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_quota.h"
  45 #include "xfs_trans_space.h"
  46 #include "xfs_buf_item.h"
  47 #include "xfs_filestream.h"
  48 #include "xfs_vnodeops.h"
  49 #include "xfs_trace.h"
  50
  51
  52 kmem_zone_t             *xfs_bmap_free_item_zone;
  53
  54 /*
  55  * Prototypes for internal bmap routines.
  56  */
  57
  58 #ifdef DEBUG
  59 STATIC void
  60 xfs_bmap_check_leaf_extents(
  61         struct xfs_btree_cur    *cur,
  62         struct xfs_inode        *ip,
  63         int                     whichfork);
  64 #else
  65 #define xfs_bmap_check_leaf_extents(cur, ip, whichfork)         do { } while (0)
  66 #endif
  67
  68
  69 /*
  70  * Called from xfs_bmap_add_attrfork to handle extents format files.
  71  */
  72 STATIC int                                      /* error */
  73 xfs_bmap_add_attrfork_extents(
  74         xfs_trans_t             *tp,            /* transaction pointer */
  75         xfs_inode_t             *ip,            /* incore inode pointer */
  76         xfs_fsblock_t           *firstblock,    /* first block allocated */
  77         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
  78         int                     *flags);        /* inode logging flags */
  79
  80 /*
  81  * Called from xfs_bmap_add_attrfork to handle local format files.
  82  */
  83 STATIC int                                      /* error */
  84 xfs_bmap_add_attrfork_local(
  85         xfs_trans_t             *tp,            /* transaction pointer */
  86         xfs_inode_t             *ip,            /* incore inode pointer */
  87         xfs_fsblock_t           *firstblock,    /* first block allocated */
  88         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
  89         int                     *flags);        /* inode logging flags */
  90
  91 /*
  92  * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
  93  * It figures out where to ask the underlying allocator to put the new extent.
  94  */
  95 STATIC int                              /* error */
  96 xfs_bmap_alloc(
  97         xfs_bmalloca_t          *ap);   /* bmap alloc argument struct */
  98
  99 /*
 100  * Transform a btree format file with only one leaf node, where the
 101  * extents list will fit in the inode, into an extents format file.
 102  * Since the file extents are already in-core, all we have to do is
 103  * give up the space for the btree root and pitch the leaf block.
 104  */
 105 STATIC int                              /* error */
 106 xfs_bmap_btree_to_extents(
 107         xfs_trans_t             *tp,    /* transaction pointer */
 108         xfs_inode_t             *ip,    /* incore inode pointer */
 109         xfs_btree_cur_t         *cur,   /* btree cursor */
 110         int                     *logflagsp, /* inode logging flags */
 111         int                     whichfork); /* data or attr fork */
 112
 113 /*
 114  * Remove the entry "free" from the free item list.  Prev points to the
 115  * previous entry, unless "free" is the head of the list.
 116  */
 117 STATIC void
 118 xfs_bmap_del_free(
 119         xfs_bmap_free_t         *flist, /* free item list header */
 120         xfs_bmap_free_item_t    *prev,  /* previous item on list, if any */
 121         xfs_bmap_free_item_t    *free); /* list item to be freed */
 122
 123 /*
 124  * Convert an extents-format file into a btree-format file.
 125  * The new file will have a root block (in the inode) and a single child block.
 126  */
 127 STATIC int                                      /* error */
 128 xfs_bmap_extents_to_btree(
 129         xfs_trans_t             *tp,            /* transaction pointer */
 130         xfs_inode_t             *ip,            /* incore inode pointer */
 131         xfs_fsblock_t           *firstblock,    /* first-block-allocated */
 132         xfs_bmap_free_t         *flist,         /* blocks freed in xaction */
 133         xfs_btree_cur_t         **curp,         /* cursor returned to caller */
 134         int                     wasdel,         /* converting a delayed alloc */
 135         int                     *logflagsp,     /* inode logging flags */
 136         int                     whichfork);     /* data or attr fork */
 137
 138 /*
 139  * Convert a local file to an extents file.
 140  * This code is sort of bogus, since the file data needs to get
 141  * logged so it won't be lost.  The bmap-level manipulations are ok, though.
 142  */
 143 STATIC int                              /* error */
 144 xfs_bmap_local_to_extents(
 145         xfs_trans_t     *tp,            /* transaction pointer */
 146         xfs_inode_t     *ip,            /* incore inode pointer */
 147         xfs_fsblock_t   *firstblock,    /* first block allocated in xaction */
 148         xfs_extlen_t    total,          /* total blocks needed by transaction */
 149         int             *logflagsp,     /* inode logging flags */
 150         int             whichfork,      /* data or attr fork */
 151         void            (*init_fn)(struct xfs_buf *bp,
 152                                    struct xfs_inode *ip,
 153                                    struct xfs_ifork *ifp));
 154
 155 /*
 156  * Search the extents list for the inode, for the extent containing bno.
 157  * If bno lies in a hole, point to the next entry.  If bno lies past eof,
 158  * *eofp will be set, and *prevp will contain the last entry (null if none).
 159  * Else, *lastxp will be set to the index of the found
 160  * entry; *gotp will contain the entry.
 161  */
 162 STATIC xfs_bmbt_rec_host_t *            /* pointer to found extent entry */
 163 xfs_bmap_search_extents(
 164         xfs_inode_t     *ip,            /* incore inode pointer */
 165         xfs_fileoff_t   bno,            /* block number searched for */
 166         int             whichfork,      /* data or attr fork */
 167         int             *eofp,          /* out: end of file found */
 168         xfs_extnum_t    *lastxp,        /* out: last extent index */
 169         xfs_bmbt_irec_t *gotp,          /* out: extent entry found */
 170         xfs_bmbt_irec_t *prevp);        /* out: previous extent entry found */
 171
 172 /*
 173  * Compute the worst-case number of indirect blocks that will be used
 174  * for ip's delayed extent of length "len".
 175  */
 176 STATIC xfs_filblks_t
 177 xfs_bmap_worst_indlen(
 178         xfs_inode_t             *ip,    /* incore inode pointer */
 179         xfs_filblks_t           len);   /* delayed extent length */
 180
 181 #ifdef DEBUG
 182 /*
 183  * Perform various validation checks on the values being returned
 184  * from xfs_bmapi().
 185  */
 186 STATIC void
 187 xfs_bmap_validate_ret(
 188         xfs_fileoff_t           bno,
 189         xfs_filblks_t           len,
 190         int                     flags,
 191         xfs_bmbt_irec_t         *mval,
 192         int                     nmap,
 193         int                     ret_nmap);
 194 #else
 195 #define xfs_bmap_validate_ret(bno,len,flags,mval,onmap,nmap)
 196 #endif /* DEBUG */
 197
 198 STATIC int
 199 xfs_bmap_count_tree(
 200         xfs_mount_t     *mp,
 201         xfs_trans_t     *tp,
 202         xfs_ifork_t     *ifp,
 203         xfs_fsblock_t   blockno,
 204         int             levelin,
 205         int             *count);
 206
 207 STATIC void
 208 xfs_bmap_count_leaves(
 209         xfs_ifork_t             *ifp,
 210         xfs_extnum_t            idx,
 211         int                     numrecs,
 212         int                     *count);
 213
 214 STATIC void
 215 xfs_bmap_disk_count_leaves(
 216         struct xfs_mount        *mp,
 217         struct xfs_btree_block  *block,
 218         int                     numrecs,
 219         int                     *count);
 220
 221 /*
 222  * Bmap internal routines.
 223  */
 224
 225 STATIC int                              /* error */
 226 xfs_bmbt_lookup_eq(
 227         struct xfs_btree_cur    *cur,
 228         xfs_fileoff_t           off,
 229         xfs_fsblock_t           bno,
 230         xfs_filblks_t           len,
 231         int                     *stat)  /* success/failure */
 232 {
 233         cur->bc_rec.b.br_startoff = off;
 234         cur->bc_rec.b.br_startblock = bno;
 235         cur->bc_rec.b.br_blockcount = len;
 236         return xfs_btree_lookup(cur, XFS_LOOKUP_EQ, stat);
 237 }
 238
 239 STATIC int                              /* error */
 240 xfs_bmbt_lookup_ge(
 241         struct xfs_btree_cur    *cur,
 242         xfs_fileoff_t           off,
 243         xfs_fsblock_t           bno,
 244         xfs_filblks_t           len,
 245         int                     *stat)  /* success/failure */
 246 {
 247         cur->bc_rec.b.br_startoff = off;
 248         cur->bc_rec.b.br_startblock = bno;
 249         cur->bc_rec.b.br_blockcount = len;
 250         return xfs_btree_lookup(cur, XFS_LOOKUP_GE, stat);
 251 }
 252
 253 /*
 254  * Check if the inode needs to be converted to btree format.
 255  */
 256 static inline bool xfs_bmap_needs_btree(struct xfs_inode *ip, int whichfork)
 257 {
 258         return XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
 259                 XFS_IFORK_NEXTENTS(ip, whichfork) >
 260                         XFS_IFORK_MAXEXT(ip, whichfork);
 261 }
 262
 263 /*
 264  * Check if the inode should be converted to extent format.
 265  */
 266 static inline bool xfs_bmap_wants_extents(struct xfs_inode *ip, int whichfork)
 267 {
 268         return XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE &&
 269                 XFS_IFORK_NEXTENTS(ip, whichfork) <=
 270                         XFS_IFORK_MAXEXT(ip, whichfork);
 271 }
 272
 273 /*
 274  * Update the record referred to by cur to the value given
 275  * by [off, bno, len, state].
 276  * This either works (return 0) or gets an EFSCORRUPTED error.
 277  */
 278 STATIC int
 279 xfs_bmbt_update(
 280         struct xfs_btree_cur    *cur,
 281         xfs_fileoff_t           off,
 282         xfs_fsblock_t           bno,
 283         xfs_filblks_t           len,
 284         xfs_exntst_t            state)
 285 {
 286         union xfs_btree_rec     rec;
 287
 288         xfs_bmbt_disk_set_allf(&rec.bmbt, off, bno, len, state);
 289         return xfs_btree_update(cur, &rec);
 290 }
 291
 292 /*
 293  * Called from xfs_bmap_add_attrfork to handle btree format files.
 294  */
 295 STATIC int                                      /* error */
 296 xfs_bmap_add_attrfork_btree(
 297         xfs_trans_t             *tp,            /* transaction pointer */
 298         xfs_inode_t             *ip,            /* incore inode pointer */
 299         xfs_fsblock_t           *firstblock,    /* first block allocated */
 300         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
 301         int                     *flags)         /* inode logging flags */
 302 {
 303         xfs_btree_cur_t         *cur;           /* btree cursor */
 304         int                     error;          /* error return value */
 305         xfs_mount_t             *mp;            /* file system mount struct */
 306         int                     stat;           /* newroot status */
 307
 308         mp = ip->i_mount;
 309         if (ip->i_df.if_broot_bytes <= XFS_IFORK_DSIZE(ip))
 310                 *flags |= XFS_ILOG_DBROOT;
 311         else {
 312                 cur = xfs_bmbt_init_cursor(mp, tp, ip, XFS_DATA_FORK);
 313                 cur->bc_private.b.flist = flist;
 314                 cur->bc_private.b.firstblock = *firstblock;
 315                 if ((error = xfs_bmbt_lookup_ge(cur, 0, 0, 0, &stat)))
 316                         goto error0;
 317                 /* must be at least one entry */
 318                 XFS_WANT_CORRUPTED_GOTO(stat == 1, error0);
 319                 if ((error = xfs_btree_new_iroot(cur, flags, &stat)))
 320                         goto error0;
 321                 if (stat == 0) {
 322                         xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
 323                         return XFS_ERROR(ENOSPC);
 324                 }
 325                 *firstblock = cur->bc_private.b.firstblock;
 326                 cur->bc_private.b.allocated = 0;
 327                 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
 328         }
 329         return 0;
 330 error0:
 331         xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
 332         return error;
 333 }
 334
 335 /*
 336  * Called from xfs_bmap_add_attrfork to handle extents format files.
 337  */
 338 STATIC int                                      /* error */
 339 xfs_bmap_add_attrfork_extents(
 340         xfs_trans_t             *tp,            /* transaction pointer */
 341         xfs_inode_t             *ip,            /* incore inode pointer */
 342         xfs_fsblock_t           *firstblock,    /* first block allocated */
 343         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
 344         int                     *flags)         /* inode logging flags */
 345 {
 346         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
 347         int                     error;          /* error return value */
 348
 349         if (ip->i_d.di_nextents * sizeof(xfs_bmbt_rec_t) <= XFS_IFORK_DSIZE(ip))
 350                 return 0;
 351         cur = NULL;
 352         error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist, &cur, 0,
 353                 flags, XFS_DATA_FORK);
 354         if (cur) {
 355                 cur->bc_private.b.allocated = 0;
 356                 xfs_btree_del_cursor(cur,
 357                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
 358         }
 359         return error;
 360 }
 361
 362 /*
 363  * Block initialisation functions for local to extent format conversion.
 364  * As these get more complex, they will be moved to the relevant files,
 365  * but for now they are too simple to worry about.
 366  */
 367 STATIC void
 368 xfs_bmap_local_to_extents_init_fn(
 369         struct xfs_buf          *bp,
 370         struct xfs_inode        *ip,
 371         struct xfs_ifork        *ifp)
 372 {
 373         bp->b_ops = &xfs_bmbt_buf_ops;
 374         memcpy(bp->b_addr, ifp->if_u1.if_data, ifp->if_bytes);
 375 }
 376
 377 STATIC void
 378 xfs_symlink_local_to_remote(
 379         struct xfs_buf          *bp,
 380         struct xfs_inode        *ip,
 381         struct xfs_ifork        *ifp)
 382 {
 383         /* remote symlink blocks are not verifiable until CRCs come along */
 384         bp->b_ops = NULL;
 385         memcpy(bp->b_addr, ifp->if_u1.if_data, ifp->if_bytes);
 386 }
 387
 388 /*
 389  * Called from xfs_bmap_add_attrfork to handle local format files. Each
 390  * different data fork content type needs a different callout to do the
 391  * conversion. Some are basic and only require special block initialisation
 392  * callouts for the data formating, others (directories) are so specialised they
 393  * handle everything themselves.
 394  *
 395  * XXX (dgc): investigate whether directory conversion can use the generic
 396  * formatting callout. It should be possible - it's just a very complex
 397  * formatter. it would also require passing the transaction through to the init
 398  * function.
 399  */
 400 STATIC int                                      /* error */
 401 xfs_bmap_add_attrfork_local(
 402         xfs_trans_t             *tp,            /* transaction pointer */
 403         xfs_inode_t             *ip,            /* incore inode pointer */
 404         xfs_fsblock_t           *firstblock,    /* first block allocated */
 405         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
 406         int                     *flags)         /* inode logging flags */
 407 {
 408         xfs_da_args_t           dargs;          /* args for dir/attr code */
 409
 410         if (ip->i_df.if_bytes <= XFS_IFORK_DSIZE(ip))
 411                 return 0;
 412
 413         if (S_ISDIR(ip->i_d.di_mode)) {
 414                 memset(&dargs, 0, sizeof(dargs));
 415                 dargs.dp = ip;
 416                 dargs.firstblock = firstblock;
 417                 dargs.flist = flist;
 418                 dargs.total = ip->i_mount->m_dirblkfsbs;
 419                 dargs.whichfork = XFS_DATA_FORK;
 420                 dargs.trans = tp;
 421                 return xfs_dir2_sf_to_block(&dargs);
 422         }
 423
 424         if (S_ISLNK(ip->i_d.di_mode))
 425                 return xfs_bmap_local_to_extents(tp, ip, firstblock, 1,
 426                                                  flags, XFS_DATA_FORK,
 427                                                  xfs_symlink_local_to_remote);
 428
 429         return xfs_bmap_local_to_extents(tp, ip, firstblock, 1, flags,
 430                                          XFS_DATA_FORK,
 431                                          xfs_bmap_local_to_extents_init_fn);
 432 }
 433
 434 /*
 435  * Convert a delayed allocation to a real allocation.
 436  */
 437 STATIC int                              /* error */
 438 xfs_bmap_add_extent_delay_real(
 439         struct xfs_bmalloca     *bma)
 440 {
 441         struct xfs_bmbt_irec    *new = &bma->got;
 442         int                     diff;   /* temp value */
 443         xfs_bmbt_rec_host_t     *ep;    /* extent entry for idx */
 444         int                     error;  /* error return value */
 445         int                     i;      /* temp state */
 446         xfs_ifork_t             *ifp;   /* inode fork pointer */
 447         xfs_fileoff_t           new_endoff;     /* end offset of new entry */
 448         xfs_bmbt_irec_t         r[3];   /* neighbor extent entries */
 449                                         /* left is 0, right is 1, prev is 2 */
 450         int                     rval=0; /* return value (logging flags) */
 451         int                     state = 0;/* state bits, accessed thru macros */
 452         xfs_filblks_t           da_new; /* new count del alloc blocks used */
 453         xfs_filblks_t           da_old; /* old count del alloc blocks used */
 454         xfs_filblks_t           temp=0; /* value for da_new calculations */
 455         xfs_filblks_t           temp2=0;/* value for da_new calculations */
 456         int                     tmp_rval;       /* partial logging flags */
 457
 458         ifp = XFS_IFORK_PTR(bma->ip, XFS_DATA_FORK);
 459
 460         ASSERT(bma->idx >= 0);
 461         ASSERT(bma->idx <= ifp->if_bytes / sizeof(struct xfs_bmbt_rec));
 462         ASSERT(!isnullstartblock(new->br_startblock));
 463         ASSERT(!bma->cur ||
 464                (bma->cur->bc_private.b.flags & XFS_BTCUR_BPRV_WASDEL));
 465
 466         XFS_STATS_INC(xs_add_exlist);
 467
 468 #define LEFT            r[0]
 469 #define RIGHT           r[1]
 470 #define PREV            r[2]
 471
 472         /*
 473          * Set up a bunch of variables to make the tests simpler.
 474          */
 475         ep = xfs_iext_get_ext(ifp, bma->idx);
 476         xfs_bmbt_get_all(ep, &PREV);
 477         new_endoff = new->br_startoff + new->br_blockcount;
 478         ASSERT(PREV.br_startoff <= new->br_startoff);
 479         ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
 480
 481         da_old = startblockval(PREV.br_startblock);
 482         da_new = 0;
 483
 484         /*
 485          * Set flags determining what part of the previous delayed allocation
 486          * extent is being replaced by a real allocation.
 487          */
 488         if (PREV.br_startoff == new->br_startoff)
 489                 state |= BMAP_LEFT_FILLING;
 490         if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
 491                 state |= BMAP_RIGHT_FILLING;
 492
 493         /*
 494          * Check and set flags if this segment has a left neighbor.
 495          * Don't set contiguous if the combined extent would be too large.
 496          */
 497         if (bma->idx > 0) {
 498                 state |= BMAP_LEFT_VALID;
 499                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx - 1), &LEFT);
 500
 501                 if (isnullstartblock(LEFT.br_startblock))
 502                         state |= BMAP_LEFT_DELAY;
 503         }
 504
 505         if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
 506             LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
 507             LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
 508             LEFT.br_state == new->br_state &&
 509             LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
 510                 state |= BMAP_LEFT_CONTIG;
 511
 512         /*
 513          * Check and set flags if this segment has a right neighbor.
 514          * Don't set contiguous if the combined extent would be too large.
 515          * Also check for all-three-contiguous being too large.
 516          */
 517         if (bma->idx < bma->ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1) {
 518                 state |= BMAP_RIGHT_VALID;
 519                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx + 1), &RIGHT);
 520
 521                 if (isnullstartblock(RIGHT.br_startblock))
 522                         state |= BMAP_RIGHT_DELAY;
 523         }
 524
 525         if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
 526             new_endoff == RIGHT.br_startoff &&
 527             new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
 528             new->br_state == RIGHT.br_state &&
 529             new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
 530             ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
 531                        BMAP_RIGHT_FILLING)) !=
 532                       (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
 533                        BMAP_RIGHT_FILLING) ||
 534              LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
 535                         <= MAXEXTLEN))
 536                 state |= BMAP_RIGHT_CONTIG;
 537
 538         error = 0;
 539         /*
 540          * Switch out based on the FILLING and CONTIG state bits.
 541          */
 542         switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
 543                          BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
 544         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
 545              BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
 546                 /*
 547                  * Filling in all of a previously delayed allocation extent.
 548                  * The left and right neighbors are both contiguous with new.
 549                  */
 550                 bma->idx--;
 551                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
 552                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
 553                         LEFT.br_blockcount + PREV.br_blockcount +
 554                         RIGHT.br_blockcount);
 555                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
 556
 557                 xfs_iext_remove(bma->ip, bma->idx + 1, 2, state);
 558                 bma->ip->i_d.di_nextents--;
 559                 if (bma->cur == NULL)
 560                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
 561                 else {
 562                         rval = XFS_ILOG_CORE;
 563                         error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff,
 564                                         RIGHT.br_startblock,
 565                                         RIGHT.br_blockcount, &i);
 566                         if (error)
 567                                 goto done;
 568                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 569                         error = xfs_btree_delete(bma->cur, &i);
 570                         if (error)
 571                                 goto done;
 572                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 573                         error = xfs_btree_decrement(bma->cur, 0, &i);
 574                         if (error)
 575                                 goto done;
 576                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 577                         error = xfs_bmbt_update(bma->cur, LEFT.br_startoff,
 578                                         LEFT.br_startblock,
 579                                         LEFT.br_blockcount +
 580                                         PREV.br_blockcount +
 581                                         RIGHT.br_blockcount, LEFT.br_state);
 582                         if (error)
 583                                 goto done;
 584                 }
 585                 break;
 586
 587         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
 588                 /*
 589                  * Filling in all of a previously delayed allocation extent.
 590                  * The left neighbor is contiguous, the right is not.
 591                  */
 592                 bma->idx--;
 593
 594                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
 595                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
 596                         LEFT.br_blockcount + PREV.br_blockcount);
 597                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
 598
 599                 xfs_iext_remove(bma->ip, bma->idx + 1, 1, state);
 600                 if (bma->cur == NULL)
 601                         rval = XFS_ILOG_DEXT;
 602                 else {
 603                         rval = 0;
 604                         error = xfs_bmbt_lookup_eq(bma->cur, LEFT.br_startoff,
 605                                         LEFT.br_startblock, LEFT.br_blockcount,
 606                                         &i);
 607                         if (error)
 608                                 goto done;
 609                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 610                         error = xfs_bmbt_update(bma->cur, LEFT.br_startoff,
 611                                         LEFT.br_startblock,
 612                                         LEFT.br_blockcount +
 613                                         PREV.br_blockcount, LEFT.br_state);
 614                         if (error)
 615                                 goto done;
 616                 }
 617                 break;
 618
 619         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
 620                 /*
 621                  * Filling in all of a previously delayed allocation extent.
 622                  * The right neighbor is contiguous, the left is not.
 623                  */
 624                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
 625                 xfs_bmbt_set_startblock(ep, new->br_startblock);
 626                 xfs_bmbt_set_blockcount(ep,
 627                         PREV.br_blockcount + RIGHT.br_blockcount);
 628                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
 629
 630                 xfs_iext_remove(bma->ip, bma->idx + 1, 1, state);
 631                 if (bma->cur == NULL)
 632                         rval = XFS_ILOG_DEXT;
 633                 else {
 634                         rval = 0;
 635                         error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff,
 636                                         RIGHT.br_startblock,
 637                                         RIGHT.br_blockcount, &i);
 638                         if (error)
 639                                 goto done;
 640                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 641                         error = xfs_bmbt_update(bma->cur, PREV.br_startoff,
 642                                         new->br_startblock,
 643                                         PREV.br_blockcount +
 644                                         RIGHT.br_blockcount, PREV.br_state);
 645                         if (error)
 646                                 goto done;
 647                 }
 648                 break;
 649
 650         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
 651                 /*
 652                  * Filling in all of a previously delayed allocation extent.
 653                  * Neither the left nor right neighbors are contiguous with
 654                  * the new one.
 655                  */
 656                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
 657                 xfs_bmbt_set_startblock(ep, new->br_startblock);
 658                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
 659
 660                 bma->ip->i_d.di_nextents++;
 661                 if (bma->cur == NULL)
 662                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
 663                 else {
 664                         rval = XFS_ILOG_CORE;
 665                         error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
 666                                         new->br_startblock, new->br_blockcount,
 667                                         &i);
 668                         if (error)
 669                                 goto done;
 670                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
 671                         bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
 672                         error = xfs_btree_insert(bma->cur, &i);
 673                         if (error)
 674                                 goto done;
 675                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 676                 }
 677                 break;
 678
 679         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
 680                 /*
 681                  * Filling in the first part of a previous delayed allocation.
 682                  * The left neighbor is contiguous.
 683                  */
 684                 trace_xfs_bmap_pre_update(bma->ip, bma->idx - 1, state, _THIS_IP_);
 685                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx - 1),
 686                         LEFT.br_blockcount + new->br_blockcount);
 687                 xfs_bmbt_set_startoff(ep,
 688                         PREV.br_startoff + new->br_blockcount);
 689                 trace_xfs_bmap_post_update(bma->ip, bma->idx - 1, state, _THIS_IP_);
 690
 691                 temp = PREV.br_blockcount - new->br_blockcount;
 692                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
 693                 xfs_bmbt_set_blockcount(ep, temp);
 694                 if (bma->cur == NULL)
 695                         rval = XFS_ILOG_DEXT;
 696                 else {
 697                         rval = 0;
 698                         error = xfs_bmbt_lookup_eq(bma->cur, LEFT.br_startoff,
 699                                         LEFT.br_startblock, LEFT.br_blockcount,
 700                                         &i);
 701                         if (error)
 702                                 goto done;
 703                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 704                         error = xfs_bmbt_update(bma->cur, LEFT.br_startoff,
 705                                         LEFT.br_startblock,
 706                                         LEFT.br_blockcount +
 707                                         new->br_blockcount,
 708                                         LEFT.br_state);
 709                         if (error)
 710                                 goto done;
 711                 }
 712                 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
 713                         startblockval(PREV.br_startblock));
 714                 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
 715                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
 716
 717                 bma->idx--;
 718                 break;
 719
 720         case BMAP_LEFT_FILLING:
 721                 /*
 722                  * Filling in the first part of a previous delayed allocation.
 723                  * The left neighbor is not contiguous.
 724                  */
 725                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
 726                 xfs_bmbt_set_startoff(ep, new_endoff);
 727                 temp = PREV.br_blockcount - new->br_blockcount;
 728                 xfs_bmbt_set_blockcount(ep, temp);
 729                 xfs_iext_insert(bma->ip, bma->idx, 1, new, state);
 730                 bma->ip->i_d.di_nextents++;
 731                 if (bma->cur == NULL)
 732                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
 733                 else {
 734                         rval = XFS_ILOG_CORE;
 735                         error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
 736                                         new->br_startblock, new->br_blockcount,
 737                                         &i);
 738                         if (error)
 739                                 goto done;
 740                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
 741                         bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
 742                         error = xfs_btree_insert(bma->cur, &i);
 743                         if (error)
 744                                 goto done;
 745                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 746                 }
 747
 748                 if (xfs_bmap_needs_btree(bma->ip, XFS_DATA_FORK)) {
 749                         error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
 750                                         bma->firstblock, bma->flist,
 751                                         &bma->cur, 1, &tmp_rval, XFS_DATA_FORK);
 752                         rval |= tmp_rval;
 753                         if (error)
 754                                 goto done;
 755                 }
 756                 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
 757                         startblockval(PREV.br_startblock) -
 758                         (bma->cur ? bma->cur->bc_private.b.allocated : 0));
 759                 ep = xfs_iext_get_ext(ifp, bma->idx + 1);
 760                 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
 761                 trace_xfs_bmap_post_update(bma->ip, bma->idx + 1, state, _THIS_IP_);
 762                 break;
 763
 764         case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
 765                 /*
 766                  * Filling in the last part of a previous delayed allocation.
 767                  * The right neighbor is contiguous with the new allocation.
 768                  */
 769                 temp = PREV.br_blockcount - new->br_blockcount;
 770                 trace_xfs_bmap_pre_update(bma->ip, bma->idx + 1, state, _THIS_IP_);
 771                 xfs_bmbt_set_blockcount(ep, temp);
 772                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, bma->idx + 1),
 773                         new->br_startoff, new->br_startblock,
 774                         new->br_blockcount + RIGHT.br_blockcount,
 775                         RIGHT.br_state);
 776                 trace_xfs_bmap_post_update(bma->ip, bma->idx + 1, state, _THIS_IP_);
 777                 if (bma->cur == NULL)
 778                         rval = XFS_ILOG_DEXT;
 779                 else {
 780                         rval = 0;
 781                         error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff,
 782                                         RIGHT.br_startblock,
 783                                         RIGHT.br_blockcount, &i);
 784                         if (error)
 785                                 goto done;
 786                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 787                         error = xfs_bmbt_update(bma->cur, new->br_startoff,
 788                                         new->br_startblock,
 789                                         new->br_blockcount +
 790                                         RIGHT.br_blockcount,
 791                                         RIGHT.br_state);
 792                         if (error)
 793                                 goto done;
 794                 }
 795
 796                 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
 797                         startblockval(PREV.br_startblock));
 798                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
 799                 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
 800                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
 801
 802                 bma->idx++;
 803                 break;
 804
 805         case BMAP_RIGHT_FILLING:
 806                 /*
 807                  * Filling in the last part of a previous delayed allocation.
 808                  * The right neighbor is not contiguous.
 809                  */
 810                 temp = PREV.br_blockcount - new->br_blockcount;
 811                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
 812                 xfs_bmbt_set_blockcount(ep, temp);
 813                 xfs_iext_insert(bma->ip, bma->idx + 1, 1, new, state);
 814                 bma->ip->i_d.di_nextents++;
 815                 if (bma->cur == NULL)
 816                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
 817                 else {
 818                         rval = XFS_ILOG_CORE;
 819                         error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
 820                                         new->br_startblock, new->br_blockcount,
 821                                         &i);
 822                         if (error)
 823                                 goto done;
 824                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
 825                         bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
 826                         error = xfs_btree_insert(bma->cur, &i);
 827                         if (error)
 828                                 goto done;
 829                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 830                 }
 831
 832                 if (xfs_bmap_needs_btree(bma->ip, XFS_DATA_FORK)) {
 833                         error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
 834                                 bma->firstblock, bma->flist, &bma->cur, 1,
 835                                 &tmp_rval, XFS_DATA_FORK);
 836                         rval |= tmp_rval;
 837                         if (error)
 838                                 goto done;
 839                 }
 840                 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
 841                         startblockval(PREV.br_startblock) -
 842                         (bma->cur ? bma->cur->bc_private.b.allocated : 0));
 843                 ep = xfs_iext_get_ext(ifp, bma->idx);
 844                 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
 845                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
 846
 847                 bma->idx++;
 848                 break;
 849
 850         case 0:
 851                 /*
 852                  * Filling in the middle part of a previous delayed allocation.
 853                  * Contiguity is impossible here.
 854                  * This case is avoided almost all the time.
 855                  *
 856                  * We start with a delayed allocation:
 857                  *
 858                  * +ddddddddddddddddddddddddddddddddddddddddddddddddddddddd+
 859                  *  PREV @ idx
 860                  *
 861                  * and we are allocating:
 862                  *                     +rrrrrrrrrrrrrrrrr+
 863                  *                            new
 864                  *
 865                  * and we set it up for insertion as:
 866                  * +ddddddddddddddddddd+rrrrrrrrrrrrrrrrr+ddddddddddddddddd+
 867                  *                            new
 868                  *  PREV @ idx          LEFT              RIGHT
 869                  *                      inserted at idx + 1
 870                  */
 871                 temp = new->br_startoff - PREV.br_startoff;
 872                 temp2 = PREV.br_startoff + PREV.br_blockcount - new_endoff;
 873                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, 0, _THIS_IP_);
 874                 xfs_bmbt_set_blockcount(ep, temp);      /* truncate PREV */
 875                 LEFT = *new;
 876                 RIGHT.br_state = PREV.br_state;
 877                 RIGHT.br_startblock = nullstartblock(
 878                                 (int)xfs_bmap_worst_indlen(bma->ip, temp2));
 879                 RIGHT.br_startoff = new_endoff;
 880                 RIGHT.br_blockcount = temp2;
 881                 /* insert LEFT (r[0]) and RIGHT (r[1]) at the same time */
 882                 xfs_iext_insert(bma->ip, bma->idx + 1, 2, &LEFT, state);
 883                 bma->ip->i_d.di_nextents++;
 884                 if (bma->cur == NULL)
 885                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
 886                 else {
 887                         rval = XFS_ILOG_CORE;
 888                         error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
 889                                         new->br_startblock, new->br_blockcount,
 890                                         &i);
 891                         if (error)
 892                                 goto done;
 893                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
 894                         bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
 895                         error = xfs_btree_insert(bma->cur, &i);
 896                         if (error)
 897                                 goto done;
 898                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 899                 }
 900
 901                 if (xfs_bmap_needs_btree(bma->ip, XFS_DATA_FORK)) {
 902                         error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
 903                                         bma->firstblock, bma->flist, &bma->cur,
 904                                         1, &tmp_rval, XFS_DATA_FORK);
 905                         rval |= tmp_rval;
 906                         if (error)
 907                                 goto done;
 908                 }
 909                 temp = xfs_bmap_worst_indlen(bma->ip, temp);
 910                 temp2 = xfs_bmap_worst_indlen(bma->ip, temp2);
 911                 diff = (int)(temp + temp2 - startblockval(PREV.br_startblock) -
 912                         (bma->cur ? bma->cur->bc_private.b.allocated : 0));
 913                 if (diff > 0) {
 914                         error = xfs_icsb_modify_counters(bma->ip->i_mount,
 915                                         XFS_SBS_FDBLOCKS,
 916                                         -((int64_t)diff), 0);
 917                         ASSERT(!error);
 918                         if (error)
 919                                 goto done;
 920                 }
 921
 922                 ep = xfs_iext_get_ext(ifp, bma->idx);
 923                 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
 924                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
 925                 trace_xfs_bmap_pre_update(bma->ip, bma->idx + 2, state, _THIS_IP_);
 926                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, bma->idx + 2),
 927                         nullstartblock((int)temp2));
 928                 trace_xfs_bmap_post_update(bma->ip, bma->idx + 2, state, _THIS_IP_);
 929
 930                 bma->idx++;
 931                 da_new = temp + temp2;
 932                 break;
 933
 934         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
 935         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
 936         case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
 937         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
 938         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
 939         case BMAP_LEFT_CONTIG:
 940         case BMAP_RIGHT_CONTIG:
 941                 /*
 942                  * These cases are all impossible.
 943                  */
 944                 ASSERT(0);
 945         }
 946
 947         /* convert to a btree if necessary */
 948         if (xfs_bmap_needs_btree(bma->ip, XFS_DATA_FORK)) {
 949                 int     tmp_logflags;   /* partial log flag return val */
 950
 951                 ASSERT(bma->cur == NULL);
 952                 error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
 953                                 bma->firstblock, bma->flist, &bma->cur,
 954                                 da_old > 0, &tmp_logflags, XFS_DATA_FORK);
 955                 bma->logflags |= tmp_logflags;
 956                 if (error)
 957                         goto done;
 958         }
 959
 960         /* adjust for changes in reserved delayed indirect blocks */
 961         if (da_old || da_new) {
 962                 temp = da_new;
 963                 if (bma->cur)
 964                         temp += bma->cur->bc_private.b.allocated;
 965                 ASSERT(temp <= da_old);
 966                 if (temp < da_old)
 967                         xfs_icsb_modify_counters(bma->ip->i_mount,
 968                                         XFS_SBS_FDBLOCKS,
 969                                         (int64_t)(da_old - temp), 0);
 970         }
 971
 972         /* clear out the allocated field, done with it now in any case. */
 973         if (bma->cur)
 974                 bma->cur->bc_private.b.allocated = 0;
 975
 976         xfs_bmap_check_leaf_extents(bma->cur, bma->ip, XFS_DATA_FORK);
 977 done:
 978         bma->logflags |= rval;
 979         return error;
 980 #undef  LEFT
 981 #undef  RIGHT
 982 #undef  PREV
 983 }
 984
 985 /*
 986  * Convert an unwritten allocation to a real allocation or vice versa.
 987  */
 988 STATIC int                              /* error */
 989 xfs_bmap_add_extent_unwritten_real(
 990         struct xfs_trans        *tp,
 991         xfs_inode_t             *ip,    /* incore inode pointer */
 992         xfs_extnum_t            *idx,   /* extent number to update/insert */
 993         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
 994         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
 995         xfs_fsblock_t           *first, /* pointer to firstblock variable */
 996         xfs_bmap_free_t         *flist, /* list of extents to be freed */
 997         int                     *logflagsp) /* inode logging flags */
 998 {
 999         xfs_btree_cur_t         *cur;   /* btree cursor */
1000         xfs_bmbt_rec_host_t     *ep;    /* extent entry for idx */
1001         int                     error;  /* error return value */
1002         int                     i;      /* temp state */
1003         xfs_ifork_t             *ifp;   /* inode fork pointer */
1004         xfs_fileoff_t           new_endoff;     /* end offset of new entry */
1005         xfs_exntst_t            newext; /* new extent state */
1006         xfs_exntst_t            oldext; /* old extent state */
1007         xfs_bmbt_irec_t         r[3];   /* neighbor extent entries */
1008                                         /* left is 0, right is 1, prev is 2 */
1009         int                     rval=0; /* return value (logging flags) */
1010         int                     state = 0;/* state bits, accessed thru macros */
1011
1012         *logflagsp = 0;
1013
1014         cur = *curp;
1015         ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
1016
1017         ASSERT(*idx >= 0);
1018         ASSERT(*idx <= ifp->if_bytes / sizeof(struct xfs_bmbt_rec));
1019         ASSERT(!isnullstartblock(new->br_startblock));
1020
1021         XFS_STATS_INC(xs_add_exlist);
1022
1023 #define LEFT            r[0]
1024 #define RIGHT           r[1]
1025 #define PREV            r[2]
1026
1027         /*
1028          * Set up a bunch of variables to make the tests simpler.
1029          */
1030         error = 0;
1031         ep = xfs_iext_get_ext(ifp, *idx);
1032         xfs_bmbt_get_all(ep, &PREV);
1033         newext = new->br_state;
1034         oldext = (newext == XFS_EXT_UNWRITTEN) ?
1035                 XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
1036         ASSERT(PREV.br_state == oldext);
1037         new_endoff = new->br_startoff + new->br_blockcount;
1038         ASSERT(PREV.br_startoff <= new->br_startoff);
1039         ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
1040
1041         /*
1042          * Set flags determining what part of the previous oldext allocation
1043          * extent is being replaced by a newext allocation.
1044          */
1045         if (PREV.br_startoff == new->br_startoff)
1046                 state |= BMAP_LEFT_FILLING;
1047         if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
1048                 state |= BMAP_RIGHT_FILLING;
1049
1050         /*
1051          * Check and set flags if this segment has a left neighbor.
1052          * Don't set contiguous if the combined extent would be too large.
1053          */
1054         if (*idx > 0) {
1055                 state |= BMAP_LEFT_VALID;
1056                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &LEFT);
1057
1058                 if (isnullstartblock(LEFT.br_startblock))
1059                         state |= BMAP_LEFT_DELAY;
1060         }
1061
1062         if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
1063             LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
1064             LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
1065             LEFT.br_state == newext &&
1066             LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
1067                 state |= BMAP_LEFT_CONTIG;
1068
1069         /*
1070          * Check and set flags if this segment has a right neighbor.
1071          * Don't set contiguous if the combined extent would be too large.
1072          * Also check for all-three-contiguous being too large.
1073          */
1074         if (*idx < ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1) {
1075                 state |= BMAP_RIGHT_VALID;
1076                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx + 1), &RIGHT);
1077                 if (isnullstartblock(RIGHT.br_startblock))
1078                         state |= BMAP_RIGHT_DELAY;
1079         }
1080
1081         if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
1082             new_endoff == RIGHT.br_startoff &&
1083             new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
1084             newext == RIGHT.br_state &&
1085             new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
1086             ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1087                        BMAP_RIGHT_FILLING)) !=
1088                       (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1089                        BMAP_RIGHT_FILLING) ||
1090              LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
1091                         <= MAXEXTLEN))
1092                 state |= BMAP_RIGHT_CONTIG;
1093
1094         /*
1095          * Switch out based on the FILLING and CONTIG state bits.
1096          */
1097         switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1098                          BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
1099         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1100              BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1101                 /*
1102                  * Setting all of a previous oldext extent to newext.
1103                  * The left and right neighbors are both contiguous with new.
1104                  */
1105                 --*idx;
1106
1107                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1108                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
1109                         LEFT.br_blockcount + PREV.br_blockcount +
1110                         RIGHT.br_blockcount);
1111                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1112
1113                 xfs_iext_remove(ip, *idx + 1, 2, state);
1114                 ip->i_d.di_nextents -= 2;
1115                 if (cur == NULL)
1116                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1117                 else {
1118                         rval = XFS_ILOG_CORE;
1119                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
1120                                         RIGHT.br_startblock,
1121                                         RIGHT.br_blockcount, &i)))
1122                                 goto done;
1123                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1124                         if ((error = xfs_btree_delete(cur, &i)))
1125                                 goto done;
1126                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1127                         if ((error = xfs_btree_decrement(cur, 0, &i)))
1128                                 goto done;
1129                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1130                         if ((error = xfs_btree_delete(cur, &i)))
1131                                 goto done;
1132                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1133                         if ((error = xfs_btree_decrement(cur, 0, &i)))
1134                                 goto done;
1135                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1136                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
1137                                 LEFT.br_startblock,
1138                                 LEFT.br_blockcount + PREV.br_blockcount +
1139                                 RIGHT.br_blockcount, LEFT.br_state)))
1140                                 goto done;
1141                 }
1142                 break;
1143
1144         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1145                 /*
1146                  * Setting all of a previous oldext extent to newext.
1147                  * The left neighbor is contiguous, the right is not.
1148                  */
1149                 --*idx;
1150
1151                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1152                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
1153                         LEFT.br_blockcount + PREV.br_blockcount);
1154                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1155
1156                 xfs_iext_remove(ip, *idx + 1, 1, state);
1157                 ip->i_d.di_nextents--;
1158                 if (cur == NULL)
1159                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1160                 else {
1161                         rval = XFS_ILOG_CORE;
1162                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1163                                         PREV.br_startblock, PREV.br_blockcount,
1164                                         &i)))
1165                                 goto done;
1166                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1167                         if ((error = xfs_btree_delete(cur, &i)))
1168                                 goto done;
1169                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1170                         if ((error = xfs_btree_decrement(cur, 0, &i)))
1171                                 goto done;
1172                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1173                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
1174                                 LEFT.br_startblock,
1175                                 LEFT.br_blockcount + PREV.br_blockcount,
1176                                 LEFT.br_state)))
1177                                 goto done;
1178                 }
1179                 break;
1180
1181         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1182                 /*
1183                  * Setting all of a previous oldext extent to newext.
1184                  * The right neighbor is contiguous, the left is not.
1185                  */
1186                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1187                 xfs_bmbt_set_blockcount(ep,
1188                         PREV.br_blockcount + RIGHT.br_blockcount);
1189                 xfs_bmbt_set_state(ep, newext);
1190                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1191                 xfs_iext_remove(ip, *idx + 1, 1, state);
1192                 ip->i_d.di_nextents--;
1193                 if (cur == NULL)
1194                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1195                 else {
1196                         rval = XFS_ILOG_CORE;
1197                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
1198                                         RIGHT.br_startblock,
1199                                         RIGHT.br_blockcount, &i)))
1200                                 goto done;
1201                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1202                         if ((error = xfs_btree_delete(cur, &i)))
1203                                 goto done;
1204                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1205                         if ((error = xfs_btree_decrement(cur, 0, &i)))
1206                                 goto done;
1207                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1208                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
1209                                 new->br_startblock,
1210                                 new->br_blockcount + RIGHT.br_blockcount,
1211                                 newext)))
1212                                 goto done;
1213                 }
1214                 break;
1215
1216         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
1217                 /*
1218                  * Setting all of a previous oldext extent to newext.
1219                  * Neither the left nor right neighbors are contiguous with
1220                  * the new one.
1221                  */
1222                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1223                 xfs_bmbt_set_state(ep, newext);
1224                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1225
1226                 if (cur == NULL)
1227                         rval = XFS_ILOG_DEXT;
1228                 else {
1229                         rval = 0;
1230                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1231                                         new->br_startblock, new->br_blockcount,
1232                                         &i)))
1233                                 goto done;
1234                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1235                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
1236                                 new->br_startblock, new->br_blockcount,
1237                                 newext)))
1238                                 goto done;
1239                 }
1240                 break;
1241
1242         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
1243                 /*
1244                  * Setting the first part of a previous oldext extent to newext.
1245                  * The left neighbor is contiguous.
1246                  */
1247                 trace_xfs_bmap_pre_update(ip, *idx - 1, state, _THIS_IP_);
1248                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx - 1),
1249                         LEFT.br_blockcount + new->br_blockcount);
1250                 xfs_bmbt_set_startoff(ep,
1251                         PREV.br_startoff + new->br_blockcount);
1252                 trace_xfs_bmap_post_update(ip, *idx - 1, state, _THIS_IP_);
1253
1254                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1255                 xfs_bmbt_set_startblock(ep,
1256                         new->br_startblock + new->br_blockcount);
1257                 xfs_bmbt_set_blockcount(ep,
1258                         PREV.br_blockcount - new->br_blockcount);
1259                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1260
1261                 --*idx;
1262
1263                 if (cur == NULL)
1264                         rval = XFS_ILOG_DEXT;
1265                 else {
1266                         rval = 0;
1267                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1268                                         PREV.br_startblock, PREV.br_blockcount,
1269                                         &i)))
1270                                 goto done;
1271                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1272                         if ((error = xfs_bmbt_update(cur,
1273                                 PREV.br_startoff + new->br_blockcount,
1274                                 PREV.br_startblock + new->br_blockcount,
1275                                 PREV.br_blockcount - new->br_blockcount,
1276                                 oldext)))
1277                                 goto done;
1278                         if ((error = xfs_btree_decrement(cur, 0, &i)))
1279                                 goto done;
1280                         error = xfs_bmbt_update(cur, LEFT.br_startoff,
1281                                 LEFT.br_startblock,
1282                                 LEFT.br_blockcount + new->br_blockcount,
1283                                 LEFT.br_state);
1284                         if (error)
1285                                 goto done;
1286                 }
1287                 break;
1288
1289         case BMAP_LEFT_FILLING:
1290                 /*
1291                  * Setting the first part of a previous oldext extent to newext.
1292                  * The left neighbor is not contiguous.
1293                  */
1294                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1295                 ASSERT(ep && xfs_bmbt_get_state(ep) == oldext);
1296                 xfs_bmbt_set_startoff(ep, new_endoff);
1297                 xfs_bmbt_set_blockcount(ep,
1298                         PREV.br_blockcount - new->br_blockcount);
1299                 xfs_bmbt_set_startblock(ep,
1300                         new->br_startblock + new->br_blockcount);
1301                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1302
1303                 xfs_iext_insert(ip, *idx, 1, new, state);
1304                 ip->i_d.di_nextents++;
1305                 if (cur == NULL)
1306                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1307                 else {
1308                         rval = XFS_ILOG_CORE;
1309                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1310                                         PREV.br_startblock, PREV.br_blockcount,
1311                                         &i)))
1312                                 goto done;
1313                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1314                         if ((error = xfs_bmbt_update(cur,
1315                                 PREV.br_startoff + new->br_blockcount,
1316                                 PREV.br_startblock + new->br_blockcount,
1317                                 PREV.br_blockcount - new->br_blockcount,
1318                                 oldext)))
1319                                 goto done;
1320                         cur->bc_rec.b = *new;
1321                         if ((error = xfs_btree_insert(cur, &i)))
1322                                 goto done;
1323                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1324                 }
1325                 break;
1326
1327         case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1328                 /*
1329                  * Setting the last part of a previous oldext extent to newext.
1330                  * The right neighbor is contiguous with the new allocation.
1331                  */
1332                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1333                 xfs_bmbt_set_blockcount(ep,
1334                         PREV.br_blockcount - new->br_blockcount);
1335                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1336
1337                 ++*idx;
1338
1339                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1340                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx),
1341                         new->br_startoff, new->br_startblock,
1342                         new->br_blockcount + RIGHT.br_blockcount, newext);
1343                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1344
1345                 if (cur == NULL)
1346                         rval = XFS_ILOG_DEXT;
1347                 else {
1348                         rval = 0;
1349                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1350                                         PREV.br_startblock,
1351                                         PREV.br_blockcount, &i)))
1352                                 goto done;
1353                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1354                         if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
1355                                 PREV.br_startblock,
1356                                 PREV.br_blockcount - new->br_blockcount,
1357                                 oldext)))
1358                                 goto done;
1359                         if ((error = xfs_btree_increment(cur, 0, &i)))
1360                                 goto done;
1361                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
1362                                 new->br_startblock,
1363                                 new->br_blockcount + RIGHT.br_blockcount,
1364                                 newext)))
1365                                 goto done;
1366                 }
1367                 break;
1368
1369         case BMAP_RIGHT_FILLING:
1370                 /*
1371                  * Setting the last part of a previous oldext extent to newext.
1372                  * The right neighbor is not contiguous.
1373                  */
1374                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1375                 xfs_bmbt_set_blockcount(ep,
1376                         PREV.br_blockcount - new->br_blockcount);
1377                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1378
1379                 ++*idx;
1380                 xfs_iext_insert(ip, *idx, 1, new, state);
1381
1382                 ip->i_d.di_nextents++;
1383                 if (cur == NULL)
1384                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1385                 else {
1386                         rval = XFS_ILOG_CORE;
1387                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1388                                         PREV.br_startblock, PREV.br_blockcount,
1389                                         &i)))
1390                                 goto done;
1391                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1392                         if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
1393                                 PREV.br_startblock,
1394                                 PREV.br_blockcount - new->br_blockcount,
1395                                 oldext)))
1396                                 goto done;
1397                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1398                                         new->br_startblock, new->br_blockcount,
1399                                         &i)))
1400                                 goto done;
1401                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1402                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
1403                         if ((error = xfs_btree_insert(cur, &i)))
1404                                 goto done;
1405                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1406                 }
1407                 break;
1408
1409         case 0:
1410                 /*
1411                  * Setting the middle part of a previous oldext extent to
1412                  * newext.  Contiguity is impossible here.
1413                  * One extent becomes three extents.
1414                  */
1415                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1416                 xfs_bmbt_set_blockcount(ep,
1417                         new->br_startoff - PREV.br_startoff);
1418                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1419
1420                 r[0] = *new;
1421                 r[1].br_startoff = new_endoff;
1422                 r[1].br_blockcount =
1423                         PREV.br_startoff + PREV.br_blockcount - new_endoff;
1424                 r[1].br_startblock = new->br_startblock + new->br_blockcount;
1425                 r[1].br_state = oldext;
1426
1427                 ++*idx;
1428                 xfs_iext_insert(ip, *idx, 2, &r[0], state);
1429
1430                 ip->i_d.di_nextents += 2;
1431                 if (cur == NULL)
1432                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1433                 else {
1434                         rval = XFS_ILOG_CORE;
1435                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1436                                         PREV.br_startblock, PREV.br_blockcount,
1437                                         &i)))
1438                                 goto done;
1439                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1440                         /* new right extent - oldext */
1441                         if ((error = xfs_bmbt_update(cur, r[1].br_startoff,
1442                                 r[1].br_startblock, r[1].br_blockcount,
1443                                 r[1].br_state)))
1444                                 goto done;
1445                         /* new left extent - oldext */
1446                         cur->bc_rec.b = PREV;
1447                         cur->bc_rec.b.br_blockcount =
1448                                 new->br_startoff - PREV.br_startoff;
1449                         if ((error = xfs_btree_insert(cur, &i)))
1450                                 goto done;
1451                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1452                         /*
1453                          * Reset the cursor to the position of the new extent
1454                          * we are about to insert as we can't trust it after
1455                          * the previous insert.
1456                          */
1457                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1458                                         new->br_startblock, new->br_blockcount,
1459                                         &i)))
1460                                 goto done;
1461                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1462                         /* new middle extent - newext */
1463                         cur->bc_rec.b.br_state = new->br_state;
1464                         if ((error = xfs_btree_insert(cur, &i)))
1465                                 goto done;
1466                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1467                 }
1468                 break;
1469
1470         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1471         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1472         case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
1473         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1474         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1475         case BMAP_LEFT_CONTIG:
1476         case BMAP_RIGHT_CONTIG:
1477                 /*
1478                  * These cases are all impossible.
1479                  */
1480                 ASSERT(0);
1481         }
1482
1483         /* convert to a btree if necessary */
1484         if (xfs_bmap_needs_btree(ip, XFS_DATA_FORK)) {
1485                 int     tmp_logflags;   /* partial log flag return val */
1486
1487                 ASSERT(cur == NULL);
1488                 error = xfs_bmap_extents_to_btree(tp, ip, first, flist, &cur,
1489                                 0, &tmp_logflags, XFS_DATA_FORK);
1490                 *logflagsp |= tmp_logflags;
1491                 if (error)
1492                         goto done;
1493         }
1494
1495         /* clear out the allocated field, done with it now in any case. */
1496         if (cur) {
1497                 cur->bc_private.b.allocated = 0;
1498                 *curp = cur;
1499         }
1500
1501         xfs_bmap_check_leaf_extents(*curp, ip, XFS_DATA_FORK);
1502 done:
1503         *logflagsp |= rval;
1504         return error;
1505 #undef  LEFT
1506 #undef  RIGHT
1507 #undef  PREV
1508 }
1509
1510 /*
1511  * Convert a hole to a delayed allocation.
1512  */
1513 STATIC void
1514 xfs_bmap_add_extent_hole_delay(
1515         xfs_inode_t             *ip,    /* incore inode pointer */
1516         xfs_extnum_t            *idx,   /* extent number to update/insert */
1517         xfs_bmbt_irec_t         *new)   /* new data to add to file extents */
1518 {
1519         xfs_ifork_t             *ifp;   /* inode fork pointer */
1520         xfs_bmbt_irec_t         left;   /* left neighbor extent entry */
1521         xfs_filblks_t           newlen=0;       /* new indirect size */
1522         xfs_filblks_t           oldlen=0;       /* old indirect size */
1523         xfs_bmbt_irec_t         right;  /* right neighbor extent entry */
1524         int                     state;  /* state bits, accessed thru macros */
1525         xfs_filblks_t           temp=0; /* temp for indirect calculations */
1526
1527         ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
1528         state = 0;
1529         ASSERT(isnullstartblock(new->br_startblock));
1530
1531         /*
1532          * Check and set flags if this segment has a left neighbor
1533          */
1534         if (*idx > 0) {
1535                 state |= BMAP_LEFT_VALID;
1536                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &left);
1537
1538                 if (isnullstartblock(left.br_startblock))
1539                         state |= BMAP_LEFT_DELAY;
1540         }
1541
1542         /*
1543          * Check and set flags if the current (right) segment exists.
1544          * If it doesn't exist, we're converting the hole at end-of-file.
1545          */
1546         if (*idx < ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t)) {
1547                 state |= BMAP_RIGHT_VALID;
1548                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx), &right);
1549
1550                 if (isnullstartblock(right.br_startblock))
1551                         state |= BMAP_RIGHT_DELAY;
1552         }
1553
1554         /*
1555          * Set contiguity flags on the left and right neighbors.
1556          * Don't let extents get too large, even if the pieces are contiguous.
1557          */
1558         if ((state & BMAP_LEFT_VALID) && (state & BMAP_LEFT_DELAY) &&
1559             left.br_startoff + left.br_blockcount == new->br_startoff &&
1560             left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
1561                 state |= BMAP_LEFT_CONTIG;
1562
1563         if ((state & BMAP_RIGHT_VALID) && (state & BMAP_RIGHT_DELAY) &&
1564             new->br_startoff + new->br_blockcount == right.br_startoff &&
1565             new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
1566             (!(state & BMAP_LEFT_CONTIG) ||
1567              (left.br_blockcount + new->br_blockcount +
1568               right.br_blockcount <= MAXEXTLEN)))
1569                 state |= BMAP_RIGHT_CONTIG;
1570
1571         /*
1572          * Switch out based on the contiguity flags.
1573          */
1574         switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
1575         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1576                 /*
1577                  * New allocation is contiguous with delayed allocations
1578                  * on the left and on the right.
1579                  * Merge all three into a single extent record.
1580                  */
1581                 --*idx;
1582                 temp = left.br_blockcount + new->br_blockcount +
1583                         right.br_blockcount;
1584
1585                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1586                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx), temp);
1587                 oldlen = startblockval(left.br_startblock) +
1588                         startblockval(new->br_startblock) +
1589                         startblockval(right.br_startblock);
1590                 newlen = xfs_bmap_worst_indlen(ip, temp);
1591                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, *idx),
1592                         nullstartblock((int)newlen));
1593                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1594
1595                 xfs_iext_remove(ip, *idx + 1, 1, state);
1596                 break;
1597
1598         case BMAP_LEFT_CONTIG:
1599                 /*
1600                  * New allocation is contiguous with a delayed allocation
1601                  * on the left.
1602                  * Merge the new allocation with the left neighbor.
1603                  */
1604                 --*idx;
1605                 temp = left.br_blockcount + new->br_blockcount;
1606
1607                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1608                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx), temp);
1609                 oldlen = startblockval(left.br_startblock) +
1610                         startblockval(new->br_startblock);
1611                 newlen = xfs_bmap_worst_indlen(ip, temp);
1612                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, *idx),
1613                         nullstartblock((int)newlen));
1614                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1615                 break;
1616
1617         case BMAP_RIGHT_CONTIG:
1618                 /*
1619                  * New allocation is contiguous with a delayed allocation
1620                  * on the right.
1621                  * Merge the new allocation with the right neighbor.
1622                  */
1623                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1624                 temp = new->br_blockcount + right.br_blockcount;
1625                 oldlen = startblockval(new->br_startblock) +
1626                         startblockval(right.br_startblock);
1627                 newlen = xfs_bmap_worst_indlen(ip, temp);
1628                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx),
1629                         new->br_startoff,
1630                         nullstartblock((int)newlen), temp, right.br_state);
1631                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1632                 break;
1633
1634         case 0:
1635                 /*
1636                  * New allocation is not contiguous with another
1637                  * delayed allocation.
1638                  * Insert a new entry.
1639                  */
1640                 oldlen = newlen = 0;
1641                 xfs_iext_insert(ip, *idx, 1, new, state);
1642                 break;
1643         }
1644         if (oldlen != newlen) {
1645                 ASSERT(oldlen > newlen);
1646                 xfs_icsb_modify_counters(ip->i_mount, XFS_SBS_FDBLOCKS,
1647                         (int64_t)(oldlen - newlen), 0);
1648                 /*
1649                  * Nothing to do for disk quota accounting here.
1650                  */
1651         }
1652 }
1653
1654 /*
1655  * Convert a hole to a real allocation.
1656  */
1657 STATIC int                              /* error */
1658 xfs_bmap_add_extent_hole_real(
1659         struct xfs_bmalloca     *bma,
1660         int                     whichfork)
1661 {
1662         struct xfs_bmbt_irec    *new = &bma->got;
1663         int                     error;  /* error return value */
1664         int                     i;      /* temp state */
1665         xfs_ifork_t             *ifp;   /* inode fork pointer */
1666         xfs_bmbt_irec_t         left;   /* left neighbor extent entry */
1667         xfs_bmbt_irec_t         right;  /* right neighbor extent entry */
1668         int                     rval=0; /* return value (logging flags) */
1669         int                     state;  /* state bits, accessed thru macros */
1670
1671         ifp = XFS_IFORK_PTR(bma->ip, whichfork);
1672
1673         ASSERT(bma->idx >= 0);
1674         ASSERT(bma->idx <= ifp->if_bytes / sizeof(struct xfs_bmbt_rec));
1675         ASSERT(!isnullstartblock(new->br_startblock));
1676         ASSERT(!bma->cur ||
1677                !(bma->cur->bc_private.b.flags & XFS_BTCUR_BPRV_WASDEL));
1678
1679         XFS_STATS_INC(xs_add_exlist);
1680
1681         state = 0;
1682         if (whichfork == XFS_ATTR_FORK)
1683                 state |= BMAP_ATTRFORK;
1684
1685         /*
1686          * Check and set flags if this segment has a left neighbor.
1687          */
1688         if (bma->idx > 0) {
1689                 state |= BMAP_LEFT_VALID;
1690                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx - 1), &left);
1691                 if (isnullstartblock(left.br_startblock))
1692                         state |= BMAP_LEFT_DELAY;
1693         }
1694
1695         /*
1696          * Check and set flags if this segment has a current value.
1697          * Not true if we're inserting into the "hole" at eof.
1698          */
1699         if (bma->idx < ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)) {
1700                 state |= BMAP_RIGHT_VALID;
1701                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx), &right);
1702                 if (isnullstartblock(right.br_startblock))
1703                         state |= BMAP_RIGHT_DELAY;
1704         }
1705
1706         /*
1707          * We're inserting a real allocation between "left" and "right".
1708          * Set the contiguity flags.  Don't let extents get too large.
1709          */
1710         if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
1711             left.br_startoff + left.br_blockcount == new->br_startoff &&
1712             left.br_startblock + left.br_blockcount == new->br_startblock &&
1713             left.br_state == new->br_state &&
1714             left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
1715                 state |= BMAP_LEFT_CONTIG;
1716
1717         if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
1718             new->br_startoff + new->br_blockcount == right.br_startoff &&
1719             new->br_startblock + new->br_blockcount == right.br_startblock &&
1720             new->br_state == right.br_state &&
1721             new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
1722             (!(state & BMAP_LEFT_CONTIG) ||
1723              left.br_blockcount + new->br_blockcount +
1724              right.br_blockcount <= MAXEXTLEN))
1725                 state |= BMAP_RIGHT_CONTIG;
1726
1727         error = 0;
1728         /*
1729          * Select which case we're in here, and implement it.
1730          */
1731         switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
1732         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1733                 /*
1734                  * New allocation is contiguous with real allocations on the
1735                  * left and on the right.
1736                  * Merge all three into a single extent record.
1737                  */
1738                 --bma->idx;
1739                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1740                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
1741                         left.br_blockcount + new->br_blockcount +
1742                         right.br_blockcount);
1743                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1744
1745                 xfs_iext_remove(bma->ip, bma->idx + 1, 1, state);
1746
1747                 XFS_IFORK_NEXT_SET(bma->ip, whichfork,
1748                         XFS_IFORK_NEXTENTS(bma->ip, whichfork) - 1);
1749                 if (bma->cur == NULL) {
1750                         rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
1751                 } else {
1752                         rval = XFS_ILOG_CORE;
1753                         error = xfs_bmbt_lookup_eq(bma->cur, right.br_startoff,
1754                                         right.br_startblock, right.br_blockcount,
1755                                         &i);
1756                         if (error)
1757                                 goto done;
1758                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1759                         error = xfs_btree_delete(bma->cur, &i);
1760                         if (error)
1761                                 goto done;
1762                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1763                         error = xfs_btree_decrement(bma->cur, 0, &i);
1764                         if (error)
1765                                 goto done;
1766                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1767                         error = xfs_bmbt_update(bma->cur, left.br_startoff,
1768                                         left.br_startblock,
1769                                         left.br_blockcount +
1770                                                 new->br_blockcount +
1771                                                 right.br_blockcount,
1772                                         left.br_state);
1773                         if (error)
1774                                 goto done;
1775                 }
1776                 break;
1777
1778         case BMAP_LEFT_CONTIG:
1779                 /*
1780                  * New allocation is contiguous with a real allocation
1781                  * on the left.
1782                  * Merge the new allocation with the left neighbor.
1783                  */
1784                 --bma->idx;
1785                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1786                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
1787                         left.br_blockcount + new->br_blockcount);
1788                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1789
1790                 if (bma->cur == NULL) {
1791                         rval = xfs_ilog_fext(whichfork);
1792                 } else {
1793                         rval = 0;
1794                         error = xfs_bmbt_lookup_eq(bma->cur, left.br_startoff,
1795                                         left.br_startblock, left.br_blockcount,
1796                                         &i);
1797                         if (error)
1798                                 goto done;
1799                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1800                         error = xfs_bmbt_update(bma->cur, left.br_startoff,
1801                                         left.br_startblock,
1802                                         left.br_blockcount +
1803                                                 new->br_blockcount,
1804                                         left.br_state);
1805                         if (error)
1806                                 goto done;
1807                 }
1808                 break;
1809
1810         case BMAP_RIGHT_CONTIG:
1811                 /*
1812                  * New allocation is contiguous with a real allocation
1813                  * on the right.
1814                  * Merge the new allocation with the right neighbor.
1815                  */
1816                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1817                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, bma->idx),
1818                         new->br_startoff, new->br_startblock,
1819                         new->br_blockcount + right.br_blockcount,
1820                         right.br_state);
1821                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1822
1823                 if (bma->cur == NULL) {
1824                         rval = xfs_ilog_fext(whichfork);
1825                 } else {
1826                         rval = 0;
1827                         error = xfs_bmbt_lookup_eq(bma->cur,
1828                                         right.br_startoff,
1829                                         right.br_startblock,
1830                                         right.br_blockcount, &i);
1831                         if (error)
1832                                 goto done;
1833                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1834                         error = xfs_bmbt_update(bma->cur, new->br_startoff,
1835                                         new->br_startblock,
1836                                         new->br_blockcount +
1837                                                 right.br_blockcount,
1838                                         right.br_state);
1839                         if (error)
1840                                 goto done;
1841                 }
1842                 break;
1843
1844         case 0:
1845                 /*
1846                  * New allocation is not contiguous with another
1847                  * real allocation.
1848                  * Insert a new entry.
1849                  */
1850                 xfs_iext_insert(bma->ip, bma->idx, 1, new, state);
1851                 XFS_IFORK_NEXT_SET(bma->ip, whichfork,
1852                         XFS_IFORK_NEXTENTS(bma->ip, whichfork) + 1);
1853                 if (bma->cur == NULL) {
1854                         rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
1855                 } else {
1856                         rval = XFS_ILOG_CORE;
1857                         error = xfs_bmbt_lookup_eq(bma->cur,
1858                                         new->br_startoff,
1859                                         new->br_startblock,
1860                                         new->br_blockcount, &i);
1861                         if (error)
1862                                 goto done;
1863                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1864                         bma->cur->bc_rec.b.br_state = new->br_state;
1865                         error = xfs_btree_insert(bma->cur, &i);
1866                         if (error)
1867                                 goto done;
1868                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1869                 }
1870                 break;
1871         }
1872
1873         /* convert to a btree if necessary */
1874         if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
1875                 int     tmp_logflags;   /* partial log flag return val */
1876
1877                 ASSERT(bma->cur == NULL);
1878                 error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
1879                                 bma->firstblock, bma->flist, &bma->cur,
1880                                 0, &tmp_logflags, whichfork);
1881                 bma->logflags |= tmp_logflags;
1882                 if (error)
1883                         goto done;
1884         }
1885
1886         /* clear out the allocated field, done with it now in any case. */
1887         if (bma->cur)
1888                 bma->cur->bc_private.b.allocated = 0;
1889
1890         xfs_bmap_check_leaf_extents(bma->cur, bma->ip, whichfork);
1891 done:
1892         bma->logflags |= rval;
1893         return error;
1894 }
1895
1896 /*
1897  * Adjust the size of the new extent based on di_extsize and rt extsize.
1898  */
1899 STATIC int
1900 xfs_bmap_extsize_align(
1901         xfs_mount_t     *mp,
1902         xfs_bmbt_irec_t *gotp,          /* next extent pointer */
1903         xfs_bmbt_irec_t *prevp,         /* previous extent pointer */
1904         xfs_extlen_t    extsz,          /* align to this extent size */
1905         int             rt,             /* is this a realtime inode? */
1906         int             eof,            /* is extent at end-of-file? */
1907         int             delay,          /* creating delalloc extent? */
1908         int             convert,        /* overwriting unwritten extent? */
1909         xfs_fileoff_t   *offp,          /* in/out: aligned offset */
1910         xfs_extlen_t    *lenp)          /* in/out: aligned length */
1911 {
1912         xfs_fileoff_t   orig_off;       /* original offset */
1913         xfs_extlen_t    orig_alen;      /* original length */
1914         xfs_fileoff_t   orig_end;       /* original off+len */
1915         xfs_fileoff_t   nexto;          /* next file offset */
1916         xfs_fileoff_t   prevo;          /* previous file offset */
1917         xfs_fileoff_t   align_off;      /* temp for offset */
1918         xfs_extlen_t    align_alen;     /* temp for length */
1919         xfs_extlen_t    temp;           /* temp for calculations */
1920
1921         if (convert)
1922                 return 0;
1923
1924         orig_off = align_off = *offp;
1925         orig_alen = align_alen = *lenp;
1926         orig_end = orig_off + orig_alen;
1927
1928         /*
1929          * If this request overlaps an existing extent, then don't
1930          * attempt to perform any additional alignment.
1931          */
1932         if (!delay && !eof &&
1933             (orig_off >= gotp->br_startoff) &&
1934             (orig_end <= gotp->br_startoff + gotp->br_blockcount)) {
1935                 return 0;
1936         }
1937
1938         /*
1939          * If the file offset is unaligned vs. the extent size
1940          * we need to align it.  This will be possible unless
1941          * the file was previously written with a kernel that didn't
1942          * perform this alignment, or if a truncate shot us in the
1943          * foot.
1944          */
1945         temp = do_mod(orig_off, extsz);
1946         if (temp) {
1947                 align_alen += temp;
1948                 align_off -= temp;
1949         }
1950         /*
1951          * Same adjustment for the end of the requested area.
1952          */
1953         if ((temp = (align_alen % extsz))) {
1954                 align_alen += extsz - temp;
1955         }
1956         /*
1957          * If the previous block overlaps with this proposed allocation
1958          * then move the start forward without adjusting the length.
1959          */
1960         if (prevp->br_startoff != NULLFILEOFF) {
1961                 if (prevp->br_startblock == HOLESTARTBLOCK)
1962                         prevo = prevp->br_startoff;
1963                 else
1964                         prevo = prevp->br_startoff + prevp->br_blockcount;
1965         } else
1966                 prevo = 0;
1967         if (align_off != orig_off && align_off < prevo)
1968                 align_off = prevo;
1969         /*
1970          * If the next block overlaps with this proposed allocation
1971          * then move the start back without adjusting the length,
1972          * but not before offset 0.
1973          * This may of course make the start overlap previous block,
1974          * and if we hit the offset 0 limit then the next block
1975          * can still overlap too.
1976          */
1977         if (!eof && gotp->br_startoff != NULLFILEOFF) {
1978                 if ((delay && gotp->br_startblock == HOLESTARTBLOCK) ||
1979                     (!delay && gotp->br_startblock == DELAYSTARTBLOCK))
1980                         nexto = gotp->br_startoff + gotp->br_blockcount;
1981                 else
1982                         nexto = gotp->br_startoff;
1983         } else
1984                 nexto = NULLFILEOFF;
1985         if (!eof &&
1986             align_off + align_alen != orig_end &&
1987             align_off + align_alen > nexto)
1988                 align_off = nexto > align_alen ? nexto - align_alen : 0;
1989         /*
1990          * If we're now overlapping the next or previous extent that
1991          * means we can't fit an extsz piece in this hole.  Just move
1992          * the start forward to the first valid spot and set
1993          * the length so we hit the end.
1994          */
1995         if (align_off != orig_off && align_off < prevo)
1996                 align_off = prevo;
1997         if (align_off + align_alen != orig_end &&
1998             align_off + align_alen > nexto &&
1999             nexto != NULLFILEOFF) {
2000                 ASSERT(nexto > prevo);
2001                 align_alen = nexto - align_off;
2002         }
2003
2004         /*
2005          * If realtime, and the result isn't a multiple of the realtime
2006          * extent size we need to remove blocks until it is.
2007          */
2008         if (rt && (temp = (align_alen % mp->m_sb.sb_rextsize))) {
2009                 /*
2010                  * We're not covering the original request, or
2011                  * we won't be able to once we fix the length.
2012                  */
2013                 if (orig_off < align_off ||
2014                     orig_end > align_off + align_alen ||
2015                     align_alen - temp < orig_alen)
2016                         return XFS_ERROR(EINVAL);
2017                 /*
2018                  * Try to fix it by moving the start up.
2019                  */
2020                 if (align_off + temp <= orig_off) {
2021                         align_alen -= temp;
2022                         align_off += temp;
2023                 }
2024                 /*
2025                  * Try to fix it by moving the end in.
2026                  */
2027                 else if (align_off + align_alen - temp >= orig_end)
2028                         align_alen -= temp;
2029                 /*
2030                  * Set the start to the minimum then trim the length.
2031                  */
2032                 else {
2033                         align_alen -= orig_off - align_off;
2034                         align_off = orig_off;
2035                         align_alen -= align_alen % mp->m_sb.sb_rextsize;
2036                 }
2037                 /*
2038                  * Result doesn't cover the request, fail it.
2039                  */
2040                 if (orig_off < align_off || orig_end > align_off + align_alen)
2041                         return XFS_ERROR(EINVAL);
2042         } else {
2043                 ASSERT(orig_off >= align_off);
2044                 ASSERT(orig_end <= align_off + align_alen);
2045         }
2046
2047 #ifdef DEBUG
2048         if (!eof && gotp->br_startoff != NULLFILEOFF)
2049                 ASSERT(align_off + align_alen <= gotp->br_startoff);
2050         if (prevp->br_startoff != NULLFILEOFF)
2051                 ASSERT(align_off >= prevp->br_startoff + prevp->br_blockcount);
2052 #endif
2053
2054         *lenp = align_alen;
2055         *offp = align_off;
2056         return 0;
2057 }
2058
2059 #define XFS_ALLOC_GAP_UNITS     4
2060
2061 STATIC void
2062 xfs_bmap_adjacent(
2063         xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
2064 {
2065         xfs_fsblock_t   adjust;         /* adjustment to block numbers */
2066         xfs_agnumber_t  fb_agno;        /* ag number of ap->firstblock */
2067         xfs_mount_t     *mp;            /* mount point structure */
2068         int             nullfb;         /* true if ap->firstblock isn't set */
2069         int             rt;             /* true if inode is realtime */
2070
2071 #define ISVALID(x,y)    \
2072         (rt ? \
2073                 (x) < mp->m_sb.sb_rblocks : \
2074                 XFS_FSB_TO_AGNO(mp, x) == XFS_FSB_TO_AGNO(mp, y) && \
2075                 XFS_FSB_TO_AGNO(mp, x) < mp->m_sb.sb_agcount && \
2076                 XFS_FSB_TO_AGBNO(mp, x) < mp->m_sb.sb_agblocks)
2077
2078         mp = ap->ip->i_mount;
2079         nullfb = *ap->firstblock == NULLFSBLOCK;
2080         rt = XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata;
2081         fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, *ap->firstblock);
2082         /*
2083          * If allocating at eof, and there's a previous real block,
2084          * try to use its last block as our starting point.
2085          */
2086         if (ap->eof && ap->prev.br_startoff != NULLFILEOFF &&
2087             !isnullstartblock(ap->prev.br_startblock) &&
2088             ISVALID(ap->prev.br_startblock + ap->prev.br_blockcount,
2089                     ap->prev.br_startblock)) {
2090                 ap->blkno = ap->prev.br_startblock + ap->prev.br_blockcount;
2091                 /*
2092                  * Adjust for the gap between prevp and us.
2093                  */
2094                 adjust = ap->offset -
2095                         (ap->prev.br_startoff + ap->prev.br_blockcount);
2096                 if (adjust &&
2097                     ISVALID(ap->blkno + adjust, ap->prev.br_startblock))
2098                         ap->blkno += adjust;
2099         }
2100         /*
2101          * If not at eof, then compare the two neighbor blocks.
2102          * Figure out whether either one gives us a good starting point,
2103          * and pick the better one.
2104          */
2105         else if (!ap->eof) {
2106                 xfs_fsblock_t   gotbno;         /* right side block number */
2107                 xfs_fsblock_t   gotdiff=0;      /* right side difference */
2108                 xfs_fsblock_t   prevbno;        /* left side block number */
2109                 xfs_fsblock_t   prevdiff=0;     /* left side difference */
2110
2111                 /*
2112                  * If there's a previous (left) block, select a requested
2113                  * start block based on it.
2114                  */
2115                 if (ap->prev.br_startoff != NULLFILEOFF &&
2116                     !isnullstartblock(ap->prev.br_startblock) &&
2117                     (prevbno = ap->prev.br_startblock +
2118                                ap->prev.br_blockcount) &&
2119                     ISVALID(prevbno, ap->prev.br_startblock)) {
2120                         /*
2121                          * Calculate gap to end of previous block.
2122                          */
2123                         adjust = prevdiff = ap->offset -
2124                                 (ap->prev.br_startoff +
2125                                  ap->prev.br_blockcount);
2126                         /*
2127                          * Figure the startblock based on the previous block's
2128                          * end and the gap size.
2129                          * Heuristic!
2130                          * If the gap is large relative to the piece we're
2131                          * allocating, or using it gives us an invalid block
2132                          * number, then just use the end of the previous block.
2133                          */
2134                         if (prevdiff <= XFS_ALLOC_GAP_UNITS * ap->length &&
2135                             ISVALID(prevbno + prevdiff,
2136                                     ap->prev.br_startblock))
2137                                 prevbno += adjust;
2138                         else
2139                                 prevdiff += adjust;
2140                         /*
2141                          * If the firstblock forbids it, can't use it,
2142                          * must use default.
2143                          */
2144                         if (!rt && !nullfb &&
2145                             XFS_FSB_TO_AGNO(mp, prevbno) != fb_agno)
2146                                 prevbno = NULLFSBLOCK;
2147                 }
2148                 /*
2149                  * No previous block or can't follow it, just default.
2150                  */
2151                 else
2152                         prevbno = NULLFSBLOCK;
2153                 /*
2154                  * If there's a following (right) block, select a requested
2155                  * start block based on it.
2156                  */
2157                 if (!isnullstartblock(ap->got.br_startblock)) {
2158                         /*
2159                          * Calculate gap to start of next block.
2160                          */
2161                         adjust = gotdiff = ap->got.br_startoff - ap->offset;
2162                         /*
2163                          * Figure the startblock based on the next block's
2164                          * start and the gap size.
2165                          */
2166                         gotbno = ap->got.br_startblock;
2167                         /*
2168                          * Heuristic!
2169                          * If the gap is large relative to the piece we're
2170                          * allocating, or using it gives us an invalid block
2171                          * number, then just use the start of the next block
2172                          * offset by our length.
2173                          */
2174                         if (gotdiff <= XFS_ALLOC_GAP_UNITS * ap->length &&
2175                             ISVALID(gotbno - gotdiff, gotbno))
2176                                 gotbno -= adjust;
2177                         else if (ISVALID(gotbno - ap->length, gotbno)) {
2178                                 gotbno -= ap->length;
2179                                 gotdiff += adjust - ap->length;
2180                         } else
2181                                 gotdiff += adjust;
2182                         /*
2183                          * If the firstblock forbids it, can't use it,
2184                          * must use default.
2185                          */
2186                         if (!rt && !nullfb &&
2187                             XFS_FSB_TO_AGNO(mp, gotbno) != fb_agno)
2188                                 gotbno = NULLFSBLOCK;
2189                 }
2190                 /*
2191                  * No next block, just default.
2192                  */
2193                 else
2194                         gotbno = NULLFSBLOCK;
2195                 /*
2196                  * If both valid, pick the better one, else the only good
2197                  * one, else ap->blkno is already set (to 0 or the inode block).
2198                  */
2199                 if (prevbno != NULLFSBLOCK && gotbno != NULLFSBLOCK)
2200                         ap->blkno = prevdiff <= gotdiff ? prevbno : gotbno;
2201                 else if (prevbno != NULLFSBLOCK)
2202                         ap->blkno = prevbno;
2203                 else if (gotbno != NULLFSBLOCK)
2204                         ap->blkno = gotbno;
2205         }
2206 #undef ISVALID
2207 }
2208
2209 STATIC int
2210 xfs_bmap_rtalloc(
2211         xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
2212 {
2213         xfs_alloctype_t atype = 0;      /* type for allocation routines */
2214         int             error;          /* error return value */
2215         xfs_mount_t     *mp;            /* mount point structure */
2216         xfs_extlen_t    prod = 0;       /* product factor for allocators */
2217         xfs_extlen_t    ralen = 0;      /* realtime allocation length */
2218         xfs_extlen_t    align;          /* minimum allocation alignment */
2219         xfs_rtblock_t   rtb;
2220
2221         mp = ap->ip->i_mount;
2222         align = xfs_get_extsz_hint(ap->ip);
2223         prod = align / mp->m_sb.sb_rextsize;
2224         error = xfs_bmap_extsize_align(mp, &ap->got, &ap->prev,
2225                                         align, 1, ap->eof, 0,
2226                                         ap->conv, &ap->offset, &ap->length);
2227         if (error)
2228                 return error;
2229         ASSERT(ap->length);
2230         ASSERT(ap->length % mp->m_sb.sb_rextsize == 0);
2231
2232         /*
2233          * If the offset & length are not perfectly aligned
2234          * then kill prod, it will just get us in trouble.
2235          */
2236         if (do_mod(ap->offset, align) || ap->length % align)
2237                 prod = 1;
2238         /*
2239          * Set ralen to be the actual requested length in rtextents.
2240          */
2241         ralen = ap->length / mp->m_sb.sb_rextsize;
2242         /*
2243          * If the old value was close enough to MAXEXTLEN that
2244          * we rounded up to it, cut it back so it's valid again.
2245          * Note that if it's a really large request (bigger than
2246          * MAXEXTLEN), we don't hear about that number, and can't
2247          * adjust the starting point to match it.
2248          */
2249         if (ralen * mp->m_sb.sb_rextsize >= MAXEXTLEN)
2250                 ralen = MAXEXTLEN / mp->m_sb.sb_rextsize;
2251
2252         /*
2253          * Lock out other modifications to the RT bitmap inode.
2254          */
2255         xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL);
2256         xfs_trans_ijoin(ap->tp, mp->m_rbmip, XFS_ILOCK_EXCL);
2257
2258         /*
2259          * If it's an allocation to an empty file at offset 0,
2260          * pick an extent that will space things out in the rt area.
2261          */
2262         if (ap->eof && ap->offset == 0) {
2263                 xfs_rtblock_t uninitialized_var(rtx); /* realtime extent no */
2264
2265                 error = xfs_rtpick_extent(mp, ap->tp, ralen, &rtx);
2266                 if (error)
2267                         return error;
2268                 ap->blkno = rtx * mp->m_sb.sb_rextsize;
2269         } else {
2270                 ap->blkno = 0;
2271         }
2272
2273         xfs_bmap_adjacent(ap);
2274
2275         /*
2276          * Realtime allocation, done through xfs_rtallocate_extent.
2277          */
2278         atype = ap->blkno == 0 ?  XFS_ALLOCTYPE_ANY_AG : XFS_ALLOCTYPE_NEAR_BNO;
2279         do_div(ap->blkno, mp->m_sb.sb_rextsize);
2280         rtb = ap->blkno;
2281         ap->length = ralen;
2282         if ((error = xfs_rtallocate_extent(ap->tp, ap->blkno, 1, ap->length,
2283                                 &ralen, atype, ap->wasdel, prod, &rtb)))
2284                 return error;
2285         if (rtb == NULLFSBLOCK && prod > 1 &&
2286             (error = xfs_rtallocate_extent(ap->tp, ap->blkno, 1,
2287                                            ap->length, &ralen, atype,
2288                                            ap->wasdel, 1, &rtb)))
2289                 return error;
2290         ap->blkno = rtb;
2291         if (ap->blkno != NULLFSBLOCK) {
2292                 ap->blkno *= mp->m_sb.sb_rextsize;
2293                 ralen *= mp->m_sb.sb_rextsize;
2294                 ap->length = ralen;
2295                 ap->ip->i_d.di_nblocks += ralen;
2296                 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
2297                 if (ap->wasdel)
2298                         ap->ip->i_delayed_blks -= ralen;
2299                 /*
2300                  * Adjust the disk quota also. This was reserved
2301                  * earlier.
2302                  */
2303                 xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
2304                         ap->wasdel ? XFS_TRANS_DQ_DELRTBCOUNT :
2305                                         XFS_TRANS_DQ_RTBCOUNT, (long) ralen);
2306         } else {
2307                 ap->length = 0;
2308         }
2309         return 0;
2310 }
2311
2312 STATIC int
2313 xfs_bmap_btalloc_nullfb(
2314         struct xfs_bmalloca     *ap,
2315         struct xfs_alloc_arg    *args,
2316         xfs_extlen_t            *blen)
2317 {
2318         struct xfs_mount        *mp = ap->ip->i_mount;
2319         struct xfs_perag        *pag;
2320         xfs_agnumber_t          ag, startag;
2321         int                     notinit = 0;
2322         int                     error;
2323
2324         if (ap->userdata && xfs_inode_is_filestream(ap->ip))
2325                 args->type = XFS_ALLOCTYPE_NEAR_BNO;
2326         else
2327                 args->type = XFS_ALLOCTYPE_START_BNO;
2328         args->total = ap->total;
2329
2330         /*
2331          * Search for an allocation group with a single extent large enough
2332          * for the request.  If one isn't found, then adjust the minimum
2333          * allocation size to the largest space found.
2334          */
2335         startag = ag = XFS_FSB_TO_AGNO(mp, args->fsbno);
2336         if (startag == NULLAGNUMBER)
2337                 startag = ag = 0;
2338
2339         pag = xfs_perag_get(mp, ag);
2340         while (*blen < args->maxlen) {
2341                 if (!pag->pagf_init) {
2342                         error = xfs_alloc_pagf_init(mp, args->tp, ag,
2343                                                     XFS_ALLOC_FLAG_TRYLOCK);
2344                         if (error) {
2345                                 xfs_perag_put(pag);
2346                                 return error;
2347                         }
2348                 }
2349
2350                 /*
2351                  * See xfs_alloc_fix_freelist...
2352                  */
2353                 if (pag->pagf_init) {
2354                         xfs_extlen_t    longest;
2355                         longest = xfs_alloc_longest_free_extent(mp, pag);
2356                         if (*blen < longest)
2357                                 *blen = longest;
2358                 } else
2359                         notinit = 1;
2360
2361                 if (xfs_inode_is_filestream(ap->ip)) {
2362                         if (*blen >= args->maxlen)
2363                                 break;
2364
2365                         if (ap->userdata) {
2366                                 /*
2367                                  * If startag is an invalid AG, we've
2368                                  * come here once before and
2369                                  * xfs_filestream_new_ag picked the
2370                                  * best currently available.
2371                                  *
2372                                  * Don't continue looping, since we
2373                                  * could loop forever.
2374                                  */
2375                                 if (startag == NULLAGNUMBER)
2376                                         break;
2377
2378                                 error = xfs_filestream_new_ag(ap, &ag);
2379                                 xfs_perag_put(pag);
2380                                 if (error)
2381                                         return error;
2382
2383                                 /* loop again to set 'blen'*/
2384                                 startag = NULLAGNUMBER;
2385                                 pag = xfs_perag_get(mp, ag);
2386                                 continue;
2387                         }
2388                 }
2389                 if (++ag == mp->m_sb.sb_agcount)
2390                         ag = 0;
2391                 if (ag == startag)
2392                         break;
2393                 xfs_perag_put(pag);
2394                 pag = xfs_perag_get(mp, ag);
2395         }
2396         xfs_perag_put(pag);
2397
2398         /*
2399          * Since the above loop did a BUF_TRYLOCK, it is
2400          * possible that there is space for this request.
2401          */
2402         if (notinit || *blen < ap->minlen)
2403                 args->minlen = ap->minlen;
2404         /*
2405          * If the best seen length is less than the request
2406          * length, use the best as the minimum.
2407          */
2408         else if (*blen < args->maxlen)
2409                 args->minlen = *blen;
2410         /*
2411          * Otherwise we've seen an extent as big as maxlen,
2412          * use that as the minimum.
2413          */
2414         else
2415                 args->minlen = args->maxlen;
2416
2417         /*
2418          * set the failure fallback case to look in the selected
2419          * AG as the stream may have moved.
2420          */
2421         if (xfs_inode_is_filestream(ap->ip))
2422                 ap->blkno = args->fsbno = XFS_AGB_TO_FSB(mp, ag, 0);
2423
2424         return 0;
2425 }
2426
2427 STATIC int
2428 xfs_bmap_btalloc(
2429         xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
2430 {
2431         xfs_mount_t     *mp;            /* mount point structure */
2432         xfs_alloctype_t atype = 0;      /* type for allocation routines */
2433         xfs_extlen_t    align;          /* minimum allocation alignment */
2434         xfs_agnumber_t  fb_agno;        /* ag number of ap->firstblock */
2435         xfs_agnumber_t  ag;
2436         xfs_alloc_arg_t args;
2437         xfs_extlen_t    blen;
2438         xfs_extlen_t    nextminlen = 0;
2439         int             nullfb;         /* true if ap->firstblock isn't set */
2440         int             isaligned;
2441         int             tryagain;
2442         int             error;
2443
2444         ASSERT(ap->length);
2445
2446         mp = ap->ip->i_mount;
2447         align = ap->userdata ? xfs_get_extsz_hint(ap->ip) : 0;
2448         if (unlikely(align)) {
2449                 error = xfs_bmap_extsize_align(mp, &ap->got, &ap->prev,
2450                                                 align, 0, ap->eof, 0, ap->conv,
2451                                                 &ap->offset, &ap->length);
2452                 ASSERT(!error);
2453                 ASSERT(ap->length);
2454         }
2455         nullfb = *ap->firstblock == NULLFSBLOCK;
2456         fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, *ap->firstblock);
2457         if (nullfb) {
2458                 if (ap->userdata && xfs_inode_is_filestream(ap->ip)) {
2459                         ag = xfs_filestream_lookup_ag(ap->ip);
2460                         ag = (ag != NULLAGNUMBER) ? ag : 0;
2461                         ap->blkno = XFS_AGB_TO_FSB(mp, ag, 0);
2462                 } else {
2463                         ap->blkno = XFS_INO_TO_FSB(mp, ap->ip->i_ino);
2464                 }
2465         } else
2466                 ap->blkno = *ap->firstblock;
2467
2468         xfs_bmap_adjacent(ap);
2469
2470         /*
2471          * If allowed, use ap->blkno; otherwise must use firstblock since
2472          * it's in the right allocation group.
2473          */
2474         if (nullfb || XFS_FSB_TO_AGNO(mp, ap->blkno) == fb_agno)
2475                 ;
2476         else
2477                 ap->blkno = *ap->firstblock;
2478         /*
2479          * Normal allocation, done through xfs_alloc_vextent.
2480          */
2481         tryagain = isaligned = 0;
2482         memset(&args, 0, sizeof(args));
2483         args.tp = ap->tp;
2484         args.mp = mp;
2485         args.fsbno = ap->blkno;
2486
2487         /* Trim the allocation back to the maximum an AG can fit. */
2488         args.maxlen = MIN(ap->length, XFS_ALLOC_AG_MAX_USABLE(mp));
2489         args.firstblock = *ap->firstblock;
2490         blen = 0;
2491         if (nullfb) {
2492                 error = xfs_bmap_btalloc_nullfb(ap, &args, &blen);
2493                 if (error)
2494                         return error;
2495         } else if (ap->flist->xbf_low) {
2496                 if (xfs_inode_is_filestream(ap->ip))
2497                         args.type = XFS_ALLOCTYPE_FIRST_AG;
2498                 else
2499                         args.type = XFS_ALLOCTYPE_START_BNO;
2500                 args.total = args.minlen = ap->minlen;
2501         } else {
2502                 args.type = XFS_ALLOCTYPE_NEAR_BNO;
2503                 args.total = ap->total;
2504                 args.minlen = ap->minlen;
2505         }
2506         /* apply extent size hints if obtained earlier */
2507         if (unlikely(align)) {
2508                 args.prod = align;
2509                 if ((args.mod = (xfs_extlen_t)do_mod(ap->offset, args.prod)))
2510                         args.mod = (xfs_extlen_t)(args.prod - args.mod);
2511         } else if (mp->m_sb.sb_blocksize >= PAGE_CACHE_SIZE) {
2512                 args.prod = 1;
2513                 args.mod = 0;
2514         } else {
2515                 args.prod = PAGE_CACHE_SIZE >> mp->m_sb.sb_blocklog;
2516                 if ((args.mod = (xfs_extlen_t)(do_mod(ap->offset, args.prod))))
2517                         args.mod = (xfs_extlen_t)(args.prod - args.mod);
2518         }
2519         /*
2520          * If we are not low on available data blocks, and the
2521          * underlying logical volume manager is a stripe, and
2522          * the file offset is zero then try to allocate data
2523          * blocks on stripe unit boundary.
2524          * NOTE: ap->aeof is only set if the allocation length
2525          * is >= the stripe unit and the allocation offset is
2526          * at the end of file.
2527          */
2528         if (!ap->flist->xbf_low && ap->aeof) {
2529                 if (!ap->offset) {
2530                         args.alignment = mp->m_dalign;
2531                         atype = args.type;
2532                         isaligned = 1;
2533                         /*
2534                          * Adjust for alignment
2535                          */
2536                         if (blen > args.alignment && blen <= args.maxlen)
2537                                 args.minlen = blen - args.alignment;
2538                         args.minalignslop = 0;
2539                 } else {
2540                         /*
2541                          * First try an exact bno allocation.
2542                          * If it fails then do a near or start bno
2543                          * allocation with alignment turned on.
2544                          */
2545                         atype = args.type;
2546                         tryagain = 1;
2547                         args.type = XFS_ALLOCTYPE_THIS_BNO;
2548                         args.alignment = 1;
2549                         /*
2550                          * Compute the minlen+alignment for the
2551                          * next case.  Set slop so that the value
2552                          * of minlen+alignment+slop doesn't go up
2553                          * between the calls.
2554                          */
2555                         if (blen > mp->m_dalign && blen <= args.maxlen)
2556                                 nextminlen = blen - mp->m_dalign;
2557                         else
2558                                 nextminlen = args.minlen;
2559                         if (nextminlen + mp->m_dalign > args.minlen + 1)
2560                                 args.minalignslop =
2561                                         nextminlen + mp->m_dalign -
2562                                         args.minlen - 1;
2563                         else
2564                                 args.minalignslop = 0;
2565                 }
2566         } else {
2567                 args.alignment = 1;
2568                 args.minalignslop = 0;
2569         }
2570         args.minleft = ap->minleft;
2571         args.wasdel = ap->wasdel;
2572         args.isfl = 0;
2573         args.userdata = ap->userdata;
2574         if ((error = xfs_alloc_vextent(&args)))
2575                 return error;
2576         if (tryagain && args.fsbno == NULLFSBLOCK) {
2577                 /*
2578                  * Exact allocation failed. Now try with alignment
2579                  * turned on.
2580                  */
2581                 args.type = atype;
2582                 args.fsbno = ap->blkno;
2583                 args.alignment = mp->m_dalign;
2584                 args.minlen = nextminlen;
2585                 args.minalignslop = 0;
2586                 isaligned = 1;
2587                 if ((error = xfs_alloc_vextent(&args)))
2588                         return error;
2589         }
2590         if (isaligned && args.fsbno == NULLFSBLOCK) {
2591                 /*
2592                  * allocation failed, so turn off alignment and
2593                  * try again.
2594                  */
2595                 args.type = atype;
2596                 args.fsbno = ap->blkno;
2597                 args.alignment = 0;
2598                 if ((error = xfs_alloc_vextent(&args)))
2599                         return error;
2600         }
2601         if (args.fsbno == NULLFSBLOCK && nullfb &&
2602             args.minlen > ap->minlen) {
2603                 args.minlen = ap->minlen;
2604                 args.type = XFS_ALLOCTYPE_START_BNO;
2605                 args.fsbno = ap->blkno;
2606                 if ((error = xfs_alloc_vextent(&args)))
2607                         return error;
2608         }
2609         if (args.fsbno == NULLFSBLOCK && nullfb) {
2610                 args.fsbno = 0;
2611                 args.type = XFS_ALLOCTYPE_FIRST_AG;
2612                 args.total = ap->minlen;
2613                 args.minleft = 0;
2614                 if ((error = xfs_alloc_vextent(&args)))
2615                         return error;
2616                 ap->flist->xbf_low = 1;
2617         }
2618         if (args.fsbno != NULLFSBLOCK) {
2619                 /*
2620                  * check the allocation happened at the same or higher AG than
2621                  * the first block that was allocated.
2622                  */
2623                 ASSERT(*ap->firstblock == NULLFSBLOCK ||
2624                        XFS_FSB_TO_AGNO(mp, *ap->firstblock) ==
2625                        XFS_FSB_TO_AGNO(mp, args.fsbno) ||
2626                        (ap->flist->xbf_low &&
2627                         XFS_FSB_TO_AGNO(mp, *ap->firstblock) <
2628                         XFS_FSB_TO_AGNO(mp, args.fsbno)));
2629
2630                 ap->blkno = args.fsbno;
2631                 if (*ap->firstblock == NULLFSBLOCK)
2632                         *ap->firstblock = args.fsbno;
2633                 ASSERT(nullfb || fb_agno == args.agno ||
2634                        (ap->flist->xbf_low && fb_agno < args.agno));
2635                 ap->length = args.len;
2636                 ap->ip->i_d.di_nblocks += args.len;
2637                 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
2638                 if (ap->wasdel)
2639                         ap->ip->i_delayed_blks -= args.len;
2640                 /*
2641                  * Adjust the disk quota also. This was reserved
2642                  * earlier.
2643                  */
2644                 xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
2645                         ap->wasdel ? XFS_TRANS_DQ_DELBCOUNT :
2646                                         XFS_TRANS_DQ_BCOUNT,
2647                         (long) args.len);
2648         } else {
2649                 ap->blkno = NULLFSBLOCK;
2650                 ap->length = 0;
2651         }
2652         return 0;
2653 }
2654
2655 /*
2656  * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
2657  * It figures out where to ask the underlying allocator to put the new extent.
2658  */
2659 STATIC int
2660 xfs_bmap_alloc(
2661         xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
2662 {
2663         if (XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata)
2664                 return xfs_bmap_rtalloc(ap);
2665         return xfs_bmap_btalloc(ap);
2666 }
2667
2668 /*
2669  * Transform a btree format file with only one leaf node, where the
2670  * extents list will fit in the inode, into an extents format file.
2671  * Since the file extents are already in-core, all we have to do is
2672  * give up the space for the btree root and pitch the leaf block.
2673  */
2674 STATIC int                              /* error */
2675 xfs_bmap_btree_to_extents(
2676         xfs_trans_t             *tp,    /* transaction pointer */
2677         xfs_inode_t             *ip,    /* incore inode pointer */
2678         xfs_btree_cur_t         *cur,   /* btree cursor */
2679         int                     *logflagsp, /* inode logging flags */
2680         int                     whichfork)  /* data or attr fork */
2681 {
2682         /* REFERENCED */
2683         struct xfs_btree_block  *cblock;/* child btree block */
2684         xfs_fsblock_t           cbno;   /* child block number */
2685         xfs_buf_t               *cbp;   /* child block's buffer */
2686         int                     error;  /* error return value */
2687         xfs_ifork_t             *ifp;   /* inode fork data */
2688         xfs_mount_t             *mp;    /* mount point structure */
2689         __be64                  *pp;    /* ptr to block address */
2690         struct xfs_btree_block  *rblock;/* root btree block */
2691
2692         mp = ip->i_mount;
2693         ifp = XFS_IFORK_PTR(ip, whichfork);
2694         ASSERT(ifp->if_flags & XFS_IFEXTENTS);
2695         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
2696         rblock = ifp->if_broot;
2697         ASSERT(be16_to_cpu(rblock->bb_level) == 1);
2698         ASSERT(be16_to_cpu(rblock->bb_numrecs) == 1);
2699         ASSERT(xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0) == 1);
2700         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, rblock, 1, ifp->if_broot_bytes);
2701         cbno = be64_to_cpu(*pp);
2702         *logflagsp = 0;
2703 #ifdef DEBUG
2704         if ((error = xfs_btree_check_lptr(cur, cbno, 1)))
2705                 return error;
2706 #endif
2707         error = xfs_btree_read_bufl(mp, tp, cbno, 0, &cbp, XFS_BMAP_BTREE_REF,
2708                                 &xfs_bmbt_buf_ops);
2709         if (error)
2710                 return error;
2711         cblock = XFS_BUF_TO_BLOCK(cbp);
2712         if ((error = xfs_btree_check_block(cur, cblock, 0, cbp)))
2713                 return error;
2714         xfs_bmap_add_free(cbno, 1, cur->bc_private.b.flist, mp);
2715         ip->i_d.di_nblocks--;
2716         xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, -1L);
2717         xfs_trans_binval(tp, cbp);
2718         if (cur->bc_bufs[0] == cbp)
2719                 cur->bc_bufs[0] = NULL;
2720         xfs_iroot_realloc(ip, -1, whichfork);
2721         ASSERT(ifp->if_broot == NULL);
2722         ASSERT((ifp->if_flags & XFS_IFBROOT) == 0);
2723         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
2724         *logflagsp = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
2725         return 0;
2726 }
2727
2728 /*
2729  * Called by xfs_bmapi to update file extent records and the btree
2730  * after removing space (or undoing a delayed allocation).
2731  */
2732 STATIC int                              /* error */
2733 xfs_bmap_del_extent(
2734         xfs_inode_t             *ip,    /* incore inode pointer */
2735         xfs_trans_t             *tp,    /* current transaction pointer */
2736         xfs_extnum_t            *idx,   /* extent number to update/delete */
2737         xfs_bmap_free_t         *flist, /* list of extents to be freed */
2738         xfs_btree_cur_t         *cur,   /* if null, not a btree */
2739         xfs_bmbt_irec_t         *del,   /* data to remove from extents */
2740         int                     *logflagsp, /* inode logging flags */
2741         int                     whichfork) /* data or attr fork */
2742 {
2743         xfs_filblks_t           da_new; /* new delay-alloc indirect blocks */
2744         xfs_filblks_t           da_old; /* old delay-alloc indirect blocks */
2745         xfs_fsblock_t           del_endblock=0; /* first block past del */
2746         xfs_fileoff_t           del_endoff;     /* first offset past del */
2747         int                     delay;  /* current block is delayed allocated */
2748         int                     do_fx;  /* free extent at end of routine */
2749         xfs_bmbt_rec_host_t     *ep;    /* current extent entry pointer */
2750         int                     error;  /* error return value */
2751         int                     flags;  /* inode logging flags */
2752         xfs_bmbt_irec_t         got;    /* current extent entry */
2753         xfs_fileoff_t           got_endoff;     /* first offset past got */
2754         int                     i;      /* temp state */
2755         xfs_ifork_t             *ifp;   /* inode fork pointer */
2756         xfs_mount_t             *mp;    /* mount structure */
2757         xfs_filblks_t           nblks;  /* quota/sb block count */
2758         xfs_bmbt_irec_t         new;    /* new record to be inserted */
2759         /* REFERENCED */
2760         uint                    qfield; /* quota field to update */
2761         xfs_filblks_t           temp;   /* for indirect length calculations */
2762         xfs_filblks_t           temp2;  /* for indirect length calculations */
2763         int                     state = 0;
2764
2765         XFS_STATS_INC(xs_del_exlist);
2766
2767         if (whichfork == XFS_ATTR_FORK)
2768                 state |= BMAP_ATTRFORK;
2769
2770         mp = ip->i_mount;
2771         ifp = XFS_IFORK_PTR(ip, whichfork);
2772         ASSERT((*idx >= 0) && (*idx < ifp->if_bytes /
2773                 (uint)sizeof(xfs_bmbt_rec_t)));
2774         ASSERT(del->br_blockcount > 0);
2775         ep = xfs_iext_get_ext(ifp, *idx);
2776         xfs_bmbt_get_all(ep, &got);
2777         ASSERT(got.br_startoff <= del->br_startoff);
2778         del_endoff = del->br_startoff + del->br_blockcount;
2779         got_endoff = got.br_startoff + got.br_blockcount;
2780         ASSERT(got_endoff >= del_endoff);
2781         delay = isnullstartblock(got.br_startblock);
2782         ASSERT(isnullstartblock(del->br_startblock) == delay);
2783         flags = 0;
2784         qfield = 0;
2785         error = 0;
2786         /*
2787          * If deleting a real allocation, must free up the disk space.
2788          */
2789         if (!delay) {
2790                 flags = XFS_ILOG_CORE;
2791                 /*
2792                  * Realtime allocation.  Free it and record di_nblocks update.
2793                  */
2794                 if (whichfork == XFS_DATA_FORK && XFS_IS_REALTIME_INODE(ip)) {
2795                         xfs_fsblock_t   bno;
2796                         xfs_filblks_t   len;
2797
2798                         ASSERT(do_mod(del->br_blockcount,
2799                                       mp->m_sb.sb_rextsize) == 0);
2800                         ASSERT(do_mod(del->br_startblock,
2801                                       mp->m_sb.sb_rextsize) == 0);
2802                         bno = del->br_startblock;
2803                         len = del->br_blockcount;
2804                         do_div(bno, mp->m_sb.sb_rextsize);
2805                         do_div(len, mp->m_sb.sb_rextsize);
2806                         error = xfs_rtfree_extent(tp, bno, (xfs_extlen_t)len);
2807                         if (error)
2808                                 goto done;
2809                         do_fx = 0;
2810                         nblks = len * mp->m_sb.sb_rextsize;
2811                         qfield = XFS_TRANS_DQ_RTBCOUNT;
2812                 }
2813                 /*
2814                  * Ordinary allocation.
2815                  */
2816                 else {
2817                         do_fx = 1;
2818                         nblks = del->br_blockcount;
2819                         qfield = XFS_TRANS_DQ_BCOUNT;
2820                 }
2821                 /*
2822                  * Set up del_endblock and cur for later.
2823                  */
2824                 del_endblock = del->br_startblock + del->br_blockcount;
2825                 if (cur) {
2826                         if ((error = xfs_bmbt_lookup_eq(cur, got.br_startoff,
2827                                         got.br_startblock, got.br_blockcount,
2828                                         &i)))
2829                                 goto done;
2830                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2831                 }
2832                 da_old = da_new = 0;
2833         } else {
2834                 da_old = startblockval(got.br_startblock);
2835                 da_new = 0;
2836                 nblks = 0;
2837                 do_fx = 0;
2838         }
2839         /*
2840          * Set flag value to use in switch statement.
2841          * Left-contig is 2, right-contig is 1.
2842          */
2843         switch (((got.br_startoff == del->br_startoff) << 1) |
2844                 (got_endoff == del_endoff)) {
2845         case 3:
2846                 /*
2847                  * Matches the whole extent.  Delete the entry.
2848                  */
2849                 xfs_iext_remove(ip, *idx, 1,
2850                                 whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0);
2851                 --*idx;
2852                 if (delay)
2853                         break;
2854
2855                 XFS_IFORK_NEXT_SET(ip, whichfork,
2856                         XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
2857                 flags |= XFS_ILOG_CORE;
2858                 if (!cur) {
2859                         flags |= xfs_ilog_fext(whichfork);
2860                         break;
2861                 }
2862                 if ((error = xfs_btree_delete(cur, &i)))
2863                         goto done;
2864                 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2865                 break;
2866
2867         case 2:
2868                 /*
2869                  * Deleting the first part of the extent.
2870                  */
2871                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2872                 xfs_bmbt_set_startoff(ep, del_endoff);
2873                 temp = got.br_blockcount - del->br_blockcount;
2874                 xfs_bmbt_set_blockcount(ep, temp);
2875                 if (delay) {
2876                         temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2877                                 da_old);
2878                         xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
2879                         trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2880                         da_new = temp;
2881                         break;
2882                 }
2883                 xfs_bmbt_set_startblock(ep, del_endblock);
2884                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2885                 if (!cur) {
2886                         flags |= xfs_ilog_fext(whichfork);
2887                         break;
2888                 }
2889                 if ((error = xfs_bmbt_update(cur, del_endoff, del_endblock,
2890                                 got.br_blockcount - del->br_blockcount,
2891                                 got.br_state)))
2892                         goto done;
2893                 break;
2894
2895         case 1:
2896                 /*
2897                  * Deleting the last part of the extent.
2898                  */
2899                 temp = got.br_blockcount - del->br_blockcount;
2900                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2901                 xfs_bmbt_set_blockcount(ep, temp);
2902                 if (delay) {
2903                         temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2904                                 da_old);
2905                         xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
2906                         trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2907                         da_new = temp;
2908                         break;
2909                 }
2910                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2911                 if (!cur) {
2912                         flags |= xfs_ilog_fext(whichfork);
2913                         break;
2914                 }
2915                 if ((error = xfs_bmbt_update(cur, got.br_startoff,
2916                                 got.br_startblock,
2917                                 got.br_blockcount - del->br_blockcount,
2918                                 got.br_state)))
2919                         goto done;
2920                 break;
2921
2922         case 0:
2923                 /*
2924                  * Deleting the middle of the extent.
2925                  */
2926                 temp = del->br_startoff - got.br_startoff;
2927                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2928                 xfs_bmbt_set_blockcount(ep, temp);
2929                 new.br_startoff = del_endoff;
2930                 temp2 = got_endoff - del_endoff;
2931                 new.br_blockcount = temp2;
2932                 new.br_state = got.br_state;
2933                 if (!delay) {
2934                         new.br_startblock = del_endblock;
2935                         flags |= XFS_ILOG_CORE;
2936                         if (cur) {
2937                                 if ((error = xfs_bmbt_update(cur,
2938                                                 got.br_startoff,
2939                                                 got.br_startblock, temp,
2940                                                 got.br_state)))
2941                                         goto done;
2942                                 if ((error = xfs_btree_increment(cur, 0, &i)))
2943                                         goto done;
2944                                 cur->bc_rec.b = new;
2945                                 error = xfs_btree_insert(cur, &i);
2946                                 if (error && error != ENOSPC)
2947                                         goto done;
2948                                 /*
2949                                  * If get no-space back from btree insert,
2950                                  * it tried a split, and we have a zero
2951                                  * block reservation.
2952                                  * Fix up our state and return the error.
2953                                  */
2954                                 if (error == ENOSPC) {
2955                                         /*
2956                                          * Reset the cursor, don't trust
2957                                          * it after any insert operation.
2958                                          */
2959                                         if ((error = xfs_bmbt_lookup_eq(cur,
2960                                                         got.br_startoff,
2961                                                         got.br_startblock,
2962                                                         temp, &i)))
2963                                                 goto done;
2964                                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2965                                         /*
2966                                          * Update the btree record back
2967                                          * to the original value.
2968                                          */
2969                                         if ((error = xfs_bmbt_update(cur,
2970                                                         got.br_startoff,
2971                                                         got.br_startblock,
2972                                                         got.br_blockcount,
2973                                                         got.br_state)))
2974                                                 goto done;
2975                                         /*
2976                                          * Reset the extent record back
2977                                          * to the original value.
2978                                          */
2979                                         xfs_bmbt_set_blockcount(ep,
2980                                                 got.br_blockcount);
2981                                         flags = 0;
2982                                         error = XFS_ERROR(ENOSPC);
2983                                         goto done;
2984                                 }
2985                                 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2986                         } else
2987                                 flags |= xfs_ilog_fext(whichfork);
2988                         XFS_IFORK_NEXT_SET(ip, whichfork,
2989                                 XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
2990                 } else {
2991                         ASSERT(whichfork == XFS_DATA_FORK);
2992                         temp = xfs_bmap_worst_indlen(ip, temp);
2993                         xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
2994                         temp2 = xfs_bmap_worst_indlen(ip, temp2);
2995                         new.br_startblock = nullstartblock((int)temp2);
2996                         da_new = temp + temp2;
2997                         while (da_new > da_old) {
2998                                 if (temp) {
2999                                         temp--;
3000                                         da_new--;
3001                                         xfs_bmbt_set_startblock(ep,
3002                                                 nullstartblock((int)temp));
3003                                 }
3004                                 if (da_new == da_old)
3005                                         break;
3006                                 if (temp2) {
3007                                         temp2--;
3008                                         da_new--;
3009                                         new.br_startblock =
3010                                                 nullstartblock((int)temp2);
3011                                 }
3012                         }
3013                 }
3014                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
3015                 xfs_iext_insert(ip, *idx + 1, 1, &new, state);
3016                 ++*idx;
3017                 break;
3018         }
3019         /*
3020          * If we need to, add to list of extents to delete.
3021          */
3022         if (do_fx)
3023                 xfs_bmap_add_free(del->br_startblock, del->br_blockcount, flist,
3024                         mp);
3025         /*
3026          * Adjust inode # blocks in the file.
3027          */
3028         if (nblks)
3029                 ip->i_d.di_nblocks -= nblks;
3030         /*
3031          * Adjust quota data.
3032          */
3033         if (qfield)
3034                 xfs_trans_mod_dquot_byino(tp, ip, qfield, (long)-nblks);
3035
3036         /*
3037          * Account for change in delayed indirect blocks.
3038          * Nothing to do for disk quota accounting here.
3039          */
3040         ASSERT(da_old >= da_new);
3041         if (da_old > da_new) {
3042                 xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
3043                         (int64_t)(da_old - da_new), 0);
3044         }
3045 done:
3046         *logflagsp = flags;
3047         return error;
3048 }
3049
3050 /*
3051  * Remove the entry "free" from the free item list.  Prev points to the
3052  * previous entry, unless "free" is the head of the list.
3053  */
3054 STATIC void
3055 xfs_bmap_del_free(
3056         xfs_bmap_free_t         *flist, /* free item list header */
3057         xfs_bmap_free_item_t    *prev,  /* previous item on list, if any */
3058         xfs_bmap_free_item_t    *free)  /* list item to be freed */
3059 {
3060         if (prev)
3061                 prev->xbfi_next = free->xbfi_next;
3062         else
3063                 flist->xbf_first = free->xbfi_next;
3064         flist->xbf_count--;
3065         kmem_zone_free(xfs_bmap_free_item_zone, free);
3066 }
3067
3068 /*
3069  * Convert an extents-format file into a btree-format file.
3070  * The new file will have a root block (in the inode) and a single child block.
3071  */
3072 STATIC int                                      /* error */
3073 xfs_bmap_extents_to_btree(
3074         xfs_trans_t             *tp,            /* transaction pointer */
3075         xfs_inode_t             *ip,            /* incore inode pointer */
3076         xfs_fsblock_t           *firstblock,    /* first-block-allocated */
3077         xfs_bmap_free_t         *flist,         /* blocks freed in xaction */
3078         xfs_btree_cur_t         **curp,         /* cursor returned to caller */
3079         int                     wasdel,         /* converting a delayed alloc */
3080         int                     *logflagsp,     /* inode logging flags */
3081         int                     whichfork)      /* data or attr fork */
3082 {
3083         struct xfs_btree_block  *ablock;        /* allocated (child) bt block */
3084         xfs_buf_t               *abp;           /* buffer for ablock */
3085         xfs_alloc_arg_t         args;           /* allocation arguments */
3086         xfs_bmbt_rec_t          *arp;           /* child record pointer */
3087         struct xfs_btree_block  *block;         /* btree root block */
3088         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
3089         xfs_bmbt_rec_host_t     *ep;            /* extent record pointer */
3090         int                     error;          /* error return value */
3091         xfs_extnum_t            i, cnt;         /* extent record index */
3092         xfs_ifork_t             *ifp;           /* inode fork pointer */
3093         xfs_bmbt_key_t          *kp;            /* root block key pointer */
3094         xfs_mount_t             *mp;            /* mount structure */
3095         xfs_extnum_t            nextents;       /* number of file extents */
3096         xfs_bmbt_ptr_t          *pp;            /* root block address pointer */
3097
3098         ifp = XFS_IFORK_PTR(ip, whichfork);
3099         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS);
3100
3101         /*
3102          * Make space in the inode incore.
3103          */
3104         xfs_iroot_realloc(ip, 1, whichfork);
3105         ifp->if_flags |= XFS_IFBROOT;
3106
3107         /*
3108          * Fill in the root.
3109          */
3110         block = ifp->if_broot;
3111         block->bb_magic = cpu_to_be32(XFS_BMAP_MAGIC);
3112         block->bb_level = cpu_to_be16(1);
3113         block->bb_numrecs = cpu_to_be16(1);
3114         block->bb_u.l.bb_leftsib = cpu_to_be64(NULLDFSBNO);
3115         block->bb_u.l.bb_rightsib = cpu_to_be64(NULLDFSBNO);
3116
3117         /*
3118          * Need a cursor.  Can't allocate until bb_level is filled in.
3119          */
3120         mp = ip->i_mount;
3121         cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
3122         cur->bc_private.b.firstblock = *firstblock;
3123         cur->bc_private.b.flist = flist;
3124         cur->bc_private.b.flags = wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
3125         /*
3126          * Convert to a btree with two levels, one record in root.
3127          */
3128         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_BTREE);
3129         memset(&args, 0, sizeof(args));
3130         args.tp = tp;
3131         args.mp = mp;
3132         args.firstblock = *firstblock;
3133         if (*firstblock == NULLFSBLOCK) {
3134                 args.type = XFS_ALLOCTYPE_START_BNO;
3135                 args.fsbno = XFS_INO_TO_FSB(mp, ip->i_ino);
3136         } else if (flist->xbf_low) {
3137                 args.type = XFS_ALLOCTYPE_START_BNO;
3138                 args.fsbno = *firstblock;
3139         } else {
3140                 args.type = XFS_ALLOCTYPE_NEAR_BNO;
3141                 args.fsbno = *firstblock;
3142         }
3143         args.minlen = args.maxlen = args.prod = 1;
3144         args.wasdel = wasdel;
3145         *logflagsp = 0;
3146         if ((error = xfs_alloc_vextent(&args))) {
3147                 xfs_iroot_realloc(ip, -1, whichfork);
3148                 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
3149                 return error;
3150         }
3151         /*
3152          * Allocation can't fail, the space was reserved.
3153          */
3154         ASSERT(args.fsbno != NULLFSBLOCK);
3155         ASSERT(*firstblock == NULLFSBLOCK ||
3156                args.agno == XFS_FSB_TO_AGNO(mp, *firstblock) ||
3157                (flist->xbf_low &&
3158                 args.agno > XFS_FSB_TO_AGNO(mp, *firstblock)));
3159         *firstblock = cur->bc_private.b.firstblock = args.fsbno;
3160         cur->bc_private.b.allocated++;
3161         ip->i_d.di_nblocks++;
3162         xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, 1L);
3163         abp = xfs_btree_get_bufl(mp, tp, args.fsbno, 0);
3164         /*
3165          * Fill in the child block.
3166          */
3167         abp->b_ops = &xfs_bmbt_buf_ops;
3168         ablock = XFS_BUF_TO_BLOCK(abp);
3169         ablock->bb_magic = cpu_to_be32(XFS_BMAP_MAGIC);
3170         ablock->bb_level = 0;
3171         ablock->bb_u.l.bb_leftsib = cpu_to_be64(NULLDFSBNO);
3172         ablock->bb_u.l.bb_rightsib = cpu_to_be64(NULLDFSBNO);
3173         arp = XFS_BMBT_REC_ADDR(mp, ablock, 1);
3174         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3175         for (cnt = i = 0; i < nextents; i++) {
3176                 ep = xfs_iext_get_ext(ifp, i);
3177                 if (!isnullstartblock(xfs_bmbt_get_startblock(ep))) {
3178                         arp->l0 = cpu_to_be64(ep->l0);
3179                         arp->l1 = cpu_to_be64(ep->l1);
3180                         arp++; cnt++;
3181                 }
3182         }
3183         ASSERT(cnt == XFS_IFORK_NEXTENTS(ip, whichfork));
3184         xfs_btree_set_numrecs(ablock, cnt);
3185
3186         /*
3187          * Fill in the root key and pointer.
3188          */
3189         kp = XFS_BMBT_KEY_ADDR(mp, block, 1);
3190         arp = XFS_BMBT_REC_ADDR(mp, ablock, 1);
3191         kp->br_startoff = cpu_to_be64(xfs_bmbt_disk_get_startoff(arp));
3192         pp = XFS_BMBT_PTR_ADDR(mp, block, 1, xfs_bmbt_get_maxrecs(cur,
3193                                                 be16_to_cpu(block->bb_level)));
3194         *pp = cpu_to_be64(args.fsbno);
3195
3196         /*
3197          * Do all this logging at the end so that
3198          * the root is at the right level.
3199          */
3200         xfs_btree_log_block(cur, abp, XFS_BB_ALL_BITS);
3201         xfs_btree_log_recs(cur, abp, 1, be16_to_cpu(ablock->bb_numrecs));
3202         ASSERT(*curp == NULL);
3203         *curp = cur;
3204         *logflagsp = XFS_ILOG_CORE | xfs_ilog_fbroot(whichfork);
3205         return 0;
3206 }
3207
3208 /*
3209  * Calculate the default attribute fork offset for newly created inodes.
3210  */
3211 uint
3212 xfs_default_attroffset(
3213         struct xfs_inode        *ip)
3214 {
3215         struct xfs_mount        *mp = ip->i_mount;
3216         uint                    offset;
3217
3218         if (mp->m_sb.sb_inodesize == 256) {
3219                 offset = XFS_LITINO(mp) -
3220                                 XFS_BMDR_SPACE_CALC(MINABTPTRS);
3221         } else {
3222                 offset = XFS_BMDR_SPACE_CALC(6 * MINABTPTRS);
3223         }
3224
3225         ASSERT(offset < XFS_LITINO(mp));
3226         return offset;
3227 }
3228
3229 /*
3230  * Helper routine to reset inode di_forkoff field when switching
3231  * attribute fork from local to extent format - we reset it where
3232  * possible to make space available for inline data fork extents.
3233  */
3234 STATIC void
3235 xfs_bmap_forkoff_reset(
3236         xfs_mount_t     *mp,
3237         xfs_inode_t     *ip,
3238         int             whichfork)
3239 {
3240         if (whichfork == XFS_ATTR_FORK &&
3241             ip->i_d.di_format != XFS_DINODE_FMT_DEV &&
3242             ip->i_d.di_format != XFS_DINODE_FMT_UUID &&
3243             ip->i_d.di_format != XFS_DINODE_FMT_BTREE) {
3244                 uint    dfl_forkoff = xfs_default_attroffset(ip) >> 3;
3245
3246                 if (dfl_forkoff > ip->i_d.di_forkoff)
3247                         ip->i_d.di_forkoff = dfl_forkoff;
3248         }
3249 }
3250
3251 /*
3252  * Convert a local file to an extents file.
3253  * This code is out of bounds for data forks of regular files,
3254  * since the file data needs to get logged so things will stay consistent.
3255  * (The bmap-level manipulations are ok, though).
3256  */
3257 STATIC int                              /* error */
3258 xfs_bmap_local_to_extents(
3259         xfs_trans_t     *tp,            /* transaction pointer */
3260         xfs_inode_t     *ip,            /* incore inode pointer */
3261         xfs_fsblock_t   *firstblock,    /* first block allocated in xaction */
3262         xfs_extlen_t    total,          /* total blocks needed by transaction */
3263         int             *logflagsp,     /* inode logging flags */
3264         int             whichfork,
3265         void            (*init_fn)(struct xfs_buf *bp,
3266                                    struct xfs_inode *ip,
3267                                    struct xfs_ifork *ifp))
3268 {
3269         int             error;          /* error return value */
3270         int             flags;          /* logging flags returned */
3271         xfs_ifork_t     *ifp;           /* inode fork pointer */
3272
3273         /*
3274          * We don't want to deal with the case of keeping inode data inline yet.
3275          * So sending the data fork of a regular inode is invalid.
3276          */
3277         ASSERT(!(S_ISREG(ip->i_d.di_mode) && whichfork == XFS_DATA_FORK));
3278         ifp = XFS_IFORK_PTR(ip, whichfork);
3279         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
3280         flags = 0;
3281         error = 0;
3282         if (ifp->if_bytes) {
3283                 xfs_alloc_arg_t args;   /* allocation arguments */
3284                 xfs_buf_t       *bp;    /* buffer for extent block */
3285                 xfs_bmbt_rec_host_t *ep;/* extent record pointer */
3286
3287                 ASSERT((ifp->if_flags &
3288                         (XFS_IFINLINE|XFS_IFEXTENTS|XFS_IFEXTIREC)) == XFS_IFINLINE);
3289                 memset(&args, 0, sizeof(args));
3290                 args.tp = tp;
3291                 args.mp = ip->i_mount;
3292                 args.firstblock = *firstblock;
3293                 /*
3294                  * Allocate a block.  We know we need only one, since the
3295                  * file currently fits in an inode.
3296                  */
3297                 if (*firstblock == NULLFSBLOCK) {
3298                         args.fsbno = XFS_INO_TO_FSB(args.mp, ip->i_ino);
3299                         args.type = XFS_ALLOCTYPE_START_BNO;
3300                 } else {
3301                         args.fsbno = *firstblock;
3302                         args.type = XFS_ALLOCTYPE_NEAR_BNO;
3303                 }
3304                 args.total = total;
3305                 args.minlen = args.maxlen = args.prod = 1;
3306                 error = xfs_alloc_vextent(&args);
3307                 if (error)
3308                         goto done;
3309
3310                 /* Can't fail, the space was reserved. */
3311                 ASSERT(args.fsbno != NULLFSBLOCK);
3312                 ASSERT(args.len == 1);
3313                 *firstblock = args.fsbno;
3314                 bp = xfs_btree_get_bufl(args.mp, tp, args.fsbno, 0);
3315
3316                 /* initialise the block and copy the data */
3317                 init_fn(bp, ip, ifp);
3318
3319                 /* account for the change in fork size and log everything */
3320                 xfs_trans_log_buf(tp, bp, 0, ifp->if_bytes - 1);
3321                 xfs_bmap_forkoff_reset(args.mp, ip, whichfork);
3322                 xfs_idata_realloc(ip, -ifp->if_bytes, whichfork);
3323                 xfs_iext_add(ifp, 0, 1);
3324                 ep = xfs_iext_get_ext(ifp, 0);
3325                 xfs_bmbt_set_allf(ep, 0, args.fsbno, 1, XFS_EXT_NORM);
3326                 trace_xfs_bmap_post_update(ip, 0,
3327                                 whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0,
3328                                 _THIS_IP_);
3329                 XFS_IFORK_NEXT_SET(ip, whichfork, 1);
3330                 ip->i_d.di_nblocks = 1;
3331                 xfs_trans_mod_dquot_byino(tp, ip,
3332                         XFS_TRANS_DQ_BCOUNT, 1L);
3333                 flags |= xfs_ilog_fext(whichfork);
3334         } else {
3335                 ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) == 0);
3336                 xfs_bmap_forkoff_reset(ip->i_mount, ip, whichfork);
3337         }
3338         ifp->if_flags &= ~XFS_IFINLINE;
3339         ifp->if_flags |= XFS_IFEXTENTS;
3340         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
3341         flags |= XFS_ILOG_CORE;
3342 done:
3343         *logflagsp = flags;
3344         return error;
3345 }
3346
3347 /*
3348  * Search the extent records for the entry containing block bno.
3349  * If bno lies in a hole, point to the next entry.  If bno lies
3350  * past eof, *eofp will be set, and *prevp will contain the last
3351  * entry (null if none).  Else, *lastxp will be set to the index
3352  * of the found entry; *gotp will contain the entry.
3353  */
3354 STATIC xfs_bmbt_rec_host_t *            /* pointer to found extent entry */
3355 xfs_bmap_search_multi_extents(
3356         xfs_ifork_t     *ifp,           /* inode fork pointer */
3357         xfs_fileoff_t   bno,            /* block number searched for */
3358         int             *eofp,          /* out: end of file found */
3359         xfs_extnum_t    *lastxp,        /* out: last extent index */
3360         xfs_bmbt_irec_t *gotp,          /* out: extent entry found */
3361         xfs_bmbt_irec_t *prevp)         /* out: previous extent entry found */
3362 {
3363         xfs_bmbt_rec_host_t *ep;                /* extent record pointer */
3364         xfs_extnum_t    lastx;          /* last extent index */
3365
3366         /*
3367          * Initialize the extent entry structure to catch access to
3368          * uninitialized br_startblock field.
3369          */
3370         gotp->br_startoff = 0xffa5a5a5a5a5a5a5LL;
3371         gotp->br_blockcount = 0xa55a5a5a5a5a5a5aLL;
3372         gotp->br_state = XFS_EXT_INVALID;
3373 #if XFS_BIG_BLKNOS
3374         gotp->br_startblock = 0xffffa5a5a5a5a5a5LL;
3375 #else
3376         gotp->br_startblock = 0xffffa5a5;
3377 #endif
3378         prevp->br_startoff = NULLFILEOFF;
3379
3380         ep = xfs_iext_bno_to_ext(ifp, bno, &lastx);
3381         if (lastx > 0) {
3382                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx - 1), prevp);
3383         }
3384         if (lastx < (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t))) {
3385                 xfs_bmbt_get_all(ep, gotp);
3386                 *eofp = 0;
3387         } else {
3388                 if (lastx > 0) {
3389                         *gotp = *prevp;
3390                 }
3391                 *eofp = 1;
3392                 ep = NULL;
3393         }
3394         *lastxp = lastx;
3395         return ep;
3396 }
3397
3398 /*
3399  * Search the extents list for the inode, for the extent containing bno.
3400  * If bno lies in a hole, point to the next entry.  If bno lies past eof,
3401  * *eofp will be set, and *prevp will contain the last entry (null if none).
3402  * Else, *lastxp will be set to the index of the found
3403  * entry; *gotp will contain the entry.
3404  */
3405 STATIC xfs_bmbt_rec_host_t *                 /* pointer to found extent entry */
3406 xfs_bmap_search_extents(
3407         xfs_inode_t     *ip,            /* incore inode pointer */
3408         xfs_fileoff_t   bno,            /* block number searched for */
3409         int             fork,           /* data or attr fork */
3410         int             *eofp,          /* out: end of file found */
3411         xfs_extnum_t    *lastxp,        /* out: last extent index */
3412         xfs_bmbt_irec_t *gotp,          /* out: extent entry found */
3413         xfs_bmbt_irec_t *prevp)         /* out: previous extent entry found */
3414 {
3415         xfs_ifork_t     *ifp;           /* inode fork pointer */
3416         xfs_bmbt_rec_host_t  *ep;            /* extent record pointer */
3417
3418         XFS_STATS_INC(xs_look_exlist);
3419         ifp = XFS_IFORK_PTR(ip, fork);
3420
3421         ep = xfs_bmap_search_multi_extents(ifp, bno, eofp, lastxp, gotp, prevp);
3422
3423         if (unlikely(!(gotp->br_startblock) && (*lastxp != NULLEXTNUM) &&
3424                      !(XFS_IS_REALTIME_INODE(ip) && fork == XFS_DATA_FORK))) {
3425                 xfs_alert_tag(ip->i_mount, XFS_PTAG_FSBLOCK_ZERO,
3426                                 "Access to block zero in inode %llu "
3427                                 "start_block: %llx start_off: %llx "
3428                                 "blkcnt: %llx extent-state: %x lastx: %x\n",
3429                         (unsigned long long)ip->i_ino,
3430                         (unsigned long long)gotp->br_startblock,
3431                         (unsigned long long)gotp->br_startoff,
3432                         (unsigned long long)gotp->br_blockcount,
3433                         gotp->br_state, *lastxp);
3434                 *lastxp = NULLEXTNUM;
3435                 *eofp = 1;
3436                 return NULL;
3437         }
3438         return ep;
3439 }
3440
3441 /*
3442  * Compute the worst-case number of indirect blocks that will be used
3443  * for ip's delayed extent of length "len".
3444  */
3445 STATIC xfs_filblks_t
3446 xfs_bmap_worst_indlen(
3447         xfs_inode_t     *ip,            /* incore inode pointer */
3448         xfs_filblks_t   len)            /* delayed extent length */
3449 {
3450         int             level;          /* btree level number */
3451         int             maxrecs;        /* maximum record count at this level */
3452         xfs_mount_t     *mp;            /* mount structure */
3453         xfs_filblks_t   rval;           /* return value */
3454
3455         mp = ip->i_mount;
3456         maxrecs = mp->m_bmap_dmxr[0];
3457         for (level = 0, rval = 0;
3458              level < XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK);
3459              level++) {
3460                 len += maxrecs - 1;
3461                 do_div(len, maxrecs);
3462                 rval += len;
3463                 if (len == 1)
3464                         return rval + XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK) -
3465                                 level - 1;
3466                 if (level == 0)
3467                         maxrecs = mp->m_bmap_dmxr[1];
3468         }
3469         return rval;
3470 }
3471
3472 /*
3473  * Convert inode from non-attributed to attributed.
3474  * Must not be in a transaction, ip must not be locked.
3475  */
3476 int                                             /* error code */
3477 xfs_bmap_add_attrfork(
3478         xfs_inode_t             *ip,            /* incore inode pointer */
3479         int                     size,           /* space new attribute needs */
3480         int                     rsvd)           /* xact may use reserved blks */
3481 {
3482         xfs_fsblock_t           firstblock;     /* 1st block/ag allocated */
3483         xfs_bmap_free_t         flist;          /* freed extent records */
3484         xfs_mount_t             *mp;            /* mount structure */
3485         xfs_trans_t             *tp;            /* transaction pointer */
3486         int                     blks;           /* space reservation */
3487         int                     version = 1;    /* superblock attr version */
3488         int                     committed;      /* xaction was committed */
3489         int                     logflags;       /* logging flags */
3490         int                     error;          /* error return value */
3491
3492         ASSERT(XFS_IFORK_Q(ip) == 0);
3493
3494         mp = ip->i_mount;
3495         ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
3496         tp = xfs_trans_alloc(mp, XFS_TRANS_ADDAFORK);
3497         blks = XFS_ADDAFORK_SPACE_RES(mp);
3498         if (rsvd)
3499                 tp->t_flags |= XFS_TRANS_RESERVE;
3500         if ((error = xfs_trans_reserve(tp, blks, XFS_ADDAFORK_LOG_RES(mp), 0,
3501                         XFS_TRANS_PERM_LOG_RES, XFS_ADDAFORK_LOG_COUNT)))
3502                 goto error0;
3503         xfs_ilock(ip, XFS_ILOCK_EXCL);
3504         error = xfs_trans_reserve_quota_nblks(tp, ip, blks, 0, rsvd ?
3505                         XFS_QMOPT_RES_REGBLKS | XFS_QMOPT_FORCE_RES :
3506                         XFS_QMOPT_RES_REGBLKS);
3507         if (error) {
3508                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3509                 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES);
3510                 return error;
3511         }
3512         if (XFS_IFORK_Q(ip))
3513                 goto error1;
3514         if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS) {
3515                 /*
3516                  * For inodes coming from pre-6.2 filesystems.
3517                  */
3518                 ASSERT(ip->i_d.di_aformat == 0);
3519                 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
3520         }
3521         ASSERT(ip->i_d.di_anextents == 0);
3522
3523         xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
3524         xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
3525
3526         switch (ip->i_d.di_format) {
3527         case XFS_DINODE_FMT_DEV:
3528                 ip->i_d.di_forkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
3529                 break;
3530         case XFS_DINODE_FMT_UUID:
3531                 ip->i_d.di_forkoff = roundup(sizeof(uuid_t), 8) >> 3;
3532                 break;
3533         case XFS_DINODE_FMT_LOCAL:
3534         case XFS_DINODE_FMT_EXTENTS:
3535         case XFS_DINODE_FMT_BTREE:
3536                 ip->i_d.di_forkoff = xfs_attr_shortform_bytesfit(ip, size);
3537                 if (!ip->i_d.di_forkoff)
3538                         ip->i_d.di_forkoff = xfs_default_attroffset(ip) >> 3;
3539                 else if (mp->m_flags & XFS_MOUNT_ATTR2)
3540                         version = 2;
3541                 break;
3542         default:
3543                 ASSERT(0);
3544                 error = XFS_ERROR(EINVAL);
3545                 goto error1;
3546         }
3547
3548         ASSERT(ip->i_afp == NULL);
3549         ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP);
3550         ip->i_afp->if_flags = XFS_IFEXTENTS;
3551         logflags = 0;
3552         xfs_bmap_init(&flist, &firstblock);
3553         switch (ip->i_d.di_format) {
3554         case XFS_DINODE_FMT_LOCAL:
3555                 error = xfs_bmap_add_attrfork_local(tp, ip, &firstblock, &flist,
3556                         &logflags);
3557                 break;
3558         case XFS_DINODE_FMT_EXTENTS:
3559                 error = xfs_bmap_add_attrfork_extents(tp, ip, &firstblock,
3560                         &flist, &logflags);
3561                 break;
3562         case XFS_DINODE_FMT_BTREE:
3563                 error = xfs_bmap_add_attrfork_btree(tp, ip, &firstblock, &flist,
3564                         &logflags);
3565                 break;
3566         default:
3567                 error = 0;
3568                 break;
3569         }
3570         if (logflags)
3571                 xfs_trans_log_inode(tp, ip, logflags);
3572         if (error)
3573                 goto error2;
3574         if (!xfs_sb_version_hasattr(&mp->m_sb) ||
3575            (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2)) {
3576                 __int64_t sbfields = 0;
3577
3578                 spin_lock(&mp->m_sb_lock);
3579                 if (!xfs_sb_version_hasattr(&mp->m_sb)) {
3580                         xfs_sb_version_addattr(&mp->m_sb);
3581                         sbfields |= XFS_SB_VERSIONNUM;
3582                 }
3583                 if (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2) {
3584                         xfs_sb_version_addattr2(&mp->m_sb);
3585                         sbfields |= (XFS_SB_VERSIONNUM | XFS_SB_FEATURES2);
3586                 }
3587                 if (sbfields) {
3588                         spin_unlock(&mp->m_sb_lock);
3589                         xfs_mod_sb(tp, sbfields);
3590                 } else
3591                         spin_unlock(&mp->m_sb_lock);
3592         }
3593
3594         error = xfs_bmap_finish(&tp, &flist, &committed);
3595         if (error)
3596                 goto error2;
3597         return xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
3598 error2:
3599         xfs_bmap_cancel(&flist);
3600 error1:
3601         xfs_iunlock(ip, XFS_ILOCK_EXCL);
3602 error0:
3603         xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
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 bma->aeof if the file (fork) is empty as any new write will be
3950  * at, or past the EOF.
3951  */
3952 STATIC int
3953 xfs_bmap_isaeof(
3954         struct xfs_bmalloca     *bma,
3955         int                     whichfork)
3956 {
3957         struct xfs_bmbt_irec    rec;
3958         int                     is_empty;
3959         int                     error;
3960
3961         bma->aeof = 0;
3962         error = xfs_bmap_last_extent(NULL, bma->ip, whichfork, &rec,
3963                                      &is_empty);
3964         if (error || is_empty)
3965                 return error;
3966
3967         /*
3968          * Check if we are allocation or past the last extent, or at least into
3969          * the last delayed allocated extent.
3970          */
3971         bma->aeof = bma->offset >= rec.br_startoff + rec.br_blockcount ||
3972                 (bma->offset >= rec.br_startoff &&
3973                  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 XFS_ISIZE(ip) == ip->i_mount->m_sb.sb_blocksize;
4051 #endif  /* !DEBUG */
4052         if (XFS_IFORK_NEXTENTS(ip, whichfork) != 1)
4053                 return 0;
4054         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
4055                 return 0;
4056         ifp = XFS_IFORK_PTR(ip, whichfork);
4057         ASSERT(ifp->if_flags & XFS_IFEXTENTS);
4058         ep = xfs_iext_get_ext(ifp, 0);
4059         xfs_bmbt_get_all(ep, &s);
4060         rval = s.br_startoff == 0 && s.br_blockcount == 1;
4061         if (rval && whichfork == XFS_DATA_FORK)
4062                 ASSERT(XFS_ISIZE(ip) == ip->i_mount->m_sb.sb_blocksize);
4063         return rval;
4064 }
4065
4066 STATIC int
4067 xfs_bmap_sanity_check(
4068         struct xfs_mount        *mp,
4069         struct xfs_buf          *bp,
4070         int                     level)
4071 {
4072         struct xfs_btree_block  *block = XFS_BUF_TO_BLOCK(bp);
4073
4074         if (block->bb_magic != cpu_to_be32(XFS_BMAP_MAGIC) ||
4075             be16_to_cpu(block->bb_level) != level ||
4076             be16_to_cpu(block->bb_numrecs) == 0 ||
4077             be16_to_cpu(block->bb_numrecs) > mp->m_bmap_dmxr[level != 0])
4078                 return 0;
4079         return 1;
4080 }
4081
4082 /*
4083  * Read in the extents to if_extents.
4084  * All inode fields are set up by caller, we just traverse the btree
4085  * and copy the records in. If the file system cannot contain unwritten
4086  * extents, the records are checked for no "state" flags.
4087  */
4088 int                                     /* error */
4089 xfs_bmap_read_extents(
4090         xfs_trans_t             *tp,    /* transaction pointer */
4091         xfs_inode_t             *ip,    /* incore inode */
4092         int                     whichfork) /* data or attr fork */
4093 {
4094         struct xfs_btree_block  *block; /* current btree block */
4095         xfs_fsblock_t           bno;    /* block # of "block" */
4096         xfs_buf_t               *bp;    /* buffer for "block" */
4097         int                     error;  /* error return value */
4098         xfs_exntfmt_t           exntf;  /* XFS_EXTFMT_NOSTATE, if checking */
4099         xfs_extnum_t            i, j;   /* index into the extents list */
4100         xfs_ifork_t             *ifp;   /* fork structure */
4101         int                     level;  /* btree level, for checking */
4102         xfs_mount_t             *mp;    /* file system mount structure */
4103         __be64                  *pp;    /* pointer to block address */
4104         /* REFERENCED */
4105         xfs_extnum_t            room;   /* number of entries there's room for */
4106
4107         bno = NULLFSBLOCK;
4108         mp = ip->i_mount;
4109         ifp = XFS_IFORK_PTR(ip, whichfork);
4110         exntf = (whichfork != XFS_DATA_FORK) ? XFS_EXTFMT_NOSTATE :
4111                                         XFS_EXTFMT_INODE(ip);
4112         block = ifp->if_broot;
4113         /*
4114          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
4115          */
4116         level = be16_to_cpu(block->bb_level);
4117         ASSERT(level > 0);
4118         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
4119         bno = be64_to_cpu(*pp);
4120         ASSERT(bno != NULLDFSBNO);
4121         ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
4122         ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
4123         /*
4124          * Go down the tree until leaf level is reached, following the first
4125          * pointer (leftmost) at each level.
4126          */
4127         while (level-- > 0) {
4128                 error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
4129                                 XFS_BMAP_BTREE_REF, &xfs_bmbt_buf_ops);
4130                 if (error)
4131                         return error;
4132                 block = XFS_BUF_TO_BLOCK(bp);
4133                 XFS_WANT_CORRUPTED_GOTO(
4134                         xfs_bmap_sanity_check(mp, bp, level),
4135                         error0);
4136                 if (level == 0)
4137                         break;
4138                 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
4139                 bno = be64_to_cpu(*pp);
4140                 XFS_WANT_CORRUPTED_GOTO(XFS_FSB_SANITY_CHECK(mp, bno), error0);
4141                 xfs_trans_brelse(tp, bp);
4142         }
4143         /*
4144          * Here with bp and block set to the leftmost leaf node in the tree.
4145          */
4146         room = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4147         i = 0;
4148         /*
4149          * Loop over all leaf nodes.  Copy information to the extent records.
4150          */
4151         for (;;) {
4152                 xfs_bmbt_rec_t  *frp;
4153                 xfs_fsblock_t   nextbno;
4154                 xfs_extnum_t    num_recs;
4155                 xfs_extnum_t    start;
4156
4157                 num_recs = xfs_btree_get_numrecs(block);
4158                 if (unlikely(i + num_recs > room)) {
4159                         ASSERT(i + num_recs <= room);
4160                         xfs_warn(ip->i_mount,
4161                                 "corrupt dinode %Lu, (btree extents).",
4162                                 (unsigned long long) ip->i_ino);
4163                         XFS_CORRUPTION_ERROR("xfs_bmap_read_extents(1)",
4164                                 XFS_ERRLEVEL_LOW, ip->i_mount, block);
4165                         goto error0;
4166                 }
4167                 XFS_WANT_CORRUPTED_GOTO(
4168                         xfs_bmap_sanity_check(mp, bp, 0),
4169                         error0);
4170                 /*
4171                  * Read-ahead the next leaf block, if any.
4172                  */
4173                 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
4174                 if (nextbno != NULLFSBLOCK)
4175                         xfs_btree_reada_bufl(mp, nextbno, 1,
4176                                              &xfs_bmbt_buf_ops);
4177                 /*
4178                  * Copy records into the extent records.
4179                  */
4180                 frp = XFS_BMBT_REC_ADDR(mp, block, 1);
4181                 start = i;
4182                 for (j = 0; j < num_recs; j++, i++, frp++) {
4183                         xfs_bmbt_rec_host_t *trp = xfs_iext_get_ext(ifp, i);
4184                         trp->l0 = be64_to_cpu(frp->l0);
4185                         trp->l1 = be64_to_cpu(frp->l1);
4186                 }
4187                 if (exntf == XFS_EXTFMT_NOSTATE) {
4188                         /*
4189                          * Check all attribute bmap btree records and
4190                          * any "older" data bmap btree records for a
4191                          * set bit in the "extent flag" position.
4192                          */
4193                         if (unlikely(xfs_check_nostate_extents(ifp,
4194                                         start, num_recs))) {
4195                                 XFS_ERROR_REPORT("xfs_bmap_read_extents(2)",
4196                                                  XFS_ERRLEVEL_LOW,
4197                                                  ip->i_mount);
4198                                 goto error0;
4199                         }
4200                 }
4201                 xfs_trans_brelse(tp, bp);
4202                 bno = nextbno;
4203                 /*
4204                  * If we've reached the end, stop.
4205                  */
4206                 if (bno == NULLFSBLOCK)
4207                         break;
4208                 error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
4209                                 XFS_BMAP_BTREE_REF, &xfs_bmbt_buf_ops);
4210                 if (error)
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
4435         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4436                 error = xfs_iread_extents(NULL, ip, whichfork);
4437                 if (error)
4438                         return error;
4439         }
4440
4441         xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got, &prev);
4442         end = bno + len;
4443         obno = bno;
4444
4445         while (bno < end && n < *nmap) {
4446                 /* Reading past eof, act as though there's a hole up to end. */
4447                 if (eof)
4448                         got.br_startoff = end;
4449                 if (got.br_startoff > bno) {
4450                         /* Reading in a hole.  */
4451                         mval->br_startoff = bno;
4452                         mval->br_startblock = HOLESTARTBLOCK;
4453                         mval->br_blockcount =
4454                                 XFS_FILBLKS_MIN(len, got.br_startoff - bno);
4455                         mval->br_state = XFS_EXT_NORM;
4456                         bno += mval->br_blockcount;
4457                         len -= mval->br_blockcount;
4458                         mval++;
4459                         n++;
4460                         continue;
4461                 }
4462
4463                 /* set up the extent map to return. */
4464                 xfs_bmapi_trim_map(mval, &got, &bno, len, obno, end, n, flags);
4465                 xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4466
4467                 /* If we're done, stop now. */
4468                 if (bno >= end || n >= *nmap)
4469                         break;
4470
4471                 /* Else go on to the next record. */
4472                 if (++lastx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t))
4473                         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx), &got);
4474                 else
4475                         eof = 1;
4476         }
4477         *nmap = n;
4478         return 0;
4479 }
4480
4481 STATIC int
4482 xfs_bmapi_reserve_delalloc(
4483         struct xfs_inode        *ip,
4484         xfs_fileoff_t           aoff,
4485         xfs_filblks_t           len,
4486         struct xfs_bmbt_irec    *got,
4487         struct xfs_bmbt_irec    *prev,
4488         xfs_extnum_t            *lastx,
4489         int                     eof)
4490 {
4491         struct xfs_mount        *mp = ip->i_mount;
4492         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
4493         xfs_extlen_t            alen;
4494         xfs_extlen_t            indlen;
4495         char                    rt = XFS_IS_REALTIME_INODE(ip);
4496         xfs_extlen_t            extsz;
4497         int                     error;
4498
4499         alen = XFS_FILBLKS_MIN(len, MAXEXTLEN);
4500         if (!eof)
4501                 alen = XFS_FILBLKS_MIN(alen, got->br_startoff - aoff);
4502
4503         /* Figure out the extent size, adjust alen */
4504         extsz = xfs_get_extsz_hint(ip);
4505         if (extsz) {
4506                 /*
4507                  * Make sure we don't exceed a single extent length when we
4508                  * align the extent by reducing length we are going to
4509                  * allocate by the maximum amount extent size aligment may
4510                  * require.
4511                  */
4512                 alen = XFS_FILBLKS_MIN(len, MAXEXTLEN - (2 * extsz - 1));
4513                 error = xfs_bmap_extsize_align(mp, got, prev, extsz, rt, eof,
4514                                                1, 0, &aoff, &alen);
4515                 ASSERT(!error);
4516         }
4517
4518         if (rt)
4519                 extsz = alen / mp->m_sb.sb_rextsize;
4520
4521         /*
4522          * Make a transaction-less quota reservation for delayed allocation
4523          * blocks.  This number gets adjusted later.  We return if we haven't
4524          * allocated blocks already inside this loop.
4525          */
4526         error = xfs_trans_reserve_quota_nblks(NULL, ip, (long)alen, 0,
4527                         rt ? XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4528         if (error)
4529                 return error;
4530
4531         /*
4532          * Split changing sb for alen and indlen since they could be coming
4533          * from different places.
4534          */
4535         indlen = (xfs_extlen_t)xfs_bmap_worst_indlen(ip, alen);
4536         ASSERT(indlen > 0);
4537
4538         if (rt) {
4539                 error = xfs_mod_incore_sb(mp, XFS_SBS_FREXTENTS,
4540                                           -((int64_t)extsz), 0);
4541         } else {
4542                 error = xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
4543                                                  -((int64_t)alen), 0);
4544         }
4545
4546         if (error)
4547                 goto out_unreserve_quota;
4548
4549         error = xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
4550                                          -((int64_t)indlen), 0);
4551         if (error)
4552                 goto out_unreserve_blocks;
4553
4554
4555         ip->i_delayed_blks += alen;
4556
4557         got->br_startoff = aoff;
4558         got->br_startblock = nullstartblock(indlen);
4559         got->br_blockcount = alen;
4560         got->br_state = XFS_EXT_NORM;
4561         xfs_bmap_add_extent_hole_delay(ip, lastx, got);
4562
4563         /*
4564          * Update our extent pointer, given that xfs_bmap_add_extent_hole_delay
4565          * might have merged it into one of the neighbouring ones.
4566          */
4567         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *lastx), got);
4568
4569         ASSERT(got->br_startoff <= aoff);
4570         ASSERT(got->br_startoff + got->br_blockcount >= aoff + alen);
4571         ASSERT(isnullstartblock(got->br_startblock));
4572         ASSERT(got->br_state == XFS_EXT_NORM);
4573         return 0;
4574
4575 out_unreserve_blocks:
4576         if (rt)
4577                 xfs_mod_incore_sb(mp, XFS_SBS_FREXTENTS, extsz, 0);
4578         else
4579                 xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS, alen, 0);
4580 out_unreserve_quota:
4581         if (XFS_IS_QUOTA_ON(mp))
4582                 xfs_trans_unreserve_quota_nblks(NULL, ip, (long)alen, 0, rt ?
4583                                 XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4584         return error;
4585 }
4586
4587 /*
4588  * Map file blocks to filesystem blocks, adding delayed allocations as needed.
4589  */
4590 int
4591 xfs_bmapi_delay(
4592         struct xfs_inode        *ip,    /* incore inode */
4593         xfs_fileoff_t           bno,    /* starting file offs. mapped */
4594         xfs_filblks_t           len,    /* length to map in file */
4595         struct xfs_bmbt_irec    *mval,  /* output: map values */
4596         int                     *nmap,  /* i/o: mval size/count */
4597         int                     flags)  /* XFS_BMAPI_... */
4598 {
4599         struct xfs_mount        *mp = ip->i_mount;
4600         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
4601         struct xfs_bmbt_irec    got;    /* current file extent record */
4602         struct xfs_bmbt_irec    prev;   /* previous file extent record */
4603         xfs_fileoff_t           obno;   /* old block number (offset) */
4604         xfs_fileoff_t           end;    /* end of mapped file region */
4605         xfs_extnum_t            lastx;  /* last useful extent number */
4606         int                     eof;    /* we've hit the end of extents */
4607         int                     n = 0;  /* current extent index */
4608         int                     error = 0;
4609
4610         ASSERT(*nmap >= 1);
4611         ASSERT(*nmap <= XFS_BMAP_MAX_NMAP);
4612         ASSERT(!(flags & ~XFS_BMAPI_ENTIRE));
4613
4614         if (unlikely(XFS_TEST_ERROR(
4615             (XFS_IFORK_FORMAT(ip, XFS_DATA_FORK) != XFS_DINODE_FMT_EXTENTS &&
4616              XFS_IFORK_FORMAT(ip, XFS_DATA_FORK) != XFS_DINODE_FMT_BTREE),
4617              mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4618                 XFS_ERROR_REPORT("xfs_bmapi_delay", XFS_ERRLEVEL_LOW, mp);
4619                 return XFS_ERROR(EFSCORRUPTED);
4620         }
4621
4622         if (XFS_FORCED_SHUTDOWN(mp))
4623                 return XFS_ERROR(EIO);
4624
4625         XFS_STATS_INC(xs_blk_mapw);
4626
4627         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4628                 error = xfs_iread_extents(NULL, ip, XFS_DATA_FORK);
4629                 if (error)
4630                         return error;
4631         }
4632
4633         xfs_bmap_search_extents(ip, bno, XFS_DATA_FORK, &eof, &lastx, &got, &prev);
4634         end = bno + len;
4635         obno = bno;
4636
4637         while (bno < end && n < *nmap) {
4638                 if (eof || got.br_startoff > bno) {
4639                         error = xfs_bmapi_reserve_delalloc(ip, bno, len, &got,
4640                                                            &prev, &lastx, eof);
4641                         if (error) {
4642                                 if (n == 0) {
4643                                         *nmap = 0;
4644                                         return error;
4645                                 }
4646                                 break;
4647                         }
4648                 }
4649
4650                 /* set up the extent map to return. */
4651                 xfs_bmapi_trim_map(mval, &got, &bno, len, obno, end, n, flags);
4652                 xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4653
4654                 /* If we're done, stop now. */
4655                 if (bno >= end || n >= *nmap)
4656                         break;
4657
4658                 /* Else go on to the next record. */
4659                 prev = got;
4660                 if (++lastx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t))
4661                         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx), &got);
4662                 else
4663                         eof = 1;
4664         }
4665
4666         *nmap = n;
4667         return 0;
4668 }
4669
4670
4671 STATIC int
4672 __xfs_bmapi_allocate(
4673         struct xfs_bmalloca     *bma)
4674 {
4675         struct xfs_mount        *mp = bma->ip->i_mount;
4676         int                     whichfork = (bma->flags & XFS_BMAPI_ATTRFORK) ?
4677                                                 XFS_ATTR_FORK : XFS_DATA_FORK;
4678         struct xfs_ifork        *ifp = XFS_IFORK_PTR(bma->ip, whichfork);
4679         int                     tmp_logflags = 0;
4680         int                     error;
4681         int                     rt;
4682
4683         ASSERT(bma->length > 0);
4684
4685         rt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(bma->ip);
4686
4687         /*
4688          * For the wasdelay case, we could also just allocate the stuff asked
4689          * for in this bmap call but that wouldn't be as good.
4690          */
4691         if (bma->wasdel) {
4692                 bma->length = (xfs_extlen_t)bma->got.br_blockcount;
4693                 bma->offset = bma->got.br_startoff;
4694                 if (bma->idx != NULLEXTNUM && bma->idx) {
4695                         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx - 1),
4696                                          &bma->prev);
4697                 }
4698         } else {
4699                 bma->length = XFS_FILBLKS_MIN(bma->length, MAXEXTLEN);
4700                 if (!bma->eof)
4701                         bma->length = XFS_FILBLKS_MIN(bma->length,
4702                                         bma->got.br_startoff - bma->offset);
4703         }
4704
4705         /*
4706          * Indicate if this is the first user data in the file, or just any
4707          * user data.
4708          */
4709         if (!(bma->flags & XFS_BMAPI_METADATA)) {
4710                 bma->userdata = (bma->offset == 0) ?
4711                         XFS_ALLOC_INITIAL_USER_DATA : XFS_ALLOC_USERDATA;
4712         }
4713
4714         bma->minlen = (bma->flags & XFS_BMAPI_CONTIG) ? bma->length : 1;
4715
4716         /*
4717          * Only want to do the alignment at the eof if it is userdata and
4718          * allocation length is larger than a stripe unit.
4719          */
4720         if (mp->m_dalign && bma->length >= mp->m_dalign &&
4721             !(bma->flags & XFS_BMAPI_METADATA) && whichfork == XFS_DATA_FORK) {
4722                 error = xfs_bmap_isaeof(bma, whichfork);
4723                 if (error)
4724                         return error;
4725         }
4726
4727         error = xfs_bmap_alloc(bma);
4728         if (error)
4729                 return error;
4730
4731         if (bma->flist->xbf_low)
4732                 bma->minleft = 0;
4733         if (bma->cur)
4734                 bma->cur->bc_private.b.firstblock = *bma->firstblock;
4735         if (bma->blkno == NULLFSBLOCK)
4736                 return 0;
4737         if ((ifp->if_flags & XFS_IFBROOT) && !bma->cur) {
4738                 bma->cur = xfs_bmbt_init_cursor(mp, bma->tp, bma->ip, whichfork);
4739                 bma->cur->bc_private.b.firstblock = *bma->firstblock;
4740                 bma->cur->bc_private.b.flist = bma->flist;
4741         }
4742         /*
4743          * Bump the number of extents we've allocated
4744          * in this call.
4745          */
4746         bma->nallocs++;
4747
4748         if (bma->cur)
4749                 bma->cur->bc_private.b.flags =
4750                         bma->wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
4751
4752         bma->got.br_startoff = bma->offset;
4753         bma->got.br_startblock = bma->blkno;
4754         bma->got.br_blockcount = bma->length;
4755         bma->got.br_state = XFS_EXT_NORM;
4756
4757         /*
4758          * A wasdelay extent has been initialized, so shouldn't be flagged
4759          * as unwritten.
4760          */
4761         if (!bma->wasdel && (bma->flags & XFS_BMAPI_PREALLOC) &&
4762             xfs_sb_version_hasextflgbit(&mp->m_sb))
4763                 bma->got.br_state = XFS_EXT_UNWRITTEN;
4764
4765         if (bma->wasdel)
4766                 error = xfs_bmap_add_extent_delay_real(bma);
4767         else
4768                 error = xfs_bmap_add_extent_hole_real(bma, whichfork);
4769
4770         bma->logflags |= tmp_logflags;
4771         if (error)
4772                 return error;
4773
4774         /*
4775          * Update our extent pointer, given that xfs_bmap_add_extent_delay_real
4776          * or xfs_bmap_add_extent_hole_real might have merged it into one of
4777          * the neighbouring ones.
4778          */
4779         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx), &bma->got);
4780
4781         ASSERT(bma->got.br_startoff <= bma->offset);
4782         ASSERT(bma->got.br_startoff + bma->got.br_blockcount >=
4783                bma->offset + bma->length);
4784         ASSERT(bma->got.br_state == XFS_EXT_NORM ||
4785                bma->got.br_state == XFS_EXT_UNWRITTEN);
4786         return 0;
4787 }
4788
4789 static void
4790 xfs_bmapi_allocate_worker(
4791         struct work_struct      *work)
4792 {
4793         struct xfs_bmalloca     *args = container_of(work,
4794                                                 struct xfs_bmalloca, work);
4795         unsigned long           pflags;
4796
4797         /* we are in a transaction context here */
4798         current_set_flags_nested(&pflags, PF_FSTRANS);
4799
4800         args->result = __xfs_bmapi_allocate(args);
4801         complete(args->done);
4802
4803         current_restore_flags_nested(&pflags, PF_FSTRANS);
4804 }
4805
4806 /*
4807  * Some allocation requests often come in with little stack to work on. Push
4808  * them off to a worker thread so there is lots of stack to use. Otherwise just
4809  * call directly to avoid the context switch overhead here.
4810  */
4811 int
4812 xfs_bmapi_allocate(
4813         struct xfs_bmalloca     *args)
4814 {
4815         DECLARE_COMPLETION_ONSTACK(done);
4816
4817         if (!args->stack_switch)
4818                 return __xfs_bmapi_allocate(args);
4819
4820
4821         args->done = &done;
4822         INIT_WORK_ONSTACK(&args->work, xfs_bmapi_allocate_worker);
4823         queue_work(xfs_alloc_wq, &args->work);
4824         wait_for_completion(&done);
4825         return args->result;
4826 }
4827
4828 STATIC int
4829 xfs_bmapi_convert_unwritten(
4830         struct xfs_bmalloca     *bma,
4831         struct xfs_bmbt_irec    *mval,
4832         xfs_filblks_t           len,
4833         int                     flags)
4834 {
4835         int                     whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4836                                                 XFS_ATTR_FORK : XFS_DATA_FORK;
4837         struct xfs_ifork        *ifp = XFS_IFORK_PTR(bma->ip, whichfork);
4838         int                     tmp_logflags = 0;
4839         int                     error;
4840
4841         /* check if we need to do unwritten->real conversion */
4842         if (mval->br_state == XFS_EXT_UNWRITTEN &&
4843             (flags & XFS_BMAPI_PREALLOC))
4844                 return 0;
4845
4846         /* check if we need to do real->unwritten conversion */
4847         if (mval->br_state == XFS_EXT_NORM &&
4848             (flags & (XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT)) !=
4849                         (XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT))
4850                 return 0;
4851
4852         /*
4853          * Modify (by adding) the state flag, if writing.
4854          */
4855         ASSERT(mval->br_blockcount <= len);
4856         if ((ifp->if_flags & XFS_IFBROOT) && !bma->cur) {
4857                 bma->cur = xfs_bmbt_init_cursor(bma->ip->i_mount, bma->tp,
4858                                         bma->ip, whichfork);
4859                 bma->cur->bc_private.b.firstblock = *bma->firstblock;
4860                 bma->cur->bc_private.b.flist = bma->flist;
4861         }
4862         mval->br_state = (mval->br_state == XFS_EXT_UNWRITTEN)
4863                                 ? XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
4864
4865         error = xfs_bmap_add_extent_unwritten_real(bma->tp, bma->ip, &bma->idx,
4866                         &bma->cur, mval, bma->firstblock, bma->flist,
4867                         &tmp_logflags);
4868         bma->logflags |= tmp_logflags;
4869         if (error)
4870                 return error;
4871
4872         /*
4873          * Update our extent pointer, given that
4874          * xfs_bmap_add_extent_unwritten_real might have merged it into one
4875          * of the neighbouring ones.
4876          */
4877         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx), &bma->got);
4878
4879         /*
4880          * We may have combined previously unwritten space with written space,
4881          * so generate another request.
4882          */
4883         if (mval->br_blockcount < len)
4884                 return EAGAIN;
4885         return 0;
4886 }
4887
4888 /*
4889  * Map file blocks to filesystem blocks, and allocate blocks or convert the
4890  * extent state if necessary.  Details behaviour is controlled by the flags
4891  * parameter.  Only allocates blocks from a single allocation group, to avoid
4892  * locking problems.
4893  *
4894  * The returned value in "firstblock" from the first call in a transaction
4895  * must be remembered and presented to subsequent calls in "firstblock".
4896  * An upper bound for the number of blocks to be allocated is supplied to
4897  * the first call in "total"; if no allocation group has that many free
4898  * blocks then the call will fail (return NULLFSBLOCK in "firstblock").
4899  */
4900 int
4901 xfs_bmapi_write(
4902         struct xfs_trans        *tp,            /* transaction pointer */
4903         struct xfs_inode        *ip,            /* incore inode */
4904         xfs_fileoff_t           bno,            /* starting file offs. mapped */
4905         xfs_filblks_t           len,            /* length to map in file */
4906         int                     flags,          /* XFS_BMAPI_... */
4907         xfs_fsblock_t           *firstblock,    /* first allocated block
4908                                                    controls a.g. for allocs */
4909         xfs_extlen_t            total,          /* total blocks needed */
4910         struct xfs_bmbt_irec    *mval,          /* output: map values */
4911         int                     *nmap,          /* i/o: mval size/count */
4912         struct xfs_bmap_free    *flist)         /* i/o: list extents to free */
4913 {
4914         struct xfs_mount        *mp = ip->i_mount;
4915         struct xfs_ifork        *ifp;
4916         struct xfs_bmalloca     bma = { 0 };    /* args for xfs_bmap_alloc */
4917         xfs_fileoff_t           end;            /* end of mapped file region */
4918         int                     eof;            /* after the end of extents */
4919         int                     error;          /* error return */
4920         int                     n;              /* current extent index */
4921         xfs_fileoff_t           obno;           /* old block number (offset) */
4922         int                     whichfork;      /* data or attr fork */
4923         char                    inhole;         /* current location is hole in file */
4924         char                    wasdelay;       /* old extent was delayed */
4925
4926 #ifdef DEBUG
4927         xfs_fileoff_t           orig_bno;       /* original block number value */
4928         int                     orig_flags;     /* original flags arg value */
4929         xfs_filblks_t           orig_len;       /* original value of len arg */
4930         struct xfs_bmbt_irec    *orig_mval;     /* original value of mval */
4931         int                     orig_nmap;      /* original value of *nmap */
4932
4933         orig_bno = bno;
4934         orig_len = len;
4935         orig_flags = flags;
4936         orig_mval = mval;
4937         orig_nmap = *nmap;
4938 #endif
4939
4940         ASSERT(*nmap >= 1);
4941         ASSERT(*nmap <= XFS_BMAP_MAX_NMAP);
4942         ASSERT(!(flags & XFS_BMAPI_IGSTATE));
4943         ASSERT(tp != NULL);
4944         ASSERT(len > 0);
4945
4946         whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4947                 XFS_ATTR_FORK : XFS_DATA_FORK;
4948
4949         if (unlikely(XFS_TEST_ERROR(
4950             (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4951              XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
4952              XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL),
4953              mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4954                 XFS_ERROR_REPORT("xfs_bmapi_write", XFS_ERRLEVEL_LOW, mp);
4955                 return XFS_ERROR(EFSCORRUPTED);
4956         }
4957
4958         if (XFS_FORCED_SHUTDOWN(mp))
4959                 return XFS_ERROR(EIO);
4960
4961         ifp = XFS_IFORK_PTR(ip, whichfork);
4962
4963         XFS_STATS_INC(xs_blk_mapw);
4964
4965         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
4966                 /*
4967                  * XXX (dgc): This assumes we are only called for inodes that
4968                  * contain content neutral data in local format. Anything that
4969                  * contains caller-specific data in local format that needs
4970                  * transformation to move to a block format needs to do the
4971                  * conversion to extent format itself.
4972                  *
4973                  * Directory data forks and attribute forks handle this
4974                  * themselves, but with the addition of metadata verifiers every
4975                  * data fork in local format now contains caller specific data
4976                  * and as such conversion through this function is likely to be
4977                  * broken.
4978                  *
4979                  * The only likely user of this branch is for remote symlinks,
4980                  * but we cannot overwrite the data fork contents of the symlink
4981                  * (EEXIST occurs higher up the stack) and so it will never go
4982                  * from local format to extent format here. Hence I don't think
4983                  * this branch is ever executed intentionally and we should
4984                  * consider removing it and asserting that xfs_bmapi_write()
4985                  * cannot be called directly on local format forks. i.e. callers
4986                  * are completely responsible for local to extent format
4987                  * conversion, not xfs_bmapi_write().
4988                  */
4989                 error = xfs_bmap_local_to_extents(tp, ip, firstblock, total,
4990                                         &bma.logflags, whichfork,
4991                                         xfs_bmap_local_to_extents_init_fn);
4992                 if (error)
4993                         goto error0;
4994         }
4995
4996         if (*firstblock == NULLFSBLOCK) {
4997                 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE)
4998                         bma.minleft = be16_to_cpu(ifp->if_broot->bb_level) + 1;
4999                 else
5000                         bma.minleft = 1;
5001         } else {
5002                 bma.minleft = 0;
5003         }
5004
5005         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
5006                 error = xfs_iread_extents(tp, ip, whichfork);
5007                 if (error)
5008                         goto error0;
5009         }
5010
5011         xfs_bmap_search_extents(ip, bno, whichfork, &eof, &bma.idx, &bma.got,
5012                                 &bma.prev);
5013         n = 0;
5014         end = bno + len;
5015         obno = bno;
5016
5017         bma.tp = tp;
5018         bma.ip = ip;
5019         bma.total = total;
5020         bma.userdata = 0;
5021         bma.flist = flist;
5022         bma.firstblock = firstblock;
5023
5024         if (flags & XFS_BMAPI_STACK_SWITCH)
5025                 bma.stack_switch = 1;
5026
5027         while (bno < end && n < *nmap) {
5028                 inhole = eof || bma.got.br_startoff > bno;
5029                 wasdelay = !inhole && isnullstartblock(bma.got.br_startblock);
5030
5031                 /*
5032                  * First, deal with the hole before the allocated space
5033                  * that we found, if any.
5034                  */
5035                 if (inhole || wasdelay) {
5036                         bma.eof = eof;
5037                         bma.conv = !!(flags & XFS_BMAPI_CONVERT);
5038                         bma.wasdel = wasdelay;
5039                         bma.offset = bno;
5040                         bma.flags = flags;
5041
5042                         /*
5043                          * There's a 32/64 bit type mismatch between the
5044                          * allocation length request (which can be 64 bits in
5045                          * length) and the bma length request, which is
5046                          * xfs_extlen_t and therefore 32 bits. Hence we have to
5047                          * check for 32-bit overflows and handle them here.
5048                          */
5049                         if (len > (xfs_filblks_t)MAXEXTLEN)
5050                                 bma.length = MAXEXTLEN;
5051                         else
5052                                 bma.length = len;
5053
5054                         ASSERT(len > 0);
5055                         ASSERT(bma.length > 0);
5056                         error = xfs_bmapi_allocate(&bma);
5057                         if (error)
5058                                 goto error0;
5059                         if (bma.blkno == NULLFSBLOCK)
5060                                 break;
5061                 }
5062
5063                 /* Deal with the allocated space we found.  */
5064                 xfs_bmapi_trim_map(mval, &bma.got, &bno, len, obno,
5065                                                         end, n, flags);
5066
5067                 /* Execute unwritten extent conversion if necessary */
5068                 error = xfs_bmapi_convert_unwritten(&bma, mval, len, flags);
5069                 if (error == EAGAIN)
5070                         continue;
5071                 if (error)
5072                         goto error0;
5073
5074                 /* update the extent map to return */
5075                 xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
5076
5077                 /*
5078                  * If we're done, stop now.  Stop when we've allocated
5079                  * XFS_BMAP_MAX_NMAP extents no matter what.  Otherwise
5080                  * the transaction may get too big.
5081                  */
5082                 if (bno >= end || n >= *nmap || bma.nallocs >= *nmap)
5083                         break;
5084
5085                 /* Else go on to the next record. */
5086                 bma.prev = bma.got;
5087                 if (++bma.idx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t)) {
5088                         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma.idx),
5089                                          &bma.got);
5090                 } else
5091                         eof = 1;
5092         }
5093         *nmap = n;
5094
5095         /*
5096          * Transform from btree to extents, give it cur.
5097          */
5098         if (xfs_bmap_wants_extents(ip, whichfork)) {
5099                 int             tmp_logflags = 0;
5100
5101                 ASSERT(bma.cur);
5102                 error = xfs_bmap_btree_to_extents(tp, ip, bma.cur,
5103                         &tmp_logflags, whichfork);
5104                 bma.logflags |= tmp_logflags;
5105                 if (error)
5106                         goto error0;
5107         }
5108
5109         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE ||
5110                XFS_IFORK_NEXTENTS(ip, whichfork) >
5111                 XFS_IFORK_MAXEXT(ip, whichfork));
5112         error = 0;
5113 error0:
5114         /*
5115          * Log everything.  Do this after conversion, there's no point in
5116          * logging the extent records if we've converted to btree format.
5117          */
5118         if ((bma.logflags & xfs_ilog_fext(whichfork)) &&
5119             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
5120                 bma.logflags &= ~xfs_ilog_fext(whichfork);
5121         else if ((bma.logflags & xfs_ilog_fbroot(whichfork)) &&
5122                  XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
5123                 bma.logflags &= ~xfs_ilog_fbroot(whichfork);
5124         /*
5125          * Log whatever the flags say, even if error.  Otherwise we might miss
5126          * detecting a case where the data is changed, there's an error,
5127          * and it's not logged so we don't shutdown when we should.
5128          */
5129         if (bma.logflags)
5130                 xfs_trans_log_inode(tp, ip, bma.logflags);
5131
5132         if (bma.cur) {
5133                 if (!error) {
5134                         ASSERT(*firstblock == NULLFSBLOCK ||
5135                                XFS_FSB_TO_AGNO(mp, *firstblock) ==
5136                                XFS_FSB_TO_AGNO(mp,
5137                                        bma.cur->bc_private.b.firstblock) ||
5138                                (flist->xbf_low &&
5139                                 XFS_FSB_TO_AGNO(mp, *firstblock) <
5140                                 XFS_FSB_TO_AGNO(mp,
5141                                         bma.cur->bc_private.b.firstblock)));
5142                         *firstblock = bma.cur->bc_private.b.firstblock;
5143                 }
5144                 xfs_btree_del_cursor(bma.cur,
5145                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5146         }
5147         if (!error)
5148                 xfs_bmap_validate_ret(orig_bno, orig_len, orig_flags, orig_mval,
5149                         orig_nmap, *nmap);
5150         return error;
5151 }
5152
5153 /*
5154  * Unmap (remove) blocks from a file.
5155  * If nexts is nonzero then the number of extents to remove is limited to
5156  * that value.  If not all extents in the block range can be removed then
5157  * *done is set.
5158  */
5159 int                                             /* error */
5160 xfs_bunmapi(
5161         xfs_trans_t             *tp,            /* transaction pointer */
5162         struct xfs_inode        *ip,            /* incore inode */
5163         xfs_fileoff_t           bno,            /* starting offset to unmap */
5164         xfs_filblks_t           len,            /* length to unmap in file */
5165         int                     flags,          /* misc flags */
5166         xfs_extnum_t            nexts,          /* number of extents max */
5167         xfs_fsblock_t           *firstblock,    /* first allocated block
5168                                                    controls a.g. for allocs */
5169         xfs_bmap_free_t         *flist,         /* i/o: list extents to free */
5170         int                     *done)          /* set if not done yet */
5171 {
5172         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
5173         xfs_bmbt_irec_t         del;            /* extent being deleted */
5174         int                     eof;            /* is deleting at eof */
5175         xfs_bmbt_rec_host_t     *ep;            /* extent record pointer */
5176         int                     error;          /* error return value */
5177         xfs_extnum_t            extno;          /* extent number in list */
5178         xfs_bmbt_irec_t         got;            /* current extent record */
5179         xfs_ifork_t             *ifp;           /* inode fork pointer */
5180         int                     isrt;           /* freeing in rt area */
5181         xfs_extnum_t            lastx;          /* last extent index used */
5182         int                     logflags;       /* transaction logging flags */
5183         xfs_extlen_t            mod;            /* rt extent offset */
5184         xfs_mount_t             *mp;            /* mount structure */
5185         xfs_extnum_t            nextents;       /* number of file extents */
5186         xfs_bmbt_irec_t         prev;           /* previous extent record */
5187         xfs_fileoff_t           start;          /* first file offset deleted */
5188         int                     tmp_logflags;   /* partial logging flags */
5189         int                     wasdel;         /* was a delayed alloc extent */
5190         int                     whichfork;      /* data or attribute fork */
5191         xfs_fsblock_t           sum;
5192
5193         trace_xfs_bunmap(ip, bno, len, flags, _RET_IP_);
5194
5195         whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
5196                 XFS_ATTR_FORK : XFS_DATA_FORK;
5197         ifp = XFS_IFORK_PTR(ip, whichfork);
5198         if (unlikely(
5199             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
5200             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
5201                 XFS_ERROR_REPORT("xfs_bunmapi", XFS_ERRLEVEL_LOW,
5202                                  ip->i_mount);
5203                 return XFS_ERROR(EFSCORRUPTED);
5204         }
5205         mp = ip->i_mount;
5206         if (XFS_FORCED_SHUTDOWN(mp))
5207                 return XFS_ERROR(EIO);
5208
5209         ASSERT(len > 0);
5210         ASSERT(nexts >= 0);
5211
5212         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5213             (error = xfs_iread_extents(tp, ip, whichfork)))
5214                 return error;
5215         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5216         if (nextents == 0) {
5217                 *done = 1;
5218                 return 0;
5219         }
5220         XFS_STATS_INC(xs_blk_unmap);
5221         isrt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
5222         start = bno;
5223         bno = start + len - 1;
5224         ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
5225                 &prev);
5226
5227         /*
5228          * Check to see if the given block number is past the end of the
5229          * file, back up to the last block if so...
5230          */
5231         if (eof) {
5232                 ep = xfs_iext_get_ext(ifp, --lastx);
5233                 xfs_bmbt_get_all(ep, &got);
5234                 bno = got.br_startoff + got.br_blockcount - 1;
5235         }
5236         logflags = 0;
5237         if (ifp->if_flags & XFS_IFBROOT) {
5238                 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
5239                 cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5240                 cur->bc_private.b.firstblock = *firstblock;
5241                 cur->bc_private.b.flist = flist;
5242                 cur->bc_private.b.flags = 0;
5243         } else
5244                 cur = NULL;
5245
5246         if (isrt) {
5247                 /*
5248                  * Synchronize by locking the bitmap inode.
5249                  */
5250                 xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL);
5251                 xfs_trans_ijoin(tp, mp->m_rbmip, XFS_ILOCK_EXCL);
5252         }
5253
5254         extno = 0;
5255         while (bno != (xfs_fileoff_t)-1 && bno >= start && lastx >= 0 &&
5256                (nexts == 0 || extno < nexts)) {
5257                 /*
5258                  * Is the found extent after a hole in which bno lives?
5259                  * Just back up to the previous extent, if so.
5260                  */
5261                 if (got.br_startoff > bno) {
5262                         if (--lastx < 0)
5263                                 break;
5264                         ep = xfs_iext_get_ext(ifp, lastx);
5265                         xfs_bmbt_get_all(ep, &got);
5266                 }
5267                 /*
5268                  * Is the last block of this extent before the range
5269                  * we're supposed to delete?  If so, we're done.
5270                  */
5271                 bno = XFS_FILEOFF_MIN(bno,
5272                         got.br_startoff + got.br_blockcount - 1);
5273                 if (bno < start)
5274                         break;
5275                 /*
5276                  * Then deal with the (possibly delayed) allocated space
5277                  * we found.
5278                  */
5279                 ASSERT(ep != NULL);
5280                 del = got;
5281                 wasdel = isnullstartblock(del.br_startblock);
5282                 if (got.br_startoff < start) {
5283                         del.br_startoff = start;
5284                         del.br_blockcount -= start - got.br_startoff;
5285                         if (!wasdel)
5286                                 del.br_startblock += start - got.br_startoff;
5287                 }
5288                 if (del.br_startoff + del.br_blockcount > bno + 1)
5289                         del.br_blockcount = bno + 1 - del.br_startoff;
5290                 sum = del.br_startblock + del.br_blockcount;
5291                 if (isrt &&
5292                     (mod = do_mod(sum, mp->m_sb.sb_rextsize))) {
5293                         /*
5294                          * Realtime extent not lined up at the end.
5295                          * The extent could have been split into written
5296                          * and unwritten pieces, or we could just be
5297                          * unmapping part of it.  But we can't really
5298                          * get rid of part of a realtime extent.
5299                          */
5300                         if (del.br_state == XFS_EXT_UNWRITTEN ||
5301                             !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5302                                 /*
5303                                  * This piece is unwritten, or we're not
5304                                  * using unwritten extents.  Skip over it.
5305                                  */
5306                                 ASSERT(bno >= mod);
5307                                 bno -= mod > del.br_blockcount ?
5308                                         del.br_blockcount : mod;
5309                                 if (bno < got.br_startoff) {
5310                                         if (--lastx >= 0)
5311                                                 xfs_bmbt_get_all(xfs_iext_get_ext(
5312                                                         ifp, lastx), &got);
5313                                 }
5314                                 continue;
5315                         }
5316                         /*
5317                          * It's written, turn it unwritten.
5318                          * This is better than zeroing it.
5319                          */
5320                         ASSERT(del.br_state == XFS_EXT_NORM);
5321                         ASSERT(xfs_trans_get_block_res(tp) > 0);
5322                         /*
5323                          * If this spans a realtime extent boundary,
5324                          * chop it back to the start of the one we end at.
5325                          */
5326                         if (del.br_blockcount > mod) {
5327                                 del.br_startoff += del.br_blockcount - mod;
5328                                 del.br_startblock += del.br_blockcount - mod;
5329                                 del.br_blockcount = mod;
5330                         }
5331                         del.br_state = XFS_EXT_UNWRITTEN;
5332                         error = xfs_bmap_add_extent_unwritten_real(tp, ip,
5333                                         &lastx, &cur, &del, firstblock, flist,
5334                                         &logflags);
5335                         if (error)
5336                                 goto error0;
5337                         goto nodelete;
5338                 }
5339                 if (isrt && (mod = do_mod(del.br_startblock, mp->m_sb.sb_rextsize))) {
5340                         /*
5341                          * Realtime extent is lined up at the end but not
5342                          * at the front.  We'll get rid of full extents if
5343                          * we can.
5344                          */
5345                         mod = mp->m_sb.sb_rextsize - mod;
5346                         if (del.br_blockcount > mod) {
5347                                 del.br_blockcount -= mod;
5348                                 del.br_startoff += mod;
5349                                 del.br_startblock += mod;
5350                         } else if ((del.br_startoff == start &&
5351                                     (del.br_state == XFS_EXT_UNWRITTEN ||
5352                                      xfs_trans_get_block_res(tp) == 0)) ||
5353                                    !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5354                                 /*
5355                                  * Can't make it unwritten.  There isn't
5356                                  * a full extent here so just skip it.
5357                                  */
5358                                 ASSERT(bno >= del.br_blockcount);
5359                                 bno -= del.br_blockcount;
5360                                 if (got.br_startoff > bno) {
5361                                         if (--lastx >= 0) {
5362                                                 ep = xfs_iext_get_ext(ifp,
5363                                                                       lastx);
5364                                                 xfs_bmbt_get_all(ep, &got);
5365                                         }
5366                                 }
5367                                 continue;
5368                         } else if (del.br_state == XFS_EXT_UNWRITTEN) {
5369                                 /*
5370                                  * This one is already unwritten.
5371                                  * It must have a written left neighbor.
5372                                  * Unwrite the killed part of that one and
5373                                  * try again.
5374                                  */
5375                                 ASSERT(lastx > 0);
5376                                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp,
5377                                                 lastx - 1), &prev);
5378                                 ASSERT(prev.br_state == XFS_EXT_NORM);
5379                                 ASSERT(!isnullstartblock(prev.br_startblock));
5380                                 ASSERT(del.br_startblock ==
5381                                        prev.br_startblock + prev.br_blockcount);
5382                                 if (prev.br_startoff < start) {
5383                                         mod = start - prev.br_startoff;
5384                                         prev.br_blockcount -= mod;
5385                                         prev.br_startblock += mod;
5386                                         prev.br_startoff = start;
5387                                 }
5388                                 prev.br_state = XFS_EXT_UNWRITTEN;
5389                                 lastx--;
5390                                 error = xfs_bmap_add_extent_unwritten_real(tp,
5391                                                 ip, &lastx, &cur, &prev,
5392                                                 firstblock, flist, &logflags);
5393                                 if (error)
5394                                         goto error0;
5395                                 goto nodelete;
5396                         } else {
5397                                 ASSERT(del.br_state == XFS_EXT_NORM);
5398                                 del.br_state = XFS_EXT_UNWRITTEN;
5399                                 error = xfs_bmap_add_extent_unwritten_real(tp,
5400                                                 ip, &lastx, &cur, &del,
5401                                                 firstblock, flist, &logflags);
5402                                 if (error)
5403                                         goto error0;
5404                                 goto nodelete;
5405                         }
5406                 }
5407                 if (wasdel) {
5408                         ASSERT(startblockval(del.br_startblock) > 0);
5409                         /* Update realtime/data freespace, unreserve quota */
5410                         if (isrt) {
5411                                 xfs_filblks_t rtexts;
5412
5413                                 rtexts = XFS_FSB_TO_B(mp, del.br_blockcount);
5414                                 do_div(rtexts, mp->m_sb.sb_rextsize);
5415                                 xfs_mod_incore_sb(mp, XFS_SBS_FREXTENTS,
5416                                                 (int64_t)rtexts, 0);
5417                                 (void)xfs_trans_reserve_quota_nblks(NULL,
5418                                         ip, -((long)del.br_blockcount), 0,
5419                                         XFS_QMOPT_RES_RTBLKS);
5420                         } else {
5421                                 xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
5422                                                 (int64_t)del.br_blockcount, 0);
5423                                 (void)xfs_trans_reserve_quota_nblks(NULL,
5424                                         ip, -((long)del.br_blockcount), 0,
5425                                         XFS_QMOPT_RES_REGBLKS);
5426                         }
5427                         ip->i_delayed_blks -= del.br_blockcount;
5428                         if (cur)
5429                                 cur->bc_private.b.flags |=
5430                                         XFS_BTCUR_BPRV_WASDEL;
5431                 } else if (cur)
5432                         cur->bc_private.b.flags &= ~XFS_BTCUR_BPRV_WASDEL;
5433                 /*
5434                  * If it's the case where the directory code is running
5435                  * with no block reservation, and the deleted block is in
5436                  * the middle of its extent, and the resulting insert
5437                  * of an extent would cause transformation to btree format,
5438                  * then reject it.  The calling code will then swap
5439                  * blocks around instead.
5440                  * We have to do this now, rather than waiting for the
5441                  * conversion to btree format, since the transaction
5442                  * will be dirty.
5443                  */
5444                 if (!wasdel && xfs_trans_get_block_res(tp) == 0 &&
5445                     XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
5446                     XFS_IFORK_NEXTENTS(ip, whichfork) >= /* Note the >= */
5447                         XFS_IFORK_MAXEXT(ip, whichfork) &&
5448                     del.br_startoff > got.br_startoff &&
5449                     del.br_startoff + del.br_blockcount <
5450                     got.br_startoff + got.br_blockcount) {
5451                         error = XFS_ERROR(ENOSPC);
5452                         goto error0;
5453                 }
5454                 error = xfs_bmap_del_extent(ip, tp, &lastx, flist, cur, &del,
5455                                 &tmp_logflags, whichfork);
5456                 logflags |= tmp_logflags;
5457                 if (error)
5458                         goto error0;
5459                 bno = del.br_startoff - 1;
5460 nodelete:
5461                 /*
5462                  * If not done go on to the next (previous) record.
5463                  */
5464                 if (bno != (xfs_fileoff_t)-1 && bno >= start) {
5465                         if (lastx >= 0) {
5466                                 ep = xfs_iext_get_ext(ifp, lastx);
5467                                 if (xfs_bmbt_get_startoff(ep) > bno) {
5468                                         if (--lastx >= 0)
5469                                                 ep = xfs_iext_get_ext(ifp,
5470                                                                       lastx);
5471                                 }
5472                                 xfs_bmbt_get_all(ep, &got);
5473                         }
5474                         extno++;
5475                 }
5476         }
5477         *done = bno == (xfs_fileoff_t)-1 || bno < start || lastx < 0;
5478
5479         /*
5480          * Convert to a btree if necessary.
5481          */
5482         if (xfs_bmap_needs_btree(ip, whichfork)) {
5483                 ASSERT(cur == NULL);
5484                 error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist,
5485                         &cur, 0, &tmp_logflags, whichfork);
5486                 logflags |= tmp_logflags;
5487                 if (error)
5488                         goto error0;
5489         }
5490         /*
5491          * transform from btree to extents, give it cur
5492          */
5493         else if (xfs_bmap_wants_extents(ip, whichfork)) {
5494                 ASSERT(cur != NULL);
5495                 error = xfs_bmap_btree_to_extents(tp, ip, cur, &tmp_logflags,
5496                         whichfork);
5497                 logflags |= tmp_logflags;
5498                 if (error)
5499                         goto error0;
5500         }
5501         /*
5502          * transform from extents to local?
5503          */
5504         error = 0;
5505 error0:
5506         /*
5507          * Log everything.  Do this after conversion, there's no point in
5508          * logging the extent records if we've converted to btree format.
5509          */
5510         if ((logflags & xfs_ilog_fext(whichfork)) &&
5511             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
5512                 logflags &= ~xfs_ilog_fext(whichfork);
5513         else if ((logflags & xfs_ilog_fbroot(whichfork)) &&
5514                  XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
5515                 logflags &= ~xfs_ilog_fbroot(whichfork);
5516         /*
5517          * Log inode even in the error case, if the transaction
5518          * is dirty we'll need to shut down the filesystem.
5519          */
5520         if (logflags)
5521                 xfs_trans_log_inode(tp, ip, logflags);
5522         if (cur) {
5523                 if (!error) {
5524                         *firstblock = cur->bc_private.b.firstblock;
5525                         cur->bc_private.b.allocated = 0;
5526                 }
5527                 xfs_btree_del_cursor(cur,
5528                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5529         }
5530         return error;
5531 }
5532
5533 /*
5534  * returns 1 for success, 0 if we failed to map the extent.
5535  */
5536 STATIC int
5537 xfs_getbmapx_fix_eof_hole(
5538         xfs_inode_t             *ip,            /* xfs incore inode pointer */
5539         struct getbmapx         *out,           /* output structure */
5540         int                     prealloced,     /* this is a file with
5541                                                  * preallocated data space */
5542         __int64_t               end,            /* last block requested */
5543         xfs_fsblock_t           startblock)
5544 {
5545         __int64_t               fixlen;
5546         xfs_mount_t             *mp;            /* file system mount point */
5547         xfs_ifork_t             *ifp;           /* inode fork pointer */
5548         xfs_extnum_t            lastx;          /* last extent pointer */
5549         xfs_fileoff_t           fileblock;
5550
5551         if (startblock == HOLESTARTBLOCK) {
5552                 mp = ip->i_mount;
5553                 out->bmv_block = -1;
5554                 fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, XFS_ISIZE(ip)));
5555                 fixlen -= out->bmv_offset;
5556                 if (prealloced && out->bmv_offset + out->bmv_length == end) {
5557                         /* Came to hole at EOF. Trim it. */
5558                         if (fixlen <= 0)
5559                                 return 0;
5560                         out->bmv_length = fixlen;
5561                 }
5562         } else {
5563                 if (startblock == DELAYSTARTBLOCK)
5564                         out->bmv_block = -2;
5565                 else
5566                         out->bmv_block = xfs_fsb_to_db(ip, startblock);
5567                 fileblock = XFS_BB_TO_FSB(ip->i_mount, out->bmv_offset);
5568                 ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
5569                 if (xfs_iext_bno_to_ext(ifp, fileblock, &lastx) &&
5570                    (lastx == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t))-1))
5571                         out->bmv_oflags |= BMV_OF_LAST;
5572         }
5573
5574         return 1;
5575 }
5576
5577 /*
5578  * Get inode's extents as described in bmv, and format for output.
5579  * Calls formatter to fill the user's buffer until all extents
5580  * are mapped, until the passed-in bmv->bmv_count slots have
5581  * been filled, or until the formatter short-circuits the loop,
5582  * if it is tracking filled-in extents on its own.
5583  */
5584 int                                             /* error code */
5585 xfs_getbmap(
5586         xfs_inode_t             *ip,
5587         struct getbmapx         *bmv,           /* user bmap structure */
5588         xfs_bmap_format_t       formatter,      /* format to user */
5589         void                    *arg)           /* formatter arg */
5590 {
5591         __int64_t               bmvend;         /* last block requested */
5592         int                     error = 0;      /* return value */
5593         __int64_t               fixlen;         /* length for -1 case */
5594         int                     i;              /* extent number */
5595         int                     lock;           /* lock state */
5596         xfs_bmbt_irec_t         *map;           /* buffer for user's data */
5597         xfs_mount_t             *mp;            /* file system mount point */
5598         int                     nex;            /* # of user extents can do */
5599         int                     nexleft;        /* # of user extents left */
5600         int                     subnex;         /* # of bmapi's can do */
5601         int                     nmap;           /* number of map entries */
5602         struct getbmapx         *out;           /* output structure */
5603         int                     whichfork;      /* data or attr fork */
5604         int                     prealloced;     /* this is a file with
5605                                                  * preallocated data space */
5606         int                     iflags;         /* interface flags */
5607         int                     bmapi_flags;    /* flags for xfs_bmapi */
5608         int                     cur_ext = 0;
5609
5610         mp = ip->i_mount;
5611         iflags = bmv->bmv_iflags;
5612         whichfork = iflags & BMV_IF_ATTRFORK ? XFS_ATTR_FORK : XFS_DATA_FORK;
5613
5614         if (whichfork == XFS_ATTR_FORK) {
5615                 if (XFS_IFORK_Q(ip)) {
5616                         if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS &&
5617                             ip->i_d.di_aformat != XFS_DINODE_FMT_BTREE &&
5618                             ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)
5619                                 return XFS_ERROR(EINVAL);
5620                 } else if (unlikely(
5621                            ip->i_d.di_aformat != 0 &&
5622                            ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS)) {
5623                         XFS_ERROR_REPORT("xfs_getbmap", XFS_ERRLEVEL_LOW,
5624                                          ip->i_mount);
5625                         return XFS_ERROR(EFSCORRUPTED);
5626                 }
5627
5628                 prealloced = 0;
5629                 fixlen = 1LL << 32;
5630         } else {
5631                 if (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS &&
5632                     ip->i_d.di_format != XFS_DINODE_FMT_BTREE &&
5633                     ip->i_d.di_format != XFS_DINODE_FMT_LOCAL)
5634                         return XFS_ERROR(EINVAL);
5635
5636                 if (xfs_get_extsz_hint(ip) ||
5637                     ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC|XFS_DIFLAG_APPEND)){
5638                         prealloced = 1;
5639                         fixlen = mp->m_super->s_maxbytes;
5640                 } else {
5641                         prealloced = 0;
5642                         fixlen = XFS_ISIZE(ip);
5643                 }
5644         }
5645
5646         if (bmv->bmv_length == -1) {
5647                 fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, fixlen));
5648                 bmv->bmv_length =
5649                         max_t(__int64_t, fixlen - bmv->bmv_offset, 0);
5650         } else if (bmv->bmv_length == 0) {
5651                 bmv->bmv_entries = 0;
5652                 return 0;
5653         } else if (bmv->bmv_length < 0) {
5654                 return XFS_ERROR(EINVAL);
5655         }
5656
5657         nex = bmv->bmv_count - 1;
5658         if (nex <= 0)
5659                 return XFS_ERROR(EINVAL);
5660         bmvend = bmv->bmv_offset + bmv->bmv_length;
5661
5662
5663         if (bmv->bmv_count > ULONG_MAX / sizeof(struct getbmapx))
5664                 return XFS_ERROR(ENOMEM);
5665         out = kmem_zalloc(bmv->bmv_count * sizeof(struct getbmapx), KM_MAYFAIL);
5666         if (!out) {
5667                 out = kmem_zalloc_large(bmv->bmv_count *
5668                                         sizeof(struct getbmapx));
5669                 if (!out)
5670                         return XFS_ERROR(ENOMEM);
5671         }
5672
5673         xfs_ilock(ip, XFS_IOLOCK_SHARED);
5674         if (whichfork == XFS_DATA_FORK && !(iflags & BMV_IF_DELALLOC)) {
5675                 if (ip->i_delayed_blks || XFS_ISIZE(ip) > ip->i_d.di_size) {
5676                         error = -filemap_write_and_wait(VFS_I(ip)->i_mapping);
5677                         if (error)
5678                                 goto out_unlock_iolock;
5679                 }
5680                 /*
5681                  * even after flushing the inode, there can still be delalloc
5682                  * blocks on the inode beyond EOF due to speculative
5683                  * preallocation. These are not removed until the release
5684                  * function is called or the inode is inactivated. Hence we
5685                  * cannot assert here that ip->i_delayed_blks == 0.
5686                  */
5687         }
5688
5689         lock = xfs_ilock_map_shared(ip);
5690
5691         /*
5692          * Don't let nex be bigger than the number of extents
5693          * we can have assuming alternating holes and real extents.
5694          */
5695         if (nex > XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1)
5696                 nex = XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1;
5697
5698         bmapi_flags = xfs_bmapi_aflag(whichfork);
5699         if (!(iflags & BMV_IF_PREALLOC))
5700                 bmapi_flags |= XFS_BMAPI_IGSTATE;
5701
5702         /*
5703          * Allocate enough space to handle "subnex" maps at a time.
5704          */
5705         error = ENOMEM;
5706         subnex = 16;
5707         map = kmem_alloc(subnex * sizeof(*map), KM_MAYFAIL | KM_NOFS);
5708         if (!map)
5709                 goto out_unlock_ilock;
5710
5711         bmv->bmv_entries = 0;
5712
5713         if (XFS_IFORK_NEXTENTS(ip, whichfork) == 0 &&
5714             (whichfork == XFS_ATTR_FORK || !(iflags & BMV_IF_DELALLOC))) {
5715                 error = 0;
5716                 goto out_free_map;
5717         }
5718
5719         nexleft = nex;
5720
5721         do {
5722                 nmap = (nexleft > subnex) ? subnex : nexleft;
5723                 error = xfs_bmapi_read(ip, XFS_BB_TO_FSBT(mp, bmv->bmv_offset),
5724                                        XFS_BB_TO_FSB(mp, bmv->bmv_length),
5725                                        map, &nmap, bmapi_flags);
5726                 if (error)
5727                         goto out_free_map;
5728                 ASSERT(nmap <= subnex);
5729
5730                 for (i = 0; i < nmap && nexleft && bmv->bmv_length; i++) {
5731                         out[cur_ext].bmv_oflags = 0;
5732                         if (map[i].br_state == XFS_EXT_UNWRITTEN)
5733                                 out[cur_ext].bmv_oflags |= BMV_OF_PREALLOC;
5734                         else if (map[i].br_startblock == DELAYSTARTBLOCK)
5735                                 out[cur_ext].bmv_oflags |= BMV_OF_DELALLOC;
5736                         out[cur_ext].bmv_offset =
5737                                 XFS_FSB_TO_BB(mp, map[i].br_startoff);
5738                         out[cur_ext].bmv_length =
5739                                 XFS_FSB_TO_BB(mp, map[i].br_blockcount);
5740                         out[cur_ext].bmv_unused1 = 0;
5741                         out[cur_ext].bmv_unused2 = 0;
5742
5743                         /*
5744                          * delayed allocation extents that start beyond EOF can
5745                          * occur due to speculative EOF allocation when the
5746                          * delalloc extent is larger than the largest freespace
5747                          * extent at conversion time. These extents cannot be
5748                          * converted by data writeback, so can exist here even
5749                          * if we are not supposed to be finding delalloc
5750                          * extents.
5751                          */
5752                         if (map[i].br_startblock == DELAYSTARTBLOCK &&
5753                             map[i].br_startoff <= XFS_B_TO_FSB(mp, XFS_ISIZE(ip)))
5754                                 ASSERT((iflags & BMV_IF_DELALLOC) != 0);
5755
5756                         if (map[i].br_startblock == HOLESTARTBLOCK &&
5757                             whichfork == XFS_ATTR_FORK) {
5758                                 /* came to the end of attribute fork */
5759                                 out[cur_ext].bmv_oflags |= BMV_OF_LAST;
5760                                 goto out_free_map;
5761                         }
5762
5763                         if (!xfs_getbmapx_fix_eof_hole(ip, &out[cur_ext],
5764                                         prealloced, bmvend,
5765                                         map[i].br_startblock))
5766                                 goto out_free_map;
5767
5768                         bmv->bmv_offset =
5769                                 out[cur_ext].bmv_offset +
5770                                 out[cur_ext].bmv_length;
5771                         bmv->bmv_length =
5772                                 max_t(__int64_t, 0, bmvend - bmv->bmv_offset);
5773
5774                         /*
5775                          * In case we don't want to return the hole,
5776                          * don't increase cur_ext so that we can reuse
5777                          * it in the next loop.
5778                          */
5779                         if ((iflags & BMV_IF_NO_HOLES) &&
5780                             map[i].br_startblock == HOLESTARTBLOCK) {
5781                                 memset(&out[cur_ext], 0, sizeof(out[cur_ext]));
5782                                 continue;
5783                         }
5784
5785                         nexleft--;
5786                         bmv->bmv_entries++;
5787                         cur_ext++;
5788                 }
5789         } while (nmap && nexleft && bmv->bmv_length);
5790
5791  out_free_map:
5792         kmem_free(map);
5793  out_unlock_ilock:
5794         xfs_iunlock_map_shared(ip, lock);
5795  out_unlock_iolock:
5796         xfs_iunlock(ip, XFS_IOLOCK_SHARED);
5797
5798         for (i = 0; i < cur_ext; i++) {
5799                 int full = 0;   /* user array is full */
5800
5801                 /* format results & advance arg */
5802                 error = formatter(&arg, &out[i], &full);
5803                 if (error || full)
5804                         break;
5805         }
5806
5807         if (is_vmalloc_addr(out))
5808                 kmem_free_large(out);
5809         else
5810                 kmem_free(out);
5811         return error;
5812 }
5813
5814 #ifdef DEBUG
5815 STATIC struct xfs_buf *
5816 xfs_bmap_get_bp(
5817         struct xfs_btree_cur    *cur,
5818         xfs_fsblock_t           bno)
5819 {
5820         struct xfs_log_item_desc *lidp;
5821         int                     i;
5822
5823         if (!cur)
5824                 return NULL;
5825
5826         for (i = 0; i < XFS_BTREE_MAXLEVELS; i++) {
5827                 if (!cur->bc_bufs[i])
5828                         break;
5829                 if (XFS_BUF_ADDR(cur->bc_bufs[i]) == bno)
5830                         return cur->bc_bufs[i];
5831         }
5832
5833         /* Chase down all the log items to see if the bp is there */
5834         list_for_each_entry(lidp, &cur->bc_tp->t_items, lid_trans) {
5835                 struct xfs_buf_log_item *bip;
5836                 bip = (struct xfs_buf_log_item *)lidp->lid_item;
5837                 if (bip->bli_item.li_type == XFS_LI_BUF &&
5838                     XFS_BUF_ADDR(bip->bli_buf) == bno)
5839                         return bip->bli_buf;
5840         }
5841
5842         return NULL;
5843 }
5844
5845 STATIC void
5846 xfs_check_block(
5847         struct xfs_btree_block  *block,
5848         xfs_mount_t             *mp,
5849         int                     root,
5850         short                   sz)
5851 {
5852         int                     i, j, dmxr;
5853         __be64                  *pp, *thispa;   /* pointer to block address */
5854         xfs_bmbt_key_t          *prevp, *keyp;
5855
5856         ASSERT(be16_to_cpu(block->bb_level) > 0);
5857
5858         prevp = NULL;
5859         for( i = 1; i <= xfs_btree_get_numrecs(block); i++) {
5860                 dmxr = mp->m_bmap_dmxr[0];
5861                 keyp = XFS_BMBT_KEY_ADDR(mp, block, i);
5862
5863                 if (prevp) {
5864                         ASSERT(be64_to_cpu(prevp->br_startoff) <
5865                                be64_to_cpu(keyp->br_startoff));
5866                 }
5867                 prevp = keyp;
5868
5869                 /*
5870                  * Compare the block numbers to see if there are dups.
5871                  */
5872                 if (root)
5873                         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, i, sz);
5874                 else
5875                         pp = XFS_BMBT_PTR_ADDR(mp, block, i, dmxr);
5876
5877                 for (j = i+1; j <= be16_to_cpu(block->bb_numrecs); j++) {
5878                         if (root)
5879                                 thispa = XFS_BMAP_BROOT_PTR_ADDR(mp, block, j, sz);
5880                         else
5881                                 thispa = XFS_BMBT_PTR_ADDR(mp, block, j, dmxr);
5882                         if (*thispa == *pp) {
5883                                 xfs_warn(mp, "%s: thispa(%d) == pp(%d) %Ld",
5884                                         __func__, j, i,
5885                                         (unsigned long long)be64_to_cpu(*thispa));
5886                                 panic("%s: ptrs are equal in node\n",
5887                                         __func__);
5888                         }
5889                 }
5890         }
5891 }
5892
5893 /*
5894  * Check that the extents for the inode ip are in the right order in all
5895  * btree leaves.
5896  */
5897
5898 STATIC void
5899 xfs_bmap_check_leaf_extents(
5900         xfs_btree_cur_t         *cur,   /* btree cursor or null */
5901         xfs_inode_t             *ip,            /* incore inode pointer */
5902         int                     whichfork)      /* data or attr fork */
5903 {
5904         struct xfs_btree_block  *block; /* current btree block */
5905         xfs_fsblock_t           bno;    /* block # of "block" */
5906         xfs_buf_t               *bp;    /* buffer for "block" */
5907         int                     error;  /* error return value */
5908         xfs_extnum_t            i=0, j; /* index into the extents list */
5909         xfs_ifork_t             *ifp;   /* fork structure */
5910         int                     level;  /* btree level, for checking */
5911         xfs_mount_t             *mp;    /* file system mount structure */
5912         __be64                  *pp;    /* pointer to block address */
5913         xfs_bmbt_rec_t          *ep;    /* pointer to current extent */
5914         xfs_bmbt_rec_t          last = {0, 0}; /* last extent in prev block */
5915         xfs_bmbt_rec_t          *nextp; /* pointer to next extent */
5916         int                     bp_release = 0;
5917
5918         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE) {
5919                 return;
5920         }
5921
5922         bno = NULLFSBLOCK;
5923         mp = ip->i_mount;
5924         ifp = XFS_IFORK_PTR(ip, whichfork);
5925         block = ifp->if_broot;
5926         /*
5927          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
5928          */
5929         level = be16_to_cpu(block->bb_level);
5930         ASSERT(level > 0);
5931         xfs_check_block(block, mp, 1, ifp->if_broot_bytes);
5932         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
5933         bno = be64_to_cpu(*pp);
5934
5935         ASSERT(bno != NULLDFSBNO);
5936         ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
5937         ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
5938
5939         /*
5940          * Go down the tree until leaf level is reached, following the first
5941          * pointer (leftmost) at each level.
5942          */
5943         while (level-- > 0) {
5944                 /* See if buf is in cur first */
5945                 bp_release = 0;
5946                 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
5947                 if (!bp) {
5948                         bp_release = 1;
5949                         error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
5950                                                 XFS_BMAP_BTREE_REF,
5951                                                 &xfs_bmbt_buf_ops);
5952                         if (error)
5953                                 goto error_norelse;
5954                 }
5955                 block = XFS_BUF_TO_BLOCK(bp);
5956                 XFS_WANT_CORRUPTED_GOTO(
5957                         xfs_bmap_sanity_check(mp, bp, level),
5958                         error0);
5959                 if (level == 0)
5960                         break;
5961
5962                 /*
5963                  * Check this block for basic sanity (increasing keys and
5964                  * no duplicate blocks).
5965                  */
5966
5967                 xfs_check_block(block, mp, 0, 0);
5968                 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
5969                 bno = be64_to_cpu(*pp);
5970                 XFS_WANT_CORRUPTED_GOTO(XFS_FSB_SANITY_CHECK(mp, bno), error0);
5971                 if (bp_release) {
5972                         bp_release = 0;
5973                         xfs_trans_brelse(NULL, bp);
5974                 }
5975         }
5976
5977         /*
5978          * Here with bp and block set to the leftmost leaf node in the tree.
5979          */
5980         i = 0;
5981
5982         /*
5983          * Loop over all leaf nodes checking that all extents are in the right order.
5984          */
5985         for (;;) {
5986                 xfs_fsblock_t   nextbno;
5987                 xfs_extnum_t    num_recs;
5988
5989
5990                 num_recs = xfs_btree_get_numrecs(block);
5991
5992                 /*
5993                  * Read-ahead the next leaf block, if any.
5994                  */
5995
5996                 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
5997
5998                 /*
5999                  * Check all the extents to make sure they are OK.
6000                  * If we had a previous block, the last entry should
6001                  * conform with the first entry in this one.
6002                  */
6003
6004                 ep = XFS_BMBT_REC_ADDR(mp, block, 1);
6005                 if (i) {
6006                         ASSERT(xfs_bmbt_disk_get_startoff(&last) +
6007                                xfs_bmbt_disk_get_blockcount(&last) <=
6008                                xfs_bmbt_disk_get_startoff(ep));
6009                 }
6010                 for (j = 1; j < num_recs; j++) {
6011                         nextp = XFS_BMBT_REC_ADDR(mp, block, j + 1);
6012                         ASSERT(xfs_bmbt_disk_get_startoff(ep) +
6013                                xfs_bmbt_disk_get_blockcount(ep) <=
6014                                xfs_bmbt_disk_get_startoff(nextp));
6015                         ep = nextp;
6016                 }
6017
6018                 last = *ep;
6019                 i += num_recs;
6020                 if (bp_release) {
6021                         bp_release = 0;
6022                         xfs_trans_brelse(NULL, bp);
6023                 }
6024                 bno = nextbno;
6025                 /*
6026                  * If we've reached the end, stop.
6027                  */
6028                 if (bno == NULLFSBLOCK)
6029                         break;
6030
6031                 bp_release = 0;
6032                 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
6033                 if (!bp) {
6034                         bp_release = 1;
6035                         error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
6036                                                 XFS_BMAP_BTREE_REF,
6037                                                 &xfs_bmbt_buf_ops);
6038                         if (error)
6039                                 goto error_norelse;
6040                 }
6041                 block = XFS_BUF_TO_BLOCK(bp);
6042         }
6043         if (bp_release) {
6044                 bp_release = 0;
6045                 xfs_trans_brelse(NULL, bp);
6046         }
6047         return;
6048
6049 error0:
6050         xfs_warn(mp, "%s: at error0", __func__);
6051         if (bp_release)
6052                 xfs_trans_brelse(NULL, bp);
6053 error_norelse:
6054         xfs_warn(mp, "%s: BAD after btree leaves for %d extents",
6055                 __func__, i);
6056         panic("%s: CORRUPTED BTREE OR SOMETHING", __func__);
6057         return;
6058 }
6059 #endif
6060
6061 /*
6062  * Count fsblocks of the given fork.
6063  */
6064 int                                             /* error */
6065 xfs_bmap_count_blocks(
6066         xfs_trans_t             *tp,            /* transaction pointer */
6067         xfs_inode_t             *ip,            /* incore inode */
6068         int                     whichfork,      /* data or attr fork */
6069         int                     *count)         /* out: count of blocks */
6070 {
6071         struct xfs_btree_block  *block; /* current btree block */
6072         xfs_fsblock_t           bno;    /* block # of "block" */
6073         xfs_ifork_t             *ifp;   /* fork structure */
6074         int                     level;  /* btree level, for checking */
6075         xfs_mount_t             *mp;    /* file system mount structure */
6076         __be64                  *pp;    /* pointer to block address */
6077
6078         bno = NULLFSBLOCK;
6079         mp = ip->i_mount;
6080         ifp = XFS_IFORK_PTR(ip, whichfork);
6081         if ( XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ) {
6082                 xfs_bmap_count_leaves(ifp, 0,
6083                         ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t),
6084                         count);
6085                 return 0;
6086         }
6087
6088         /*
6089          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
6090          */
6091         block = ifp->if_broot;
6092         level = be16_to_cpu(block->bb_level);
6093         ASSERT(level > 0);
6094         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
6095         bno = be64_to_cpu(*pp);
6096         ASSERT(bno != NULLDFSBNO);
6097         ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
6098         ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
6099
6100         if (unlikely(xfs_bmap_count_tree(mp, tp, ifp, bno, level, count) < 0)) {
6101                 XFS_ERROR_REPORT("xfs_bmap_count_blocks(2)", XFS_ERRLEVEL_LOW,
6102                                  mp);
6103                 return XFS_ERROR(EFSCORRUPTED);
6104         }
6105
6106         return 0;
6107 }
6108
6109 /*
6110  * Recursively walks each level of a btree
6111  * to count total fsblocks is use.
6112  */
6113 STATIC int                                     /* error */
6114 xfs_bmap_count_tree(
6115         xfs_mount_t     *mp,            /* file system mount point */
6116         xfs_trans_t     *tp,            /* transaction pointer */
6117         xfs_ifork_t     *ifp,           /* inode fork pointer */
6118         xfs_fsblock_t   blockno,        /* file system block number */
6119         int             levelin,        /* level in btree */
6120         int             *count)         /* Count of blocks */
6121 {
6122         int                     error;
6123         xfs_buf_t               *bp, *nbp;
6124         int                     level = levelin;
6125         __be64                  *pp;
6126         xfs_fsblock_t           bno = blockno;
6127         xfs_fsblock_t           nextbno;
6128         struct xfs_btree_block  *block, *nextblock;
6129         int                     numrecs;
6130
6131         error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp, XFS_BMAP_BTREE_REF,
6132                                                 &xfs_bmbt_buf_ops);
6133         if (error)
6134                 return error;
6135         *count += 1;
6136         block = XFS_BUF_TO_BLOCK(bp);
6137
6138         if (--level) {
6139                 /* Not at node above leaves, count this level of nodes */
6140                 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
6141                 while (nextbno != NULLFSBLOCK) {
6142                         error = xfs_btree_read_bufl(mp, tp, nextbno, 0, &nbp,
6143                                                 XFS_BMAP_BTREE_REF,
6144                                                 &xfs_bmbt_buf_ops);
6145                         if (error)
6146                                 return error;
6147                         *count += 1;
6148                         nextblock = XFS_BUF_TO_BLOCK(nbp);
6149                         nextbno = be64_to_cpu(nextblock->bb_u.l.bb_rightsib);
6150                         xfs_trans_brelse(tp, nbp);
6151                 }
6152
6153                 /* Dive to the next level */
6154                 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
6155                 bno = be64_to_cpu(*pp);
6156                 if (unlikely((error =
6157                      xfs_bmap_count_tree(mp, tp, ifp, bno, level, count)) < 0)) {
6158                         xfs_trans_brelse(tp, bp);
6159                         XFS_ERROR_REPORT("xfs_bmap_count_tree(1)",
6160                                          XFS_ERRLEVEL_LOW, mp);
6161                         return XFS_ERROR(EFSCORRUPTED);
6162                 }
6163                 xfs_trans_brelse(tp, bp);
6164         } else {
6165                 /* count all level 1 nodes and their leaves */
6166                 for (;;) {
6167                         nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
6168                         numrecs = be16_to_cpu(block->bb_numrecs);
6169                         xfs_bmap_disk_count_leaves(mp, block, numrecs, count);
6170                         xfs_trans_brelse(tp, bp);
6171                         if (nextbno == NULLFSBLOCK)
6172                                 break;
6173                         bno = nextbno;
6174                         error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
6175                                                 XFS_BMAP_BTREE_REF,
6176                                                 &xfs_bmbt_buf_ops);
6177                         if (error)
6178                                 return error;
6179                         *count += 1;
6180                         block = XFS_BUF_TO_BLOCK(bp);
6181                 }
6182         }
6183         return 0;
6184 }
6185
6186 /*
6187  * Count leaf blocks given a range of extent records.
6188  */
6189 STATIC void
6190 xfs_bmap_count_leaves(
6191         xfs_ifork_t             *ifp,
6192         xfs_extnum_t            idx,
6193         int                     numrecs,
6194         int                     *count)
6195 {
6196         int             b;
6197
6198         for (b = 0; b < numrecs; b++) {
6199                 xfs_bmbt_rec_host_t *frp = xfs_iext_get_ext(ifp, idx + b);
6200                 *count += xfs_bmbt_get_blockcount(frp);
6201         }
6202 }
6203
6204 /*
6205  * Count leaf blocks given a range of extent records originally
6206  * in btree format.
6207  */
6208 STATIC void
6209 xfs_bmap_disk_count_leaves(
6210         struct xfs_mount        *mp,
6211         struct xfs_btree_block  *block,
6212         int                     numrecs,
6213         int                     *count)
6214 {
6215         int             b;
6216         xfs_bmbt_rec_t  *frp;
6217
6218         for (b = 1; b <= numrecs; b++) {
6219                 frp = XFS_BMBT_REC_ADDR(mp, block, b);
6220                 *count += xfs_bmbt_disk_get_blockcount(frp);
6221         }
6222 }
6223
6224 /*
6225  * dead simple method of punching delalyed allocation blocks from a range in
6226  * the inode. Walks a block at a time so will be slow, but is only executed in
6227  * rare error cases so the overhead is not critical. This will alays punch out
6228  * both the start and end blocks, even if the ranges only partially overlap
6229  * them, so it is up to the caller to ensure that partial blocks are not
6230  * passed in.
6231  */
6232 int
6233 xfs_bmap_punch_delalloc_range(
6234         struct xfs_inode        *ip,
6235         xfs_fileoff_t           start_fsb,
6236         xfs_fileoff_t           length)
6237 {
6238         xfs_fileoff_t           remaining = length;
6239         int                     error = 0;
6240
6241         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
6242
6243         do {
6244                 int             done;
6245                 xfs_bmbt_irec_t imap;
6246                 int             nimaps = 1;
6247                 xfs_fsblock_t   firstblock;
6248                 xfs_bmap_free_t flist;
6249
6250                 /*
6251                  * Map the range first and check that it is a delalloc extent
6252                  * before trying to unmap the range. Otherwise we will be
6253                  * trying to remove a real extent (which requires a
6254                  * transaction) or a hole, which is probably a bad idea...
6255                  */
6256                 error = xfs_bmapi_read(ip, start_fsb, 1, &imap, &nimaps,
6257                                        XFS_BMAPI_ENTIRE);
6258
6259                 if (error) {
6260                         /* something screwed, just bail */
6261                         if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
6262                                 xfs_alert(ip->i_mount,
6263                         "Failed delalloc mapping lookup ino %lld fsb %lld.",
6264                                                 ip->i_ino, start_fsb);
6265                         }
6266                         break;
6267                 }
6268                 if (!nimaps) {
6269                         /* nothing there */
6270                         goto next_block;
6271                 }
6272                 if (imap.br_startblock != DELAYSTARTBLOCK) {
6273                         /* been converted, ignore */
6274                         goto next_block;
6275                 }
6276                 WARN_ON(imap.br_blockcount == 0);
6277
6278                 /*
6279                  * Note: while we initialise the firstblock/flist pair, they
6280                  * should never be used because blocks should never be
6281                  * allocated or freed for a delalloc extent and hence we need
6282                  * don't cancel or finish them after the xfs_bunmapi() call.
6283                  */
6284                 xfs_bmap_init(&flist, &firstblock);
6285                 error = xfs_bunmapi(NULL, ip, start_fsb, 1, 0, 1, &firstblock,
6286                                         &flist, &done);
6287                 if (error)
6288                         break;
6289
6290                 ASSERT(!flist.xbf_count && !flist.xbf_first);
6291 next_block:
6292                 start_fsb++;
6293                 remaining--;
6294         } while(remaining > 0);
6295
6296         return error;
6297 }
6298
6299 /*
6300  * Convert the given file system block to a disk block.  We have to treat it
6301  * differently based on whether the file is a real time file or not, because the
6302  * bmap code does.
6303  */
6304 xfs_daddr_t
6305 xfs_fsb_to_db(struct xfs_inode *ip, xfs_fsblock_t fsb)
6306 {
6307         return (XFS_IS_REALTIME_INODE(ip) ? \
6308                  (xfs_daddr_t)XFS_FSB_TO_BB((ip)->i_mount, (fsb)) : \
6309                  XFS_FSB_TO_DADDR((ip)->i_mount, (fsb)));
6310 }