xfs_bmbt_update(cur,
2930                                                 got.br_startoff,
2931                                                 got.br_startblock, temp,
2932                                                 got.br_state)))
2933                                         goto done;
2934                                 if ((error = xfs_btree_increment(cur, 0, &i)))
2935                                         goto done;
2936                                 cur->bc_rec.b = new;
2937                                 error = xfs_btree_insert(cur, &i);
2938                                 if (error && error != ENOSPC)
2939                                         goto done;
2940                                 /*
2941                                  * If get no-space back from btree insert,
2942                                  * it tried a split, and we have a zero
2943                                  * block reservation.
2944                                  * Fix up our state and return the error.
2945                                  */
2946                                 if (error == ENOSPC) {
2947                                         /*
2948                                          * Reset the cursor, don't trust
2949                                          * it after any insert operation.
2950                                          */
2951                                         if ((error = xfs_bmbt_lookup_eq(cur,
2952                                                         got.br_startoff,
2953                                                         got.br_startblock,
2954                                                         temp, &i)))
2955                                                 goto done;
2956                                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2957                                         /*
2958                                          * Update the btree record back
2959                                          * to the original value.
2960                                          */
2961                                         if ((error = xfs_bmbt_update(cur,
2962                                                         got.br_startoff,
2963                                                         got.br_startblock,
2964                                                         got.br_blockcount,
2965                                                         got.br_state)))
2966                                                 goto done;
2967                                         /*
2968                                          * Reset the extent record back
2969                                          * to the original value.
2970                                          */
2971                                         xfs_bmbt_set_blockcount(ep,
2972                                                 got.br_blockcount);
2973                                         flags = 0;
2974                                         error = XFS_ERROR(ENOSPC);
2975                                         goto done;
2976                                 }
2977                                 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2978                         } else
2979                                 flags |= xfs_ilog_fext(whichfork);
2980                         XFS_IFORK_NEXT_SET(ip, whichfork,
2981                                 XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
2982                 } else {
2983                         ASSERT(whichfork == XFS_DATA_FORK);
2984                         temp = xfs_bmap_worst_indlen(ip, temp);
2985                         xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
2986                         temp2 = xfs_bmap_worst_indlen(ip, temp2);
2987                         new.br_startblock = nullstartblock((int)temp2);
2988                         da_new = temp + temp2;
2989                         while (da_new > da_old) {
2990                                 if (temp) {
2991                                         temp--;
2992                                         da_new--;
2993                                         xfs_bmbt_set_startblock(ep,
2994                                                 nullstartblock((int)temp));
2995                                 }
2996                                 if (da_new == da_old)
2997                                         break;
2998                                 if (temp2) {
2999                                         temp2--;
3000                                         da_new--;
3001                                         new.br_startblock =
3002                                                 nullstartblock((int)temp2);
3003                                 }
3004                         }
3005                 }
3006                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
3007                 xfs_iext_insert(ip, *idx + 1, 1, &new, state);
3008                 ++*idx;
3009                 break;
3010         }
3011         /*
3012          * If we need to, add to list of extents to delete.
3013          */
3014         if (do_fx)
3015                 xfs_bmap_add_free(del->br_startblock, del->br_blockcount, flist,
3016                         mp);
3017         /*
3018          * Adjust inode # blocks in the file.
3019          */
3020         if (nblks)
3021                 ip->i_d.di_nblocks -= nblks;
3022         /*
3023          * Adjust quota data.
3024          */
3025         if (qfield)
3026                 xfs_trans_mod_dquot_byino(tp, ip, qfield, (long)-nblks);
3027
3028         /*
3029          * Account for change in delayed indirect blocks.
3030          * Nothing to do for disk quota accounting here.
3031          */
3032         ASSERT(da_old >= da_new);
3033         if (da_old > da_new) {
3034                 xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
3035                         (int64_t)(da_old - da_new), 0);
3036         }
3037 done:
3038         *logflagsp = flags;
3039         return error;
3040 }
3041
3042 /*
3043  * Remove the entry "free" from the free item list.  Prev points to the
3044  * previous entry, unless "free" is the head of the list.
3045  */
3046 STATIC void
3047 xfs_bmap_del_free(
3048         xfs_bmap_free_t         *flist, /* free item list header */
3049         xfs_bmap_free_item_t    *prev,  /* previous item on list, if any */
3050         xfs_bmap_free_item_t    *free)  /* list item to be freed */
3051 {
3052         if (prev)
3053                 prev->xbfi_next = free->xbfi_next;
3054         else
3055                 flist->xbf_first = free->xbfi_next;
3056         flist->xbf_count--;
3057         kmem_zone_free(xfs_bmap_free_item_zone, free);
3058 }
3059
3060 /*
3061  * Convert an extents-format file into a btree-format file.
3062  * The new file will have a root block (in the inode) and a single child block.
3063  */
3064 STATIC int                                      /* error */
3065 xfs_bmap_extents_to_btree(
3066         xfs_trans_t             *tp,            /* transaction pointer */
3067         xfs_inode_t             *ip,            /* incore inode pointer */
3068         xfs_fsblock_t           *firstblock,    /* first-block-allocated */
3069         xfs_bmap_free_t         *flist,         /* blocks freed in xaction */
3070         xfs_btree_cur_t         **curp,         /* cursor returned to caller */
3071         int                     wasdel,         /* converting a delayed alloc */
3072         int                     *logflagsp,     /* inode logging flags */
3073         int                     whichfork)      /* data or attr fork */
3074 {
3075         struct xfs_btree_block  *ablock;        /* allocated (child) bt block */
3076         xfs_buf_t               *abp;           /* buffer for ablock */
3077         xfs_alloc_arg_t         args;           /* allocation arguments */
3078         xfs_bmbt_rec_t          *arp;           /* child record pointer */
3079         struct xfs_btree_block  *block;         /* btree root block */
3080         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
3081         xfs_bmbt_rec_host_t     *ep;            /* extent record pointer */
3082         int                     error;          /* error return value */
3083         xfs_extnum_t            i, cnt;         /* extent record index */
3084         xfs_ifork_t             *ifp;           /* inode fork pointer */
3085         xfs_bmbt_key_t          *kp;            /* root block key pointer */
3086         xfs_mount_t             *mp;            /* mount structure */
3087         xfs_extnum_t            nextents;       /* number of file extents */
3088         xfs_bmbt_ptr_t          *pp;            /* root block address pointer */
3089
3090         ifp = XFS_IFORK_PTR(ip, whichfork);
3091         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS);
3092         ASSERT(ifp->if_ext_max ==
3093                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
3094         /*
3095          * Make space in the inode incore.
3096          */
3097         xfs_iroot_realloc(ip, 1, whichfork);
3098         ifp->if_flags |= XFS_IFBROOT;
3099
3100         /*
3101          * Fill in the root.
3102          */
3103         block = ifp->if_broot;
3104         block->bb_magic = cpu_to_be32(XFS_BMAP_MAGIC);
3105         block->bb_level = cpu_to_be16(1);
3106         block->bb_numrecs = cpu_to_be16(1);
3107         block->bb_u.l.bb_leftsib = cpu_to_be64(NULLDFSBNO);
3108         block->bb_u.l.bb_rightsib = cpu_to_be64(NULLDFSBNO);
3109
3110         /*
3111          * Need a cursor.  Can't allocate until bb_level is filled in.
3112          */
3113         mp = ip->i_mount;
3114         cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
3115         cur->bc_private.b.firstblock = *firstblock;
3116         cur->bc_private.b.flist = flist;
3117         cur->bc_private.b.flags = wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
3118         /*
3119          * Convert to a btree with two levels, one record in root.
3120          */
3121         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_BTREE);
3122         args.tp = tp;
3123         args.mp = mp;
3124         args.firstblock = *firstblock;
3125         if (*firstblock == NULLFSBLOCK) {
3126                 args.type = XFS_ALLOCTYPE_START_BNO;
3127                 args.fsbno = XFS_INO_TO_FSB(mp, ip->i_ino);
3128         } else if (flist->xbf_low) {
3129                 args.type = XFS_ALLOCTYPE_START_BNO;
3130                 args.fsbno = *firstblock;
3131         } else {
3132                 args.type = XFS_ALLOCTYPE_NEAR_BNO;
3133                 args.fsbno = *firstblock;
3134         }
3135         args.minlen = args.maxlen = args.prod = 1;
3136         args.total = args.minleft = args.alignment = args.mod = args.isfl =
3137                 args.minalignslop = 0;
3138         args.wasdel = wasdel;
3139         *logflagsp = 0;
3140         if ((error = xfs_alloc_vextent(&args))) {
3141                 xfs_iroot_realloc(ip, -1, whichfork);
3142                 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
3143                 return error;
3144         }
3145         /*
3146          * Allocation can't fail, the space was reserved.
3147          */
3148         ASSERT(args.fsbno != NULLFSBLOCK);
3149         ASSERT(*firstblock == NULLFSBLOCK ||
3150                args.agno == XFS_FSB_TO_AGNO(mp, *firstblock) ||
3151                (flist->xbf_low &&
3152                 args.agno > XFS_FSB_TO_AGNO(mp, *firstblock)));
3153         *firstblock = cur->bc_private.b.firstblock = args.fsbno;
3154         cur->bc_private.b.allocated++;
3155         ip->i_d.di_nblocks++;
3156         xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, 1L);
3157         abp = xfs_btree_get_bufl(mp, tp, args.fsbno, 0);
3158         /*
3159          * Fill in the child block.
3160          */
3161         ablock = XFS_BUF_TO_BLOCK(abp);
3162         ablock->bb_magic = cpu_to_be32(XFS_BMAP_MAGIC);
3163         ablock->bb_level = 0;
3164         ablock->bb_u.l.bb_leftsib = cpu_to_be64(NULLDFSBNO);
3165         ablock->bb_u.l.bb_rightsib = cpu_to_be64(NULLDFSBNO);
3166         arp = XFS_BMBT_REC_ADDR(mp, ablock, 1);
3167         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3168         for (cnt = i = 0; i < nextents; i++) {
3169                 ep = xfs_iext_get_ext(ifp, i);
3170                 if (!isnullstartblock(xfs_bmbt_get_startblock(ep))) {
3171                         arp->l0 = cpu_to_be64(ep->l0);
3172                         arp->l1 = cpu_to_be64(ep->l1);
3173                         arp++; cnt++;
3174                 }
3175         }
3176         ASSERT(cnt == XFS_IFORK_NEXTENTS(ip, whichfork));
3177         xfs_btree_set_numrecs(ablock, cnt);
3178
3179         /*
3180          * Fill in the root key and pointer.
3181          */
3182         kp = XFS_BMBT_KEY_ADDR(mp, block, 1);
3183         arp = XFS_BMBT_REC_ADDR(mp, ablock, 1);
3184         kp->br_startoff = cpu_to_be64(xfs_bmbt_disk_get_startoff(arp));
3185         pp = XFS_BMBT_PTR_ADDR(mp, block, 1, xfs_bmbt_get_maxrecs(cur,
3186                                                 be16_to_cpu(block->bb_level)));
3187         *pp = cpu_to_be64(args.fsbno);
3188
3189         /*
3190          * Do all this logging at the end so that
3191          * the root is at the right level.
3192          */
3193         xfs_btree_log_block(cur, abp, XFS_BB_ALL_BITS);
3194         xfs_btree_log_recs(cur, abp, 1, be16_to_cpu(ablock->bb_numrecs));
3195         ASSERT(*curp == NULL);
3196         *curp = cur;
3197         *logflagsp = XFS_ILOG_CORE | xfs_ilog_fbroot(whichfork);
3198         return 0;
3199 }
3200
3201 /*
3202  * Calculate the default attribute fork offset for newly created inodes.
3203  */
3204 uint
3205 xfs_default_attroffset(
3206         struct xfs_inode        *ip)
3207 {
3208         struct xfs_mount        *mp = ip->i_mount;
3209         uint                    offset;
3210
3211         if (mp->m_sb.sb_inodesize == 256) {
3212                 offset = XFS_LITINO(mp) -
3213                                 XFS_BMDR_SPACE_CALC(MINABTPTRS);
3214         } else {
3215                 offset = XFS_BMDR_SPACE_CALC(6 * MINABTPTRS);
3216         }
3217
3218         ASSERT(offset < XFS_LITINO(mp));
3219         return offset;
3220 }
3221
3222 /*
3223  * Helper routine to reset inode di_forkoff field when switching
3224  * attribute fork from local to extent format - we reset it where
3225  * possible to make space available for inline data fork extents.
3226  */
3227 STATIC void
3228 xfs_bmap_forkoff_reset(
3229         xfs_mount_t     *mp,
3230         xfs_inode_t     *ip,
3231         int             whichfork)
3232 {
3233         if (whichfork == XFS_ATTR_FORK &&
3234             ip->i_d.di_format != XFS_DINODE_FMT_DEV &&
3235             ip->i_d.di_format != XFS_DINODE_FMT_UUID &&
3236             ip->i_d.di_format != XFS_DINODE_FMT_BTREE) {
3237                 uint    dfl_forkoff = xfs_default_attroffset(ip) >> 3;
3238
3239                 if (dfl_forkoff > ip->i_d.di_forkoff) {
3240                         ip->i_d.di_forkoff = dfl_forkoff;
3241                         ip->i_df.if_ext_max =
3242                                 XFS_IFORK_DSIZE(ip) / sizeof(xfs_bmbt_rec_t);
3243                         ip->i_afp->if_ext_max =
3244                                 XFS_IFORK_ASIZE(ip) / sizeof(xfs_bmbt_rec_t);
3245                 }
3246         }
3247 }
3248
3249 /*
3250  * Convert a local file to an extents file.
3251  * This code is out of bounds for data forks of regular files,
3252  * since the file data needs to get logged so things will stay consistent.
3253  * (The bmap-level manipulations are ok, though).
3254  */
3255 STATIC int                              /* error */
3256 xfs_bmap_local_to_extents(
3257         xfs_trans_t     *tp,            /* transaction pointer */
3258         xfs_inode_t     *ip,            /* incore inode pointer */
3259         xfs_fsblock_t   *firstblock,    /* first block allocated in xaction */
3260         xfs_extlen_t    total,          /* total blocks needed by transaction */
3261         int             *logflagsp,     /* inode logging flags */
3262         int             whichfork)      /* data or attr fork */
3263 {
3264         int             error;          /* error return value */
3265         int             flags;          /* logging flags returned */
3266         xfs_ifork_t     *ifp;           /* inode fork pointer */
3267
3268         /*
3269          * We don't want to deal with the case of keeping inode data inline yet.
3270          * So sending the data fork of a regular inode is invalid.
3271          */
3272         ASSERT(!(S_ISREG(ip->i_d.di_mode) && whichfork == XFS_DATA_FORK));
3273         ifp = XFS_IFORK_PTR(ip, whichfork);
3274         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
3275         flags = 0;
3276         error = 0;
3277         if (ifp->if_bytes) {
3278                 xfs_alloc_arg_t args;   /* allocation arguments */
3279                 xfs_buf_t       *bp;    /* buffer for extent block */
3280                 xfs_bmbt_rec_host_t *ep;/* extent record pointer */
3281
3282                 args.tp = tp;
3283                 args.mp = ip->i_mount;
3284                 args.firstblock = *firstblock;
3285                 ASSERT((ifp->if_flags &
3286                         (XFS_IFINLINE|XFS_IFEXTENTS|XFS_IFEXTIREC)) == XFS_IFINLINE);
3287                 /*
3288                  * Allocate a block.  We know we need only one, since the
3289                  * file currently fits in an inode.
3290                  */
3291                 if (*firstblock == NULLFSBLOCK) {
3292                         args.fsbno = XFS_INO_TO_FSB(args.mp, ip->i_ino);
3293                         args.type = XFS_ALLOCTYPE_START_BNO;
3294                 } else {
3295                         args.fsbno = *firstblock;
3296                         args.type = XFS_ALLOCTYPE_NEAR_BNO;
3297                 }
3298                 args.total = total;
3299                 args.mod = args.minleft = args.alignment = args.wasdel =
3300                         args.isfl = args.minalignslop = 0;
3301                 args.minlen = args.maxlen = args.prod = 1;
3302                 if ((error = xfs_alloc_vextent(&args)))
3303                         goto done;
3304                 /*
3305                  * Can't fail, the space was reserved.
3306                  */
3307                 ASSERT(args.fsbno != NULLFSBLOCK);
3308                 ASSERT(args.len == 1);
3309                 *firstblock = args.fsbno;
3310                 bp = xfs_btree_get_bufl(args.mp, tp, args.fsbno, 0);
3311                 memcpy(bp->b_addr, ifp->if_u1.if_data, ifp->if_bytes);
3312                 xfs_trans_log_buf(tp, bp, 0, ifp->if_bytes - 1);
3313                 xfs_bmap_forkoff_reset(args.mp, ip, whichfork);
3314                 xfs_idata_realloc(ip, -ifp->if_bytes, whichfork);
3315                 xfs_iext_add(ifp, 0, 1);
3316                 ep = xfs_iext_get_ext(ifp, 0);
3317                 xfs_bmbt_set_allf(ep, 0, args.fsbno, 1, XFS_EXT_NORM);
3318                 trace_xfs_bmap_post_update(ip, 0,
3319                                 whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0,
3320                                 _THIS_IP_);
3321                 XFS_IFORK_NEXT_SET(ip, whichfork, 1);
3322                 ip->i_d.di_nblocks = 1;
3323                 xfs_trans_mod_dquot_byino(tp, ip,
3324                         XFS_TRANS_DQ_BCOUNT, 1L);
3325                 flags |= xfs_ilog_fext(whichfork);
3326         } else {
3327                 ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) == 0);
3328                 xfs_bmap_forkoff_reset(ip->i_mount, ip, whichfork);
3329         }
3330         ifp->if_flags &= ~XFS_IFINLINE;
3331         ifp->if_flags |= XFS_IFEXTENTS;
3332         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
3333         flags |= XFS_ILOG_CORE;
3334 done:
3335         *logflagsp = flags;
3336         return error;
3337 }
3338
3339 /*
3340  * Search the extent records for the entry containing block bno.
3341  * If bno lies in a hole, point to the next entry.  If bno lies
3342  * past eof, *eofp will be set, and *prevp will contain the last
3343  * entry (null if none).  Else, *lastxp will be set to the index
3344  * of the found entry; *gotp will contain the entry.
3345  */
3346 STATIC xfs_bmbt_rec_host_t *            /* pointer to found extent entry */
3347 xfs_bmap_search_multi_extents(
3348         xfs_ifork_t     *ifp,           /* inode fork pointer */
3349         xfs_fileoff_t   bno,            /* block number searched for */
3350         int             *eofp,          /* out: end of file found */
3351         xfs_extnum_t    *lastxp,        /* out: last extent index */
3352         xfs_bmbt_irec_t *gotp,          /* out: extent entry found */
3353         xfs_bmbt_irec_t *prevp)         /* out: previous extent entry found */
3354 {
3355         xfs_bmbt_rec_host_t *ep;                /* extent record pointer */
3356         xfs_extnum_t    lastx;          /* last extent index */
3357
3358         /*
3359          * Initialize the extent entry structure to catch access to
3360          * uninitialized br_startblock field.
3361          */
3362         gotp->br_startoff = 0xffa5a5a5a5a5a5a5LL;
3363         gotp->br_blockcount = 0xa55a5a5a5a5a5a5aLL;
3364         gotp->br_state = XFS_EXT_INVALID;
3365 #if XFS_BIG_BLKNOS
3366         gotp->br_startblock = 0xffffa5a5a5a5a5a5LL;
3367 #else
3368         gotp->br_startblock = 0xffffa5a5;
3369 #endif
3370         prevp->br_startoff = NULLFILEOFF;
3371
3372         ep = xfs_iext_bno_to_ext(ifp, bno, &lastx);
3373         if (lastx > 0) {
3374                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx - 1), prevp);
3375         }
3376         if (lastx < (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t))) {
3377                 xfs_bmbt_get_all(ep, gotp);
3378                 *eofp = 0;
3379         } else {
3380                 if (lastx > 0) {
3381                         *gotp = *prevp;
3382                 }
3383                 *eofp = 1;
3384                 ep = NULL;
3385         }
3386         *lastxp = lastx;
3387         return ep;
3388 }
3389
3390 /*
3391  * Search the extents list for the inode, for the extent containing bno.
3392  * If bno lies in a hole, point to the next entry.  If bno lies past eof,
3393  * *eofp will be set, and *prevp will contain the last entry (null if none).
3394  * Else, *lastxp will be set to the index of the found
3395  * entry; *gotp will contain the entry.
3396  */
3397 STATIC xfs_bmbt_rec_host_t *                 /* pointer to found extent entry */
3398 xfs_bmap_search_extents(
3399         xfs_inode_t     *ip,            /* incore inode pointer */
3400         xfs_fileoff_t   bno,            /* block number searched for */
3401         int             fork,           /* data or attr fork */
3402         int             *eofp,          /* out: end of file found */
3403         xfs_extnum_t    *lastxp,        /* out: last extent index */
3404         xfs_bmbt_irec_t *gotp,          /* out: extent entry found */
3405         xfs_bmbt_irec_t *prevp)         /* out: previous extent entry found */
3406 {
3407         xfs_ifork_t     *ifp;           /* inode fork pointer */
3408         xfs_bmbt_rec_host_t  *ep;            /* extent record pointer */
3409
3410         XFS_STATS_INC(xs_look_exlist);
3411         ifp = XFS_IFORK_PTR(ip, fork);
3412
3413         ep = xfs_bmap_search_multi_extents(ifp, bno, eofp, lastxp, gotp, prevp);
3414
3415         if (unlikely(!(gotp->br_startblock) && (*lastxp != NULLEXTNUM) &&
3416                      !(XFS_IS_REALTIME_INODE(ip) && fork == XFS_DATA_FORK))) {
3417                 xfs_alert_tag(ip->i_mount, XFS_PTAG_FSBLOCK_ZERO,
3418                                 "Access to block zero in inode %llu "
3419                                 "start_block: %llx start_off: %llx "
3420                                 "blkcnt: %llx extent-state: %x lastx: %x\n",
3421                         (unsigned long long)ip->i_ino,
3422                         (unsigned long long)gotp->br_startblock,
3423                         (unsigned long long)gotp->br_startoff,
3424                         (unsigned long long)gotp->br_blockcount,
3425                         gotp->br_state, *lastxp);
3426                 *lastxp = NULLEXTNUM;
3427                 *eofp = 1;
3428                 return NULL;
3429         }
3430         return ep;
3431 }
3432
3433 /*
3434  * Compute the worst-case number of indirect blocks that will be used
3435  * for ip's delayed extent of length "len".
3436  */
3437 STATIC xfs_filblks_t
3438 xfs_bmap_worst_indlen(
3439         xfs_inode_t     *ip,            /* incore inode pointer */
3440         xfs_filblks_t   len)            /* delayed extent length */
3441 {
3442         int             level;          /* btree level number */
3443         int             maxrecs;        /* maximum record count at this level */
3444         xfs_mount_t     *mp;            /* mount structure */
3445         xfs_filblks_t   rval;           /* return value */
3446
3447         mp = ip->i_mount;
3448         maxrecs = mp->m_bmap_dmxr[0];
3449         for (level = 0, rval = 0;
3450              level < XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK);
3451              level++) {
3452                 len += maxrecs - 1;
3453                 do_div(len, maxrecs);
3454                 rval += len;
3455                 if (len == 1)
3456                         return rval + XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK) -
3457                                 level - 1;
3458                 if (level == 0)
3459                         maxrecs = mp->m_bmap_dmxr[1];
3460         }
3461         return rval;
3462 }
3463
3464 /*
3465  * Convert inode from non-attributed to attributed.
3466  * Must not be in a transaction, ip must not be locked.
3467  */
3468 int                                             /* error code */
3469 xfs_bmap_add_attrfork(
3470         xfs_inode_t             *ip,            /* incore inode pointer */
3471         int                     size,           /* space new attribute needs */
3472         int                     rsvd)           /* xact may use reserved blks */
3473 {
3474         xfs_fsblock_t           firstblock;     /* 1st block/ag allocated */
3475         xfs_bmap_free_t         flist;          /* freed extent records */
3476         xfs_mount_t             *mp;            /* mount structure */
3477         xfs_trans_t             *tp;            /* transaction pointer */
3478         int                     blks;           /* space reservation */
3479         int                     version = 1;    /* superblock attr version */
3480         int                     committed;      /* xaction was committed */
3481         int                     logflags;       /* logging flags */
3482         int                     error;          /* error return value */
3483
3484         ASSERT(XFS_IFORK_Q(ip) == 0);
3485         ASSERT(ip->i_df.if_ext_max ==
3486                XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t));
3487
3488         mp = ip->i_mount;
3489         ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
3490         tp = xfs_trans_alloc(mp, XFS_TRANS_ADDAFORK);
3491         blks = XFS_ADDAFORK_SPACE_RES(mp);
3492         if (rsvd)
3493                 tp->t_flags |= XFS_TRANS_RESERVE;
3494         if ((error = xfs_trans_reserve(tp, blks, XFS_ADDAFORK_LOG_RES(mp), 0,
3495                         XFS_TRANS_PERM_LOG_RES, XFS_ADDAFORK_LOG_COUNT)))
3496                 goto error0;
3497         xfs_ilock(ip, XFS_ILOCK_EXCL);
3498         error = xfs_trans_reserve_quota_nblks(tp, ip, blks, 0, rsvd ?
3499                         XFS_QMOPT_RES_REGBLKS | XFS_QMOPT_FORCE_RES :
3500                         XFS_QMOPT_RES_REGBLKS);
3501         if (error) {
3502                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3503                 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES);
3504                 return error;
3505         }
3506         if (XFS_IFORK_Q(ip))
3507                 goto error1;
3508         if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS) {
3509                 /*
3510                  * For inodes coming from pre-6.2 filesystems.
3511                  */
3512                 ASSERT(ip->i_d.di_aformat == 0);
3513                 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
3514         }
3515         ASSERT(ip->i_d.di_anextents == 0);
3516
3517         xfs_trans_ijoin_ref(tp, ip, XFS_ILOCK_EXCL);
3518         xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
3519
3520         switch (ip->i_d.di_format) {
3521         case XFS_DINODE_FMT_DEV:
3522                 ip->i_d.di_forkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
3523                 break;
3524         case XFS_DINODE_FMT_UUID:
3525                 ip->i_d.di_forkoff = roundup(sizeof(uuid_t), 8) >> 3;
3526                 break;
3527         case XFS_DINODE_FMT_LOCAL:
3528         case XFS_DINODE_FMT_EXTENTS:
3529         case XFS_DINODE_FMT_BTREE:
3530                 ip->i_d.di_forkoff = xfs_attr_shortform_bytesfit(ip, size);
3531                 if (!ip->i_d.di_forkoff)
3532                         ip->i_d.di_forkoff = xfs_default_attroffset(ip) >> 3;
3533                 else if (mp->m_flags & XFS_MOUNT_ATTR2)
3534                         version = 2;
3535                 break;
3536         default:
3537                 ASSERT(0);
3538                 error = XFS_ERROR(EINVAL);
3539                 goto error1;
3540         }
3541         ip->i_df.if_ext_max =
3542                 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
3543         ASSERT(ip->i_afp == NULL);
3544         ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP);
3545         ip->i_afp->if_ext_max =
3546                 XFS_IFORK_ASIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
3547         ip->i_afp->if_flags = XFS_IFEXTENTS;
3548         logflags = 0;
3549         xfs_bmap_init(&flist, &firstblock);
3550         switch (ip->i_d.di_format) {
3551         case XFS_DINODE_FMT_LOCAL:
3552                 error = xfs_bmap_add_attrfork_local(tp, ip, &firstblock, &flist,
3553                         &logflags);
3554                 break;
3555         case XFS_DINODE_FMT_EXTENTS:
3556                 error = xfs_bmap_add_attrfork_extents(tp, ip, &firstblock,
3557                         &flist, &logflags);
3558                 break;
3559         case XFS_DINODE_FMT_BTREE:
3560                 error = xfs_bmap_add_attrfork_btree(tp, ip, &firstblock, &flist,
3561                         &logflags);
3562                 break;
3563         default:
3564                 error = 0;
3565                 break;
3566         }
3567         if (logflags)
3568                 xfs_trans_log_inode(tp, ip, logflags);
3569         if (error)
3570                 goto error2;
3571         if (!xfs_sb_version_hasattr(&mp->m_sb) ||
3572            (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2)) {
3573                 __int64_t sbfields = 0;
3574
3575                 spin_lock(&mp->m_sb_lock);
3576                 if (!xfs_sb_version_hasattr(&mp->m_sb)) {
3577                         xfs_sb_version_addattr(&mp->m_sb);
3578                         sbfields |= XFS_SB_VERSIONNUM;
3579                 }
3580                 if (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2) {
3581                         xfs_sb_version_addattr2(&mp->m_sb);
3582                         sbfields |= (XFS_SB_VERSIONNUM | XFS_SB_FEATURES2);
3583                 }
3584                 if (sbfields) {
3585                         spin_unlock(&mp->m_sb_lock);
3586                         xfs_mod_sb(tp, sbfields);
3587                 } else
3588                         spin_unlock(&mp->m_sb_lock);
3589         }
3590         if ((error = xfs_bmap_finish(&tp, &flist, &committed)))
3591                 goto error2;
3592         error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
3593         ASSERT(ip->i_df.if_ext_max ==
3594                XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t));
3595         return error;
3596 error2:
3597         xfs_bmap_cancel(&flist);
3598 error1:
3599         xfs_iunlock(ip, XFS_ILOCK_EXCL);
3600 error0:
3601         xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
3602         ASSERT(ip->i_df.if_ext_max ==
3603                XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t));
3604         return error;
3605 }
3606
3607 /*
3608  * Add the extent to the list of extents to be free at transaction end.
3609  * The list is maintained sorted (by block number).
3610  */
3611 /* ARGSUSED */
3612 void
3613 xfs_bmap_add_free(
3614         xfs_fsblock_t           bno,            /* fs block number of extent */
3615         xfs_filblks_t           len,            /* length of extent */
3616         xfs_bmap_free_t         *flist,         /* list of extents */
3617         xfs_mount_t             *mp)            /* mount point structure */
3618 {
3619         xfs_bmap_free_item_t    *cur;           /* current (next) element */
3620         xfs_bmap_free_item_t    *new;           /* new element */
3621         xfs_bmap_free_item_t    *prev;          /* previous element */
3622 #ifdef DEBUG
3623         xfs_agnumber_t          agno;
3624         xfs_agblock_t           agbno;
3625
3626         ASSERT(bno != NULLFSBLOCK);
3627         ASSERT(len > 0);
3628         ASSERT(len <= MAXEXTLEN);
3629         ASSERT(!isnullstartblock(bno));
3630         agno = XFS_FSB_TO_AGNO(mp, bno);
3631         agbno = XFS_FSB_TO_AGBNO(mp, bno);
3632         ASSERT(agno < mp->m_sb.sb_agcount);
3633         ASSERT(agbno < mp->m_sb.sb_agblocks);
3634         ASSERT(len < mp->m_sb.sb_agblocks);
3635         ASSERT(agbno + len <= mp->m_sb.sb_agblocks);
3636 #endif
3637         ASSERT(xfs_bmap_free_item_zone != NULL);
3638         new = kmem_zone_alloc(xfs_bmap_free_item_zone, KM_SLEEP);
3639         new->xbfi_startblock = bno;
3640         new->xbfi_blockcount = (xfs_extlen_t)len;
3641         for (prev = NULL, cur = flist->xbf_first;
3642              cur != NULL;
3643              prev = cur, cur = cur->xbfi_next) {
3644                 if (cur->xbfi_startblock >= bno)
3645                         break;
3646         }
3647         if (prev)
3648                 prev->xbfi_next = new;
3649         else
3650                 flist->xbf_first = new;
3651         new->xbfi_next = cur;
3652         flist->xbf_count++;
3653 }
3654
3655 /*
3656  * Compute and fill in the value of the maximum depth of a bmap btree
3657  * in this filesystem.  Done once, during mount.
3658  */
3659 void
3660 xfs_bmap_compute_maxlevels(
3661         xfs_mount_t     *mp,            /* file system mount structure */
3662         int             whichfork)      /* data or attr fork */
3663 {
3664         int             level;          /* btree level */
3665         uint            maxblocks;      /* max blocks at this level */
3666         uint            maxleafents;    /* max leaf entries possible */
3667         int             maxrootrecs;    /* max records in root block */
3668         int             minleafrecs;    /* min records in leaf block */
3669         int             minnoderecs;    /* min records in node block */
3670         int             sz;             /* root block size */
3671
3672         /*
3673          * The maximum number of extents in a file, hence the maximum
3674          * number of leaf entries, is controlled by the type of di_nextents
3675          * (a signed 32-bit number, xfs_extnum_t), or by di_anextents
3676          * (a signed 16-bit number, xfs_aextnum_t).
3677          *
3678          * Note that we can no longer assume that if we are in ATTR1 that
3679          * the fork offset of all the inodes will be
3680          * (xfs_default_attroffset(ip) >> 3) because we could have mounted
3681          * with ATTR2 and then mounted back with ATTR1, keeping the
3682          * di_forkoff's fixed but probably at various positions. Therefore,
3683          * for both ATTR1 and ATTR2 we have to assume the worst case scenario
3684          * of a minimum size available.
3685          */
3686         if (whichfork == XFS_DATA_FORK) {
3687                 maxleafents = MAXEXTNUM;
3688                 sz = XFS_BMDR_SPACE_CALC(MINDBTPTRS);
3689         } else {
3690                 maxleafents = MAXAEXTNUM;
3691                 sz = XFS_BMDR_SPACE_CALC(MINABTPTRS);
3692         }
3693         maxrootrecs = xfs_bmdr_maxrecs(mp, sz, 0);
3694         minleafrecs = mp->m_bmap_dmnr[0];
3695         minnoderecs = mp->m_bmap_dmnr[1];
3696         maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs;
3697         for (level = 1; maxblocks > 1; level++) {
3698                 if (maxblocks <= maxrootrecs)
3699                         maxblocks = 1;
3700                 else
3701                         maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs;
3702         }
3703         mp->m_bm_maxlevels[whichfork] = level;
3704 }
3705
3706 /*
3707  * Routine to be called at transaction's end by xfs_bmapi, xfs_bunmapi
3708  * caller.  Frees all the extents that need freeing, which must be done
3709  * last due to locking considerations.  We never free any extents in
3710  * the first transaction.
3711  *
3712  * Return 1 if the given transaction was committed and a new one
3713  * started, and 0 otherwise in the committed parameter.
3714  */
3715 int                                             /* error */
3716 xfs_bmap_finish(
3717         xfs_trans_t             **tp,           /* transaction pointer addr */
3718         xfs_bmap_free_t         *flist,         /* i/o: list extents to free */
3719         int                     *committed)     /* xact committed or not */
3720 {
3721         xfs_efd_log_item_t      *efd;           /* extent free data */
3722         xfs_efi_log_item_t      *efi;           /* extent free intention */
3723         int                     error;          /* error return value */
3724         xfs_bmap_free_item_t    *free;          /* free extent item */
3725         unsigned int            logres;         /* new log reservation */
3726         unsigned int            logcount;       /* new log count */
3727         xfs_mount_t             *mp;            /* filesystem mount structure */
3728         xfs_bmap_free_item_t    *next;          /* next item on free list */
3729         xfs_trans_t             *ntp;           /* new transaction pointer */
3730
3731         ASSERT((*tp)->t_flags & XFS_TRANS_PERM_LOG_RES);
3732         if (flist->xbf_count == 0) {
3733                 *committed = 0;
3734                 return 0;
3735         }
3736         ntp = *tp;
3737         efi = xfs_trans_get_efi(ntp, flist->xbf_count);
3738         for (free = flist->xbf_first; free; free = free->xbfi_next)
3739                 xfs_trans_log_efi_extent(ntp, efi, free->xbfi_startblock,
3740                         free->xbfi_blockcount);
3741         logres = ntp->t_log_res;
3742         logcount = ntp->t_log_count;
3743         ntp = xfs_trans_dup(*tp);
3744         error = xfs_trans_commit(*tp, 0);
3745         *tp = ntp;
3746         *committed = 1;
3747         /*
3748          * We have a new transaction, so we should return committed=1,
3749          * even though we're returning an error.
3750          */
3751         if (error)
3752                 return error;
3753
3754         /*
3755          * transaction commit worked ok so we can drop the extra ticket
3756          * reference that we gained in xfs_trans_dup()
3757          */
3758         xfs_log_ticket_put(ntp->t_ticket);
3759
3760         if ((error = xfs_trans_reserve(ntp, 0, logres, 0, XFS_TRANS_PERM_LOG_RES,
3761                         logcount)))
3762                 return error;
3763         efd = xfs_trans_get_efd(ntp, efi, flist->xbf_count);
3764         for (free = flist->xbf_first; free != NULL; free = next) {
3765                 next = free->xbfi_next;
3766                 if ((error = xfs_free_extent(ntp, free->xbfi_startblock,
3767                                 free->xbfi_blockcount))) {
3768                         /*
3769                          * The bmap free list will be cleaned up at a
3770                          * higher level.  The EFI will be canceled when
3771                          * this transaction is aborted.
3772                          * Need to force shutdown here to make sure it
3773                          * happens, since this transaction may not be
3774                          * dirty yet.
3775                          */
3776                         mp = ntp->t_mountp;
3777                         if (!XFS_FORCED_SHUTDOWN(mp))
3778                                 xfs_force_shutdown(mp,
3779                                                    (error == EFSCORRUPTED) ?
3780                                                    SHUTDOWN_CORRUPT_INCORE :
3781                                                    SHUTDOWN_META_IO_ERROR);
3782                         return error;
3783                 }
3784                 xfs_trans_log_efd_extent(ntp, efd, free->xbfi_startblock,
3785                         free->xbfi_blockcount);
3786                 xfs_bmap_del_free(flist, NULL, free);
3787         }
3788         return 0;
3789 }
3790
3791 /*
3792  * Free up any items left in the list.
3793  */
3794 void
3795 xfs_bmap_cancel(
3796         xfs_bmap_free_t         *flist) /* list of bmap_free_items */
3797 {
3798         xfs_bmap_free_item_t    *free;  /* free list item */
3799         xfs_bmap_free_item_t    *next;
3800
3801         if (flist->xbf_count == 0)
3802                 return;
3803         ASSERT(flist->xbf_first != NULL);
3804         for (free = flist->xbf_first; free; free = next) {
3805                 next = free->xbfi_next;
3806                 xfs_bmap_del_free(flist, NULL, free);
3807         }
3808         ASSERT(flist->xbf_count == 0);
3809 }
3810
3811 /*
3812  * Returns the file-relative block number of the first unused block(s)
3813  * in the file with at least "len" logically contiguous blocks free.
3814  * This is the lowest-address hole if the file has holes, else the first block
3815  * past the end of file.
3816  * Return 0 if the file is currently local (in-inode).
3817  */
3818 int                                             /* error */
3819 xfs_bmap_first_unused(
3820         xfs_trans_t     *tp,                    /* transaction pointer */
3821         xfs_inode_t     *ip,                    /* incore inode */
3822         xfs_extlen_t    len,                    /* size of hole to find */
3823         xfs_fileoff_t   *first_unused,          /* unused block */
3824         int             whichfork)              /* data or attr fork */
3825 {
3826         int             error;                  /* error return value */
3827         int             idx;                    /* extent record index */
3828         xfs_ifork_t     *ifp;                   /* inode fork pointer */
3829         xfs_fileoff_t   lastaddr;               /* last block number seen */
3830         xfs_fileoff_t   lowest;                 /* lowest useful block */
3831         xfs_fileoff_t   max;                    /* starting useful block */
3832         xfs_fileoff_t   off;                    /* offset for this block */
3833         xfs_extnum_t    nextents;               /* number of extent entries */
3834
3835         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE ||
3836                XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ||
3837                XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
3838         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
3839                 *first_unused = 0;
3840                 return 0;
3841         }
3842         ifp = XFS_IFORK_PTR(ip, whichfork);
3843         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
3844             (error = xfs_iread_extents(tp, ip, whichfork)))
3845                 return error;
3846         lowest = *first_unused;
3847         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3848         for (idx = 0, lastaddr = 0, max = lowest; idx < nextents; idx++) {
3849                 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, idx);
3850                 off = xfs_bmbt_get_startoff(ep);
3851                 /*
3852                  * See if the hole before this extent will work.
3853                  */
3854                 if (off >= lowest + len && off - max >= len) {
3855                         *first_unused = max;
3856                         return 0;
3857                 }
3858                 lastaddr = off + xfs_bmbt_get_blockcount(ep);
3859                 max = XFS_FILEOFF_MAX(lastaddr, lowest);
3860         }
3861         *first_unused = max;
3862         return 0;
3863 }
3864
3865 /*
3866  * Returns the file-relative block number of the last block + 1 before
3867  * last_block (input value) in the file.
3868  * This is not based on i_size, it is based on the extent records.
3869  * Returns 0 for local files, as they do not have extent records.
3870  */
3871 int                                             /* error */
3872 xfs_bmap_last_before(
3873         xfs_trans_t     *tp,                    /* transaction pointer */
3874         xfs_inode_t     *ip,                    /* incore inode */
3875         xfs_fileoff_t   *last_block,            /* last block */
3876         int             whichfork)              /* data or attr fork */
3877 {
3878         xfs_fileoff_t   bno;                    /* input file offset */
3879         int             eof;                    /* hit end of file */
3880         xfs_bmbt_rec_host_t *ep;                /* pointer to last extent */
3881         int             error;                  /* error return value */
3882         xfs_bmbt_irec_t got;                    /* current extent value */
3883         xfs_ifork_t     *ifp;                   /* inode fork pointer */
3884         xfs_extnum_t    lastx;                  /* last extent used */
3885         xfs_bmbt_irec_t prev;                   /* previous extent value */
3886
3887         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
3888             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
3889             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL)
3890                return XFS_ERROR(EIO);
3891         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
3892                 *last_block = 0;
3893                 return 0;
3894         }
3895         ifp = XFS_IFORK_PTR(ip, whichfork);
3896         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
3897             (error = xfs_iread_extents(tp, ip, whichfork)))
3898                 return error;
3899         bno = *last_block - 1;
3900         ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
3901                 &prev);
3902         if (eof || xfs_bmbt_get_startoff(ep) > bno) {
3903                 if (prev.br_startoff == NULLFILEOFF)
3904                         *last_block = 0;
3905                 else
3906                         *last_block = prev.br_startoff + prev.br_blockcount;
3907         }
3908         /*
3909          * Otherwise *last_block is already the right answer.
3910          */
3911         return 0;
3912 }
3913
3914 STATIC int
3915 xfs_bmap_last_extent(
3916         struct xfs_trans        *tp,
3917         struct xfs_inode        *ip,
3918         int                     whichfork,
3919         struct xfs_bmbt_irec    *rec,
3920         int                     *is_empty)
3921 {
3922         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, whichfork);
3923         int                     error;
3924         int                     nextents;
3925
3926         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
3927                 error = xfs_iread_extents(tp, ip, whichfork);
3928                 if (error)
3929                         return error;
3930         }
3931
3932         nextents = ifp->if_bytes / sizeof(xfs_bmbt_rec_t);
3933         if (nextents == 0) {
3934                 *is_empty = 1;
3935                 return 0;
3936         }
3937
3938         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, nextents - 1), rec);
3939         *is_empty = 0;
3940         return 0;
3941 }
3942
3943 /*
3944  * Check the last inode extent to determine whether this allocation will result
3945  * in blocks being allocated at the end of the file. When we allocate new data
3946  * blocks at the end of the file which do not start at the previous data block,
3947  * we will try to align the new blocks at stripe unit boundaries.
3948  *
3949  * Returns 0 in *aeof if the file (fork) is empty as any new write will be at,
3950  * or past the EOF.
3951  */
3952 STATIC int
3953 xfs_bmap_isaeof(
3954         struct xfs_inode        *ip,
3955         xfs_fileoff_t           off,
3956         int                     whichfork,
3957         char                    *aeof)
3958 {
3959         struct xfs_bmbt_irec    rec;
3960         int                     is_empty;
3961         int                     error;
3962
3963         *aeof = 0;
3964         error = xfs_bmap_last_extent(NULL, ip, whichfork, &rec, &is_empty);
3965         if (error || is_empty)
3966                 return error;
3967
3968         /*
3969          * Check if we are allocation or past the last extent, or at least into
3970          * the last delayed allocated extent.
3971          */
3972         *aeof = off >= rec.br_startoff + rec.br_blockcount ||
3973                 (off >= rec.br_startoff && isnullstartblock(rec.br_startblock));
3974         return 0;
3975 }
3976
3977 /*
3978  * Check if the endoff is outside the last extent. If so the caller will grow
3979  * the allocation to a stripe unit boundary.  All offsets are considered outside
3980  * the end of file for an empty fork, so 1 is returned in *eof in that case.
3981  */
3982 int
3983 xfs_bmap_eof(
3984         struct xfs_inode        *ip,
3985         xfs_fileoff_t           endoff,
3986         int                     whichfork,
3987         int                     *eof)
3988 {
3989         struct xfs_bmbt_irec    rec;
3990         int                     error;
3991
3992         error = xfs_bmap_last_extent(NULL, ip, whichfork, &rec, eof);
3993         if (error || *eof)
3994                 return error;
3995
3996         *eof = endoff >= rec.br_startoff + rec.br_blockcount;
3997         return 0;
3998 }
3999
4000 /*
4001  * Returns the file-relative block number of the first block past eof in
4002  * the file.  This is not based on i_size, it is based on the extent records.
4003  * Returns 0 for local files, as they do not have extent records.
4004  */
4005 int
4006 xfs_bmap_last_offset(
4007         struct xfs_trans        *tp,
4008         struct xfs_inode        *ip,
4009         xfs_fileoff_t           *last_block,
4010         int                     whichfork)
4011 {
4012         struct xfs_bmbt_irec    rec;
4013         int                     is_empty;
4014         int                     error;
4015
4016         *last_block = 0;
4017
4018         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL)
4019                 return 0;
4020
4021         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
4022             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
4023                return XFS_ERROR(EIO);
4024
4025         error = xfs_bmap_last_extent(NULL, ip, whichfork, &rec, &is_empty);
4026         if (error || is_empty)
4027                 return error;
4028
4029         *last_block = rec.br_startoff + rec.br_blockcount;
4030         return 0;
4031 }
4032
4033 /*
4034  * Returns whether the selected fork of the inode has exactly one
4035  * block or not.  For the data fork we check this matches di_size,
4036  * implying the file's range is 0..bsize-1.
4037  */
4038 int                                     /* 1=>1 block, 0=>otherwise */
4039 xfs_bmap_one_block(
4040         xfs_inode_t     *ip,            /* incore inode */
4041         int             whichfork)      /* data or attr fork */
4042 {
4043         xfs_bmbt_rec_host_t *ep;        /* ptr to fork's extent */
4044         xfs_ifork_t     *ifp;           /* inode fork pointer */
4045         int             rval;           /* return value */
4046         xfs_bmbt_irec_t s;              /* internal version of extent */
4047
4048 #ifndef DEBUG
4049         if (whichfork == XFS_DATA_FORK) {
4050                 return S_ISREG(ip->i_d.di_mode) ?
4051                         (ip->i_size == ip->i_mount->m_sb.sb_blocksize) :
4052                         (ip->i_d.di_size == ip->i_mount->m_sb.sb_blocksize);
4053         }
4054 #endif  /* !DEBUG */
4055         if (XFS_IFORK_NEXTENTS(ip, whichfork) != 1)
4056                 return 0;
4057         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
4058                 return 0;
4059         ifp = XFS_IFORK_PTR(ip, whichfork);
4060         ASSERT(ifp->if_flags & XFS_IFEXTENTS);
4061         ep = xfs_iext_get_ext(ifp, 0);
4062         xfs_bmbt_get_all(ep, &s);
4063         rval = s.br_startoff == 0 && s.br_blockcount == 1;
4064         if (rval && whichfork == XFS_DATA_FORK)
4065                 ASSERT(ip->i_size == ip->i_mount->m_sb.sb_blocksize);
4066         return rval;
4067 }
4068
4069 STATIC int
4070 xfs_bmap_sanity_check(
4071         struct xfs_mount        *mp,
4072         struct xfs_buf          *bp,
4073         int                     level)
4074 {
4075         struct xfs_btree_block  *block = XFS_BUF_TO_BLOCK(bp);
4076
4077         if (block->bb_magic != cpu_to_be32(XFS_BMAP_MAGIC) ||
4078             be16_to_cpu(block->bb_level) != level ||
4079             be16_to_cpu(block->bb_numrecs) == 0 ||
4080             be16_to_cpu(block->bb_numrecs) > mp->m_bmap_dmxr[level != 0])
4081                 return 0;
4082         return 1;
4083 }
4084
4085 /*
4086  * Read in the extents to if_extents.
4087  * All inode fields are set up by caller, we just traverse the btree
4088  * and copy the records in. If the file system cannot contain unwritten
4089  * extents, the records are checked for no "state" flags.
4090  */
4091 int                                     /* error */
4092 xfs_bmap_read_extents(
4093         xfs_trans_t             *tp,    /* transaction pointer */
4094         xfs_inode_t             *ip,    /* incore inode */
4095         int                     whichfork) /* data or attr fork */
4096 {
4097         struct xfs_btree_block  *block; /* current btree block */
4098         xfs_fsblock_t           bno;    /* block # of "block" */
4099         xfs_buf_t               *bp;    /* buffer for "block" */
4100         int                     error;  /* error return value */
4101         xfs_exntfmt_t           exntf;  /* XFS_EXTFMT_NOSTATE, if checking */
4102         xfs_extnum_t            i, j;   /* index into the extents list */
4103         xfs_ifork_t             *ifp;   /* fork structure */
4104         int                     level;  /* btree level, for checking */
4105         xfs_mount_t             *mp;    /* file system mount structure */
4106         __be64                  *pp;    /* pointer to block address */
4107         /* REFERENCED */
4108         xfs_extnum_t            room;   /* number of entries there's room for */
4109
4110         bno = NULLFSBLOCK;
4111         mp = ip->i_mount;
4112         ifp = XFS_IFORK_PTR(ip, whichfork);
4113         exntf = (whichfork != XFS_DATA_FORK) ? XFS_EXTFMT_NOSTATE :
4114                                         XFS_EXTFMT_INODE(ip);
4115         block = ifp->if_broot;
4116         /*
4117          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
4118          */
4119         level = be16_to_cpu(block->bb_level);
4120         ASSERT(level > 0);
4121         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
4122         bno = be64_to_cpu(*pp);
4123         ASSERT(bno != NULLDFSBNO);
4124         ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
4125         ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
4126         /*
4127          * Go down the tree until leaf level is reached, following the first
4128          * pointer (leftmost) at each level.
4129          */
4130         while (level-- > 0) {
4131                 if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
4132                                 XFS_BMAP_BTREE_REF)))
4133                         return error;
4134                 block = XFS_BUF_TO_BLOCK(bp);
4135                 XFS_WANT_CORRUPTED_GOTO(
4136                         xfs_bmap_sanity_check(mp, bp, level),
4137                         error0);
4138                 if (level == 0)
4139                         break;
4140                 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
4141                 bno = be64_to_cpu(*pp);
4142                 XFS_WANT_CORRUPTED_GOTO(XFS_FSB_SANITY_CHECK(mp, bno), error0);
4143                 xfs_trans_brelse(tp, bp);
4144         }
4145         /*
4146          * Here with bp and block set to the leftmost leaf node in the tree.
4147          */
4148         room = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4149         i = 0;
4150         /*
4151          * Loop over all leaf nodes.  Copy information to the extent records.
4152          */
4153         for (;;) {
4154                 xfs_bmbt_rec_t  *frp;
4155                 xfs_fsblock_t   nextbno;
4156                 xfs_extnum_t    num_recs;
4157                 xfs_extnum_t    start;
4158
4159                 num_recs = xfs_btree_get_numrecs(block);
4160                 if (unlikely(i + num_recs > room)) {
4161                         ASSERT(i + num_recs <= room);
4162                         xfs_warn(ip->i_mount,
4163                                 "corrupt dinode %Lu, (btree extents).",
4164                                 (unsigned long long) ip->i_ino);
4165                         XFS_CORRUPTION_ERROR("xfs_bmap_read_extents(1)",
4166                                 XFS_ERRLEVEL_LOW, ip->i_mount, block);
4167                         goto error0;
4168                 }
4169                 XFS_WANT_CORRUPTED_GOTO(
4170                         xfs_bmap_sanity_check(mp, bp, 0),
4171                         error0);
4172                 /*
4173                  * Read-ahead the next leaf block, if any.
4174                  */
4175                 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
4176                 if (nextbno != NULLFSBLOCK)
4177                         xfs_btree_reada_bufl(mp, nextbno, 1);
4178                 /*
4179                  * Copy records into the extent records.
4180                  */
4181                 frp = XFS_BMBT_REC_ADDR(mp, block, 1);
4182                 start = i;
4183                 for (j = 0; j < num_recs; j++, i++, frp++) {
4184                         xfs_bmbt_rec_host_t *trp = xfs_iext_get_ext(ifp, i);
4185                         trp->l0 = be64_to_cpu(frp->l0);
4186                         trp->l1 = be64_to_cpu(frp->l1);
4187                 }
4188                 if (exntf == XFS_EXTFMT_NOSTATE) {
4189                         /*
4190                          * Check all attribute bmap btree records and
4191                          * any "older" data bmap btree records for a
4192                          * set bit in the "extent flag" position.
4193                          */
4194                         if (unlikely(xfs_check_nostate_extents(ifp,
4195                                         start, num_recs))) {
4196                                 XFS_ERROR_REPORT("xfs_bmap_read_extents(2)",
4197                                                  XFS_ERRLEVEL_LOW,
4198                                                  ip->i_mount);
4199                                 goto error0;
4200                         }
4201                 }
4202                 xfs_trans_brelse(tp, bp);
4203                 bno = nextbno;
4204                 /*
4205                  * If we've reached the end, stop.
4206                  */
4207                 if (bno == NULLFSBLOCK)
4208                         break;
4209                 if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
4210                                 XFS_BMAP_BTREE_REF)))
4211                         return error;
4212                 block = XFS_BUF_TO_BLOCK(bp);
4213         }
4214         ASSERT(i == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)));
4215         ASSERT(i == XFS_IFORK_NEXTENTS(ip, whichfork));
4216         XFS_BMAP_TRACE_EXLIST(ip, i, whichfork);
4217         return 0;
4218 error0:
4219         xfs_trans_brelse(tp, bp);
4220         return XFS_ERROR(EFSCORRUPTED);
4221 }
4222
4223 #ifdef DEBUG
4224 /*
4225  * Add bmap trace insert entries for all the contents of the extent records.
4226  */
4227 void
4228 xfs_bmap_trace_exlist(
4229         xfs_inode_t     *ip,            /* incore inode pointer */
4230         xfs_extnum_t    cnt,            /* count of entries in the list */
4231         int             whichfork,      /* data or attr fork */
4232         unsigned long   caller_ip)
4233 {
4234         xfs_extnum_t    idx;            /* extent record index */
4235         xfs_ifork_t     *ifp;           /* inode fork pointer */
4236         int             state = 0;
4237
4238         if (whichfork == XFS_ATTR_FORK)
4239                 state |= BMAP_ATTRFORK;
4240
4241         ifp = XFS_IFORK_PTR(ip, whichfork);
4242         ASSERT(cnt == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)));
4243         for (idx = 0; idx < cnt; idx++)
4244                 trace_xfs_extlist(ip, idx, whichfork, caller_ip);
4245 }
4246
4247 /*
4248  * Validate that the bmbt_irecs being returned from bmapi are valid
4249  * given the callers original parameters.  Specifically check the
4250  * ranges of the returned irecs to ensure that they only extent beyond
4251  * the given parameters if the XFS_BMAPI_ENTIRE flag was set.
4252  */
4253 STATIC void
4254 xfs_bmap_validate_ret(
4255         xfs_fileoff_t           bno,
4256         xfs_filblks_t           len,
4257         int                     flags,
4258         xfs_bmbt_irec_t         *mval,
4259         int                     nmap,
4260         int                     ret_nmap)
4261 {
4262         int                     i;              /* index to map values */
4263
4264         ASSERT(ret_nmap <= nmap);
4265
4266         for (i = 0; i < ret_nmap; i++) {
4267                 ASSERT(mval[i].br_blockcount > 0);
4268                 if (!(flags & XFS_BMAPI_ENTIRE)) {
4269                         ASSERT(mval[i].br_startoff >= bno);
4270                         ASSERT(mval[i].br_blockcount <= len);
4271                         ASSERT(mval[i].br_startoff + mval[i].br_blockcount <=
4272                                bno + len);
4273                 } else {
4274                         ASSERT(mval[i].br_startoff < bno + len);
4275                         ASSERT(mval[i].br_startoff + mval[i].br_blockcount >
4276                                bno);
4277                 }
4278                 ASSERT(i == 0 ||
4279                        mval[i - 1].br_startoff + mval[i - 1].br_blockcount ==
4280                        mval[i].br_startoff);
4281                 ASSERT(mval[i].br_startblock != DELAYSTARTBLOCK &&
4282                        mval[i].br_startblock != HOLESTARTBLOCK);
4283                 ASSERT(mval[i].br_state == XFS_EXT_NORM ||
4284                        mval[i].br_state == XFS_EXT_UNWRITTEN);
4285         }
4286 }
4287 #endif /* DEBUG */
4288
4289
4290 /*
4291  * Trim the returned map to the required bounds
4292  */
4293 STATIC void
4294 xfs_bmapi_trim_map(
4295         struct xfs_bmbt_irec    *mval,
4296         struct xfs_bmbt_irec    *got,
4297         xfs_fileoff_t           *bno,
4298         xfs_filblks_t           len,
4299         xfs_fileoff_t           obno,
4300         xfs_fileoff_t           end,
4301         int                     n,
4302         int                     flags)
4303 {
4304         if ((flags & XFS_BMAPI_ENTIRE) ||
4305             got->br_startoff + got->br_blockcount <= obno) {
4306                 *mval = *got;
4307                 if (isnullstartblock(got->br_startblock))
4308                         mval->br_startblock = DELAYSTARTBLOCK;
4309                 return;
4310         }
4311
4312         if (obno > *bno)
4313                 *bno = obno;
4314         ASSERT((*bno >= obno) || (n == 0));
4315         ASSERT(*bno < end);
4316         mval->br_startoff = *bno;
4317         if (isnullstartblock(got->br_startblock))
4318                 mval->br_startblock = DELAYSTARTBLOCK;
4319         else
4320                 mval->br_startblock = got->br_startblock +
4321                                         (*bno - got->br_startoff);
4322         /*
4323          * Return the minimum of what we got and what we asked for for
4324          * the length.  We can use the len variable here because it is
4325          * modified below and we could have been there before coming
4326          * here if the first part of the allocation didn't overlap what
4327          * was asked for.
4328          */
4329         mval->br_blockcount = XFS_FILBLKS_MIN(end - *bno,
4330                         got->br_blockcount - (*bno - got->br_startoff));
4331         mval->br_state = got->br_state;
4332         ASSERT(mval->br_blockcount <= len);
4333         return;
4334 }
4335
4336 /*
4337  * Update and validate the extent map to return
4338  */
4339 STATIC void
4340 xfs_bmapi_update_map(
4341         struct xfs_bmbt_irec    **map,
4342         xfs_fileoff_t           *bno,
4343         xfs_filblks_t           *len,
4344         xfs_fileoff_t           obno,
4345         xfs_fileoff_t           end,
4346         int                     *n,
4347         int                     flags)
4348 {
4349         xfs_bmbt_irec_t *mval = *map;
4350
4351         ASSERT((flags & XFS_BMAPI_ENTIRE) ||
4352                ((mval->br_startoff + mval->br_blockcount) <= end));
4353         ASSERT((flags & XFS_BMAPI_ENTIRE) || (mval->br_blockcount <= *len) ||
4354                (mval->br_startoff < obno));
4355
4356         *bno = mval->br_startoff + mval->br_blockcount;
4357         *len = end - *bno;
4358         if (*n > 0 && mval->br_startoff == mval[-1].br_startoff) {
4359                 /* update previous map with new information */
4360                 ASSERT(mval->br_startblock == mval[-1].br_startblock);
4361                 ASSERT(mval->br_blockcount > mval[-1].br_blockcount);
4362                 ASSERT(mval->br_state == mval[-1].br_state);
4363                 mval[-1].br_blockcount = mval->br_blockcount;
4364                 mval[-1].br_state = mval->br_state;
4365         } else if (*n > 0 && mval->br_startblock != DELAYSTARTBLOCK &&
4366                    mval[-1].br_startblock != DELAYSTARTBLOCK &&
4367                    mval[-1].br_startblock != HOLESTARTBLOCK &&
4368                    mval->br_startblock == mval[-1].br_startblock +
4369                                           mval[-1].br_blockcount &&
4370                    ((flags & XFS_BMAPI_IGSTATE) ||
4371                         mval[-1].br_state == mval->br_state)) {
4372                 ASSERT(mval->br_startoff ==
4373                        mval[-1].br_startoff + mval[-1].br_blockcount);
4374                 mval[-1].br_blockcount += mval->br_blockcount;
4375         } else if (*n > 0 &&
4376                    mval->br_startblock == DELAYSTARTBLOCK &&
4377                    mval[-1].br_startblock == DELAYSTARTBLOCK &&
4378                    mval->br_startoff ==
4379                    mval[-1].br_startoff + mval[-1].br_blockcount) {
4380                 mval[-1].br_blockcount += mval->br_blockcount;
4381                 mval[-1].br_state = mval->br_state;
4382         } else if (!((*n == 0) &&
4383                      ((mval->br_startoff + mval->br_blockcount) <=
4384                       obno))) {
4385                 mval++;
4386                 (*n)++;
4387         }
4388         *map = mval;
4389 }
4390
4391 /*
4392  * Map file blocks to filesystem blocks without allocation.
4393  */
4394 int
4395 xfs_bmapi_read(
4396         struct xfs_inode        *ip,
4397         xfs_fileoff_t           bno,
4398         xfs_filblks_t           len,
4399         struct xfs_bmbt_irec    *mval,
4400         int                     *nmap,
4401         int                     flags)
4402 {
4403         struct xfs_mount        *mp = ip->i_mount;
4404         struct xfs_ifork        *ifp;
4405         struct xfs_bmbt_irec    got;
4406         struct xfs_bmbt_irec    prev;
4407         xfs_fileoff_t           obno;
4408         xfs_fileoff_t           end;
4409         xfs_extnum_t            lastx;
4410         int                     error;
4411         int                     eof;
4412         int                     n = 0;
4413         int                     whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4414                                                 XFS_ATTR_FORK : XFS_DATA_FORK;
4415
4416         ASSERT(*nmap >= 1);
4417         ASSERT(!(flags & ~(XFS_BMAPI_ATTRFORK|XFS_BMAPI_ENTIRE|
4418                            XFS_BMAPI_IGSTATE)));
4419
4420         if (unlikely(XFS_TEST_ERROR(
4421             (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4422              XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
4423              mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4424                 XFS_ERROR_REPORT("xfs_bmapi_read", XFS_ERRLEVEL_LOW, mp);
4425                 return XFS_ERROR(EFSCORRUPTED);
4426         }
4427
4428         if (XFS_FORCED_SHUTDOWN(mp))
4429                 return XFS_ERROR(EIO);
4430
4431         XFS_STATS_INC(xs_blk_mapr);
4432
4433         ifp = XFS_IFORK_PTR(ip, whichfork);
4434         ASSERT(ifp->if_ext_max ==
4435                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
4436
4437         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4438                 error = xfs_iread_extents(NULL, ip, whichfork);
4439                 if (error)
4440                         return error;
4441         }
4442
4443         xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got, &prev);
4444         end = bno + len;
4445         obno = bno;
4446
4447         while (bno < end && n < *nmap) {
4448                 /* Reading past eof, act as though there's a hole up to end. */
4449                 if (eof)
4450                         got.br_startoff = end;
4451                 if (got.br_startoff > bno) {
4452                         /* Reading in a hole.  */
4453                         mval->br_startoff = bno;
4454                         mval->br_startblock = HOLESTARTBLOCK;
4455                         mval->br_blockcount =
4456                                 XFS_FILBLKS_MIN(len, got.br_startoff - bno);
4457                         mval->br_state = XFS_EXT_NORM;
4458                         bno += mval->br_blockcount;
4459                         len -= mval->br_blockcount;
4460                         mval++;
4461                         n++;
4462                         continue;
4463                 }
4464
4465                 /* set up the extent map to return. */
4466                 xfs_bmapi_trim_map(mval, &got, &bno, len, obno, end, n, flags);
4467                 xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4468
4469                 /* If we're done, stop now. */
4470                 if (bno >= end || n >= *nmap)
4471                         break;
4472
4473                 /* Else go on to the next record. */
4474                 if (++lastx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t))
4475                         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx), &got);
4476                 else
4477                         eof = 1;
4478         }
4479         *nmap = n;
4480         return 0;
4481 }
4482
4483 STATIC int
4484 xfs_bmapi_reserve_delalloc(
4485         struct xfs_inode        *ip,
4486         xfs_fileoff_t           aoff,
4487         xfs_filblks_t           len,
4488         struct xfs_bmbt_irec    *got,
4489         struct xfs_bmbt_irec    *prev,
4490         xfs_extnum_t            *lastx,
4491         int                     eof)
4492 {
4493         struct xfs_mount        *mp = ip->i_mount;
4494         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
4495         xfs_extlen_t            alen;
4496         xfs_extlen_t            indlen;
4497         char                    rt = XFS_IS_REALTIME_INODE(ip);
4498         xfs_extlen_t            extsz;
4499         int                     error;
4500
4501         alen = XFS_FILBLKS_MIN(len, MAXEXTLEN);
4502         if (!eof)
4503                 alen = XFS_FILBLKS_MIN(alen, got->br_startoff - aoff);
4504
4505         /* Figure out the extent size, adjust alen */
4506         extsz = xfs_get_extsz_hint(ip);
4507         if (extsz) {
4508                 /*
4509                  * Make sure we don't exceed a single extent length when we
4510                  * align the extent by reducing length we are going to
4511                  * allocate by the maximum amount extent size aligment may
4512                  * require.
4513                  */
4514                 alen = XFS_FILBLKS_MIN(len, MAXEXTLEN - (2 * extsz - 1));
4515                 error = xfs_bmap_extsize_align(mp, got, prev, extsz, rt, eof,
4516                                                1, 0, &aoff, &alen);
4517                 ASSERT(!error);
4518         }
4519
4520         if (rt)
4521                 extsz = alen / mp->m_sb.sb_rextsize;
4522
4523         /*
4524          * Make a transaction-less quota reservation for delayed allocation
4525          * blocks.  This number gets adjusted later.  We return if we haven't
4526          * allocated blocks already inside this loop.
4527          */
4528         error = xfs_trans_reserve_quota_nblks(NULL, ip, (long)alen, 0,
4529                         rt ? XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4530         if (error)
4531                 return error;
4532
4533         /*
4534          * Split changing sb for alen and indlen since they could be coming
4535          * from different places.
4536          */
4537         indlen = (xfs_extlen_t)xfs_bmap_worst_indlen(ip, alen);
4538         ASSERT(indlen > 0);
4539
4540         if (rt) {
4541                 error = xfs_mod_incore_sb(mp, XFS_SBS_FREXTENTS,
4542                                           -((int64_t)extsz), 0);
4543         } else {
4544                 error = xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
4545                                                  -((int64_t)alen), 0);
4546         }
4547
4548         if (error)
4549                 goto out_unreserve_quota;
4550
4551         error = xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
4552                                          -((int64_t)indlen), 0);
4553         if (error)
4554                 goto out_unreserve_blocks;
4555
4556
4557         ip->i_delayed_blks += alen;
4558
4559         got->br_startoff = aoff;
4560         got->br_startblock = nullstartblock(indlen);
4561         got->br_blockcount = alen;
4562         got->br_state = XFS_EXT_NORM;
4563         xfs_bmap_add_extent_hole_delay(ip, lastx, got);
4564
4565         /*
4566          * Update our extent pointer, given that xfs_bmap_add_extent_hole_delay
4567          * might have merged it into one of the neighbouring ones.
4568          */
4569         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *lastx), got);
4570
4571         ASSERT(got->br_startoff <= aoff);
4572         ASSERT(got->br_startoff + got->br_blockcount >= aoff + alen);
4573         ASSERT(isnullstartblock(got->br_startblock));
4574         ASSERT(got->br_state == XFS_EXT_NORM);
4575         return 0;
4576
4577 out_unreserve_blocks:
4578         if (rt)
4579                 xfs_mod_incore_sb(mp, XFS_SBS_FREXTENTS, extsz, 0);
4580         else
4581                 xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS, alen, 0);
4582 out_unreserve_quota:
4583         if (XFS_IS_QUOTA_ON(mp))
4584                 xfs_trans_unreserve_quota_nblks(NULL, ip, alen, 0, rt ?
4585                                 XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4586         return error;
4587 }
4588
4589 /*
4590  * Map file blocks to filesystem blocks, adding delayed allocations as needed.
4591  */
4592 int
4593 xfs_bmapi_delay(
4594         struct xfs_inode        *ip,    /* incore inode */
4595         xfs_fileoff_t           bno,    /* starting file offs. mapped */
4596         xfs_filblks_t           len,    /* length to map in file */
4597         struct xfs_bmbt_irec    *mval,  /* output: map values */
4598         int                     *nmap,  /* i/o: mval size/count */
4599         int                     flags)  /* XFS_BMAPI_... */
4600 {
4601         struct xfs_mount        *mp = ip->i_mount;
4602         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
4603         struct xfs_bmbt_irec    got;    /* current file extent record */
4604         struct xfs_bmbt_irec    prev;   /* previous file extent record */
4605         xfs_fileoff_t           obno;   /* old block number (offset) */
4606         xfs_fileoff_t           end;    /* end of mapped file region */
4607         xfs_extnum_t            lastx;  /* last useful extent number */
4608         int                     eof;    /* we've hit the end of extents */
4609         int                     n = 0;  /* current extent index */
4610         int                     error = 0;
4611
4612         ASSERT(*nmap >= 1);
4613         ASSERT(*nmap <= XFS_BMAP_MAX_NMAP);
4614         ASSERT(!(flags & ~XFS_BMAPI_ENTIRE));
4615
4616         if (unlikely(XFS_TEST_ERROR(
4617             (XFS_IFORK_FORMAT(ip, XFS_DATA_FORK) != XFS_DINODE_FMT_EXTENTS &&
4618              XFS_IFORK_FORMAT(ip, XFS_DATA_FORK) != XFS_DINODE_FMT_BTREE),
4619              mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4620                 XFS_ERROR_REPORT("xfs_bmapi_delay", XFS_ERRLEVEL_LOW, mp);
4621                 return XFS_ERROR(EFSCORRUPTED);
4622         }
4623
4624         if (XFS_FORCED_SHUTDOWN(mp))
4625                 return XFS_ERROR(EIO);
4626
4627         XFS_STATS_INC(xs_blk_mapw);
4628
4629         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4630                 error = xfs_iread_extents(NULL, ip, XFS_DATA_FORK);
4631                 if (error)
4632                         return error;
4633         }
4634
4635         xfs_bmap_search_extents(ip, bno, XFS_DATA_FORK, &eof, &lastx, &got, &prev);
4636         end = bno + len;
4637         obno = bno;
4638
4639         while (bno < end && n < *nmap) {
4640                 if (eof || got.br_startoff > bno) {
4641                         error = xfs_bmapi_reserve_delalloc(ip, bno, len, &got,
4642                                                            &prev, &lastx, eof);
4643                         if (error) {
4644                                 if (n == 0) {
4645                                         *nmap = 0;
4646                                         return error;
4647                                 }
4648                                 break;
4649                         }
4650                 }
4651
4652                 /* set up the extent map to return. */
4653                 xfs_bmapi_trim_map(mval, &got, &bno, len, obno, end, n, flags);
4654                 xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4655
4656                 /* If we're done, stop now. */
4657                 if (bno >= end || n >= *nmap)
4658                         break;
4659
4660                 /* Else go on to the next record. */
4661                 prev = got;
4662                 if (++lastx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t))
4663                         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx), &got);
4664                 else
4665                         eof = 1;
4666         }
4667
4668         *nmap = n;
4669         return 0;
4670 }
4671
4672
4673 STATIC int
4674 xfs_bmapi_allocate(
4675         struct xfs_bmalloca     *bma,
4676         xfs_extnum_t            *lastx,
4677         struct xfs_btree_cur    **cur,
4678         xfs_fsblock_t           *firstblock,
4679         struct xfs_bmap_free    *flist,
4680         int                     flags,
4681         int                     *nallocs,
4682         int                     *logflags)
4683 {
4684         struct xfs_mount        *mp = bma->ip->i_mount;
4685         int                     whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4686                                                 XFS_ATTR_FORK : XFS_DATA_FORK;
4687         struct xfs_ifork        *ifp = XFS_IFORK_PTR(bma->ip, whichfork);
4688         xfs_fsblock_t           abno;
4689         xfs_extlen_t            alen;
4690         xfs_fileoff_t           aoff;
4691         int                     error;
4692         int                     rt;
4693
4694         rt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(bma->ip);
4695
4696         /*
4697          * For the wasdelay case, we could also just allocate the stuff asked
4698          * for in this bmap call but that wouldn't be as good.
4699          */
4700         if (bma->wasdel) {
4701                 alen = (xfs_extlen_t)bma->gotp->br_blockcount;
4702                 aoff = bma->gotp->br_startoff;
4703                 if (*lastx != NULLEXTNUM && *lastx) {
4704                         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *lastx - 1),
4705                                          bma->prevp);
4706                 }
4707         } else {
4708                 alen = (xfs_extlen_t)XFS_FILBLKS_MIN(bma->alen, MAXEXTLEN);
4709                 if (!bma->eof)
4710                         alen = (xfs_extlen_t)XFS_FILBLKS_MIN(alen,
4711                                         bma->gotp->br_startoff - bma->off);
4712                 aoff = bma->off;
4713         }
4714
4715         /*
4716          * Indicate if this is the first user data in the file, or just any
4717          * user data.
4718          */
4719         if (!(flags & XFS_BMAPI_METADATA)) {
4720                 bma->userdata = (aoff == 0) ?
4721                         XFS_ALLOC_INITIAL_USER_DATA : XFS_ALLOC_USERDATA;
4722         }
4723
4724         /*
4725          * Fill in changeable bma fields.
4726          */
4727         bma->alen = alen;
4728         bma->off = aoff;
4729         bma->firstblock = *firstblock;
4730         bma->minlen = (flags & XFS_BMAPI_CONTIG) ? alen : 1;
4731         bma->low = flist->xbf_low;
4732         bma->aeof = 0;
4733
4734         /*
4735          * Only want to do the alignment at the eof if it is userdata and
4736          * allocation length is larger than a stripe unit.
4737          */
4738         if (mp->m_dalign && alen >= mp->m_dalign &&
4739             !(flags & XFS_BMAPI_METADATA) && whichfork == XFS_DATA_FORK) {
4740                 error = xfs_bmap_isaeof(bma->ip, aoff, whichfork, &bma->aeof);
4741                 if (error)
4742                         return error;
4743         }
4744
4745         error = xfs_bmap_alloc(bma);
4746         if (error)
4747                 return error;
4748
4749         /*
4750          * Copy out result fields.
4751          */
4752         abno = bma->rval;
4753         flist->xbf_low = bma->low;
4754         alen = bma->alen;
4755         aoff = bma->off;
4756         ASSERT(*firstblock == NULLFSBLOCK ||
4757                XFS_FSB_TO_AGNO(mp, *firstblock) ==
4758                XFS_FSB_TO_AGNO(mp, bma->firstblock) ||
4759                (flist->xbf_low &&
4760                 XFS_FSB_TO_AGNO(mp, *firstblock) <
4761                         XFS_FSB_TO_AGNO(mp, bma->firstblock)));
4762         *firstblock = bma->firstblock;
4763         if (*cur)
4764                 (*cur)->bc_private.b.firstblock = *firstblock;
4765         if (abno == NULLFSBLOCK)
4766                 return 0;
4767         if ((ifp->if_flags & XFS_IFBROOT) && !*cur) {
4768                 (*cur) = xfs_bmbt_init_cursor(mp, bma->tp, bma->ip, whichfork);
4769                 (*cur)->bc_private.b.firstblock = *firstblock;
4770                 (*cur)->bc_private.b.flist = flist;
4771         }
4772         /*
4773          * Bump the number of extents we've allocated
4774          * in this call.
4775          */
4776         (*nallocs)++;
4777
4778         if (*cur)
4779                 (*cur)->bc_private.b.flags =
4780                         bma->wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
4781
4782         bma->gotp->br_startoff = aoff;
4783         bma->gotp->br_startblock = abno;
4784         bma->gotp->br_blockcount = alen;
4785         bma->gotp->br_state = XFS_EXT_NORM;
4786
4787         /*
4788          * A wasdelay extent has been initialized, so shouldn't be flagged
4789          * as unwritten.
4790          */
4791         if (!bma->wasdel && (flags & XFS_BMAPI_PREALLOC) &&
4792             xfs_sb_version_hasextflgbit(&mp->m_sb))
4793                 bma->gotp->br_state = XFS_EXT_UNWRITTEN;
4794
4795         error = xfs_bmap_add_extent(bma->tp, bma->ip, lastx, cur, bma->gotp,
4796                                     firstblock, flist, logflags, whichfork);
4797         if (error)
4798                 return error;
4799
4800         /*
4801          * Update our extent pointer, given that xfs_bmap_add_extent  might
4802          * have merged it into one of the neighbouring ones.
4803          */
4804         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *lastx), bma->gotp);
4805
4806         ASSERT(bma->gotp->br_startoff <= aoff);
4807         ASSERT(bma->gotp->br_startoff + bma->gotp->br_blockcount >=
4808                 aoff + alen);
4809         ASSERT(bma->gotp->br_state == XFS_EXT_NORM ||
4810                bma->gotp->br_state == XFS_EXT_UNWRITTEN);
4811         return 0;
4812 }
4813
4814 STATIC int
4815 xfs_bmapi_convert_unwritten(
4816         struct xfs_bmalloca     *bma,
4817         struct xfs_bmbt_irec    *mval,
4818         xfs_filblks_t           len,
4819         xfs_extnum_t            *lastx,
4820         struct xfs_btree_cur    **cur,
4821         xfs_fsblock_t           *firstblock,
4822         struct xfs_bmap_free    *flist,
4823         int                     flags,
4824         int                     *logflags)
4825 {
4826         int                     whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4827                                                 XFS_ATTR_FORK : XFS_DATA_FORK;
4828         struct xfs_ifork        *ifp = XFS_IFORK_PTR(bma->ip, whichfork);
4829         int                     error;
4830
4831         *logflags = 0;
4832
4833         /* check if we need to do unwritten->real conversion */
4834         if (mval->br_state == XFS_EXT_UNWRITTEN &&
4835             (flags & XFS_BMAPI_PREALLOC))
4836                 return 0;
4837
4838         /* check if we need to do real->unwritten conversion */
4839         if (mval->br_state == XFS_EXT_NORM &&
4840             (flags & (XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT)) !=
4841                         (XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT))
4842                 return 0;
4843
4844         /*
4845          * Modify (by adding) the state flag, if writing.
4846          */
4847         ASSERT(mval->br_blockcount <= len);
4848         if ((ifp->if_flags & XFS_IFBROOT) && !*cur) {
4849                 *cur = xfs_bmbt_init_cursor(bma->ip->i_mount, bma->tp,
4850                                         bma->ip, whichfork);
4851                 (*cur)->bc_private.b.firstblock = *firstblock;
4852                 (*cur)->bc_private.b.flist = flist;
4853         }
4854         mval->br_state = (mval->br_state == XFS_EXT_UNWRITTEN)
4855                                 ? XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
4856
4857         error = xfs_bmap_add_extent(bma->tp, bma->ip, lastx, cur, mval,
4858                                 firstblock, flist, logflags, whichfork);
4859         if (error)
4860                 return error;
4861
4862         /*
4863          * Update our extent pointer, given that xfs_bmap_add_extent  might
4864          * have merged it into one of the neighbouring ones.
4865          */
4866         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *lastx), bma->gotp);
4867
4868         /*
4869          * We may have combined previously unwritten space with written space,
4870          * so generate another request.
4871          */
4872         if (mval->br_blockcount < len)
4873                 return EAGAIN;
4874         return 0;
4875 }
4876
4877 /*
4878  * Map file blocks to filesystem blocks, and allocate blocks or convert the
4879  * extent state if necessary.  Details behaviour is controlled by the flags
4880  * parameter.  Only allocates blocks from a single allocation group, to avoid
4881  * locking problems.
4882  *
4883  * The returned value in "firstblock" from the first call in a transaction
4884  * must be remembered and presented to subsequent calls in "firstblock".
4885  * An upper bound for the number of blocks to be allocated is supplied to
4886  * the first call in "total"; if no allocation group has that many free
4887  * blocks then the call will fail (return NULLFSBLOCK in "firstblock").
4888  */
4889 int
4890 xfs_bmapi_write(
4891         struct xfs_trans        *tp,            /* transaction pointer */
4892         struct xfs_inode        *ip,            /* incore inode */
4893         xfs_fileoff_t           bno,            /* starting file offs. mapped */
4894         xfs_filblks_t           len,            /* length to map in file */
4895         int                     flags,          /* XFS_BMAPI_... */
4896         xfs_fsblock_t           *firstblock,    /* first allocated block
4897                                                    controls a.g. for allocs */
4898         xfs_extlen_t            total,          /* total blocks needed */
4899         struct xfs_bmbt_irec    *mval,          /* output: map values */
4900         int                     *nmap,          /* i/o: mval size/count */
4901         struct xfs_bmap_free    *flist)         /* i/o: list extents to free */
4902 {
4903         struct xfs_mount        *mp = ip->i_mount;
4904         struct xfs_ifork        *ifp;
4905         struct xfs_bmalloca     bma = { 0 };    /* args for xfs_bmap_alloc */
4906         struct xfs_btree_cur    *cur;           /* bmap btree cursor */
4907         xfs_fileoff_t           end;            /* end of mapped file region */
4908         int                     eof;            /* after the end of extents */
4909         int                     error;          /* error return */
4910         struct xfs_bmbt_irec    got;            /* current file extent record */
4911         xfs_extnum_t            lastx;          /* last useful extent number */
4912         int                     logflags;       /* flags for transaction logging */
4913         xfs_extlen_t            minleft;        /* min blocks left after allocation */
4914         int                     n;              /* current extent index */
4915         int                     nallocs;        /* number of extents alloc'd */
4916         xfs_fileoff_t           obno;           /* old block number (offset) */
4917         struct xfs_bmbt_irec    prev;           /* previous file extent record */
4918         int                     tmp_logflags;   /* temp flags holder */
4919         int                     whichfork;      /* data or attr fork */
4920         char                    inhole;         /* current location is hole in file */
4921         char                    wasdelay;       /* old extent was delayed */
4922
4923 #ifdef DEBUG
4924         xfs_fileoff_t           orig_bno;       /* original block number value */
4925         int                     orig_flags;     /* original flags arg value */
4926         xfs_filblks_t           orig_len;       /* original value of len arg */
4927         struct xfs_bmbt_irec    *orig_mval;     /* original value of mval */
4928         int                     orig_nmap;      /* original value of *nmap */
4929
4930         orig_bno = bno;
4931         orig_len = len;
4932         orig_flags = flags;
4933         orig_mval = mval;
4934         orig_nmap = *nmap;
4935 #endif
4936
4937         ASSERT(*nmap >= 1);
4938         ASSERT(*nmap <= XFS_BMAP_MAX_NMAP);
4939         ASSERT(!(flags & XFS_BMAPI_IGSTATE));
4940         ASSERT(tp != NULL);
4941
4942         whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4943                 XFS_ATTR_FORK : XFS_DATA_FORK;
4944
4945         if (unlikely(XFS_TEST_ERROR(
4946             (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4947              XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
4948              XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL),
4949              mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4950                 XFS_ERROR_REPORT("xfs_bmapi_write", XFS_ERRLEVEL_LOW, mp);
4951                 return XFS_ERROR(EFSCORRUPTED);
4952         }
4953
4954         if (XFS_FORCED_SHUTDOWN(mp))
4955                 return XFS_ERROR(EIO);
4956
4957         ifp = XFS_IFORK_PTR(ip, whichfork);
4958         ASSERT(ifp->if_ext_max ==
4959                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
4960
4961         XFS_STATS_INC(xs_blk_mapw);
4962
4963         logflags = 0;
4964         nallocs = 0;
4965         cur = NULL;
4966
4967         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
4968                 error = xfs_bmap_local_to_extents(tp, ip, firstblock, total,
4969                                                   &logflags, whichfork);
4970                 if (error)
4971                         goto error0;
4972         }
4973
4974         if (*firstblock == NULLFSBLOCK) {
4975                 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE)
4976                         minleft = be16_to_cpu(ifp->if_broot->bb_level) + 1;
4977                 else
4978                         minleft = 1;
4979         } else {
4980                 minleft = 0;
4981         }
4982
4983         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4984                 error = xfs_iread_extents(tp, ip, whichfork);
4985                 if (error)
4986                         goto error0;
4987         }
4988
4989         xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got, &prev);
4990         n = 0;
4991         end = bno + len;
4992         obno = bno;
4993
4994         bma.tp = tp;
4995         bma.ip = ip;
4996         bma.prevp = &prev;
4997         bma.gotp = &got;
4998         bma.total = total;
4999         bma.userdata = 0;
5000
5001         while (bno < end && n < *nmap) {
5002                 inhole = eof || got.br_startoff > bno;
5003                 wasdelay = !inhole && isnullstartblock(got.br_startblock);
5004
5005                 /*
5006                  * First, deal with the hole before the allocated space
5007                  * that we found, if any.
5008                  */
5009                 if (inhole || wasdelay) {
5010                         bma.eof = eof;
5011                         bma.conv = !!(flags & XFS_BMAPI_CONVERT);
5012                         bma.wasdel = wasdelay;
5013                         bma.alen = len;
5014                         bma.off = bno;
5015                         bma.minleft = minleft;
5016
5017                         error = xfs_bmapi_allocate(&bma, &lastx, &cur,
5018                                         firstblock, flist, flags, &nallocs,
5019                                         &tmp_logflags);
5020                         logflags |= tmp_logflags;
5021                         if (error)
5022                                 goto error0;
5023                         if (flist && flist->xbf_low)
5024                                 minleft = 0;
5025                         if (bma.rval == NULLFSBLOCK)
5026                                 break;
5027                 }
5028
5029                 /* Deal with the allocated space we found.  */
5030                 xfs_bmapi_trim_map(mval, &got, &bno, len, obno, end, n, flags);
5031
5032                 /* Execute unwritten extent conversion if necessary */
5033                 error = xfs_bmapi_convert_unwritten(&bma, mval, len, &lastx,
5034                                                     &cur, firstblock, flist,
5035                                                     flags, &tmp_logflags);
5036                 logflags |= tmp_logflags;
5037                 if (error == EAGAIN)
5038                         continue;
5039                 if (error)
5040                         goto error0;
5041
5042                 /* update the extent map to return */
5043                 xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
5044
5045                 /*
5046                  * If we're done, stop now.  Stop when we've allocated
5047                  * XFS_BMAP_MAX_NMAP extents no matter what.  Otherwise
5048                  * the transaction may get too big.
5049                  */
5050                 if (bno >= end || n >= *nmap || nallocs >= *nmap)
5051                         break;
5052
5053                 /* Else go on to the next record. */
5054                 prev = got;
5055                 if (++lastx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t))
5056                         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx), &got);
5057                 else
5058                         eof = 1;
5059         }
5060         *nmap = n;
5061
5062         /*
5063          * Transform from btree to extents, give it cur.
5064          */
5065         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE &&
5066             XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max) {
5067                 ASSERT(cur);
5068                 error = xfs_bmap_btree_to_extents(tp, ip, cur,
5069                         &tmp_logflags, whichfork);
5070                 logflags |= tmp_logflags;
5071                 if (error)
5072                         goto error0;
5073         }
5074         ASSERT(ifp->if_ext_max ==
5075                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5076         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE ||
5077                XFS_IFORK_NEXTENTS(ip, whichfork) > ifp->if_ext_max);
5078         error = 0;
5079 error0:
5080         /*
5081          * Log everything.  Do this after conversion, there's no point in
5082          * logging the extent records if we've converted to btree format.
5083          */
5084         if ((logflags & xfs_ilog_fext(whichfork)) &&
5085             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
5086                 logflags &= ~xfs_ilog_fext(whichfork);
5087         else if ((logflags & xfs_ilog_fbroot(whichfork)) &&
5088                  XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
5089                 logflags &= ~xfs_ilog_fbroot(whichfork);
5090         /*
5091          * Log whatever the flags say, even if error.  Otherwise we might miss
5092          * detecting a case where the data is changed, there's an error,
5093          * and it's not logged so we don't shutdown when we should.
5094          */
5095         if (logflags)
5096                 xfs_trans_log_inode(tp, ip, logflags);
5097
5098         if (cur) {
5099                 if (!error) {
5100                         ASSERT(*firstblock == NULLFSBLOCK ||
5101                                XFS_FSB_TO_AGNO(mp, *firstblock) ==
5102                                XFS_FSB_TO_AGNO(mp,
5103                                        cur->bc_private.b.firstblock) ||
5104                                (flist->xbf_low &&
5105                                 XFS_FSB_TO_AGNO(mp, *firstblock) <
5106                                 XFS_FSB_TO_AGNO(mp,
5107                                         cur->bc_private.b.firstblock)));
5108                         *firstblock = cur->bc_private.b.firstblock;
5109                 }
5110                 xfs_btree_del_cursor(cur,
5111                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5112         }
5113         if (!error)
5114                 xfs_bmap_validate_ret(orig_bno, orig_len, orig_flags, orig_mval,
5115                         orig_nmap, *nmap);
5116         return error;
5117 }
5118
5119 /*
5120  * Unmap (remove) blocks from a file.
5121  * If nexts is nonzero then the number of extents to remove is limited to
5122  * that value.  If not all extents in the block range can be removed then
5123  * *done is set.
5124  */
5125 int                                             /* error */
5126 xfs_bunmapi(
5127         xfs_trans_t             *tp,            /* transaction pointer */
5128         struct xfs_inode        *ip,            /* incore inode */
5129         xfs_fileoff_t           bno,            /* starting offset to unmap */
5130         xfs_filblks_t           len,            /* length to unmap in file */
5131         int                     flags,          /* misc flags */
5132         xfs_extnum_t            nexts,          /* number of extents max */
5133         xfs_fsblock_t           *firstblock,    /* first allocated block
5134                                                    controls a.g. for allocs */
5135         xfs_bmap_free_t         *flist,         /* i/o: list extents to free */
5136         int                     *done)          /* set if not done yet */
5137 {
5138         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
5139         xfs_bmbt_irec_t         del;            /* extent being deleted */
5140         int                     eof;            /* is deleting at eof */
5141         xfs_bmbt_rec_host_t     *ep;            /* extent record pointer */
5142         int                     error;          /* error return value */
5143         xfs_extnum_t            extno;          /* extent number in list */
5144         xfs_bmbt_irec_t         got;            /* current extent record */
5145         xfs_ifork_t             *ifp;           /* inode fork pointer */
5146         int                     isrt;           /* freeing in rt area */
5147         xfs_extnum_t            lastx;          /* last extent index used */
5148         int                     logflags;       /* transaction logging flags */
5149         xfs_extlen_t            mod;            /* rt extent offset */
5150         xfs_mount_t             *mp;            /* mount structure */
5151         xfs_extnum_t            nextents;       /* number of file extents */
5152         xfs_bmbt_irec_t         prev;           /* previous extent record */
5153         xfs_fileoff_t           start;          /* first file offset deleted */
5154         int                     tmp_logflags;   /* partial logging flags */
5155         int                     wasdel;         /* was a delayed alloc extent */
5156         int                     whichfork;      /* data or attribute fork */
5157         xfs_fsblock_t           sum;
5158
5159         trace_xfs_bunmap(ip, bno, len, flags, _RET_IP_);
5160
5161         whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
5162                 XFS_ATTR_FORK : XFS_DATA_FORK;
5163         ifp = XFS_IFORK_PTR(ip, whichfork);
5164         if (unlikely(
5165             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
5166             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
5167                 XFS_ERROR_REPORT("xfs_bunmapi", XFS_ERRLEVEL_LOW,
5168                                  ip->i_mount);
5169                 return XFS_ERROR(EFSCORRUPTED);
5170         }
5171         mp = ip->i_mount;
5172         if (XFS_FORCED_SHUTDOWN(mp))
5173                 return XFS_ERROR(EIO);
5174
5175         ASSERT(len > 0);
5176         ASSERT(nexts >= 0);
5177         ASSERT(ifp->if_ext_max ==
5178                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5179         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5180             (error = xfs_iread_extents(tp, ip, whichfork)))
5181                 return error;
5182         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5183         if (nextents == 0) {
5184                 *done = 1;
5185                 return 0;
5186         }
5187         XFS_STATS_INC(xs_blk_unmap);
5188         isrt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
5189         start = bno;
5190         bno = start + len - 1;
5191         ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
5192                 &prev);
5193
5194         /*
5195          * Check to see if the given block number is past the end of the
5196          * file, back up to the last block if so...
5197          */
5198         if (eof) {
5199                 ep = xfs_iext_get_ext(ifp, --lastx);
5200                 xfs_bmbt_get_all(ep, &got);
5201                 bno = got.br_startoff + got.br_blockcount - 1;
5202         }
5203         logflags = 0;
5204         if (ifp->if_flags & XFS_IFBROOT) {
5205                 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
5206                 cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5207                 cur->bc_private.b.firstblock = *firstblock;
5208                 cur->bc_private.b.flist = flist;
5209                 cur->bc_private.b.flags = 0;
5210         } else
5211                 cur = NULL;
5212         extno = 0;
5213         while (bno != (xfs_fileoff_t)-1 && bno >= start && lastx >= 0 &&
5214                (nexts == 0 || extno < nexts)) {
5215                 /*
5216                  * Is the found extent after a hole in which bno lives?
5217                  * Just back up to the previous extent, if so.
5218                  */
5219                 if (got.br_startoff > bno) {
5220                         if (--lastx < 0)
5221                                 break;
5222                         ep = xfs_iext_get_ext(ifp, lastx);
5223                         xfs_bmbt_get_all(ep, &got);
5224                 }
5225                 /*
5226                  * Is the last block of this extent before the range
5227                  * we're supposed to delete?  If so, we're done.
5228                  */
5229                 bno = XFS_FILEOFF_MIN(bno,
5230                         got.br_startoff + got.br_blockcount - 1);
5231                 if (bno < start)
5232                         break;
5233                 /*
5234                  * Then deal with the (possibly delayed) allocated space
5235                  * we found.
5236                  */
5237                 ASSERT(ep != NULL);
5238                 del = got;
5239                 wasdel = isnullstartblock(del.br_startblock);
5240                 if (got.br_startoff < start) {
5241                         del.br_startoff = start;
5242                         del.br_blockcount -= start - got.br_startoff;
5243                         if (!wasdel)
5244                                 del.br_startblock += start - got.br_startoff;
5245                 }
5246                 if (del.br_startoff + del.br_blockcount > bno + 1)
5247                         del.br_blockcount = bno + 1 - del.br_startoff;
5248                 sum = del.br_startblock + del.br_blockcount;
5249                 if (isrt &&
5250                     (mod = do_mod(sum, mp->m_sb.sb_rextsize))) {
5251                         /*
5252                          * Realtime extent not lined up at the end.
5253                          * The extent could have been split into written
5254                          * and unwritten pieces, or we could just be
5255                          * unmapping part of it.  But we can't really
5256                          * get rid of part of a realtime extent.
5257                          */
5258                         if (del.br_state == XFS_EXT_UNWRITTEN ||
5259                             !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5260                                 /*
5261                                  * This piece is unwritten, or we're not
5262                                  * using unwritten extents.  Skip over it.
5263                                  */
5264                                 ASSERT(bno >= mod);
5265                                 bno -= mod > del.br_blockcount ?
5266                                         del.br_blockcount : mod;
5267                                 if (bno < got.br_startoff) {
5268                                         if (--lastx >= 0)
5269                                                 xfs_bmbt_get_all(xfs_iext_get_ext(
5270                                                         ifp, lastx), &got);
5271                                 }
5272                                 continue;
5273                         }
5274                         /*
5275                          * It's written, turn it unwritten.
5276                          * This is better than zeroing it.
5277                          */
5278                         ASSERT(del.br_state == XFS_EXT_NORM);
5279                         ASSERT(xfs_trans_get_block_res(tp) > 0);
5280                         /*
5281                          * If this spans a realtime extent boundary,
5282                          * chop it back to the start of the one we end at.
5283                          */
5284                         if (del.br_blockcount > mod) {
5285                                 del.br_startoff += del.br_blockcount - mod;
5286                                 del.br_startblock += del.br_blockcount - mod;
5287                                 del.br_blockcount = mod;
5288                         }
5289                         del.br_state = XFS_EXT_UNWRITTEN;
5290                         error = xfs_bmap_add_extent(tp, ip, &lastx, &cur, &del,
5291                                 firstblock, flist, &logflags,
5292                                 XFS_DATA_FORK);
5293                         if (error)
5294                                 goto error0;
5295                         goto nodelete;
5296                 }
5297                 if (isrt && (mod = do_mod(del.br_startblock, mp->m_sb.sb_rextsize))) {
5298                         /*
5299                          * Realtime extent is lined up at the end but not
5300                          * at the front.  We'll get rid of full extents if
5301                          * we can.
5302                          */
5303                         mod = mp->m_sb.sb_rextsize - mod;
5304                         if (del.br_blockcount > mod) {
5305                                 del.br_blockcount -= mod;
5306                                 del.br_startoff += mod;
5307                                 del.br_startblock += mod;
5308                         } else if ((del.br_startoff == start &&
5309                                     (del.br_state == XFS_EXT_UNWRITTEN ||
5310                                      xfs_trans_get_block_res(tp) == 0)) ||
5311                                    !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5312                                 /*
5313                                  * Can't make it unwritten.  There isn't
5314                                  * a full extent here so just skip it.
5315                                  */
5316                                 ASSERT(bno >= del.br_blockcount);
5317                                 bno -= del.br_blockcount;
5318                                 if (got.br_startoff > bno) {
5319                                         if (--lastx >= 0) {
5320                                                 ep = xfs_iext_get_ext(ifp,
5321                                                                       lastx);
5322                                                 xfs_bmbt_get_all(ep, &got);
5323                                         }
5324                                 }
5325                                 continue;
5326                         } else if (del.br_state == XFS_EXT_UNWRITTEN) {
5327                                 /*
5328                                  * This one is already unwritten.
5329                                  * It must have a written left neighbor.
5330                                  * Unwrite the killed part of that one and
5331                                  * try again.
5332                                  */
5333                                 ASSERT(lastx > 0);
5334                                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp,
5335                                                 lastx - 1), &prev);
5336                                 ASSERT(prev.br_state == XFS_EXT_NORM);
5337                                 ASSERT(!isnullstartblock(prev.br_startblock));
5338                                 ASSERT(del.br_startblock ==
5339                                        prev.br_startblock + prev.br_blockcount);
5340                                 if (prev.br_startoff < start) {
5341                                         mod = start - prev.br_startoff;
5342                                         prev.br_blockcount -= mod;
5343                                         prev.br_startblock += mod;
5344                                         prev.br_startoff = start;
5345                                 }
5346                                 prev.br_state = XFS_EXT_UNWRITTEN;
5347                                 lastx--;
5348                                 error = xfs_bmap_add_extent(tp, ip, &lastx,
5349                                                 &cur, &prev, firstblock, flist,
5350                                                 &logflags, XFS_DATA_FORK);
5351                                 if (error)
5352                                         goto error0;
5353                                 goto nodelete;
5354                         } else {
5355                                 ASSERT(del.br_state == XFS_EXT_NORM);
5356                                 del.br_state = XFS_EXT_UNWRITTEN;
5357                                 error = xfs_bmap_add_extent(tp, ip, &lastx,
5358                                                 &cur, &del, firstblock, flist,
5359                                                 &logflags, XFS_DATA_FORK);
5360                                 if (error)
5361                                         goto error0;
5362                                 goto nodelete;
5363                         }
5364                 }
5365                 if (wasdel) {
5366                         ASSERT(startblockval(del.br_startblock) > 0);
5367                         /* Update realtime/data freespace, unreserve quota */
5368                         if (isrt) {
5369                                 xfs_filblks_t rtexts;
5370
5371                                 rtexts = XFS_FSB_TO_B(mp, del.br_blockcount);
5372                                 do_div(rtexts, mp->m_sb.sb_rextsize);
5373                                 xfs_mod_incore_sb(mp, XFS_SBS_FREXTENTS,
5374                                                 (int64_t)rtexts, 0);
5375                                 (void)xfs_trans_reserve_quota_nblks(NULL,
5376                                         ip, -((long)del.br_blockcount), 0,
5377                                         XFS_QMOPT_RES_RTBLKS);
5378                         } else {
5379                                 xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
5380                                                 (int64_t)del.br_blockcount, 0);
5381                                 (void)xfs_trans_reserve_quota_nblks(NULL,
5382                                         ip, -((long)del.br_blockcount), 0,
5383                                         XFS_QMOPT_RES_REGBLKS);
5384                         }
5385                         ip->i_delayed_blks -= del.br_blockcount;
5386                         if (cur)
5387                                 cur->bc_private.b.flags |=
5388                                         XFS_BTCUR_BPRV_WASDEL;
5389                 } else if (cur)
5390                         cur->bc_private.b.flags &= ~XFS_BTCUR_BPRV_WASDEL;
5391                 /*
5392                  * If it's the case where the directory code is running
5393                  * with no block reservation, and the deleted block is in
5394                  * the middle of its extent, and the resulting insert
5395                  * of an extent would cause transformation to btree format,
5396                  * then reject it.  The calling code will then swap
5397                  * blocks around instead.
5398                  * We have to do this now, rather than waiting for the
5399                  * conversion to btree format, since the transaction
5400                  * will be dirty.
5401                  */
5402                 if (!wasdel && xfs_trans_get_block_res(tp) == 0 &&
5403                     XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
5404                     XFS_IFORK_NEXTENTS(ip, whichfork) >= ifp->if_ext_max &&
5405                     del.br_startoff > got.br_startoff &&
5406                     del.br_startoff + del.br_blockcount <
5407                     got.br_startoff + got.br_blockcount) {
5408                         error = XFS_ERROR(ENOSPC);
5409                         goto error0;
5410                 }
5411                 error = xfs_bmap_del_extent(ip, tp, &lastx, flist, cur, &del,
5412                                 &tmp_logflags, whichfork);
5413                 logflags |= tmp_logflags;
5414                 if (error)
5415                         goto error0;
5416                 bno = del.br_startoff - 1;
5417 nodelete:
5418                 /*
5419                  * If not done go on to the next (previous) record.
5420                  */
5421                 if (bno != (xfs_fileoff_t)-1 && bno >= start) {
5422                         if (lastx >= 0) {
5423                                 ep = xfs_iext_get_ext(ifp, lastx);
5424                                 if (xfs_bmbt_get_startoff(ep) > bno) {
5425                                         if (--lastx >= 0)
5426                                                 ep = xfs_iext_get_ext(ifp,
5427                                                                       lastx);
5428                                 }
5429                                 xfs_bmbt_get_all(ep, &got);
5430                         }
5431                         extno++;
5432                 }
5433         }
5434         *done = bno == (xfs_fileoff_t)-1 || bno < start || lastx < 0;
5435         ASSERT(ifp->if_ext_max ==
5436                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5437         /*
5438          * Convert to a btree if necessary.
5439          */
5440         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
5441             XFS_IFORK_NEXTENTS(ip, whichfork) > ifp->if_ext_max) {
5442                 ASSERT(cur == NULL);
5443                 error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist,
5444                         &cur, 0, &tmp_logflags, whichfork);
5445                 logflags |= tmp_logflags;
5446                 if (error)
5447                         goto error0;
5448         }
5449         /*
5450          * transform from btree to extents, give it cur
5451          */
5452         else if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE &&
5453                  XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max) {
5454                 ASSERT(cur != NULL);
5455                 error = xfs_bmap_btree_to_extents(tp, ip, cur, &tmp_logflags,
5456                         whichfork);
5457                 logflags |= tmp_logflags;
5458                 if (error)
5459                         goto error0;
5460         }
5461         /*
5462          * transform from extents to local?
5463          */
5464         ASSERT(ifp->if_ext_max ==
5465                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5466         error = 0;
5467 error0:
5468         /*
5469          * Log everything.  Do this after conversion, there's no point in
5470          * logging the extent records if we've converted to btree format.
5471          */
5472         if ((logflags & xfs_ilog_fext(whichfork)) &&
5473             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
5474                 logflags &= ~xfs_ilog_fext(whichfork);
5475         else if ((logflags & xfs_ilog_fbroot(whichfork)) &&
5476                  XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
5477                 logflags &= ~xfs_ilog_fbroot(whichfork);
5478         /*
5479          * Log inode even in the error case, if the transaction
5480          * is dirty we'll need to shut down the filesystem.
5481          */
5482         if (logflags)
5483                 xfs_trans_log_inode(tp, ip, logflags);
5484         if (cur) {
5485                 if (!error) {
5486                         *firstblock = cur->bc_private.b.firstblock;
5487                         cur->bc_private.b.allocated = 0;
5488                 }
5489                 xfs_btree_del_cursor(cur,
5490                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5491         }
5492         return error;
5493 }
5494
5495 /*
5496  * returns 1 for success, 0 if we failed to map the extent.
5497  */
5498 STATIC int
5499 xfs_getbmapx_fix_eof_hole(
5500         xfs_inode_t             *ip,            /* xfs incore inode pointer */
5501         struct getbmapx         *out,           /* output structure */
5502         int                     prealloced,     /* this is a file with
5503                                                  * preallocated data space */
5504         __int64_t               end,            /* last block requested */
5505         xfs_fsblock_t           startblock)
5506 {
5507         __int64_t               fixlen;
5508         xfs_mount_t             *mp;            /* file system mount point */
5509         xfs_ifork_t             *ifp;           /* inode fork pointer */
5510         xfs_extnum_t            lastx;          /* last extent pointer */
5511         xfs_fileoff_t           fileblock;
5512
5513         if (startblock == HOLESTARTBLOCK) {
5514                 mp = ip->i_mount;
5515                 out->bmv_block = -1;
5516                 fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, ip->i_size));
5517                 fixlen -= out->bmv_offset;
5518                 if (prealloced && out->bmv_offset + out->bmv_length == end) {
5519                         /* Came to hole at EOF. Trim it. */
5520                         if (fixlen <= 0)
5521                                 return 0;
5522                         out->bmv_length = fixlen;
5523                 }
5524         } else {
5525                 if (startblock == DELAYSTARTBLOCK)
5526                         out->bmv_block = -2;
5527                 else
5528                         out->bmv_block = xfs_fsb_to_db(ip, startblock);
5529                 fileblock = XFS_BB_TO_FSB(ip->i_mount, out->bmv_offset);
5530                 ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
5531                 if (xfs_iext_bno_to_ext(ifp, fileblock, &lastx) &&
5532                    (lastx == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t))-1))
5533                         out->bmv_oflags |= BMV_OF_LAST;
5534         }
5535
5536         return 1;
5537 }
5538
5539 /*
5540  * Get inode's extents as described in bmv, and format for output.
5541  * Calls formatter to fill the user's buffer until all extents
5542  * are mapped, until the passed-in bmv->bmv_count slots have
5543  * been filled, or until the formatter short-circuits the loop,
5544  * if it is tracking filled-in extents on its own.
5545  */
5546 int                                             /* error code */
5547 xfs_getbmap(
5548         xfs_inode_t             *ip,
5549         struct getbmapx         *bmv,           /* user bmap structure */
5550         xfs_bmap_format_t       formatter,      /* format to user */
5551         void                    *arg)           /* formatter arg */
5552 {
5553         __int64_t               bmvend;         /* last block requested */
5554         int                     error = 0;      /* return value */
5555         __int64_t               fixlen;         /* length for -1 case */
5556         int                     i;              /* extent number */
5557         int                     lock;           /* lock state */
5558         xfs_bmbt_irec_t         *map;           /* buffer for user's data */
5559         xfs_mount_t             *mp;            /* file system mount point */
5560         int                     nex;            /* # of user extents can do */
5561         int                     nexleft;        /* # of user extents left */
5562         int                     subnex;         /* # of bmapi's can do */
5563         int                     nmap;           /* number of map entries */
5564         struct getbmapx         *out;           /* output structure */
5565         int                     whichfork;      /* data or attr fork */
5566         int                     prealloced;     /* this is a file with
5567                                                  * preallocated data space */
5568         int                     iflags;         /* interface flags */
5569         int                     bmapi_flags;    /* flags for xfs_bmapi */
5570         int                     cur_ext = 0;
5571
5572         mp = ip->i_mount;
5573         iflags = bmv->bmv_iflags;
5574         whichfork = iflags & BMV_IF_ATTRFORK ? XFS_ATTR_FORK : XFS_DATA_FORK;
5575
5576         if (whichfork == XFS_ATTR_FORK) {
5577                 if (XFS_IFORK_Q(ip)) {
5578                         if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS &&
5579                             ip->i_d.di_aformat != XFS_DINODE_FMT_BTREE &&
5580                             ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)
5581                                 return XFS_ERROR(EINVAL);
5582                 } else if (unlikely(
5583                            ip->i_d.di_aformat != 0 &&
5584                            ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS)) {
5585                         XFS_ERROR_REPORT("xfs_getbmap", XFS_ERRLEVEL_LOW,
5586                                          ip->i_mount);
5587                         return XFS_ERROR(EFSCORRUPTED);
5588                 }
5589
5590                 prealloced = 0;
5591                 fixlen = 1LL << 32;
5592         } else {
5593                 if (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS &&
5594                     ip->i_d.di_format != XFS_DINODE_FMT_BTREE &&
5595                     ip->i_d.di_format != XFS_DINODE_FMT_LOCAL)
5596                         return XFS_ERROR(EINVAL);
5597
5598                 if (xfs_get_extsz_hint(ip) ||
5599                     ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC|XFS_DIFLAG_APPEND)){
5600                         prealloced = 1;
5601                         fixlen = XFS_MAXIOFFSET(mp);
5602                 } else {
5603                         prealloced = 0;
5604                         fixlen = ip->i_size;
5605                 }
5606         }
5607
5608         if (bmv->bmv_length == -1) {
5609                 fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, fixlen));
5610                 bmv->bmv_length =
5611                         max_t(__int64_t, fixlen - bmv->bmv_offset, 0);
5612         } else if (bmv->bmv_length == 0) {
5613                 bmv->bmv_entries = 0;
5614                 return 0;
5615         } else if (bmv->bmv_length < 0) {
5616                 return XFS_ERROR(EINVAL);
5617         }
5618
5619         nex = bmv->bmv_count - 1;
5620         if (nex <= 0)
5621                 return XFS_ERROR(EINVAL);
5622         bmvend = bmv->bmv_offset + bmv->bmv_length;
5623
5624
5625         if (bmv->bmv_count > ULONG_MAX / sizeof(struct getbmapx))
5626                 return XFS_ERROR(ENOMEM);
5627         out = kmem_zalloc(bmv->bmv_count * sizeof(struct getbmapx), KM_MAYFAIL);
5628         if (!out)
5629                 return XFS_ERROR(ENOMEM);
5630
5631         xfs_ilock(ip, XFS_IOLOCK_SHARED);
5632         if (whichfork == XFS_DATA_FORK && !(iflags & BMV_IF_DELALLOC)) {
5633                 if (ip->i_delayed_blks || ip->i_size > ip->i_d.di_size) {
5634                         error = xfs_flush_pages(ip, 0, -1, 0, FI_REMAPF);
5635                         if (error)
5636                                 goto out_unlock_iolock;
5637                 }
5638                 /*
5639                  * even after flushing the inode, there can still be delalloc
5640                  * blocks on the inode beyond EOF due to speculative
5641                  * preallocation. These are not removed until the release
5642                  * function is called or the inode is inactivated. Hence we
5643                  * cannot assert here that ip->i_delayed_blks == 0.
5644                  */
5645         }
5646
5647         lock = xfs_ilock_map_shared(ip);
5648
5649         /*
5650          * Don't let nex be bigger than the number of extents
5651          * we can have assuming alternating holes and real extents.
5652          */
5653         if (nex > XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1)
5654                 nex = XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1;
5655
5656         bmapi_flags = xfs_bmapi_aflag(whichfork);
5657         if (!(iflags & BMV_IF_PREALLOC))
5658                 bmapi_flags |= XFS_BMAPI_IGSTATE;
5659
5660         /*
5661          * Allocate enough space to handle "subnex" maps at a time.
5662          */
5663         error = ENOMEM;
5664         subnex = 16;
5665         map = kmem_alloc(subnex * sizeof(*map), KM_MAYFAIL | KM_NOFS);
5666         if (!map)
5667                 goto out_unlock_ilock;
5668
5669         bmv->bmv_entries = 0;
5670
5671         if (XFS_IFORK_NEXTENTS(ip, whichfork) == 0 &&
5672             (whichfork == XFS_ATTR_FORK || !(iflags & BMV_IF_DELALLOC))) {
5673                 error = 0;
5674                 goto out_free_map;
5675         }
5676
5677         nexleft = nex;
5678
5679         do {
5680                 nmap = (nexleft > subnex) ? subnex : nexleft;
5681                 error = xfs_bmapi_read(ip, XFS_BB_TO_FSBT(mp, bmv->bmv_offset),
5682                                        XFS_BB_TO_FSB(mp, bmv->bmv_length),
5683                                        map, &nmap, bmapi_flags);
5684                 if (error)
5685                         goto out_free_map;
5686                 ASSERT(nmap <= subnex);
5687
5688                 for (i = 0; i < nmap && nexleft && bmv->bmv_length; i++) {
5689                         out[cur_ext].bmv_oflags = 0;
5690                         if (map[i].br_state == XFS_EXT_UNWRITTEN)
5691                                 out[cur_ext].bmv_oflags |= BMV_OF_PREALLOC;
5692                         else if (map[i].br_startblock == DELAYSTARTBLOCK)
5693                                 out[cur_ext].bmv_oflags |= BMV_OF_DELALLOC;
5694                         out[cur_ext].bmv_offset =
5695                                 XFS_FSB_TO_BB(mp, map[i].br_startoff);
5696                         out[cur_ext].bmv_length =
5697                                 XFS_FSB_TO_BB(mp, map[i].br_blockcount);
5698                         out[cur_ext].bmv_unused1 = 0;
5699                         out[cur_ext].bmv_unused2 = 0;
5700                         ASSERT(((iflags & BMV_IF_DELALLOC) != 0) ||
5701                               (map[i].br_startblock != DELAYSTARTBLOCK));
5702                         if (map[i].br_startblock == HOLESTARTBLOCK &&
5703                             whichfork == XFS_ATTR_FORK) {
5704                                 /* came to the end of attribute fork */
5705                                 out[cur_ext].bmv_oflags |= BMV_OF_LAST;
5706                                 goto out_free_map;
5707                         }
5708
5709                         if (!xfs_getbmapx_fix_eof_hole(ip, &out[cur_ext],
5710                                         prealloced, bmvend,
5711                                         map[i].br_startblock))
5712                                 goto out_free_map;
5713
5714                         bmv->bmv_offset =
5715                                 out[cur_ext].bmv_offset +
5716                                 out[cur_ext].bmv_length;
5717                         bmv->bmv_length =
5718                                 max_t(__int64_t, 0, bmvend - bmv->bmv_offset);
5719
5720                         /*
5721                          * In case we don't want to return the hole,
5722                          * don't increase cur_ext so that we can reuse
5723                          * it in the next loop.
5724                          */
5725                         if ((iflags & BMV_IF_NO_HOLES) &&
5726                             map[i].br_startblock == HOLESTARTBLOCK) {
5727                                 memset(&out[cur_ext], 0, sizeof(out[cur_ext]));
5728                                 continue;
5729                         }
5730
5731                         nexleft--;
5732                         bmv->bmv_entries++;
5733                         cur_ext++;
5734                 }
5735         } while (nmap && nexleft && bmv->bmv_length);
5736
5737  out_free_map:
5738         kmem_free(map);
5739  out_unlock_ilock:
5740         xfs_iunlock_map_shared(ip, lock);
5741  out_unlock_iolock:
5742         xfs_iunlock(ip, XFS_IOLOCK_SHARED);
5743
5744         for (i = 0; i < cur_ext; i++) {
5745                 int full = 0;   /* user array is full */
5746
5747                 /* format results & advance arg */
5748                 error = formatter(&arg, &out[i], &full);
5749                 if (error || full)
5750                         break;
5751         }
5752
5753         kmem_free(out);
5754         return error;
5755 }
5756
5757 #ifdef DEBUG
5758 STATIC struct xfs_buf *
5759 xfs_bmap_get_bp(
5760         struct xfs_btree_cur    *cur,
5761         xfs_fsblock_t           bno)
5762 {
5763         struct xfs_log_item_desc *lidp;
5764         int                     i;
5765
5766         if (!cur)
5767                 return NULL;
5768
5769         for (i = 0; i < XFS_BTREE_MAXLEVELS; i++) {
5770                 if (!cur->bc_bufs[i])
5771                         break;
5772                 if (XFS_BUF_ADDR(cur->bc_bufs[i]) == bno)
5773                         return cur->bc_bufs[i];
5774         }
5775
5776         /* Chase down all the log items to see if the bp is there */
5777         list_for_each_entry(lidp, &cur->bc_tp->t_items, lid_trans) {
5778                 struct xfs_buf_log_item *bip;
5779                 bip = (struct xfs_buf_log_item *)lidp->lid_item;
5780                 if (bip->bli_item.li_type == XFS_LI_BUF &&
5781                     XFS_BUF_ADDR(bip->bli_buf) == bno)
5782                         return bip->bli_buf;
5783         }
5784
5785         return NULL;
5786 }
5787
5788 STATIC void
5789 xfs_check_block(
5790         struct xfs_btree_block  *block,
5791         xfs_mount_t             *mp,
5792         int                     root,
5793         short                   sz)
5794 {
5795         int                     i, j, dmxr;
5796         __be64                  *pp, *thispa;   /* pointer to block address */
5797         xfs_bmbt_key_t          *prevp, *keyp;
5798
5799         ASSERT(be16_to_cpu(block->bb_level) > 0);
5800
5801         prevp = NULL;
5802         for( i = 1; i <= xfs_btree_get_numrecs(block); i++) {
5803                 dmxr = mp->m_bmap_dmxr[0];
5804                 keyp = XFS_BMBT_KEY_ADDR(mp, block, i);
5805
5806                 if (prevp) {
5807                         ASSERT(be64_to_cpu(prevp->br_startoff) <
5808                                be64_to_cpu(keyp->br_startoff));
5809                 }
5810                 prevp = keyp;
5811
5812                 /*
5813                  * Compare the block numbers to see if there are dups.
5814                  */
5815                 if (root)
5816                         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, i, sz);
5817                 else
5818                         pp = XFS_BMBT_PTR_ADDR(mp, block, i, dmxr);
5819
5820                 for (j = i+1; j <= be16_to_cpu(block->bb_numrecs); j++) {
5821                         if (root)
5822                                 thispa = XFS_BMAP_BROOT_PTR_ADDR(mp, block, j, sz);
5823                         else
5824                                 thispa = XFS_BMBT_PTR_ADDR(mp, block, j, dmxr);
5825                         if (*thispa == *pp) {
5826                                 xfs_warn(mp, "%s: thispa(%d) == pp(%d) %Ld",
5827                                         __func__, j, i,
5828                                         (unsigned long long)be64_to_cpu(*thispa));
5829                                 panic("%s: ptrs are equal in node\n",
5830                                         __func__);
5831                         }
5832                 }
5833         }
5834 }
5835
5836 /*
5837  * Check that the extents for the inode ip are in the right order in all
5838  * btree leaves.
5839  */
5840
5841 STATIC void
5842 xfs_bmap_check_leaf_extents(
5843         xfs_btree_cur_t         *cur,   /* btree cursor or null */
5844         xfs_inode_t             *ip,            /* incore inode pointer */
5845         int                     whichfork)      /* data or attr fork */
5846 {
5847         struct xfs_btree_block  *block; /* current btree block */
5848         xfs_fsblock_t           bno;    /* block # of "block" */
5849         xfs_buf_t               *bp;    /* buffer for "block" */
5850         int                     error;  /* error return value */
5851         xfs_extnum_t            i=0, j; /* index into the extents list */
5852         xfs_ifork_t             *ifp;   /* fork structure */
5853         int                     level;  /* btree level, for checking */
5854         xfs_mount_t             *mp;    /* file system mount structure */
5855         __be64                  *pp;    /* pointer to block address */
5856         xfs_bmbt_rec_t          *ep;    /* pointer to current extent */
5857         xfs_bmbt_rec_t          last = {0, 0}; /* last extent in prev block */
5858         xfs_bmbt_rec_t          *nextp; /* pointer to next extent */
5859         int                     bp_release = 0;
5860
5861         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE) {
5862                 return;
5863         }
5864
5865         bno = NULLFSBLOCK;
5866         mp = ip->i_mount;
5867         ifp = XFS_IFORK_PTR(ip, whichfork);
5868         block = ifp->if_broot;
5869         /*
5870          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
5871          */
5872         level = be16_to_cpu(block->bb_level);
5873         ASSERT(level > 0);
5874         xfs_check_block(block, mp, 1, ifp->if_broot_bytes);
5875         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
5876         bno = be64_to_cpu(*pp);
5877
5878         ASSERT(bno != NULLDFSBNO);
5879         ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
5880         ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
5881
5882         /*
5883          * Go down the tree until leaf level is reached, following the first
5884          * pointer (leftmost) at each level.
5885          */
5886         while (level-- > 0) {
5887                 /* See if buf is in cur first */
5888                 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
5889                 if (bp) {
5890                         bp_release = 0;
5891                 } else {
5892                         bp_release = 1;
5893                 }
5894                 if (!bp && (error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
5895                                 XFS_BMAP_BTREE_REF)))
5896                         goto error_norelse;
5897                 block = XFS_BUF_TO_BLOCK(bp);
5898                 XFS_WANT_CORRUPTED_GOTO(
5899                         xfs_bmap_sanity_check(mp, bp, level),
5900                         error0);
5901                 if (level == 0)
5902                         break;
5903
5904                 /*
5905                  * Check this block for basic sanity (increasing keys and
5906                  * no duplicate blocks).
5907                  */
5908
5909                 xfs_check_block(block, mp, 0, 0);
5910                 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
5911                 bno = be64_to_cpu(*pp);
5912                 XFS_WANT_CORRUPTED_GOTO(XFS_FSB_SANITY_CHECK(mp, bno), error0);
5913                 if (bp_release) {
5914                         bp_release = 0;
5915                         xfs_trans_brelse(NULL, bp);
5916                 }
5917         }
5918
5919         /*
5920          * Here with bp and block set to the leftmost leaf node in the tree.
5921          */
5922         i = 0;
5923
5924         /*
5925          * Loop over all leaf nodes checking that all extents are in the right order.
5926          */
5927         for (;;) {
5928                 xfs_fsblock_t   nextbno;
5929                 xfs_extnum_t    num_recs;
5930
5931
5932                 num_recs = xfs_btree_get_numrecs(block);
5933
5934                 /*
5935                  * Read-ahead the next leaf block, if any.
5936                  */
5937
5938                 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
5939
5940                 /*
5941                  * Check all the extents to make sure they are OK.
5942                  * If we had a previous block, the last entry should
5943                  * conform with the first entry in this one.
5944                  */
5945
5946                 ep = XFS_BMBT_REC_ADDR(mp, block, 1);
5947                 if (i) {
5948                         ASSERT(xfs_bmbt_disk_get_startoff(&last) +
5949                                xfs_bmbt_disk_get_blockcount(&last) <=
5950                                xfs_bmbt_disk_get_startoff(ep));
5951                 }
5952                 for (j = 1; j < num_recs; j++) {
5953                         nextp = XFS_BMBT_REC_ADDR(mp, block, j + 1);
5954                         ASSERT(xfs_bmbt_disk_get_startoff(ep) +
5955                                xfs_bmbt_disk_get_blockcount(ep) <=
5956                                xfs_bmbt_disk_get_startoff(nextp));
5957                         ep = nextp;
5958                 }
5959
5960                 last = *ep;
5961                 i += num_recs;
5962                 if (bp_release) {
5963                         bp_release = 0;
5964                         xfs_trans_brelse(NULL, bp);
5965                 }
5966                 bno = nextbno;
5967                 /*
5968                  * If we've reached the end, stop.
5969                  */
5970                 if (bno == NULLFSBLOCK)
5971                         break;
5972
5973                 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
5974                 if (bp) {
5975                         bp_release = 0;
5976                 } else {
5977                         bp_release = 1;
5978                 }
5979                 if (!bp && (error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
5980                                 XFS_BMAP_BTREE_REF)))
5981                         goto error_norelse;
5982                 block = XFS_BUF_TO_BLOCK(bp);
5983         }
5984         if (bp_release) {
5985                 bp_release = 0;
5986                 xfs_trans_brelse(NULL, bp);
5987         }
5988         return;
5989
5990 error0:
5991         xfs_warn(mp, "%s: at error0", __func__);
5992         if (bp_release)
5993                 xfs_trans_brelse(NULL, bp);
5994 error_norelse:
5995         xfs_warn(mp, "%s: BAD after btree leaves for %d extents",
5996                 __func__, i);
5997         panic("%s: CORRUPTED BTREE OR SOMETHING", __func__);
5998         return;
5999 }
6000 #endif
6001
6002 /*
6003  * Count fsblocks of the given fork.
6004  */
6005 int                                             /* error */
6006 xfs_bmap_count_blocks(
6007         xfs_trans_t             *tp,            /* transaction pointer */
6008         xfs_inode_t             *ip,            /* incore inode */
6009         int                     whichfork,      /* data or attr fork */
6010         int                     *count)         /* out: count of blocks */
6011 {
6012         struct xfs_btree_block  *block; /* current btree block */
6013         xfs_fsblock_t           bno;    /* block # of "block" */
6014         xfs_ifork_t             *ifp;   /* fork structure */
6015         int                     level;  /* btree level, for checking */
6016         xfs_mount_t             *mp;    /* file system mount structure */
6017         __be64                  *pp;    /* pointer to block address */
6018
6019         bno = NULLFSBLOCK;
6020         mp = ip->i_mount;
6021         ifp = XFS_IFORK_PTR(ip, whichfork);
6022         if ( XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ) {
6023                 xfs_bmap_count_leaves(ifp, 0,
6024                         ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t),
6025                         count);
6026                 return 0;
6027         }
6028
6029         /*
6030          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
6031          */
6032         block = ifp->if_broot;
6033         level = be16_to_cpu(block->bb_level);
6034         ASSERT(level > 0);
6035         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
6036         bno = be64_to_cpu(*pp);
6037         ASSERT(bno != NULLDFSBNO);
6038         ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
6039         ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
6040
6041         if (unlikely(xfs_bmap_count_tree(mp, tp, ifp, bno, level, count) < 0)) {
6042                 XFS_ERROR_REPORT("xfs_bmap_count_blocks(2)", XFS_ERRLEVEL_LOW,
6043                                  mp);
6044                 return XFS_ERROR(EFSCORRUPTED);
6045         }
6046
6047         return 0;
6048 }
6049
6050 /*
6051  * Recursively walks each level of a btree
6052  * to count total fsblocks is use.
6053  */
6054 STATIC int                                     /* error */
6055 xfs_bmap_count_tree(
6056         xfs_mount_t     *mp,            /* file system mount point */
6057         xfs_trans_t     *tp,            /* transaction pointer */
6058         xfs_ifork_t     *ifp,           /* inode fork pointer */
6059         xfs_fsblock_t   blockno,        /* file system block number */
6060         int             levelin,        /* level in btree */
6061         int             *count)         /* Count of blocks */
6062 {
6063         int                     error;
6064         xfs_buf_t               *bp, *nbp;
6065         int                     level = levelin;
6066         __be64                  *pp;
6067         xfs_fsblock_t           bno = blockno;
6068         xfs_fsblock_t           nextbno;
6069         struct xfs_btree_block  *block, *nextblock;
6070         int                     numrecs;
6071
6072         if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp, XFS_BMAP_BTREE_REF)))
6073                 return error;
6074         *count += 1;
6075         block = XFS_BUF_TO_BLOCK(bp);
6076
6077         if (--level) {
6078                 /* Not at node above leaves, count this level of nodes */
6079                 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
6080                 while (nextbno != NULLFSBLOCK) {
6081                         if ((error = xfs_btree_read_bufl(mp, tp, nextbno,
6082                                 0, &nbp, XFS_BMAP_BTREE_REF)))
6083                                 return error;
6084                         *count += 1;
6085                         nextblock = XFS_BUF_TO_BLOCK(nbp);
6086                         nextbno = be64_to_cpu(nextblock->bb_u.l.bb_rightsib);
6087                         xfs_trans_brelse(tp, nbp);
6088                 }
6089
6090                 /* Dive to the next level */
6091                 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
6092                 bno = be64_to_cpu(*pp);
6093                 if (unlikely((error =
6094                      xfs_bmap_count_tree(mp, tp, ifp, bno, level, count)) < 0)) {
6095                         xfs_trans_brelse(tp, bp);
6096                         XFS_ERROR_REPORT("xfs_bmap_count_tree(1)",
6097                                          XFS_ERRLEVEL_LOW, mp);
6098                         return XFS_ERROR(EFSCORRUPTED);
6099                 }
6100                 xfs_trans_brelse(tp, bp);
6101         } else {
6102                 /* count all level 1 nodes and their leaves */
6103                 for (;;) {
6104                         nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
6105                         numrecs = be16_to_cpu(block->bb_numrecs);
6106                         xfs_bmap_disk_count_leaves(mp, block, numrecs, count);
6107                         xfs_trans_brelse(tp, bp);
6108                         if (nextbno == NULLFSBLOCK)
6109                                 break;
6110                         bno = nextbno;
6111                         if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
6112                                 XFS_BMAP_BTREE_REF)))
6113                                 return error;
6114                         *count += 1;
6115                         block = XFS_BUF_TO_BLOCK(bp);
6116                 }
6117         }
6118         return 0;
6119 }
6120
6121 /*
6122  * Count leaf blocks given a range of extent records.
6123  */
6124 STATIC void
6125 xfs_bmap_count_leaves(
6126         xfs_ifork_t             *ifp,
6127         xfs_extnum_t            idx,
6128         int                     numrecs,
6129         int                     *count)
6130 {
6131         int             b;
6132
6133         for (b = 0; b < numrecs; b++) {
6134                 xfs_bmbt_rec_host_t *frp = xfs_iext_get_ext(ifp, idx + b);
6135                 *count += xfs_bmbt_get_blockcount(frp);
6136         }
6137 }
6138
6139 /*
6140  * Count leaf blocks given a range of extent records originally
6141  * in btree format.
6142  */
6143 STATIC void
6144 xfs_bmap_disk_count_leaves(
6145         struct xfs_mount        *mp,
6146         struct xfs_btree_block  *block,
6147         int                     numrecs,
6148         int                     *count)
6149 {
6150         int             b;
6151         xfs_bmbt_rec_t  *frp;
6152
6153         for (b = 1; b <= numrecs; b++) {
6154                 frp = XFS_BMBT_REC_ADDR(mp, block, b);
6155                 *count += xfs_bmbt_disk_get_blockcount(frp);
6156         }
6157 }
6158
6159 /*
6160  * dead simple method of punching delalyed allocation blocks from a range in
6161  * the inode. Walks a block at a time so will be slow, but is only executed in
6162  * rare error cases so the overhead is not critical. This will alays punch out
6163  * both the start and end blocks, even if the ranges only partially overlap
6164  * them, so it is up to the caller to ensure that partial blocks are not
6165  * passed in.
6166  */
6167 int
6168 xfs_bmap_punch_delalloc_range(
6169         struct xfs_inode        *ip,
6170         xfs_fileoff_t           start_fsb,
6171         xfs_fileoff_t           length)
6172 {
6173         xfs_fileoff_t           remaining = length;
6174         int                     error = 0;
6175
6176         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
6177
6178         do {
6179                 int             done;
6180                 xfs_bmbt_irec_t imap;
6181                 int             nimaps = 1;
6182                 xfs_fsblock_t   firstblock;
6183                 xfs_bmap_free_t flist;
6184
6185                 /*
6186                  * Map the range first and check that it is a delalloc extent
6187                  * before trying to unmap the range. Otherwise we will be
6188                  * trying to remove a real extent (which requires a
6189                  * transaction) or a hole, which is probably a bad idea...
6190                  */
6191                 error = xfs_bmapi_read(ip, start_fsb, 1, &imap, &nimaps,
6192                                        XFS_BMAPI_ENTIRE);
6193
6194                 if (error) {
6195                         /* something screwed, just bail */
6196                         if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
6197                                 xfs_alert(ip->i_mount,
6198                         "Failed delalloc mapping lookup ino %lld fsb %lld.",
6199                                                 ip->i_ino, start_fsb);
6200                         }
6201                         break;
6202                 }
6203                 if (!nimaps) {
6204                         /* nothing there */
6205                         goto next_block;
6206                 }
6207                 if (imap.br_startblock != DELAYSTARTBLOCK) {
6208                         /* been converted, ignore */
6209                         goto next_block;
6210                 }
6211                 WARN_ON(imap.br_blockcount == 0);
6212
6213                 /*
6214                  * Note: while we initialise the firstblock/flist pair, they
6215                  * should never be used because blocks should never be
6216                  * allocated or freed for a delalloc extent and hence we need
6217                  * don't cancel or finish them after the xfs_bunmapi() call.
6218                  */
6219                 xfs_bmap_init(&flist, &firstblock);
6220                 error = xfs_bunmapi(NULL, ip, start_fsb, 1, 0, 1, &firstblock,
6221                                         &flist, &done);
6222                 if (error)
6223                         break;
6224
6225                 ASSERT(!flist.xbf_count && !flist.xbf_first);
6226 next_block:
6227                 start_fsb++;
6228                 remaining--;
6229         } while(remaining > 0);
6230
6231         return error;
6232 }