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
 152 /*
 153  * Search the extents list for the inode, for the extent containing bno.
 154  * If bno lies in a hole, point to the next entry.  If bno lies past eof,
 155  * *eofp will be set, and *prevp will contain the last entry (null if none).
 156  * Else, *lastxp will be set to the index of the found
 157  * entry; *gotp will contain the entry.
 158  */
 159 STATIC xfs_bmbt_rec_host_t *            /* pointer to found extent entry */
 160 xfs_bmap_search_extents(
 161         xfs_inode_t     *ip,            /* incore inode pointer */
 162         xfs_fileoff_t   bno,            /* block number searched for */
 163         int             whichfork,      /* data or attr fork */
 164         int             *eofp,          /* out: end of file found */
 165         xfs_extnum_t    *lastxp,        /* out: last extent index */
 166         xfs_bmbt_irec_t *gotp,          /* out: extent entry found */
 167         xfs_bmbt_irec_t *prevp);        /* out: previous extent entry found */
 168
 169 /*
 170  * Compute the worst-case number of indirect blocks that will be used
 171  * for ip's delayed extent of length "len".
 172  */
 173 STATIC xfs_filblks_t
 174 xfs_bmap_worst_indlen(
 175         xfs_inode_t             *ip,    /* incore inode pointer */
 176         xfs_filblks_t           len);   /* delayed extent length */
 177
 178 #ifdef DEBUG
 179 /*
 180  * Perform various validation checks on the values being returned
 181  * from xfs_bmapi().
 182  */
 183 STATIC void
 184 xfs_bmap_validate_ret(
 185         xfs_fileoff_t           bno,
 186         xfs_filblks_t           len,
 187         int                     flags,
 188         xfs_bmbt_irec_t         *mval,
 189         int                     nmap,
 190         int                     ret_nmap);
 191 #else
 192 #define xfs_bmap_validate_ret(bno,len,flags,mval,onmap,nmap)
 193 #endif /* DEBUG */
 194
 195 STATIC int
 196 xfs_bmap_count_tree(
 197         xfs_mount_t     *mp,
 198         xfs_trans_t     *tp,
 199         xfs_ifork_t     *ifp,
 200         xfs_fsblock_t   blockno,
 201         int             levelin,
 202         int             *count);
 203
 204 STATIC void
 205 xfs_bmap_count_leaves(
 206         xfs_ifork_t             *ifp,
 207         xfs_extnum_t            idx,
 208         int                     numrecs,
 209         int                     *count);
 210
 211 STATIC void
 212 xfs_bmap_disk_count_leaves(
 213         struct xfs_mount        *mp,
 214         struct xfs_btree_block  *block,
 215         int                     numrecs,
 216         int                     *count);
 217
 218 /*
 219  * Bmap internal routines.
 220  */
 221
 222 STATIC int                              /* error */
 223 xfs_bmbt_lookup_eq(
 224         struct xfs_btree_cur    *cur,
 225         xfs_fileoff_t           off,
 226         xfs_fsblock_t           bno,
 227         xfs_filblks_t           len,
 228         int                     *stat)  /* success/failure */
 229 {
 230         cur->bc_rec.b.br_startoff = off;
 231         cur->bc_rec.b.br_startblock = bno;
 232         cur->bc_rec.b.br_blockcount = len;
 233         return xfs_btree_lookup(cur, XFS_LOOKUP_EQ, stat);
 234 }
 235
 236 STATIC int                              /* error */
 237 xfs_bmbt_lookup_ge(
 238         struct xfs_btree_cur    *cur,
 239         xfs_fileoff_t           off,
 240         xfs_fsblock_t           bno,
 241         xfs_filblks_t           len,
 242         int                     *stat)  /* success/failure */
 243 {
 244         cur->bc_rec.b.br_startoff = off;
 245         cur->bc_rec.b.br_startblock = bno;
 246         cur->bc_rec.b.br_blockcount = len;
 247         return xfs_btree_lookup(cur, XFS_LOOKUP_GE, stat);
 248 }
 249
 250 /*
 251  * Check if the inode needs to be converted to btree format.
 252  */
 253 static inline bool xfs_bmap_needs_btree(struct xfs_inode *ip, int whichfork)
 254 {
 255         return XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
 256                 XFS_IFORK_NEXTENTS(ip, whichfork) >
 257                         XFS_IFORK_MAXEXT(ip, whichfork);
 258 }
 259
 260 /*
 261  * Check if the inode should be converted to extent format.
 262  */
 263 static inline bool xfs_bmap_wants_extents(struct xfs_inode *ip, int whichfork)
 264 {
 265         return XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE &&
 266                 XFS_IFORK_NEXTENTS(ip, whichfork) <=
 267                         XFS_IFORK_MAXEXT(ip, whichfork);
 268 }
 269
 270 /*
 271  * Update the record referred to by cur to the value given
 272  * by [off, bno, len, state].
 273  * This either works (return 0) or gets an EFSCORRUPTED error.
 274  */
 275 STATIC int
 276 xfs_bmbt_update(
 277         struct xfs_btree_cur    *cur,
 278         xfs_fileoff_t           off,
 279         xfs_fsblock_t           bno,
 280         xfs_filblks_t           len,
 281         xfs_exntst_t            state)
 282 {
 283         union xfs_btree_rec     rec;
 284
 285         xfs_bmbt_disk_set_allf(&rec.bmbt, off, bno, len, state);
 286         return xfs_btree_update(cur, &rec);
 287 }
 288
 289 /*
 290  * Called from xfs_bmap_add_attrfork to handle btree format files.
 291  */
 292 STATIC int                                      /* error */
 293 xfs_bmap_add_attrfork_btree(
 294         xfs_trans_t             *tp,            /* transaction pointer */
 295         xfs_inode_t             *ip,            /* incore inode pointer */
 296         xfs_fsblock_t           *firstblock,    /* first block allocated */
 297         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
 298         int                     *flags)         /* inode logging flags */
 299 {
 300         xfs_btree_cur_t         *cur;           /* btree cursor */
 301         int                     error;          /* error return value */
 302         xfs_mount_t             *mp;            /* file system mount struct */
 303         int                     stat;           /* newroot status */
 304
 305         mp = ip->i_mount;
 306         if (ip->i_df.if_broot_bytes <= XFS_IFORK_DSIZE(ip))
 307                 *flags |= XFS_ILOG_DBROOT;
 308         else {
 309                 cur = xfs_bmbt_init_cursor(mp, tp, ip, XFS_DATA_FORK);
 310                 cur->bc_private.b.flist = flist;
 311                 cur->bc_private.b.firstblock = *firstblock;
 312                 if ((error = xfs_bmbt_lookup_ge(cur, 0, 0, 0, &stat)))
 313                         goto error0;
 314                 /* must be at least one entry */
 315                 XFS_WANT_CORRUPTED_GOTO(stat == 1, error0);
 316                 if ((error = xfs_btree_new_iroot(cur, flags, &stat)))
 317                         goto error0;
 318                 if (stat == 0) {
 319                         xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
 320                         return XFS_ERROR(ENOSPC);
 321                 }
 322                 *firstblock = cur->bc_private.b.firstblock;
 323                 cur->bc_private.b.allocated = 0;
 324                 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
 325         }
 326         return 0;
 327 error0:
 328         xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
 329         return error;
 330 }
 331
 332 /*
 333  * Called from xfs_bmap_add_attrfork to handle extents format files.
 334  */
 335 STATIC int                                      /* error */
 336 xfs_bmap_add_attrfork_extents(
 337         xfs_trans_t             *tp,            /* transaction pointer */
 338         xfs_inode_t             *ip,            /* incore inode pointer */
 339         xfs_fsblock_t           *firstblock,    /* first block allocated */
 340         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
 341         int                     *flags)         /* inode logging flags */
 342 {
 343         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
 344         int                     error;          /* error return value */
 345
 346         if (ip->i_d.di_nextents * sizeof(xfs_bmbt_rec_t) <= XFS_IFORK_DSIZE(ip))
 347                 return 0;
 348         cur = NULL;
 349         error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist, &cur, 0,
 350                 flags, XFS_DATA_FORK);
 351         if (cur) {
 352                 cur->bc_private.b.allocated = 0;
 353                 xfs_btree_del_cursor(cur,
 354                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
 355         }
 356         return error;
 357 }
 358
 359 /*
 360  * Called from xfs_bmap_add_attrfork to handle local format files.
 361  */
 362 STATIC int                                      /* error */
 363 xfs_bmap_add_attrfork_local(
 364         xfs_trans_t             *tp,            /* transaction pointer */
 365         xfs_inode_t             *ip,            /* incore inode pointer */
 366         xfs_fsblock_t           *firstblock,    /* first block allocated */
 367         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
 368         int                     *flags)         /* inode logging flags */
 369 {
 370         xfs_da_args_t           dargs;          /* args for dir/attr code */
 371         int                     error;          /* error return value */
 372         xfs_mount_t             *mp;            /* mount structure pointer */
 373
 374         if (ip->i_df.if_bytes <= XFS_IFORK_DSIZE(ip))
 375                 return 0;
 376         if (S_ISDIR(ip->i_d.di_mode)) {
 377                 mp = ip->i_mount;
 378                 memset(&dargs, 0, sizeof(dargs));
 379                 dargs.dp = ip;
 380                 dargs.firstblock = firstblock;
 381                 dargs.flist = flist;
 382                 dargs.total = mp->m_dirblkfsbs;
 383                 dargs.whichfork = XFS_DATA_FORK;
 384                 dargs.trans = tp;
 385                 error = xfs_dir2_sf_to_block(&dargs);
 386         } else
 387                 error = xfs_bmap_local_to_extents(tp, ip, firstblock, 1, flags,
 388                         XFS_DATA_FORK);
 389         return error;
 390 }
 391
 392 /*
 393  * Convert a delayed allocation to a real allocation.
 394  */
 395 STATIC int                              /* error */
 396 xfs_bmap_add_extent_delay_real(
 397         struct xfs_bmalloca     *bma)
 398 {
 399         struct xfs_bmbt_irec    *new = &bma->got;
 400         int                     diff;   /* temp value */
 401         xfs_bmbt_rec_host_t     *ep;    /* extent entry for idx */
 402         int                     error;  /* error return value */
 403         int                     i;      /* temp state */
 404         xfs_ifork_t             *ifp;   /* inode fork pointer */
 405         xfs_fileoff_t           new_endoff;     /* end offset of new entry */
 406         xfs_bmbt_irec_t         r[3];   /* neighbor extent entries */
 407                                         /* left is 0, right is 1, prev is 2 */
 408         int                     rval=0; /* return value (logging flags) */
 409         int                     state = 0;/* state bits, accessed thru macros */
 410         xfs_filblks_t           da_new; /* new count del alloc blocks used */
 411         xfs_filblks_t           da_old; /* old count del alloc blocks used */
 412         xfs_filblks_t           temp=0; /* value for da_new calculations */
 413         xfs_filblks_t           temp2=0;/* value for da_new calculations */
 414         int                     tmp_rval;       /* partial logging flags */
 415
 416         ifp = XFS_IFORK_PTR(bma->ip, XFS_DATA_FORK);
 417
 418         ASSERT(bma->idx >= 0);
 419         ASSERT(bma->idx <= ifp->if_bytes / sizeof(struct xfs_bmbt_rec));
 420         ASSERT(!isnullstartblock(new->br_startblock));
 421         ASSERT(!bma->cur ||
 422                (bma->cur->bc_private.b.flags & XFS_BTCUR_BPRV_WASDEL));
 423
 424         XFS_STATS_INC(xs_add_exlist);
 425
 426 #define LEFT            r[0]
 427 #define RIGHT           r[1]
 428 #define PREV            r[2]
 429
 430         /*
 431          * Set up a bunch of variables to make the tests simpler.
 432          */
 433         ep = xfs_iext_get_ext(ifp, bma->idx);
 434         xfs_bmbt_get_all(ep, &PREV);
 435         new_endoff = new->br_startoff + new->br_blockcount;
 436         ASSERT(PREV.br_startoff <= new->br_startoff);
 437         ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
 438
 439         da_old = startblockval(PREV.br_startblock);
 440         da_new = 0;
 441
 442         /*
 443          * Set flags determining what part of the previous delayed allocation
 444          * extent is being replaced by a real allocation.
 445          */
 446         if (PREV.br_startoff == new->br_startoff)
 447                 state |= BMAP_LEFT_FILLING;
 448         if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
 449                 state |= BMAP_RIGHT_FILLING;
 450
 451         /*
 452          * Check and set flags if this segment has a left neighbor.
 453          * Don't set contiguous if the combined extent would be too large.
 454          */
 455         if (bma->idx > 0) {
 456                 state |= BMAP_LEFT_VALID;
 457                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx - 1), &LEFT);
 458
 459                 if (isnullstartblock(LEFT.br_startblock))
 460                         state |= BMAP_LEFT_DELAY;
 461         }
 462
 463         if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
 464             LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
 465             LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
 466             LEFT.br_state == new->br_state &&
 467             LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
 468                 state |= BMAP_LEFT_CONTIG;
 469
 470         /*
 471          * Check and set flags if this segment has a right neighbor.
 472          * Don't set contiguous if the combined extent would be too large.
 473          * Also check for all-three-contiguous being too large.
 474          */
 475         if (bma->idx < bma->ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1) {
 476                 state |= BMAP_RIGHT_VALID;
 477                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx + 1), &RIGHT);
 478
 479                 if (isnullstartblock(RIGHT.br_startblock))
 480                         state |= BMAP_RIGHT_DELAY;
 481         }
 482
 483         if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
 484             new_endoff == RIGHT.br_startoff &&
 485             new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
 486             new->br_state == RIGHT.br_state &&
 487             new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
 488             ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
 489                        BMAP_RIGHT_FILLING)) !=
 490                       (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
 491                        BMAP_RIGHT_FILLING) ||
 492              LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
 493                         <= MAXEXTLEN))
 494                 state |= BMAP_RIGHT_CONTIG;
 495
 496         error = 0;
 497         /*
 498          * Switch out based on the FILLING and CONTIG state bits.
 499          */
 500         switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
 501                          BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
 502         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
 503              BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
 504                 /*
 505                  * Filling in all of a previously delayed allocation extent.
 506                  * The left and right neighbors are both contiguous with new.
 507                  */
 508                 bma->idx--;
 509                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
 510                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
 511                         LEFT.br_blockcount + PREV.br_blockcount +
 512                         RIGHT.br_blockcount);
 513                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
 514
 515                 xfs_iext_remove(bma->ip, bma->idx + 1, 2, state);
 516                 bma->ip->i_d.di_nextents--;
 517                 if (bma->cur == NULL)
 518                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
 519                 else {
 520                         rval = XFS_ILOG_CORE;
 521                         error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff,
 522                                         RIGHT.br_startblock,
 523                                         RIGHT.br_blockcount, &i);
 524                         if (error)
 525                                 goto done;
 526                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 527                         error = xfs_btree_delete(bma->cur, &i);
 528                         if (error)
 529                                 goto done;
 530                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 531                         error = xfs_btree_decrement(bma->cur, 0, &i);
 532                         if (error)
 533                                 goto done;
 534                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 535                         error = xfs_bmbt_update(bma->cur, LEFT.br_startoff,
 536                                         LEFT.br_startblock,
 537                                         LEFT.br_blockcount +
 538                                         PREV.br_blockcount +
 539                                         RIGHT.br_blockcount, LEFT.br_state);
 540                         if (error)
 541                                 goto done;
 542                 }
 543                 break;
 544
 545         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
 546                 /*
 547                  * Filling in all of a previously delayed allocation extent.
 548                  * The left neighbor is contiguous, the right is not.
 549                  */
 550                 bma->idx--;
 551
 552                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
 553                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
 554                         LEFT.br_blockcount + PREV.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, 1, state);
 558                 if (bma->cur == NULL)
 559                         rval = XFS_ILOG_DEXT;
 560                 else {
 561                         rval = 0;
 562                         error = xfs_bmbt_lookup_eq(bma->cur, LEFT.br_startoff,
 563                                         LEFT.br_startblock, LEFT.br_blockcount,
 564                                         &i);
 565                         if (error)
 566                                 goto done;
 567                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 568                         error = xfs_bmbt_update(bma->cur, LEFT.br_startoff,
 569                                         LEFT.br_startblock,
 570                                         LEFT.br_blockcount +
 571                                         PREV.br_blockcount, LEFT.br_state);
 572                         if (error)
 573                                 goto done;
 574                 }
 575                 break;
 576
 577         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
 578                 /*
 579                  * Filling in all of a previously delayed allocation extent.
 580                  * The right neighbor is contiguous, the left is not.
 581                  */
 582                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
 583                 xfs_bmbt_set_startblock(ep, new->br_startblock);
 584                 xfs_bmbt_set_blockcount(ep,
 585                         PREV.br_blockcount + RIGHT.br_blockcount);
 586                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
 587
 588                 xfs_iext_remove(bma->ip, bma->idx + 1, 1, state);
 589                 if (bma->cur == NULL)
 590                         rval = XFS_ILOG_DEXT;
 591                 else {
 592                         rval = 0;
 593                         error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff,
 594                                         RIGHT.br_startblock,
 595                                         RIGHT.br_blockcount, &i);
 596                         if (error)
 597                                 goto done;
 598                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 599                         error = xfs_bmbt_update(bma->cur, PREV.br_startoff,
 600                                         new->br_startblock,
 601                                         PREV.br_blockcount +
 602                                         RIGHT.br_blockcount, PREV.br_state);
 603                         if (error)
 604                                 goto done;
 605                 }
 606                 break;
 607
 608         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
 609                 /*
 610                  * Filling in all of a previously delayed allocation extent.
 611                  * Neither the left nor right neighbors are contiguous with
 612                  * the new one.
 613                  */
 614                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
 615                 xfs_bmbt_set_startblock(ep, new->br_startblock);
 616                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
 617
 618                 bma->ip->i_d.di_nextents++;
 619                 if (bma->cur == NULL)
 620                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
 621                 else {
 622                         rval = XFS_ILOG_CORE;
 623                         error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
 624                                         new->br_startblock, new->br_blockcount,
 625                                         &i);
 626                         if (error)
 627                                 goto done;
 628                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
 629                         bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
 630                         error = xfs_btree_insert(bma->cur, &i);
 631                         if (error)
 632                                 goto done;
 633                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 634                 }
 635                 break;
 636
 637         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
 638                 /*
 639                  * Filling in the first part of a previous delayed allocation.
 640                  * The left neighbor is contiguous.
 641                  */
 642                 trace_xfs_bmap_pre_update(bma->ip, bma->idx - 1, state, _THIS_IP_);
 643                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx - 1),
 644                         LEFT.br_blockcount + new->br_blockcount);
 645                 xfs_bmbt_set_startoff(ep,
 646                         PREV.br_startoff + new->br_blockcount);
 647                 trace_xfs_bmap_post_update(bma->ip, bma->idx - 1, state, _THIS_IP_);
 648
 649                 temp = PREV.br_blockcount - new->br_blockcount;
 650                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
 651                 xfs_bmbt_set_blockcount(ep, temp);
 652                 if (bma->cur == NULL)
 653                         rval = XFS_ILOG_DEXT;
 654                 else {
 655                         rval = 0;
 656                         error = xfs_bmbt_lookup_eq(bma->cur, LEFT.br_startoff,
 657                                         LEFT.br_startblock, LEFT.br_blockcount,
 658                                         &i);
 659                         if (error)
 660                                 goto done;
 661                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 662                         error = xfs_bmbt_update(bma->cur, LEFT.br_startoff,
 663                                         LEFT.br_startblock,
 664                                         LEFT.br_blockcount +
 665                                         new->br_blockcount,
 666                                         LEFT.br_state);
 667                         if (error)
 668                                 goto done;
 669                 }
 670                 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
 671                         startblockval(PREV.br_startblock));
 672                 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
 673                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
 674
 675                 bma->idx--;
 676                 break;
 677
 678         case BMAP_LEFT_FILLING:
 679                 /*
 680                  * Filling in the first part of a previous delayed allocation.
 681                  * The left neighbor is not contiguous.
 682                  */
 683                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
 684                 xfs_bmbt_set_startoff(ep, new_endoff);
 685                 temp = PREV.br_blockcount - new->br_blockcount;
 686                 xfs_bmbt_set_blockcount(ep, temp);
 687                 xfs_iext_insert(bma->ip, bma->idx, 1, new, state);
 688                 bma->ip->i_d.di_nextents++;
 689                 if (bma->cur == NULL)
 690                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
 691                 else {
 692                         rval = XFS_ILOG_CORE;
 693                         error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
 694                                         new->br_startblock, new->br_blockcount,
 695                                         &i);
 696                         if (error)
 697                                 goto done;
 698                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
 699                         bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
 700                         error = xfs_btree_insert(bma->cur, &i);
 701                         if (error)
 702                                 goto done;
 703                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 704                 }
 705
 706                 if (xfs_bmap_needs_btree(bma->ip, XFS_DATA_FORK)) {
 707                         error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
 708                                         bma->firstblock, bma->flist,
 709                                         &bma->cur, 1, &tmp_rval, XFS_DATA_FORK);
 710                         rval |= tmp_rval;
 711                         if (error)
 712                                 goto done;
 713                 }
 714                 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
 715                         startblockval(PREV.br_startblock) -
 716                         (bma->cur ? bma->cur->bc_private.b.allocated : 0));
 717                 ep = xfs_iext_get_ext(ifp, bma->idx + 1);
 718                 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
 719                 trace_xfs_bmap_post_update(bma->ip, bma->idx + 1, state, _THIS_IP_);
 720                 break;
 721
 722         case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
 723                 /*
 724                  * Filling in the last part of a previous delayed allocation.
 725                  * The right neighbor is contiguous with the new allocation.
 726                  */
 727                 temp = PREV.br_blockcount - new->br_blockcount;
 728                 trace_xfs_bmap_pre_update(bma->ip, bma->idx + 1, state, _THIS_IP_);
 729                 xfs_bmbt_set_blockcount(ep, temp);
 730                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, bma->idx + 1),
 731                         new->br_startoff, new->br_startblock,
 732                         new->br_blockcount + RIGHT.br_blockcount,
 733                         RIGHT.br_state);
 734                 trace_xfs_bmap_post_update(bma->ip, bma->idx + 1, state, _THIS_IP_);
 735                 if (bma->cur == NULL)
 736                         rval = XFS_ILOG_DEXT;
 737                 else {
 738                         rval = 0;
 739                         error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff,
 740                                         RIGHT.br_startblock,
 741                                         RIGHT.br_blockcount, &i);
 742                         if (error)
 743                                 goto done;
 744                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 745                         error = xfs_bmbt_update(bma->cur, new->br_startoff,
 746                                         new->br_startblock,
 747                                         new->br_blockcount +
 748                                         RIGHT.br_blockcount,
 749                                         RIGHT.br_state);
 750                         if (error)
 751                                 goto done;
 752                 }
 753
 754                 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
 755                         startblockval(PREV.br_startblock));
 756                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
 757                 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
 758                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
 759
 760                 bma->idx++;
 761                 break;
 762
 763         case BMAP_RIGHT_FILLING:
 764                 /*
 765                  * Filling in the last part of a previous delayed allocation.
 766                  * The right neighbor is not contiguous.
 767                  */
 768                 temp = PREV.br_blockcount - new->br_blockcount;
 769                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
 770                 xfs_bmbt_set_blockcount(ep, temp);
 771                 xfs_iext_insert(bma->ip, bma->idx + 1, 1, new, state);
 772                 bma->ip->i_d.di_nextents++;
 773                 if (bma->cur == NULL)
 774                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
 775                 else {
 776                         rval = XFS_ILOG_CORE;
 777                         error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
 778                                         new->br_startblock, new->br_blockcount,
 779                                         &i);
 780                         if (error)
 781                                 goto done;
 782                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
 783                         bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
 784                         error = xfs_btree_insert(bma->cur, &i);
 785                         if (error)
 786                                 goto done;
 787                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 788                 }
 789
 790                 if (xfs_bmap_needs_btree(bma->ip, XFS_DATA_FORK)) {
 791                         error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
 792                                 bma->firstblock, bma->flist, &bma->cur, 1,
 793                                 &tmp_rval, XFS_DATA_FORK);
 794                         rval |= tmp_rval;
 795                         if (error)
 796                                 goto done;
 797                 }
 798                 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
 799                         startblockval(PREV.br_startblock) -
 800                         (bma->cur ? bma->cur->bc_private.b.allocated : 0));
 801                 ep = xfs_iext_get_ext(ifp, bma->idx);
 802                 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
 803                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
 804
 805                 bma->idx++;
 806                 break;
 807
 808         case 0:
 809                 /*
 810                  * Filling in the middle part of a previous delayed allocation.
 811                  * Contiguity is impossible here.
 812                  * This case is avoided almost all the time.
 813                  *
 814                  * We start with a delayed allocation:
 815                  *
 816                  * +ddddddddddddddddddddddddddddddddddddddddddddddddddddddd+
 817                  *  PREV @ idx
 818                  *
 819                  * and we are allocating:
 820                  *                     +rrrrrrrrrrrrrrrrr+
 821                  *                            new
 822                  *
 823                  * and we set it up for insertion as:
 824                  * +ddddddddddddddddddd+rrrrrrrrrrrrrrrrr+ddddddddddddddddd+
 825                  *                            new
 826                  *  PREV @ idx          LEFT              RIGHT
 827                  *                      inserted at idx + 1
 828                  */
 829                 temp = new->br_startoff - PREV.br_startoff;
 830                 temp2 = PREV.br_startoff + PREV.br_blockcount - new_endoff;
 831                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, 0, _THIS_IP_);
 832                 xfs_bmbt_set_blockcount(ep, temp);      /* truncate PREV */
 833                 LEFT = *new;
 834                 RIGHT.br_state = PREV.br_state;
 835                 RIGHT.br_startblock = nullstartblock(
 836                                 (int)xfs_bmap_worst_indlen(bma->ip, temp2));
 837                 RIGHT.br_startoff = new_endoff;
 838                 RIGHT.br_blockcount = temp2;
 839                 /* insert LEFT (r[0]) and RIGHT (r[1]) at the same time */
 840                 xfs_iext_insert(bma->ip, bma->idx + 1, 2, &LEFT, state);
 841                 bma->ip->i_d.di_nextents++;
 842                 if (bma->cur == NULL)
 843                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
 844                 else {
 845                         rval = XFS_ILOG_CORE;
 846                         error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
 847                                         new->br_startblock, new->br_blockcount,
 848                                         &i);
 849                         if (error)
 850                                 goto done;
 851                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
 852                         bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
 853                         error = xfs_btree_insert(bma->cur, &i);
 854                         if (error)
 855                                 goto done;
 856                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
 857                 }
 858
 859                 if (xfs_bmap_needs_btree(bma->ip, XFS_DATA_FORK)) {
 860                         error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
 861                                         bma->firstblock, bma->flist, &bma->cur,
 862                                         1, &tmp_rval, XFS_DATA_FORK);
 863                         rval |= tmp_rval;
 864                         if (error)
 865                                 goto done;
 866                 }
 867                 temp = xfs_bmap_worst_indlen(bma->ip, temp);
 868                 temp2 = xfs_bmap_worst_indlen(bma->ip, temp2);
 869                 diff = (int)(temp + temp2 - startblockval(PREV.br_startblock) -
 870                         (bma->cur ? bma->cur->bc_private.b.allocated : 0));
 871                 if (diff > 0) {
 872                         error = xfs_icsb_modify_counters(bma->ip->i_mount,
 873                                         XFS_SBS_FDBLOCKS,
 874                                         -((int64_t)diff), 0);
 875                         ASSERT(!error);
 876                         if (error)
 877                                 goto done;
 878                 }
 879
 880                 ep = xfs_iext_get_ext(ifp, bma->idx);
 881                 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
 882                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
 883                 trace_xfs_bmap_pre_update(bma->ip, bma->idx + 2, state, _THIS_IP_);
 884                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, bma->idx + 2),
 885                         nullstartblock((int)temp2));
 886                 trace_xfs_bmap_post_update(bma->ip, bma->idx + 2, state, _THIS_IP_);
 887
 888                 bma->idx++;
 889                 da_new = temp + temp2;
 890                 break;
 891
 892         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
 893         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
 894         case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
 895         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
 896         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
 897         case BMAP_LEFT_CONTIG:
 898         case BMAP_RIGHT_CONTIG:
 899                 /*
 900                  * These cases are all impossible.
 901                  */
 902                 ASSERT(0);
 903         }
 904
 905         /* convert to a btree if necessary */
 906         if (xfs_bmap_needs_btree(bma->ip, XFS_DATA_FORK)) {
 907                 int     tmp_logflags;   /* partial log flag return val */
 908
 909                 ASSERT(bma->cur == NULL);
 910                 error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
 911                                 bma->firstblock, bma->flist, &bma->cur,
 912                                 da_old > 0, &tmp_logflags, XFS_DATA_FORK);
 913                 bma->logflags |= tmp_logflags;
 914                 if (error)
 915                         goto done;
 916         }
 917
 918         /* adjust for changes in reserved delayed indirect blocks */
 919         if (da_old || da_new) {
 920                 temp = da_new;
 921                 if (bma->cur)
 922                         temp += bma->cur->bc_private.b.allocated;
 923                 ASSERT(temp <= da_old);
 924                 if (temp < da_old)
 925                         xfs_icsb_modify_counters(bma->ip->i_mount,
 926                                         XFS_SBS_FDBLOCKS,
 927                                         (int64_t)(da_old - temp), 0);
 928         }
 929
 930         /* clear out the allocated field, done with it now in any case. */
 931         if (bma->cur)
 932                 bma->cur->bc_private.b.allocated = 0;
 933
 934         xfs_bmap_check_leaf_extents(bma->cur, bma->ip, XFS_DATA_FORK);
 935 done:
 936         bma->logflags |= rval;
 937         return error;
 938 #undef  LEFT
 939 #undef  RIGHT
 940 #undef  PREV
 941 }
 942
 943 /*
 944  * Convert an unwritten allocation to a real allocation or vice versa.
 945  */
 946 STATIC int                              /* error */
 947 xfs_bmap_add_extent_unwritten_real(
 948         struct xfs_trans        *tp,
 949         xfs_inode_t             *ip,    /* incore inode pointer */
 950         xfs_extnum_t            *idx,   /* extent number to update/insert */
 951         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
 952         xfs_bmbt_irec_t         *new,   /* new data to add to file extents */
 953         xfs_fsblock_t           *first, /* pointer to firstblock variable */
 954         xfs_bmap_free_t         *flist, /* list of extents to be freed */
 955         int                     *logflagsp) /* inode logging flags */
 956 {
 957         xfs_btree_cur_t         *cur;   /* btree cursor */
 958         xfs_bmbt_rec_host_t     *ep;    /* extent entry for idx */
 959         int                     error;  /* error return value */
 960         int                     i;      /* temp state */
 961         xfs_ifork_t             *ifp;   /* inode fork pointer */
 962         xfs_fileoff_t           new_endoff;     /* end offset of new entry */
 963         xfs_exntst_t            newext; /* new extent state */
 964         xfs_exntst_t            oldext; /* old extent state */
 965         xfs_bmbt_irec_t         r[3];   /* neighbor extent entries */
 966                                         /* left is 0, right is 1, prev is 2 */
 967         int                     rval=0; /* return value (logging flags) */
 968         int                     state = 0;/* state bits, accessed thru macros */
 969
 970         *logflagsp = 0;
 971
 972         cur = *curp;
 973         ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
 974
 975         ASSERT(*idx >= 0);
 976         ASSERT(*idx <= ifp->if_bytes / sizeof(struct xfs_bmbt_rec));
 977         ASSERT(!isnullstartblock(new->br_startblock));
 978
 979         XFS_STATS_INC(xs_add_exlist);
 980
 981 #define LEFT            r[0]
 982 #define RIGHT           r[1]
 983 #define PREV            r[2]
 984
 985         /*
 986          * Set up a bunch of variables to make the tests simpler.
 987          */
 988         error = 0;
 989         ep = xfs_iext_get_ext(ifp, *idx);
 990         xfs_bmbt_get_all(ep, &PREV);
 991         newext = new->br_state;
 992         oldext = (newext == XFS_EXT_UNWRITTEN) ?
 993                 XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
 994         ASSERT(PREV.br_state == oldext);
 995         new_endoff = new->br_startoff + new->br_blockcount;
 996         ASSERT(PREV.br_startoff <= new->br_startoff);
 997         ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
 998
 999         /*
1000          * Set flags determining what part of the previous oldext allocation
1001          * extent is being replaced by a newext allocation.
1002          */
1003         if (PREV.br_startoff == new->br_startoff)
1004                 state |= BMAP_LEFT_FILLING;
1005         if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
1006                 state |= BMAP_RIGHT_FILLING;
1007
1008         /*
1009          * Check and set flags if this segment has a left neighbor.
1010          * Don't set contiguous if the combined extent would be too large.
1011          */
1012         if (*idx > 0) {
1013                 state |= BMAP_LEFT_VALID;
1014                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &LEFT);
1015
1016                 if (isnullstartblock(LEFT.br_startblock))
1017                         state |= BMAP_LEFT_DELAY;
1018         }
1019
1020         if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
1021             LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
1022             LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
1023             LEFT.br_state == newext &&
1024             LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
1025                 state |= BMAP_LEFT_CONTIG;
1026
1027         /*
1028          * Check and set flags if this segment has a right neighbor.
1029          * Don't set contiguous if the combined extent would be too large.
1030          * Also check for all-three-contiguous being too large.
1031          */
1032         if (*idx < ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1) {
1033                 state |= BMAP_RIGHT_VALID;
1034                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx + 1), &RIGHT);
1035                 if (isnullstartblock(RIGHT.br_startblock))
1036                         state |= BMAP_RIGHT_DELAY;
1037         }
1038
1039         if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
1040             new_endoff == RIGHT.br_startoff &&
1041             new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
1042             newext == RIGHT.br_state &&
1043             new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
1044             ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1045                        BMAP_RIGHT_FILLING)) !=
1046                       (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1047                        BMAP_RIGHT_FILLING) ||
1048              LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
1049                         <= MAXEXTLEN))
1050                 state |= BMAP_RIGHT_CONTIG;
1051
1052         /*
1053          * Switch out based on the FILLING and CONTIG state bits.
1054          */
1055         switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1056                          BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
1057         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1058              BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1059                 /*
1060                  * Setting all of a previous oldext extent to newext.
1061                  * The left and right neighbors are both contiguous with new.
1062                  */
1063                 --*idx;
1064
1065                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1066                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
1067                         LEFT.br_blockcount + PREV.br_blockcount +
1068                         RIGHT.br_blockcount);
1069                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1070
1071                 xfs_iext_remove(ip, *idx + 1, 2, state);
1072                 ip->i_d.di_nextents -= 2;
1073                 if (cur == NULL)
1074                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1075                 else {
1076                         rval = XFS_ILOG_CORE;
1077                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
1078                                         RIGHT.br_startblock,
1079                                         RIGHT.br_blockcount, &i)))
1080                                 goto done;
1081                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1082                         if ((error = xfs_btree_delete(cur, &i)))
1083                                 goto done;
1084                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1085                         if ((error = xfs_btree_decrement(cur, 0, &i)))
1086                                 goto done;
1087                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1088                         if ((error = xfs_btree_delete(cur, &i)))
1089                                 goto done;
1090                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1091                         if ((error = xfs_btree_decrement(cur, 0, &i)))
1092                                 goto done;
1093                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1094                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
1095                                 LEFT.br_startblock,
1096                                 LEFT.br_blockcount + PREV.br_blockcount +
1097                                 RIGHT.br_blockcount, LEFT.br_state)))
1098                                 goto done;
1099                 }
1100                 break;
1101
1102         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1103                 /*
1104                  * Setting all of a previous oldext extent to newext.
1105                  * The left neighbor is contiguous, the right is not.
1106                  */
1107                 --*idx;
1108
1109                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1110                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
1111                         LEFT.br_blockcount + PREV.br_blockcount);
1112                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1113
1114                 xfs_iext_remove(ip, *idx + 1, 1, state);
1115                 ip->i_d.di_nextents--;
1116                 if (cur == NULL)
1117                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1118                 else {
1119                         rval = XFS_ILOG_CORE;
1120                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1121                                         PREV.br_startblock, PREV.br_blockcount,
1122                                         &i)))
1123                                 goto done;
1124                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1125                         if ((error = xfs_btree_delete(cur, &i)))
1126                                 goto done;
1127                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1128                         if ((error = xfs_btree_decrement(cur, 0, &i)))
1129                                 goto done;
1130                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1131                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
1132                                 LEFT.br_startblock,
1133                                 LEFT.br_blockcount + PREV.br_blockcount,
1134                                 LEFT.br_state)))
1135                                 goto done;
1136                 }
1137                 break;
1138
1139         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1140                 /*
1141                  * Setting all of a previous oldext extent to newext.
1142                  * The right neighbor is contiguous, the left is not.
1143                  */
1144                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1145                 xfs_bmbt_set_blockcount(ep,
1146                         PREV.br_blockcount + RIGHT.br_blockcount);
1147                 xfs_bmbt_set_state(ep, newext);
1148                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1149                 xfs_iext_remove(ip, *idx + 1, 1, state);
1150                 ip->i_d.di_nextents--;
1151                 if (cur == NULL)
1152                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1153                 else {
1154                         rval = XFS_ILOG_CORE;
1155                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
1156                                         RIGHT.br_startblock,
1157                                         RIGHT.br_blockcount, &i)))
1158                                 goto done;
1159                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1160                         if ((error = xfs_btree_delete(cur, &i)))
1161                                 goto done;
1162                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1163                         if ((error = xfs_btree_decrement(cur, 0, &i)))
1164                                 goto done;
1165                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1166                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
1167                                 new->br_startblock,
1168                                 new->br_blockcount + RIGHT.br_blockcount,
1169                                 newext)))
1170                                 goto done;
1171                 }
1172                 break;
1173
1174         case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
1175                 /*
1176                  * Setting all of a previous oldext extent to newext.
1177                  * Neither the left nor right neighbors are contiguous with
1178                  * the new one.
1179                  */
1180                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1181                 xfs_bmbt_set_state(ep, newext);
1182                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1183
1184                 if (cur == NULL)
1185                         rval = XFS_ILOG_DEXT;
1186                 else {
1187                         rval = 0;
1188                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1189                                         new->br_startblock, new->br_blockcount,
1190                                         &i)))
1191                                 goto done;
1192                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1193                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
1194                                 new->br_startblock, new->br_blockcount,
1195                                 newext)))
1196                                 goto done;
1197                 }
1198                 break;
1199
1200         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
1201                 /*
1202                  * Setting the first part of a previous oldext extent to newext.
1203                  * The left neighbor is contiguous.
1204                  */
1205                 trace_xfs_bmap_pre_update(ip, *idx - 1, state, _THIS_IP_);
1206                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx - 1),
1207                         LEFT.br_blockcount + new->br_blockcount);
1208                 xfs_bmbt_set_startoff(ep,
1209                         PREV.br_startoff + new->br_blockcount);
1210                 trace_xfs_bmap_post_update(ip, *idx - 1, state, _THIS_IP_);
1211
1212                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1213                 xfs_bmbt_set_startblock(ep,
1214                         new->br_startblock + new->br_blockcount);
1215                 xfs_bmbt_set_blockcount(ep,
1216                         PREV.br_blockcount - new->br_blockcount);
1217                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1218
1219                 --*idx;
1220
1221                 if (cur == NULL)
1222                         rval = XFS_ILOG_DEXT;
1223                 else {
1224                         rval = 0;
1225                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1226                                         PREV.br_startblock, PREV.br_blockcount,
1227                                         &i)))
1228                                 goto done;
1229                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1230                         if ((error = xfs_bmbt_update(cur,
1231                                 PREV.br_startoff + new->br_blockcount,
1232                                 PREV.br_startblock + new->br_blockcount,
1233                                 PREV.br_blockcount - new->br_blockcount,
1234                                 oldext)))
1235                                 goto done;
1236                         if ((error = xfs_btree_decrement(cur, 0, &i)))
1237                                 goto done;
1238                         error = xfs_bmbt_update(cur, LEFT.br_startoff,
1239                                 LEFT.br_startblock,
1240                                 LEFT.br_blockcount + new->br_blockcount,
1241                                 LEFT.br_state);
1242                         if (error)
1243                                 goto done;
1244                 }
1245                 break;
1246
1247         case BMAP_LEFT_FILLING:
1248                 /*
1249                  * Setting the first part of a previous oldext extent to newext.
1250                  * The left neighbor is not contiguous.
1251                  */
1252                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1253                 ASSERT(ep && xfs_bmbt_get_state(ep) == oldext);
1254                 xfs_bmbt_set_startoff(ep, new_endoff);
1255                 xfs_bmbt_set_blockcount(ep,
1256                         PREV.br_blockcount - new->br_blockcount);
1257                 xfs_bmbt_set_startblock(ep,
1258                         new->br_startblock + new->br_blockcount);
1259                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1260
1261                 xfs_iext_insert(ip, *idx, 1, new, state);
1262                 ip->i_d.di_nextents++;
1263                 if (cur == NULL)
1264                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1265                 else {
1266                         rval = XFS_ILOG_CORE;
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                         cur->bc_rec.b = *new;
1279                         if ((error = xfs_btree_insert(cur, &i)))
1280                                 goto done;
1281                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1282                 }
1283                 break;
1284
1285         case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1286                 /*
1287                  * Setting the last part of a previous oldext extent to newext.
1288                  * The right neighbor is contiguous with the new allocation.
1289                  */
1290                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1291                 xfs_bmbt_set_blockcount(ep,
1292                         PREV.br_blockcount - new->br_blockcount);
1293                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1294
1295                 ++*idx;
1296
1297                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1298                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx),
1299                         new->br_startoff, new->br_startblock,
1300                         new->br_blockcount + RIGHT.br_blockcount, newext);
1301                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1302
1303                 if (cur == NULL)
1304                         rval = XFS_ILOG_DEXT;
1305                 else {
1306                         rval = 0;
1307                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1308                                         PREV.br_startblock,
1309                                         PREV.br_blockcount, &i)))
1310                                 goto done;
1311                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1312                         if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
1313                                 PREV.br_startblock,
1314                                 PREV.br_blockcount - new->br_blockcount,
1315                                 oldext)))
1316                                 goto done;
1317                         if ((error = xfs_btree_increment(cur, 0, &i)))
1318                                 goto done;
1319                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
1320                                 new->br_startblock,
1321                                 new->br_blockcount + RIGHT.br_blockcount,
1322                                 newext)))
1323                                 goto done;
1324                 }
1325                 break;
1326
1327         case BMAP_RIGHT_FILLING:
1328                 /*
1329                  * Setting the last part of a previous oldext extent to newext.
1330                  * The right neighbor is not contiguous.
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                 xfs_iext_insert(ip, *idx, 1, new, state);
1339
1340                 ip->i_d.di_nextents++;
1341                 if (cur == NULL)
1342                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1343                 else {
1344                         rval = XFS_ILOG_CORE;
1345                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1346                                         PREV.br_startblock, PREV.br_blockcount,
1347                                         &i)))
1348                                 goto done;
1349                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1350                         if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
1351                                 PREV.br_startblock,
1352                                 PREV.br_blockcount - new->br_blockcount,
1353                                 oldext)))
1354                                 goto done;
1355                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1356                                         new->br_startblock, new->br_blockcount,
1357                                         &i)))
1358                                 goto done;
1359                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1360                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
1361                         if ((error = xfs_btree_insert(cur, &i)))
1362                                 goto done;
1363                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1364                 }
1365                 break;
1366
1367         case 0:
1368                 /*
1369                  * Setting the middle part of a previous oldext extent to
1370                  * newext.  Contiguity is impossible here.
1371                  * One extent becomes three extents.
1372                  */
1373                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1374                 xfs_bmbt_set_blockcount(ep,
1375                         new->br_startoff - PREV.br_startoff);
1376                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1377
1378                 r[0] = *new;
1379                 r[1].br_startoff = new_endoff;
1380                 r[1].br_blockcount =
1381                         PREV.br_startoff + PREV.br_blockcount - new_endoff;
1382                 r[1].br_startblock = new->br_startblock + new->br_blockcount;
1383                 r[1].br_state = oldext;
1384
1385                 ++*idx;
1386                 xfs_iext_insert(ip, *idx, 2, &r[0], state);
1387
1388                 ip->i_d.di_nextents += 2;
1389                 if (cur == NULL)
1390                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1391                 else {
1392                         rval = XFS_ILOG_CORE;
1393                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1394                                         PREV.br_startblock, PREV.br_blockcount,
1395                                         &i)))
1396                                 goto done;
1397                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1398                         /* new right extent - oldext */
1399                         if ((error = xfs_bmbt_update(cur, r[1].br_startoff,
1400                                 r[1].br_startblock, r[1].br_blockcount,
1401                                 r[1].br_state)))
1402                                 goto done;
1403                         /* new left extent - oldext */
1404                         cur->bc_rec.b = PREV;
1405                         cur->bc_rec.b.br_blockcount =
1406                                 new->br_startoff - PREV.br_startoff;
1407                         if ((error = xfs_btree_insert(cur, &i)))
1408                                 goto done;
1409                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1410                         /*
1411                          * Reset the cursor to the position of the new extent
1412                          * we are about to insert as we can't trust it after
1413                          * the previous insert.
1414                          */
1415                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1416                                         new->br_startblock, new->br_blockcount,
1417                                         &i)))
1418                                 goto done;
1419                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1420                         /* new middle extent - newext */
1421                         cur->bc_rec.b.br_state = new->br_state;
1422                         if ((error = xfs_btree_insert(cur, &i)))
1423                                 goto done;
1424                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1425                 }
1426                 break;
1427
1428         case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1429         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1430         case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
1431         case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1432         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1433         case BMAP_LEFT_CONTIG:
1434         case BMAP_RIGHT_CONTIG:
1435                 /*
1436                  * These cases are all impossible.
1437                  */
1438                 ASSERT(0);
1439         }
1440
1441         /* convert to a btree if necessary */
1442         if (xfs_bmap_needs_btree(ip, XFS_DATA_FORK)) {
1443                 int     tmp_logflags;   /* partial log flag return val */
1444
1445                 ASSERT(cur == NULL);
1446                 error = xfs_bmap_extents_to_btree(tp, ip, first, flist, &cur,
1447                                 0, &tmp_logflags, XFS_DATA_FORK);
1448                 *logflagsp |= tmp_logflags;
1449                 if (error)
1450                         goto done;
1451         }
1452
1453         /* clear out the allocated field, done with it now in any case. */
1454         if (cur) {
1455                 cur->bc_private.b.allocated = 0;
1456                 *curp = cur;
1457         }
1458
1459         xfs_bmap_check_leaf_extents(*curp, ip, XFS_DATA_FORK);
1460 done:
1461         *logflagsp |= rval;
1462         return error;
1463 #undef  LEFT
1464 #undef  RIGHT
1465 #undef  PREV
1466 }
1467
1468 /*
1469  * Convert a hole to a delayed allocation.
1470  */
1471 STATIC void
1472 xfs_bmap_add_extent_hole_delay(
1473         xfs_inode_t             *ip,    /* incore inode pointer */
1474         xfs_extnum_t            *idx,   /* extent number to update/insert */
1475         xfs_bmbt_irec_t         *new)   /* new data to add to file extents */
1476 {
1477         xfs_ifork_t             *ifp;   /* inode fork pointer */
1478         xfs_bmbt_irec_t         left;   /* left neighbor extent entry */
1479         xfs_filblks_t           newlen=0;       /* new indirect size */
1480         xfs_filblks_t           oldlen=0;       /* old indirect size */
1481         xfs_bmbt_irec_t         right;  /* right neighbor extent entry */
1482         int                     state;  /* state bits, accessed thru macros */
1483         xfs_filblks_t           temp=0; /* temp for indirect calculations */
1484
1485         ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
1486         state = 0;
1487         ASSERT(isnullstartblock(new->br_startblock));
1488
1489         /*
1490          * Check and set flags if this segment has a left neighbor
1491          */
1492         if (*idx > 0) {
1493                 state |= BMAP_LEFT_VALID;
1494                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &left);
1495
1496                 if (isnullstartblock(left.br_startblock))
1497                         state |= BMAP_LEFT_DELAY;
1498         }
1499
1500         /*
1501          * Check and set flags if the current (right) segment exists.
1502          * If it doesn't exist, we're converting the hole at end-of-file.
1503          */
1504         if (*idx < ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t)) {
1505                 state |= BMAP_RIGHT_VALID;
1506                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx), &right);
1507
1508                 if (isnullstartblock(right.br_startblock))
1509                         state |= BMAP_RIGHT_DELAY;
1510         }
1511
1512         /*
1513          * Set contiguity flags on the left and right neighbors.
1514          * Don't let extents get too large, even if the pieces are contiguous.
1515          */
1516         if ((state & BMAP_LEFT_VALID) && (state & BMAP_LEFT_DELAY) &&
1517             left.br_startoff + left.br_blockcount == new->br_startoff &&
1518             left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
1519                 state |= BMAP_LEFT_CONTIG;
1520
1521         if ((state & BMAP_RIGHT_VALID) && (state & BMAP_RIGHT_DELAY) &&
1522             new->br_startoff + new->br_blockcount == right.br_startoff &&
1523             new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
1524             (!(state & BMAP_LEFT_CONTIG) ||
1525              (left.br_blockcount + new->br_blockcount +
1526               right.br_blockcount <= MAXEXTLEN)))
1527                 state |= BMAP_RIGHT_CONTIG;
1528
1529         /*
1530          * Switch out based on the contiguity flags.
1531          */
1532         switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
1533         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1534                 /*
1535                  * New allocation is contiguous with delayed allocations
1536                  * on the left and on the right.
1537                  * Merge all three into a single extent record.
1538                  */
1539                 --*idx;
1540                 temp = left.br_blockcount + new->br_blockcount +
1541                         right.br_blockcount;
1542
1543                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1544                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx), temp);
1545                 oldlen = startblockval(left.br_startblock) +
1546                         startblockval(new->br_startblock) +
1547                         startblockval(right.br_startblock);
1548                 newlen = xfs_bmap_worst_indlen(ip, temp);
1549                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, *idx),
1550                         nullstartblock((int)newlen));
1551                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1552
1553                 xfs_iext_remove(ip, *idx + 1, 1, state);
1554                 break;
1555
1556         case BMAP_LEFT_CONTIG:
1557                 /*
1558                  * New allocation is contiguous with a delayed allocation
1559                  * on the left.
1560                  * Merge the new allocation with the left neighbor.
1561                  */
1562                 --*idx;
1563                 temp = left.br_blockcount + new->br_blockcount;
1564
1565                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1566                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx), temp);
1567                 oldlen = startblockval(left.br_startblock) +
1568                         startblockval(new->br_startblock);
1569                 newlen = xfs_bmap_worst_indlen(ip, temp);
1570                 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, *idx),
1571                         nullstartblock((int)newlen));
1572                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1573                 break;
1574
1575         case BMAP_RIGHT_CONTIG:
1576                 /*
1577                  * New allocation is contiguous with a delayed allocation
1578                  * on the right.
1579                  * Merge the new allocation with the right neighbor.
1580                  */
1581                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
1582                 temp = new->br_blockcount + right.br_blockcount;
1583                 oldlen = startblockval(new->br_startblock) +
1584                         startblockval(right.br_startblock);
1585                 newlen = xfs_bmap_worst_indlen(ip, temp);
1586                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx),
1587                         new->br_startoff,
1588                         nullstartblock((int)newlen), temp, right.br_state);
1589                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
1590                 break;
1591
1592         case 0:
1593                 /*
1594                  * New allocation is not contiguous with another
1595                  * delayed allocation.
1596                  * Insert a new entry.
1597                  */
1598                 oldlen = newlen = 0;
1599                 xfs_iext_insert(ip, *idx, 1, new, state);
1600                 break;
1601         }
1602         if (oldlen != newlen) {
1603                 ASSERT(oldlen > newlen);
1604                 xfs_icsb_modify_counters(ip->i_mount, XFS_SBS_FDBLOCKS,
1605                         (int64_t)(oldlen - newlen), 0);
1606                 /*
1607                  * Nothing to do for disk quota accounting here.
1608                  */
1609         }
1610 }
1611
1612 /*
1613  * Convert a hole to a real allocation.
1614  */
1615 STATIC int                              /* error */
1616 xfs_bmap_add_extent_hole_real(
1617         struct xfs_bmalloca     *bma,
1618         int                     whichfork)
1619 {
1620         struct xfs_bmbt_irec    *new = &bma->got;
1621         int                     error;  /* error return value */
1622         int                     i;      /* temp state */
1623         xfs_ifork_t             *ifp;   /* inode fork pointer */
1624         xfs_bmbt_irec_t         left;   /* left neighbor extent entry */
1625         xfs_bmbt_irec_t         right;  /* right neighbor extent entry */
1626         int                     rval=0; /* return value (logging flags) */
1627         int                     state;  /* state bits, accessed thru macros */
1628
1629         ifp = XFS_IFORK_PTR(bma->ip, whichfork);
1630
1631         ASSERT(bma->idx >= 0);
1632         ASSERT(bma->idx <= ifp->if_bytes / sizeof(struct xfs_bmbt_rec));
1633         ASSERT(!isnullstartblock(new->br_startblock));
1634         ASSERT(!bma->cur ||
1635                !(bma->cur->bc_private.b.flags & XFS_BTCUR_BPRV_WASDEL));
1636
1637         XFS_STATS_INC(xs_add_exlist);
1638
1639         state = 0;
1640         if (whichfork == XFS_ATTR_FORK)
1641                 state |= BMAP_ATTRFORK;
1642
1643         /*
1644          * Check and set flags if this segment has a left neighbor.
1645          */
1646         if (bma->idx > 0) {
1647                 state |= BMAP_LEFT_VALID;
1648                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx - 1), &left);
1649                 if (isnullstartblock(left.br_startblock))
1650                         state |= BMAP_LEFT_DELAY;
1651         }
1652
1653         /*
1654          * Check and set flags if this segment has a current value.
1655          * Not true if we're inserting into the "hole" at eof.
1656          */
1657         if (bma->idx < ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)) {
1658                 state |= BMAP_RIGHT_VALID;
1659                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx), &right);
1660                 if (isnullstartblock(right.br_startblock))
1661                         state |= BMAP_RIGHT_DELAY;
1662         }
1663
1664         /*
1665          * We're inserting a real allocation between "left" and "right".
1666          * Set the contiguity flags.  Don't let extents get too large.
1667          */
1668         if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
1669             left.br_startoff + left.br_blockcount == new->br_startoff &&
1670             left.br_startblock + left.br_blockcount == new->br_startblock &&
1671             left.br_state == new->br_state &&
1672             left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
1673                 state |= BMAP_LEFT_CONTIG;
1674
1675         if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
1676             new->br_startoff + new->br_blockcount == right.br_startoff &&
1677             new->br_startblock + new->br_blockcount == right.br_startblock &&
1678             new->br_state == right.br_state &&
1679             new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
1680             (!(state & BMAP_LEFT_CONTIG) ||
1681              left.br_blockcount + new->br_blockcount +
1682              right.br_blockcount <= MAXEXTLEN))
1683                 state |= BMAP_RIGHT_CONTIG;
1684
1685         error = 0;
1686         /*
1687          * Select which case we're in here, and implement it.
1688          */
1689         switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
1690         case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
1691                 /*
1692                  * New allocation is contiguous with real allocations on the
1693                  * left and on the right.
1694                  * Merge all three into a single extent record.
1695                  */
1696                 --bma->idx;
1697                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1698                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
1699                         left.br_blockcount + new->br_blockcount +
1700                         right.br_blockcount);
1701                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1702
1703                 xfs_iext_remove(bma->ip, bma->idx + 1, 1, state);
1704
1705                 XFS_IFORK_NEXT_SET(bma->ip, whichfork,
1706                         XFS_IFORK_NEXTENTS(bma->ip, whichfork) - 1);
1707                 if (bma->cur == NULL) {
1708                         rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
1709                 } else {
1710                         rval = XFS_ILOG_CORE;
1711                         error = xfs_bmbt_lookup_eq(bma->cur, right.br_startoff,
1712                                         right.br_startblock, right.br_blockcount,
1713                                         &i);
1714                         if (error)
1715                                 goto done;
1716                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1717                         error = xfs_btree_delete(bma->cur, &i);
1718                         if (error)
1719                                 goto done;
1720                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1721                         error = xfs_btree_decrement(bma->cur, 0, &i);
1722                         if (error)
1723                                 goto done;
1724                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1725                         error = xfs_bmbt_update(bma->cur, left.br_startoff,
1726                                         left.br_startblock,
1727                                         left.br_blockcount +
1728                                                 new->br_blockcount +
1729                                                 right.br_blockcount,
1730                                         left.br_state);
1731                         if (error)
1732                                 goto done;
1733                 }
1734                 break;
1735
1736         case BMAP_LEFT_CONTIG:
1737                 /*
1738                  * New allocation is contiguous with a real allocation
1739                  * on the left.
1740                  * Merge the new allocation with the left neighbor.
1741                  */
1742                 --bma->idx;
1743                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1744                 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
1745                         left.br_blockcount + new->br_blockcount);
1746                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1747
1748                 if (bma->cur == NULL) {
1749                         rval = xfs_ilog_fext(whichfork);
1750                 } else {
1751                         rval = 0;
1752                         error = xfs_bmbt_lookup_eq(bma->cur, left.br_startoff,
1753                                         left.br_startblock, left.br_blockcount,
1754                                         &i);
1755                         if (error)
1756                                 goto done;
1757                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1758                         error = xfs_bmbt_update(bma->cur, left.br_startoff,
1759                                         left.br_startblock,
1760                                         left.br_blockcount +
1761                                                 new->br_blockcount,
1762                                         left.br_state);
1763                         if (error)
1764                                 goto done;
1765                 }
1766                 break;
1767
1768         case BMAP_RIGHT_CONTIG:
1769                 /*
1770                  * New allocation is contiguous with a real allocation
1771                  * on the right.
1772                  * Merge the new allocation with the right neighbor.
1773                  */
1774                 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1775                 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, bma->idx),
1776                         new->br_startoff, new->br_startblock,
1777                         new->br_blockcount + right.br_blockcount,
1778                         right.br_state);
1779                 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1780
1781                 if (bma->cur == NULL) {
1782                         rval = xfs_ilog_fext(whichfork);
1783                 } else {
1784                         rval = 0;
1785                         error = xfs_bmbt_lookup_eq(bma->cur,
1786                                         right.br_startoff,
1787                                         right.br_startblock,
1788                                         right.br_blockcount, &i);
1789                         if (error)
1790                                 goto done;
1791                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1792                         error = xfs_bmbt_update(bma->cur, new->br_startoff,
1793                                         new->br_startblock,
1794                                         new->br_blockcount +
1795                                                 right.br_blockcount,
1796                                         right.br_state);
1797                         if (error)
1798                                 goto done;
1799                 }
1800                 break;
1801
1802         case 0:
1803                 /*
1804                  * New allocation is not contiguous with another
1805                  * real allocation.
1806                  * Insert a new entry.
1807                  */
1808                 xfs_iext_insert(bma->ip, bma->idx, 1, new, state);
1809                 XFS_IFORK_NEXT_SET(bma->ip, whichfork,
1810                         XFS_IFORK_NEXTENTS(bma->ip, whichfork) + 1);
1811                 if (bma->cur == NULL) {
1812                         rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
1813                 } else {
1814                         rval = XFS_ILOG_CORE;
1815                         error = xfs_bmbt_lookup_eq(bma->cur,
1816                                         new->br_startoff,
1817                                         new->br_startblock,
1818                                         new->br_blockcount, &i);
1819                         if (error)
1820                                 goto done;
1821                         XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1822                         bma->cur->bc_rec.b.br_state = new->br_state;
1823                         error = xfs_btree_insert(bma->cur, &i);
1824                         if (error)
1825                                 goto done;
1826                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1827                 }
1828                 break;
1829         }
1830
1831         /* convert to a btree if necessary */
1832         if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
1833                 int     tmp_logflags;   /* partial log flag return val */
1834
1835                 ASSERT(bma->cur == NULL);
1836                 error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
1837                                 bma->firstblock, bma->flist, &bma->cur,
1838                                 0, &tmp_logflags, whichfork);
1839                 bma->logflags |= tmp_logflags;
1840                 if (error)
1841                         goto done;
1842         }
1843
1844         /* clear out the allocated field, done with it now in any case. */
1845         if (bma->cur)
1846                 bma->cur->bc_private.b.allocated = 0;
1847
1848         xfs_bmap_check_leaf_extents(bma->cur, bma->ip, whichfork);
1849 done:
1850         bma->logflags |= rval;
1851         return error;
1852 }
1853
1854 /*
1855  * Adjust the size of the new extent based on di_extsize and rt extsize.
1856  */
1857 STATIC int
1858 xfs_bmap_extsize_align(
1859         xfs_mount_t     *mp,
1860         xfs_bmbt_irec_t *gotp,          /* next extent pointer */
1861         xfs_bmbt_irec_t *prevp,         /* previous extent pointer */
1862         xfs_extlen_t    extsz,          /* align to this extent size */
1863         int             rt,             /* is this a realtime inode? */
1864         int             eof,            /* is extent at end-of-file? */
1865         int             delay,          /* creating delalloc extent? */
1866         int             convert,        /* overwriting unwritten extent? */
1867         xfs_fileoff_t   *offp,          /* in/out: aligned offset */
1868         xfs_extlen_t    *lenp)          /* in/out: aligned length */
1869 {
1870         xfs_fileoff_t   orig_off;       /* original offset */
1871         xfs_extlen_t    orig_alen;      /* original length */
1872         xfs_fileoff_t   orig_end;       /* original off+len */
1873         xfs_fileoff_t   nexto;          /* next file offset */
1874         xfs_fileoff_t   prevo;          /* previous file offset */
1875         xfs_fileoff_t   align_off;      /* temp for offset */
1876         xfs_extlen_t    align_alen;     /* temp for length */
1877         xfs_extlen_t    temp;           /* temp for calculations */
1878
1879         if (convert)
1880                 return 0;
1881
1882         orig_off = align_off = *offp;
1883         orig_alen = align_alen = *lenp;
1884         orig_end = orig_off + orig_alen;
1885
1886         /*
1887          * If this request overlaps an existing extent, then don't
1888          * attempt to perform any additional alignment.
1889          */
1890         if (!delay && !eof &&
1891             (orig_off >= gotp->br_startoff) &&
1892             (orig_end <= gotp->br_startoff + gotp->br_blockcount)) {
1893                 return 0;
1894         }
1895
1896         /*
1897          * If the file offset is unaligned vs. the extent size
1898          * we need to align it.  This will be possible unless
1899          * the file was previously written with a kernel that didn't
1900          * perform this alignment, or if a truncate shot us in the
1901          * foot.
1902          */
1903         temp = do_mod(orig_off, extsz);
1904         if (temp) {
1905                 align_alen += temp;
1906                 align_off -= temp;
1907         }
1908         /*
1909          * Same adjustment for the end of the requested area.
1910          */
1911         if ((temp = (align_alen % extsz))) {
1912                 align_alen += extsz - temp;
1913         }
1914         /*
1915          * If the previous block overlaps with this proposed allocation
1916          * then move the start forward without adjusting the length.
1917          */
1918         if (prevp->br_startoff != NULLFILEOFF) {
1919                 if (prevp->br_startblock == HOLESTARTBLOCK)
1920                         prevo = prevp->br_startoff;
1921                 else
1922                         prevo = prevp->br_startoff + prevp->br_blockcount;
1923         } else
1924                 prevo = 0;
1925         if (align_off != orig_off && align_off < prevo)
1926                 align_off = prevo;
1927         /*
1928          * If the next block overlaps with this proposed allocation
1929          * then move the start back without adjusting the length,
1930          * but not before offset 0.
1931          * This may of course make the start overlap previous block,
1932          * and if we hit the offset 0 limit then the next block
1933          * can still overlap too.
1934          */
1935         if (!eof && gotp->br_startoff != NULLFILEOFF) {
1936                 if ((delay && gotp->br_startblock == HOLESTARTBLOCK) ||
1937                     (!delay && gotp->br_startblock == DELAYSTARTBLOCK))
1938                         nexto = gotp->br_startoff + gotp->br_blockcount;
1939                 else
1940                         nexto = gotp->br_startoff;
1941         } else
1942                 nexto = NULLFILEOFF;
1943         if (!eof &&
1944             align_off + align_alen != orig_end &&
1945             align_off + align_alen > nexto)
1946                 align_off = nexto > align_alen ? nexto - align_alen : 0;
1947         /*
1948          * If we're now overlapping the next or previous extent that
1949          * means we can't fit an extsz piece in this hole.  Just move
1950          * the start forward to the first valid spot and set
1951          * the length so we hit the end.
1952          */
1953         if (align_off != orig_off && align_off < prevo)
1954                 align_off = prevo;
1955         if (align_off + align_alen != orig_end &&
1956             align_off + align_alen > nexto &&
1957             nexto != NULLFILEOFF) {
1958                 ASSERT(nexto > prevo);
1959                 align_alen = nexto - align_off;
1960         }
1961
1962         /*
1963          * If realtime, and the result isn't a multiple of the realtime
1964          * extent size we need to remove blocks until it is.
1965          */
1966         if (rt && (temp = (align_alen % mp->m_sb.sb_rextsize))) {
1967                 /*
1968                  * We're not covering the original request, or
1969                  * we won't be able to once we fix the length.
1970                  */
1971                 if (orig_off < align_off ||
1972                     orig_end > align_off + align_alen ||
1973                     align_alen - temp < orig_alen)
1974                         return XFS_ERROR(EINVAL);
1975                 /*
1976                  * Try to fix it by moving the start up.
1977                  */
1978                 if (align_off + temp <= orig_off) {
1979                         align_alen -= temp;
1980                         align_off += temp;
1981                 }
1982                 /*
1983                  * Try to fix it by moving the end in.
1984                  */
1985                 else if (align_off + align_alen - temp >= orig_end)
1986                         align_alen -= temp;
1987                 /*
1988                  * Set the start to the minimum then trim the length.
1989                  */
1990                 else {
1991                         align_alen -= orig_off - align_off;
1992                         align_off = orig_off;
1993                         align_alen -= align_alen % mp->m_sb.sb_rextsize;
1994                 }
1995                 /*
1996                  * Result doesn't cover the request, fail it.
1997                  */
1998                 if (orig_off < align_off || orig_end > align_off + align_alen)
1999                         return XFS_ERROR(EINVAL);
2000         } else {
2001                 ASSERT(orig_off >= align_off);
2002                 ASSERT(orig_end <= align_off + align_alen);
2003         }
2004
2005 #ifdef DEBUG
2006         if (!eof && gotp->br_startoff != NULLFILEOFF)
2007                 ASSERT(align_off + align_alen <= gotp->br_startoff);
2008         if (prevp->br_startoff != NULLFILEOFF)
2009                 ASSERT(align_off >= prevp->br_startoff + prevp->br_blockcount);
2010 #endif
2011
2012         *lenp = align_alen;
2013         *offp = align_off;
2014         return 0;
2015 }
2016
2017 #define XFS_ALLOC_GAP_UNITS     4
2018
2019 STATIC void
2020 xfs_bmap_adjacent(
2021         xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
2022 {
2023         xfs_fsblock_t   adjust;         /* adjustment to block numbers */
2024         xfs_agnumber_t  fb_agno;        /* ag number of ap->firstblock */
2025         xfs_mount_t     *mp;            /* mount point structure */
2026         int             nullfb;         /* true if ap->firstblock isn't set */
2027         int             rt;             /* true if inode is realtime */
2028
2029 #define ISVALID(x,y)    \
2030         (rt ? \
2031                 (x) < mp->m_sb.sb_rblocks : \
2032                 XFS_FSB_TO_AGNO(mp, x) == XFS_FSB_TO_AGNO(mp, y) && \
2033                 XFS_FSB_TO_AGNO(mp, x) < mp->m_sb.sb_agcount && \
2034                 XFS_FSB_TO_AGBNO(mp, x) < mp->m_sb.sb_agblocks)
2035
2036         mp = ap->ip->i_mount;
2037         nullfb = *ap->firstblock == NULLFSBLOCK;
2038         rt = XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata;
2039         fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, *ap->firstblock);
2040         /*
2041          * If allocating at eof, and there's a previous real block,
2042          * try to use its last block as our starting point.
2043          */
2044         if (ap->eof && ap->prev.br_startoff != NULLFILEOFF &&
2045             !isnullstartblock(ap->prev.br_startblock) &&
2046             ISVALID(ap->prev.br_startblock + ap->prev.br_blockcount,
2047                     ap->prev.br_startblock)) {
2048                 ap->blkno = ap->prev.br_startblock + ap->prev.br_blockcount;
2049                 /*
2050                  * Adjust for the gap between prevp and us.
2051                  */
2052                 adjust = ap->offset -
2053                         (ap->prev.br_startoff + ap->prev.br_blockcount);
2054                 if (adjust &&
2055                     ISVALID(ap->blkno + adjust, ap->prev.br_startblock))
2056                         ap->blkno += adjust;
2057         }
2058         /*
2059          * If not at eof, then compare the two neighbor blocks.
2060          * Figure out whether either one gives us a good starting point,
2061          * and pick the better one.
2062          */
2063         else if (!ap->eof) {
2064                 xfs_fsblock_t   gotbno;         /* right side block number */
2065                 xfs_fsblock_t   gotdiff=0;      /* right side difference */
2066                 xfs_fsblock_t   prevbno;        /* left side block number */
2067                 xfs_fsblock_t   prevdiff=0;     /* left side difference */
2068
2069                 /*
2070                  * If there's a previous (left) block, select a requested
2071                  * start block based on it.
2072                  */
2073                 if (ap->prev.br_startoff != NULLFILEOFF &&
2074                     !isnullstartblock(ap->prev.br_startblock) &&
2075                     (prevbno = ap->prev.br_startblock +
2076                                ap->prev.br_blockcount) &&
2077                     ISVALID(prevbno, ap->prev.br_startblock)) {
2078                         /*
2079                          * Calculate gap to end of previous block.
2080                          */
2081                         adjust = prevdiff = ap->offset -
2082                                 (ap->prev.br_startoff +
2083                                  ap->prev.br_blockcount);
2084                         /*
2085                          * Figure the startblock based on the previous block's
2086                          * end and the gap size.
2087                          * Heuristic!
2088                          * If the gap is large relative to the piece we're
2089                          * allocating, or using it gives us an invalid block
2090                          * number, then just use the end of the previous block.
2091                          */
2092                         if (prevdiff <= XFS_ALLOC_GAP_UNITS * ap->length &&
2093                             ISVALID(prevbno + prevdiff,
2094                                     ap->prev.br_startblock))
2095                                 prevbno += adjust;
2096                         else
2097                                 prevdiff += adjust;
2098                         /*
2099                          * If the firstblock forbids it, can't use it,
2100                          * must use default.
2101                          */
2102                         if (!rt && !nullfb &&
2103                             XFS_FSB_TO_AGNO(mp, prevbno) != fb_agno)
2104                                 prevbno = NULLFSBLOCK;
2105                 }
2106                 /*
2107                  * No previous block or can't follow it, just default.
2108                  */
2109                 else
2110                         prevbno = NULLFSBLOCK;
2111                 /*
2112                  * If there's a following (right) block, select a requested
2113                  * start block based on it.
2114                  */
2115                 if (!isnullstartblock(ap->got.br_startblock)) {
2116                         /*
2117                          * Calculate gap to start of next block.
2118                          */
2119                         adjust = gotdiff = ap->got.br_startoff - ap->offset;
2120                         /*
2121                          * Figure the startblock based on the next block's
2122                          * start and the gap size.
2123                          */
2124                         gotbno = ap->got.br_startblock;
2125                         /*
2126                          * Heuristic!
2127                          * If the gap is large relative to the piece we're
2128                          * allocating, or using it gives us an invalid block
2129                          * number, then just use the start of the next block
2130                          * offset by our length.
2131                          */
2132                         if (gotdiff <= XFS_ALLOC_GAP_UNITS * ap->length &&
2133                             ISVALID(gotbno - gotdiff, gotbno))
2134                                 gotbno -= adjust;
2135                         else if (ISVALID(gotbno - ap->length, gotbno)) {
2136                                 gotbno -= ap->length;
2137                                 gotdiff += adjust - ap->length;
2138                         } else
2139                                 gotdiff += 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, gotbno) != fb_agno)
2146                                 gotbno = NULLFSBLOCK;
2147                 }
2148                 /*
2149                  * No next block, just default.
2150                  */
2151                 else
2152                         gotbno = NULLFSBLOCK;
2153                 /*
2154                  * If both valid, pick the better one, else the only good
2155                  * one, else ap->blkno is already set (to 0 or the inode block).
2156                  */
2157                 if (prevbno != NULLFSBLOCK && gotbno != NULLFSBLOCK)
2158                         ap->blkno = prevdiff <= gotdiff ? prevbno : gotbno;
2159                 else if (prevbno != NULLFSBLOCK)
2160                         ap->blkno = prevbno;
2161                 else if (gotbno != NULLFSBLOCK)
2162                         ap->blkno = gotbno;
2163         }
2164 #undef ISVALID
2165 }
2166
2167 STATIC int
2168 xfs_bmap_rtalloc(
2169         xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
2170 {
2171         xfs_alloctype_t atype = 0;      /* type for allocation routines */
2172         int             error;          /* error return value */
2173         xfs_mount_t     *mp;            /* mount point structure */
2174         xfs_extlen_t    prod = 0;       /* product factor for allocators */
2175         xfs_extlen_t    ralen = 0;      /* realtime allocation length */
2176         xfs_extlen_t    align;          /* minimum allocation alignment */
2177         xfs_rtblock_t   rtb;
2178
2179         mp = ap->ip->i_mount;
2180         align = xfs_get_extsz_hint(ap->ip);
2181         prod = align / mp->m_sb.sb_rextsize;
2182         error = xfs_bmap_extsize_align(mp, &ap->got, &ap->prev,
2183                                         align, 1, ap->eof, 0,
2184                                         ap->conv, &ap->offset, &ap->length);
2185         if (error)
2186                 return error;
2187         ASSERT(ap->length);
2188         ASSERT(ap->length % mp->m_sb.sb_rextsize == 0);
2189
2190         /*
2191          * If the offset & length are not perfectly aligned
2192          * then kill prod, it will just get us in trouble.
2193          */
2194         if (do_mod(ap->offset, align) || ap->length % align)
2195                 prod = 1;
2196         /*
2197          * Set ralen to be the actual requested length in rtextents.
2198          */
2199         ralen = ap->length / mp->m_sb.sb_rextsize;
2200         /*
2201          * If the old value was close enough to MAXEXTLEN that
2202          * we rounded up to it, cut it back so it's valid again.
2203          * Note that if it's a really large request (bigger than
2204          * MAXEXTLEN), we don't hear about that number, and can't
2205          * adjust the starting point to match it.
2206          */
2207         if (ralen * mp->m_sb.sb_rextsize >= MAXEXTLEN)
2208                 ralen = MAXEXTLEN / mp->m_sb.sb_rextsize;
2209
2210         /*
2211          * Lock out other modifications to the RT bitmap inode.
2212          */
2213         xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL);
2214         xfs_trans_ijoin(ap->tp, mp->m_rbmip, XFS_ILOCK_EXCL);
2215
2216         /*
2217          * If it's an allocation to an empty file at offset 0,
2218          * pick an extent that will space things out in the rt area.
2219          */
2220         if (ap->eof && ap->offset == 0) {
2221                 xfs_rtblock_t uninitialized_var(rtx); /* realtime extent no */
2222
2223                 error = xfs_rtpick_extent(mp, ap->tp, ralen, &rtx);
2224                 if (error)
2225                         return error;
2226                 ap->blkno = rtx * mp->m_sb.sb_rextsize;
2227         } else {
2228                 ap->blkno = 0;
2229         }
2230
2231         xfs_bmap_adjacent(ap);
2232
2233         /*
2234          * Realtime allocation, done through xfs_rtallocate_extent.
2235          */
2236         atype = ap->blkno == 0 ?  XFS_ALLOCTYPE_ANY_AG : XFS_ALLOCTYPE_NEAR_BNO;
2237         do_div(ap->blkno, mp->m_sb.sb_rextsize);
2238         rtb = ap->blkno;
2239         ap->length = ralen;
2240         if ((error = xfs_rtallocate_extent(ap->tp, ap->blkno, 1, ap->length,
2241                                 &ralen, atype, ap->wasdel, prod, &rtb)))
2242                 return error;
2243         if (rtb == NULLFSBLOCK && prod > 1 &&
2244             (error = xfs_rtallocate_extent(ap->tp, ap->blkno, 1,
2245                                            ap->length, &ralen, atype,
2246                                            ap->wasdel, 1, &rtb)))
2247                 return error;
2248         ap->blkno = rtb;
2249         if (ap->blkno != NULLFSBLOCK) {
2250                 ap->blkno *= mp->m_sb.sb_rextsize;
2251                 ralen *= mp->m_sb.sb_rextsize;
2252                 ap->length = ralen;
2253                 ap->ip->i_d.di_nblocks += ralen;
2254                 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
2255                 if (ap->wasdel)
2256                         ap->ip->i_delayed_blks -= ralen;
2257                 /*
2258                  * Adjust the disk quota also. This was reserved
2259                  * earlier.
2260                  */
2261                 xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
2262                         ap->wasdel ? XFS_TRANS_DQ_DELRTBCOUNT :
2263                                         XFS_TRANS_DQ_RTBCOUNT, (long) ralen);
2264         } else {
2265                 ap->length = 0;
2266         }
2267         return 0;
2268 }
2269
2270 STATIC int
2271 xfs_bmap_btalloc_nullfb(
2272         struct xfs_bmalloca     *ap,
2273         struct xfs_alloc_arg    *args,
2274         xfs_extlen_t            *blen)
2275 {
2276         struct xfs_mount        *mp = ap->ip->i_mount;
2277         struct xfs_perag        *pag;
2278         xfs_agnumber_t          ag, startag;
2279         int                     notinit = 0;
2280         int                     error;
2281
2282         if (ap->userdata && xfs_inode_is_filestream(ap->ip))
2283                 args->type = XFS_ALLOCTYPE_NEAR_BNO;
2284         else
2285                 args->type = XFS_ALLOCTYPE_START_BNO;
2286         args->total = ap->total;
2287
2288         /*
2289          * Search for an allocation group with a single extent large enough
2290          * for the request.  If one isn't found, then adjust the minimum
2291          * allocation size to the largest space found.
2292          */
2293         startag = ag = XFS_FSB_TO_AGNO(mp, args->fsbno);
2294         if (startag == NULLAGNUMBER)
2295                 startag = ag = 0;
2296
2297         pag = xfs_perag_get(mp, ag);
2298         while (*blen < args->maxlen) {
2299                 if (!pag->pagf_init) {
2300                         error = xfs_alloc_pagf_init(mp, args->tp, ag,
2301                                                     XFS_ALLOC_FLAG_TRYLOCK);
2302                         if (error) {
2303                                 xfs_perag_put(pag);
2304                                 return error;
2305                         }
2306                 }
2307
2308                 /*
2309                  * See xfs_alloc_fix_freelist...
2310                  */
2311                 if (pag->pagf_init) {
2312                         xfs_extlen_t    longest;
2313                         longest = xfs_alloc_longest_free_extent(mp, pag);
2314                         if (*blen < longest)
2315                                 *blen = longest;
2316                 } else
2317                         notinit = 1;
2318
2319                 if (xfs_inode_is_filestream(ap->ip)) {
2320                         if (*blen >= args->maxlen)
2321                                 break;
2322
2323                         if (ap->userdata) {
2324                                 /*
2325                                  * If startag is an invalid AG, we've
2326                                  * come here once before and
2327                                  * xfs_filestream_new_ag picked the
2328                                  * best currently available.
2329                                  *
2330                                  * Don't continue looping, since we
2331                                  * could loop forever.
2332                                  */
2333                                 if (startag == NULLAGNUMBER)
2334                                         break;
2335
2336                                 error = xfs_filestream_new_ag(ap, &ag);
2337                                 xfs_perag_put(pag);
2338                                 if (error)
2339                                         return error;
2340
2341                                 /* loop again to set 'blen'*/
2342                                 startag = NULLAGNUMBER;
2343                                 pag = xfs_perag_get(mp, ag);
2344                                 continue;
2345                         }
2346                 }
2347                 if (++ag == mp->m_sb.sb_agcount)
2348                         ag = 0;
2349                 if (ag == startag)
2350                         break;
2351                 xfs_perag_put(pag);
2352                 pag = xfs_perag_get(mp, ag);
2353         }
2354         xfs_perag_put(pag);
2355
2356         /*
2357          * Since the above loop did a BUF_TRYLOCK, it is
2358          * possible that there is space for this request.
2359          */
2360         if (notinit || *blen < ap->minlen)
2361                 args->minlen = ap->minlen;
2362         /*
2363          * If the best seen length is less than the request
2364          * length, use the best as the minimum.
2365          */
2366         else if (*blen < args->maxlen)
2367                 args->minlen = *blen;
2368         /*
2369          * Otherwise we've seen an extent as big as maxlen,
2370          * use that as the minimum.
2371          */
2372         else
2373                 args->minlen = args->maxlen;
2374
2375         /*
2376          * set the failure fallback case to look in the selected
2377          * AG as the stream may have moved.
2378          */
2379         if (xfs_inode_is_filestream(ap->ip))
2380                 ap->blkno = args->fsbno = XFS_AGB_TO_FSB(mp, ag, 0);
2381
2382         return 0;
2383 }
2384
2385 STATIC int
2386 xfs_bmap_btalloc(
2387         xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
2388 {
2389         xfs_mount_t     *mp;            /* mount point structure */
2390         xfs_alloctype_t atype = 0;      /* type for allocation routines */
2391         xfs_extlen_t    align;          /* minimum allocation alignment */
2392         xfs_agnumber_t  fb_agno;        /* ag number of ap->firstblock */
2393         xfs_agnumber_t  ag;
2394         xfs_alloc_arg_t args;
2395         xfs_extlen_t    blen;
2396         xfs_extlen_t    nextminlen = 0;
2397         int             nullfb;         /* true if ap->firstblock isn't set */
2398         int             isaligned;
2399         int             tryagain;
2400         int             error;
2401
2402         ASSERT(ap->length);
2403
2404         mp = ap->ip->i_mount;
2405         align = ap->userdata ? xfs_get_extsz_hint(ap->ip) : 0;
2406         if (unlikely(align)) {
2407                 error = xfs_bmap_extsize_align(mp, &ap->got, &ap->prev,
2408                                                 align, 0, ap->eof, 0, ap->conv,
2409                                                 &ap->offset, &ap->length);
2410                 ASSERT(!error);
2411                 ASSERT(ap->length);
2412         }
2413         nullfb = *ap->firstblock == NULLFSBLOCK;
2414         fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, *ap->firstblock);
2415         if (nullfb) {
2416                 if (ap->userdata && xfs_inode_is_filestream(ap->ip)) {
2417                         ag = xfs_filestream_lookup_ag(ap->ip);
2418                         ag = (ag != NULLAGNUMBER) ? ag : 0;
2419                         ap->blkno = XFS_AGB_TO_FSB(mp, ag, 0);
2420                 } else {
2421                         ap->blkno = XFS_INO_TO_FSB(mp, ap->ip->i_ino);
2422                 }
2423         } else
2424                 ap->blkno = *ap->firstblock;
2425
2426         xfs_bmap_adjacent(ap);
2427
2428         /*
2429          * If allowed, use ap->blkno; otherwise must use firstblock since
2430          * it's in the right allocation group.
2431          */
2432         if (nullfb || XFS_FSB_TO_AGNO(mp, ap->blkno) == fb_agno)
2433                 ;
2434         else
2435                 ap->blkno = *ap->firstblock;
2436         /*
2437          * Normal allocation, done through xfs_alloc_vextent.
2438          */
2439         tryagain = isaligned = 0;
2440         memset(&args, 0, sizeof(args));
2441         args.tp = ap->tp;
2442         args.mp = mp;
2443         args.fsbno = ap->blkno;
2444
2445         /* Trim the allocation back to the maximum an AG can fit. */
2446         args.maxlen = MIN(ap->length, XFS_ALLOC_AG_MAX_USABLE(mp));
2447         args.firstblock = *ap->firstblock;
2448         blen = 0;
2449         if (nullfb) {
2450                 error = xfs_bmap_btalloc_nullfb(ap, &args, &blen);
2451                 if (error)
2452                         return error;
2453         } else if (ap->flist->xbf_low) {
2454                 if (xfs_inode_is_filestream(ap->ip))
2455                         args.type = XFS_ALLOCTYPE_FIRST_AG;
2456                 else
2457                         args.type = XFS_ALLOCTYPE_START_BNO;
2458                 args.total = args.minlen = ap->minlen;
2459         } else {
2460                 args.type = XFS_ALLOCTYPE_NEAR_BNO;
2461                 args.total = ap->total;
2462                 args.minlen = ap->minlen;
2463         }
2464         /* apply extent size hints if obtained earlier */
2465         if (unlikely(align)) {
2466                 args.prod = align;
2467                 if ((args.mod = (xfs_extlen_t)do_mod(ap->offset, args.prod)))
2468                         args.mod = (xfs_extlen_t)(args.prod - args.mod);
2469         } else if (mp->m_sb.sb_blocksize >= PAGE_CACHE_SIZE) {
2470                 args.prod = 1;
2471                 args.mod = 0;
2472         } else {
2473                 args.prod = PAGE_CACHE_SIZE >> mp->m_sb.sb_blocklog;
2474                 if ((args.mod = (xfs_extlen_t)(do_mod(ap->offset, args.prod))))
2475                         args.mod = (xfs_extlen_t)(args.prod - args.mod);
2476         }
2477         /*
2478          * If we are not low on available data blocks, and the
2479          * underlying logical volume manager is a stripe, and
2480          * the file offset is zero then try to allocate data
2481          * blocks on stripe unit boundary.
2482          * NOTE: ap->aeof is only set if the allocation length
2483          * is >= the stripe unit and the allocation offset is
2484          * at the end of file.
2485          */
2486         if (!ap->flist->xbf_low && ap->aeof) {
2487                 if (!ap->offset) {
2488                         args.alignment = mp->m_dalign;
2489                         atype = args.type;
2490                         isaligned = 1;
2491                         /*
2492                          * Adjust for alignment
2493                          */
2494                         if (blen > args.alignment && blen <= args.maxlen)
2495                                 args.minlen = blen - args.alignment;
2496                         args.minalignslop = 0;
2497                 } else {
2498                         /*
2499                          * First try an exact bno allocation.
2500                          * If it fails then do a near or start bno
2501                          * allocation with alignment turned on.
2502                          */
2503                         atype = args.type;
2504                         tryagain = 1;
2505                         args.type = XFS_ALLOCTYPE_THIS_BNO;
2506                         args.alignment = 1;
2507                         /*
2508                          * Compute the minlen+alignment for the
2509                          * next case.  Set slop so that the value
2510                          * of minlen+alignment+slop doesn't go up
2511                          * between the calls.
2512                          */
2513                         if (blen > mp->m_dalign && blen <= args.maxlen)
2514                                 nextminlen = blen - mp->m_dalign;
2515                         else
2516                                 nextminlen = args.minlen;
2517                         if (nextminlen + mp->m_dalign > args.minlen + 1)
2518                                 args.minalignslop =
2519                                         nextminlen + mp->m_dalign -
2520                                         args.minlen - 1;
2521                         else
2522                                 args.minalignslop = 0;
2523                 }
2524         } else {
2525                 args.alignment = 1;
2526                 args.minalignslop = 0;
2527         }
2528         args.minleft = ap->minleft;
2529         args.wasdel = ap->wasdel;
2530         args.isfl = 0;
2531         args.userdata = ap->userdata;
2532         if ((error = xfs_alloc_vextent(&args)))
2533                 return error;
2534         if (tryagain && args.fsbno == NULLFSBLOCK) {
2535                 /*
2536                  * Exact allocation failed. Now try with alignment
2537                  * turned on.
2538                  */
2539                 args.type = atype;
2540                 args.fsbno = ap->blkno;
2541                 args.alignment = mp->m_dalign;
2542                 args.minlen = nextminlen;
2543                 args.minalignslop = 0;
2544                 isaligned = 1;
2545                 if ((error = xfs_alloc_vextent(&args)))
2546                         return error;
2547         }
2548         if (isaligned && args.fsbno == NULLFSBLOCK) {
2549                 /*
2550                  * allocation failed, so turn off alignment and
2551                  * try again.
2552                  */
2553                 args.type = atype;
2554                 args.fsbno = ap->blkno;
2555                 args.alignment = 0;
2556                 if ((error = xfs_alloc_vextent(&args)))
2557                         return error;
2558         }
2559         if (args.fsbno == NULLFSBLOCK && nullfb &&
2560             args.minlen > ap->minlen) {
2561                 args.minlen = ap->minlen;
2562                 args.type = XFS_ALLOCTYPE_START_BNO;
2563                 args.fsbno = ap->blkno;
2564                 if ((error = xfs_alloc_vextent(&args)))
2565                         return error;
2566         }
2567         if (args.fsbno == NULLFSBLOCK && nullfb) {
2568                 args.fsbno = 0;
2569                 args.type = XFS_ALLOCTYPE_FIRST_AG;
2570                 args.total = ap->minlen;
2571                 args.minleft = 0;
2572                 if ((error = xfs_alloc_vextent(&args)))
2573                         return error;
2574                 ap->flist->xbf_low = 1;
2575         }
2576         if (args.fsbno != NULLFSBLOCK) {
2577                 /*
2578                  * check the allocation happened at the same or higher AG than
2579                  * the first block that was allocated.
2580                  */
2581                 ASSERT(*ap->firstblock == NULLFSBLOCK ||
2582                        XFS_FSB_TO_AGNO(mp, *ap->firstblock) ==
2583                        XFS_FSB_TO_AGNO(mp, args.fsbno) ||
2584                        (ap->flist->xbf_low &&
2585                         XFS_FSB_TO_AGNO(mp, *ap->firstblock) <
2586                         XFS_FSB_TO_AGNO(mp, args.fsbno)));
2587
2588                 ap->blkno = args.fsbno;
2589                 if (*ap->firstblock == NULLFSBLOCK)
2590                         *ap->firstblock = args.fsbno;
2591                 ASSERT(nullfb || fb_agno == args.agno ||
2592                        (ap->flist->xbf_low && fb_agno < args.agno));
2593                 ap->length = args.len;
2594                 ap->ip->i_d.di_nblocks += args.len;
2595                 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
2596                 if (ap->wasdel)
2597                         ap->ip->i_delayed_blks -= args.len;
2598                 /*
2599                  * Adjust the disk quota also. This was reserved
2600                  * earlier.
2601                  */
2602                 xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
2603                         ap->wasdel ? XFS_TRANS_DQ_DELBCOUNT :
2604                                         XFS_TRANS_DQ_BCOUNT,
2605                         (long) args.len);
2606         } else {
2607                 ap->blkno = NULLFSBLOCK;
2608                 ap->length = 0;
2609         }
2610         return 0;
2611 }
2612
2613 /*
2614  * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
2615  * It figures out where to ask the underlying allocator to put the new extent.
2616  */
2617 STATIC int
2618 xfs_bmap_alloc(
2619         xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
2620 {
2621         if (XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata)
2622                 return xfs_bmap_rtalloc(ap);
2623         return xfs_bmap_btalloc(ap);
2624 }
2625
2626 /*
2627  * Transform a btree format file with only one leaf node, where the
2628  * extents list will fit in the inode, into an extents format file.
2629  * Since the file extents are already in-core, all we have to do is
2630  * give up the space for the btree root and pitch the leaf block.
2631  */
2632 STATIC int                              /* error */
2633 xfs_bmap_btree_to_extents(
2634         xfs_trans_t             *tp,    /* transaction pointer */
2635         xfs_inode_t             *ip,    /* incore inode pointer */
2636         xfs_btree_cur_t         *cur,   /* btree cursor */
2637         int                     *logflagsp, /* inode logging flags */
2638         int                     whichfork)  /* data or attr fork */
2639 {
2640         /* REFERENCED */
2641         struct xfs_btree_block  *cblock;/* child btree block */
2642         xfs_fsblock_t           cbno;   /* child block number */
2643         xfs_buf_t               *cbp;   /* child block's buffer */
2644         int                     error;  /* error return value */
2645         xfs_ifork_t             *ifp;   /* inode fork data */
2646         xfs_mount_t             *mp;    /* mount point structure */
2647         __be64                  *pp;    /* ptr to block address */
2648         struct xfs_btree_block  *rblock;/* root btree block */
2649
2650         mp = ip->i_mount;
2651         ifp = XFS_IFORK_PTR(ip, whichfork);
2652         ASSERT(ifp->if_flags & XFS_IFEXTENTS);
2653         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
2654         rblock = ifp->if_broot;
2655         ASSERT(be16_to_cpu(rblock->bb_level) == 1);
2656         ASSERT(be16_to_cpu(rblock->bb_numrecs) == 1);
2657         ASSERT(xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0) == 1);
2658         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, rblock, 1, ifp->if_broot_bytes);
2659         cbno = be64_to_cpu(*pp);
2660         *logflagsp = 0;
2661 #ifdef DEBUG
2662         if ((error = xfs_btree_check_lptr(cur, cbno, 1)))
2663                 return error;
2664 #endif
2665         error = xfs_btree_read_bufl(mp, tp, cbno, 0, &cbp, XFS_BMAP_BTREE_REF,
2666                                 &xfs_bmbt_buf_ops);
2667         if (error)
2668                 return error;
2669         cblock = XFS_BUF_TO_BLOCK(cbp);
2670         if ((error = xfs_btree_check_block(cur, cblock, 0, cbp)))
2671                 return error;
2672         xfs_bmap_add_free(cbno, 1, cur->bc_private.b.flist, mp);
2673         ip->i_d.di_nblocks--;
2674         xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, -1L);
2675         xfs_trans_binval(tp, cbp);
2676         if (cur->bc_bufs[0] == cbp)
2677                 cur->bc_bufs[0] = NULL;
2678         xfs_iroot_realloc(ip, -1, whichfork);
2679         ASSERT(ifp->if_broot == NULL);
2680         ASSERT((ifp->if_flags & XFS_IFBROOT) == 0);
2681         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
2682         *logflagsp = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
2683         return 0;
2684 }
2685
2686 /*
2687  * Called by xfs_bmapi to update file extent records and the btree
2688  * after removing space (or undoing a delayed allocation).
2689  */
2690 STATIC int                              /* error */
2691 xfs_bmap_del_extent(
2692         xfs_inode_t             *ip,    /* incore inode pointer */
2693         xfs_trans_t             *tp,    /* current transaction pointer */
2694         xfs_extnum_t            *idx,   /* extent number to update/delete */
2695         xfs_bmap_free_t         *flist, /* list of extents to be freed */
2696         xfs_btree_cur_t         *cur,   /* if null, not a btree */
2697         xfs_bmbt_irec_t         *del,   /* data to remove from extents */
2698         int                     *logflagsp, /* inode logging flags */
2699         int                     whichfork) /* data or attr fork */
2700 {
2701         xfs_filblks_t           da_new; /* new delay-alloc indirect blocks */
2702         xfs_filblks_t           da_old; /* old delay-alloc indirect blocks */
2703         xfs_fsblock_t           del_endblock=0; /* first block past del */
2704         xfs_fileoff_t           del_endoff;     /* first offset past del */
2705         int                     delay;  /* current block is delayed allocated */
2706         int                     do_fx;  /* free extent at end of routine */
2707         xfs_bmbt_rec_host_t     *ep;    /* current extent entry pointer */
2708         int                     error;  /* error return value */
2709         int                     flags;  /* inode logging flags */
2710         xfs_bmbt_irec_t         got;    /* current extent entry */
2711         xfs_fileoff_t           got_endoff;     /* first offset past got */
2712         int                     i;      /* temp state */
2713         xfs_ifork_t             *ifp;   /* inode fork pointer */
2714         xfs_mount_t             *mp;    /* mount structure */
2715         xfs_filblks_t           nblks;  /* quota/sb block count */
2716         xfs_bmbt_irec_t         new;    /* new record to be inserted */
2717         /* REFERENCED */
2718         uint                    qfield; /* quota field to update */
2719         xfs_filblks_t           temp;   /* for indirect length calculations */
2720         xfs_filblks_t           temp2;  /* for indirect length calculations */
2721         int                     state = 0;
2722
2723         XFS_STATS_INC(xs_del_exlist);
2724
2725         if (whichfork == XFS_ATTR_FORK)
2726                 state |= BMAP_ATTRFORK;
2727
2728         mp = ip->i_mount;
2729         ifp = XFS_IFORK_PTR(ip, whichfork);
2730         ASSERT((*idx >= 0) && (*idx < ifp->if_bytes /
2731                 (uint)sizeof(xfs_bmbt_rec_t)));
2732         ASSERT(del->br_blockcount > 0);
2733         ep = xfs_iext_get_ext(ifp, *idx);
2734         xfs_bmbt_get_all(ep, &got);
2735         ASSERT(got.br_startoff <= del->br_startoff);
2736         del_endoff = del->br_startoff + del->br_blockcount;
2737         got_endoff = got.br_startoff + got.br_blockcount;
2738         ASSERT(got_endoff >= del_endoff);
2739         delay = isnullstartblock(got.br_startblock);
2740         ASSERT(isnullstartblock(del->br_startblock) == delay);
2741         flags = 0;
2742         qfield = 0;
2743         error = 0;
2744         /*
2745          * If deleting a real allocation, must free up the disk space.
2746          */
2747         if (!delay) {
2748                 flags = XFS_ILOG_CORE;
2749                 /*
2750                  * Realtime allocation.  Free it and record di_nblocks update.
2751                  */
2752                 if (whichfork == XFS_DATA_FORK && XFS_IS_REALTIME_INODE(ip)) {
2753                         xfs_fsblock_t   bno;
2754                         xfs_filblks_t   len;
2755
2756                         ASSERT(do_mod(del->br_blockcount,
2757                                       mp->m_sb.sb_rextsize) == 0);
2758                         ASSERT(do_mod(del->br_startblock,
2759                                       mp->m_sb.sb_rextsize) == 0);
2760                         bno = del->br_startblock;
2761                         len = del->br_blockcount;
2762                         do_div(bno, mp->m_sb.sb_rextsize);
2763                         do_div(len, mp->m_sb.sb_rextsize);
2764                         error = xfs_rtfree_extent(tp, bno, (xfs_extlen_t)len);
2765                         if (error)
2766                                 goto done;
2767                         do_fx = 0;
2768                         nblks = len * mp->m_sb.sb_rextsize;
2769                         qfield = XFS_TRANS_DQ_RTBCOUNT;
2770                 }
2771                 /*
2772                  * Ordinary allocation.
2773                  */
2774                 else {
2775                         do_fx = 1;
2776                         nblks = del->br_blockcount;
2777                         qfield = XFS_TRANS_DQ_BCOUNT;
2778                 }
2779                 /*
2780                  * Set up del_endblock and cur for later.
2781                  */
2782                 del_endblock = del->br_startblock + del->br_blockcount;
2783                 if (cur) {
2784                         if ((error = xfs_bmbt_lookup_eq(cur, got.br_startoff,
2785                                         got.br_startblock, got.br_blockcount,
2786                                         &i)))
2787                                 goto done;
2788                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2789                 }
2790                 da_old = da_new = 0;
2791         } else {
2792                 da_old = startblockval(got.br_startblock);
2793                 da_new = 0;
2794                 nblks = 0;
2795                 do_fx = 0;
2796         }
2797         /*
2798          * Set flag value to use in switch statement.
2799          * Left-contig is 2, right-contig is 1.
2800          */
2801         switch (((got.br_startoff == del->br_startoff) << 1) |
2802                 (got_endoff == del_endoff)) {
2803         case 3:
2804                 /*
2805                  * Matches the whole extent.  Delete the entry.
2806                  */
2807                 xfs_iext_remove(ip, *idx, 1,
2808                                 whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0);
2809                 --*idx;
2810                 if (delay)
2811                         break;
2812
2813                 XFS_IFORK_NEXT_SET(ip, whichfork,
2814                         XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
2815                 flags |= XFS_ILOG_CORE;
2816                 if (!cur) {
2817                         flags |= xfs_ilog_fext(whichfork);
2818                         break;
2819                 }
2820                 if ((error = xfs_btree_delete(cur, &i)))
2821                         goto done;
2822                 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2823                 break;
2824
2825         case 2:
2826                 /*
2827                  * Deleting the first part of the extent.
2828                  */
2829                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2830                 xfs_bmbt_set_startoff(ep, del_endoff);
2831                 temp = got.br_blockcount - del->br_blockcount;
2832                 xfs_bmbt_set_blockcount(ep, temp);
2833                 if (delay) {
2834                         temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2835                                 da_old);
2836                         xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
2837                         trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2838                         da_new = temp;
2839                         break;
2840                 }
2841                 xfs_bmbt_set_startblock(ep, del_endblock);
2842                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2843                 if (!cur) {
2844                         flags |= xfs_ilog_fext(whichfork);
2845                         break;
2846                 }
2847                 if ((error = xfs_bmbt_update(cur, del_endoff, del_endblock,
2848                                 got.br_blockcount - del->br_blockcount,
2849                                 got.br_state)))
2850                         goto done;
2851                 break;
2852
2853         case 1:
2854                 /*
2855                  * Deleting the last part of the extent.
2856                  */
2857                 temp = got.br_blockcount - del->br_blockcount;
2858                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2859                 xfs_bmbt_set_blockcount(ep, temp);
2860                 if (delay) {
2861                         temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2862                                 da_old);
2863                         xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
2864                         trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2865                         da_new = temp;
2866                         break;
2867                 }
2868                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2869                 if (!cur) {
2870                         flags |= xfs_ilog_fext(whichfork);
2871                         break;
2872                 }
2873                 if ((error = xfs_bmbt_update(cur, got.br_startoff,
2874                                 got.br_startblock,
2875                                 got.br_blockcount - del->br_blockcount,
2876                                 got.br_state)))
2877                         goto done;
2878                 break;
2879
2880         case 0:
2881                 /*
2882                  * Deleting the middle of the extent.
2883                  */
2884                 temp = del->br_startoff - got.br_startoff;
2885                 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2886                 xfs_bmbt_set_blockcount(ep, temp);
2887                 new.br_startoff = del_endoff;
2888                 temp2 = got_endoff - del_endoff;
2889                 new.br_blockcount = temp2;
2890                 new.br_state = got.br_state;
2891                 if (!delay) {
2892                         new.br_startblock = del_endblock;
2893                         flags |= XFS_ILOG_CORE;
2894                         if (cur) {
2895                                 if ((error = xfs_bmbt_update(cur,
2896                                                 got.br_startoff,
2897                                                 got.br_startblock, temp,
2898                                                 got.br_state)))
2899                                         goto done;
2900                                 if ((error = xfs_btree_increment(cur, 0, &i)))
2901                                         goto done;
2902                                 cur->bc_rec.b = new;
2903                                 error = xfs_btree_insert(cur, &i);
2904                                 if (error && error != ENOSPC)
2905                                         goto done;
2906                                 /*
2907                                  * If get no-space back from btree insert,
2908                                  * it tried a split, and we have a zero
2909                                  * block reservation.
2910                                  * Fix up our state and return the error.
2911                                  */
2912                                 if (error == ENOSPC) {
2913                                         /*
2914                                          * Reset the cursor, don't trust
2915                                          * it after any insert operation.
2916                                          */
2917                                         if ((error = xfs_bmbt_lookup_eq(cur,
2918                                                         got.br_startoff,
2919                                                         got.br_startblock,
2920                                                         temp, &i)))
2921                                                 goto done;
2922                                         XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2923                                         /*
2924                                          * Update the btree record back
2925                                          * to the original value.
2926                                          */
2927                                         if ((error = xfs_bmbt_update(cur,
2928                                                         got.br_startoff,
2929                                                         got.br_startblock,
2930                                                         got.br_blockcount,
2931                                                         got.br_state)))
2932                                                 goto done;
2933                                         /*
2934                                          * Reset the extent record back
2935                                          * to the original value.
2936                                          */
2937                                         xfs_bmbt_set_blockcount(ep,
2938                                                 got.br_blockcount);
2939                                         flags = 0;
2940                                         error = XFS_ERROR(ENOSPC);
2941                                         goto done;
2942                                 }
2943                                 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2944                         } else
2945                                 flags |= xfs_ilog_fext(whichfork);
2946                         XFS_IFORK_NEXT_SET(ip, whichfork,
2947                                 XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
2948                 } else {
2949                         ASSERT(whichfork == XFS_DATA_FORK);
2950                         temp = xfs_bmap_worst_indlen(ip, temp);
2951                         xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
2952                         temp2 = xfs_bmap_worst_indlen(ip, temp2);
2953                         new.br_startblock = nullstartblock((int)temp2);
2954                         da_new = temp + temp2;
2955                         while (da_new > da_old) {
2956                                 if (temp) {
2957                                         temp--;
2958                                         da_new--;
2959                                         xfs_bmbt_set_startblock(ep,
2960                                                 nullstartblock((int)temp));
2961                                 }
2962                                 if (da_new == da_old)
2963                                         break;
2964                                 if (temp2) {
2965                                         temp2--;
2966                                         da_new--;
2967                                         new.br_startblock =
2968                                                 nullstartblock((int)temp2);
2969                                 }
2970                         }
2971                 }
2972                 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2973                 xfs_iext_insert(ip, *idx + 1, 1, &new, state);
2974                 ++*idx;
2975                 break;
2976         }
2977         /*
2978          * If we need to, add to list of extents to delete.
2979          */
2980         if (do_fx)
2981                 xfs_bmap_add_free(del->br_startblock, del->br_blockcount, flist,
2982                         mp);
2983         /*
2984          * Adjust inode # blocks in the file.
2985          */
2986         if (nblks)
2987                 ip->i_d.di_nblocks -= nblks;
2988         /*
2989          * Adjust quota data.
2990          */
2991         if (qfield)
2992                 xfs_trans_mod_dquot_byino(tp, ip, qfield, (long)-nblks);
2993
2994         /*
2995          * Account for change in delayed indirect blocks.
2996          * Nothing to do for disk quota accounting here.
2997          */
2998         ASSERT(da_old >= da_new);
2999         if (da_old > da_new) {
3000                 xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
3001                         (int64_t)(da_old - da_new), 0);
3002         }
3003 done:
3004         *logflagsp = flags;
3005         return error;
3006 }
3007
3008 /*
3009  * Remove the entry "free" from the free item list.  Prev points to the
3010  * previous entry, unless "free" is the head of the list.
3011  */
3012 STATIC void
3013 xfs_bmap_del_free(
3014         xfs_bmap_free_t         *flist, /* free item list header */
3015         xfs_bmap_free_item_t    *prev,  /* previous item on list, if any */
3016         xfs_bmap_free_item_t    *free)  /* list item to be freed */
3017 {
3018         if (prev)
3019                 prev->xbfi_next = free->xbfi_next;
3020         else
3021                 flist->xbf_first = free->xbfi_next;
3022         flist->xbf_count--;
3023         kmem_zone_free(xfs_bmap_free_item_zone, free);
3024 }
3025
3026 /*
3027  * Convert an extents-format file into a btree-format file.
3028  * The new file will have a root block (in the inode) and a single child block.
3029  */
3030 STATIC int                                      /* error */
3031 xfs_bmap_extents_to_btree(
3032         xfs_trans_t             *tp,            /* transaction pointer */
3033         xfs_inode_t             *ip,            /* incore inode pointer */
3034         xfs_fsblock_t           *firstblock,    /* first-block-allocated */
3035         xfs_bmap_free_t         *flist,         /* blocks freed in xaction */
3036         xfs_btree_cur_t         **curp,         /* cursor returned to caller */
3037         int                     wasdel,         /* converting a delayed alloc */
3038         int                     *logflagsp,     /* inode logging flags */
3039         int                     whichfork)      /* data or attr fork */
3040 {
3041         struct xfs_btree_block  *ablock;        /* allocated (child) bt block */
3042         xfs_buf_t               *abp;           /* buffer for ablock */
3043         xfs_alloc_arg_t         args;           /* allocation arguments */
3044         xfs_bmbt_rec_t          *arp;           /* child record pointer */
3045         struct xfs_btree_block  *block;         /* btree root block */
3046         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
3047         xfs_bmbt_rec_host_t     *ep;            /* extent record pointer */
3048         int                     error;          /* error return value */
3049         xfs_extnum_t            i, cnt;         /* extent record index */
3050         xfs_ifork_t             *ifp;           /* inode fork pointer */
3051         xfs_bmbt_key_t          *kp;            /* root block key pointer */
3052         xfs_mount_t             *mp;            /* mount structure */
3053         xfs_extnum_t            nextents;       /* number of file extents */
3054         xfs_bmbt_ptr_t          *pp;            /* root block address pointer */
3055
3056         ifp = XFS_IFORK_PTR(ip, whichfork);
3057         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS);
3058
3059         /*
3060          * Make space in the inode incore.
3061          */
3062         xfs_iroot_realloc(ip, 1, whichfork);
3063         ifp->if_flags |= XFS_IFBROOT;
3064
3065         /*
3066          * Fill in the root.
3067          */
3068         block = ifp->if_broot;
3069         block->bb_magic = cpu_to_be32(XFS_BMAP_MAGIC);
3070         block->bb_level = cpu_to_be16(1);
3071         block->bb_numrecs = cpu_to_be16(1);
3072         block->bb_u.l.bb_leftsib = cpu_to_be64(NULLDFSBNO);
3073         block->bb_u.l.bb_rightsib = cpu_to_be64(NULLDFSBNO);
3074
3075         /*
3076          * Need a cursor.  Can't allocate until bb_level is filled in.
3077          */
3078         mp = ip->i_mount;
3079         cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
3080         cur->bc_private.b.firstblock = *firstblock;
3081         cur->bc_private.b.flist = flist;
3082         cur->bc_private.b.flags = wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
3083         /*
3084          * Convert to a btree with two levels, one record in root.
3085          */
3086         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_BTREE);
3087         memset(&args, 0, sizeof(args));
3088         args.tp = tp;
3089         args.mp = mp;
3090         args.firstblock = *firstblock;
3091         if (*firstblock == NULLFSBLOCK) {
3092                 args.type = XFS_ALLOCTYPE_START_BNO;
3093                 args.fsbno = XFS_INO_TO_FSB(mp, ip->i_ino);
3094         } else if (flist->xbf_low) {
3095                 args.type = XFS_ALLOCTYPE_START_BNO;
3096                 args.fsbno = *firstblock;
3097         } else {
3098                 args.type = XFS_ALLOCTYPE_NEAR_BNO;
3099                 args.fsbno = *firstblock;
3100         }
3101         args.minlen = args.maxlen = args.prod = 1;
3102         args.total = args.minleft = args.alignment = args.mod = args.isfl =
3103                 args.minalignslop = 0;
3104         args.wasdel = wasdel;
3105         *logflagsp = 0;
3106         if ((error = xfs_alloc_vextent(&args))) {
3107                 xfs_iroot_realloc(ip, -1, whichfork);
3108                 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
3109                 return error;
3110         }
3111         /*
3112          * Allocation can't fail, the space was reserved.
3113          */
3114         ASSERT(args.fsbno != NULLFSBLOCK);
3115         ASSERT(*firstblock == NULLFSBLOCK ||
3116                args.agno == XFS_FSB_TO_AGNO(mp, *firstblock) ||
3117                (flist->xbf_low &&
3118                 args.agno > XFS_FSB_TO_AGNO(mp, *firstblock)));
3119         *firstblock = cur->bc_private.b.firstblock = args.fsbno;
3120         cur->bc_private.b.allocated++;
3121         ip->i_d.di_nblocks++;
3122         xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, 1L);
3123         abp = xfs_btree_get_bufl(mp, tp, args.fsbno, 0);
3124         /*
3125          * Fill in the child block.
3126          */
3127         abp->b_ops = &xfs_bmbt_buf_ops;
3128         ablock = XFS_BUF_TO_BLOCK(abp);
3129         ablock->bb_magic = cpu_to_be32(XFS_BMAP_MAGIC);
3130         ablock->bb_level = 0;
3131         ablock->bb_u.l.bb_leftsib = cpu_to_be64(NULLDFSBNO);
3132         ablock->bb_u.l.bb_rightsib = cpu_to_be64(NULLDFSBNO);
3133         arp = XFS_BMBT_REC_ADDR(mp, ablock, 1);
3134         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3135         for (cnt = i = 0; i < nextents; i++) {
3136                 ep = xfs_iext_get_ext(ifp, i);
3137                 if (!isnullstartblock(xfs_bmbt_get_startblock(ep))) {
3138                         arp->l0 = cpu_to_be64(ep->l0);
3139                         arp->l1 = cpu_to_be64(ep->l1);
3140                         arp++; cnt++;
3141                 }
3142         }
3143         ASSERT(cnt == XFS_IFORK_NEXTENTS(ip, whichfork));
3144         xfs_btree_set_numrecs(ablock, cnt);
3145
3146         /*
3147          * Fill in the root key and pointer.
3148          */
3149         kp = XFS_BMBT_KEY_ADDR(mp, block, 1);
3150         arp = XFS_BMBT_REC_ADDR(mp, ablock, 1);
3151         kp->br_startoff = cpu_to_be64(xfs_bmbt_disk_get_startoff(arp));
3152         pp = XFS_BMBT_PTR_ADDR(mp, block, 1, xfs_bmbt_get_maxrecs(cur,
3153                                                 be16_to_cpu(block->bb_level)));
3154         *pp = cpu_to_be64(args.fsbno);
3155
3156         /*
3157          * Do all this logging at the end so that
3158          * the root is at the right level.
3159          */
3160         xfs_btree_log_block(cur, abp, XFS_BB_ALL_BITS);
3161         xfs_btree_log_recs(cur, abp, 1, be16_to_cpu(ablock->bb_numrecs));
3162         ASSERT(*curp == NULL);
3163         *curp = cur;
3164         *logflagsp = XFS_ILOG_CORE | xfs_ilog_fbroot(whichfork);
3165         return 0;
3166 }
3167
3168 /*
3169  * Calculate the default attribute fork offset for newly created inodes.
3170  */
3171 uint
3172 xfs_default_attroffset(
3173         struct xfs_inode        *ip)
3174 {
3175         struct xfs_mount        *mp = ip->i_mount;
3176         uint                    offset;
3177
3178         if (mp->m_sb.sb_inodesize == 256) {
3179                 offset = XFS_LITINO(mp) -
3180                                 XFS_BMDR_SPACE_CALC(MINABTPTRS);
3181         } else {
3182                 offset = XFS_BMDR_SPACE_CALC(6 * MINABTPTRS);
3183         }
3184
3185         ASSERT(offset < XFS_LITINO(mp));
3186         return offset;
3187 }
3188
3189 /*
3190  * Helper routine to reset inode di_forkoff field when switching
3191  * attribute fork from local to extent format - we reset it where
3192  * possible to make space available for inline data fork extents.
3193  */
3194 STATIC void
3195 xfs_bmap_forkoff_reset(
3196         xfs_mount_t     *mp,
3197         xfs_inode_t     *ip,
3198         int             whichfork)
3199 {
3200         if (whichfork == XFS_ATTR_FORK &&
3201             ip->i_d.di_format != XFS_DINODE_FMT_DEV &&
3202             ip->i_d.di_format != XFS_DINODE_FMT_UUID &&
3203             ip->i_d.di_format != XFS_DINODE_FMT_BTREE) {
3204                 uint    dfl_forkoff = xfs_default_attroffset(ip) >> 3;
3205
3206                 if (dfl_forkoff > ip->i_d.di_forkoff)
3207                         ip->i_d.di_forkoff = dfl_forkoff;
3208         }
3209 }
3210
3211 /*
3212  * Convert a local file to an extents file.
3213  * This code is out of bounds for data forks of regular files,
3214  * since the file data needs to get logged so things will stay consistent.
3215  * (The bmap-level manipulations are ok, though).
3216  */
3217 STATIC int                              /* error */
3218 xfs_bmap_local_to_extents(
3219         xfs_trans_t     *tp,            /* transaction pointer */
3220         xfs_inode_t     *ip,            /* incore inode pointer */
3221         xfs_fsblock_t   *firstblock,    /* first block allocated in xaction */
3222         xfs_extlen_t    total,          /* total blocks needed by transaction */
3223         int             *logflagsp,     /* inode logging flags */
3224         int             whichfork)      /* data or attr fork */
3225 {
3226         int             error;          /* error return value */
3227         int             flags;          /* logging flags returned */
3228         xfs_ifork_t     *ifp;           /* inode fork pointer */
3229
3230         /*
3231          * We don't want to deal with the case of keeping inode data inline yet.
3232          * So sending the data fork of a regular inode is invalid.
3233          */
3234         ASSERT(!(S_ISREG(ip->i_d.di_mode) && whichfork == XFS_DATA_FORK));
3235         ifp = XFS_IFORK_PTR(ip, whichfork);
3236         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
3237         flags = 0;
3238         error = 0;
3239         if (ifp->if_bytes) {
3240                 xfs_alloc_arg_t args;   /* allocation arguments */
3241                 xfs_buf_t       *bp;    /* buffer for extent block */
3242                 xfs_bmbt_rec_host_t *ep;/* extent record pointer */
3243
3244                 memset(&args, 0, sizeof(args));
3245                 args.tp = tp;
3246                 args.mp = ip->i_mount;
3247                 args.firstblock = *firstblock;
3248                 ASSERT((ifp->if_flags &
3249                         (XFS_IFINLINE|XFS_IFEXTENTS|XFS_IFEXTIREC)) == XFS_IFINLINE);
3250                 /*
3251                  * Allocate a block.  We know we need only one, since the
3252                  * file currently fits in an inode.
3253                  */
3254                 if (*firstblock == NULLFSBLOCK) {
3255                         args.fsbno = XFS_INO_TO_FSB(args.mp, ip->i_ino);
3256                         args.type = XFS_ALLOCTYPE_START_BNO;
3257                 } else {
3258                         args.fsbno = *firstblock;
3259                         args.type = XFS_ALLOCTYPE_NEAR_BNO;
3260                 }
3261                 args.total = total;
3262                 args.mod = args.minleft = args.alignment = args.wasdel =
3263                         args.isfl = args.minalignslop = 0;
3264                 args.minlen = args.maxlen = args.prod = 1;
3265                 if ((error = xfs_alloc_vextent(&args)))
3266                         goto done;
3267                 /*
3268                  * Can't fail, the space was reserved.
3269                  */
3270                 ASSERT(args.fsbno != NULLFSBLOCK);
3271                 ASSERT(args.len == 1);
3272                 *firstblock = args.fsbno;
3273                 bp = xfs_btree_get_bufl(args.mp, tp, args.fsbno, 0);
3274                 bp->b_ops = &xfs_bmbt_buf_ops;
3275                 memcpy(bp->b_addr, ifp->if_u1.if_data, ifp->if_bytes);
3276                 xfs_trans_log_buf(tp, bp, 0, ifp->if_bytes - 1);
3277                 xfs_bmap_forkoff_reset(args.mp, ip, whichfork);
3278                 xfs_idata_realloc(ip, -ifp->if_bytes, whichfork);
3279                 xfs_iext_add(ifp, 0, 1);
3280                 ep = xfs_iext_get_ext(ifp, 0);
3281                 xfs_bmbt_set_allf(ep, 0, args.fsbno, 1, XFS_EXT_NORM);
3282                 trace_xfs_bmap_post_update(ip, 0,
3283                                 whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0,
3284                                 _THIS_IP_);
3285                 XFS_IFORK_NEXT_SET(ip, whichfork, 1);
3286                 ip->i_d.di_nblocks = 1;
3287                 xfs_trans_mod_dquot_byino(tp, ip,
3288                         XFS_TRANS_DQ_BCOUNT, 1L);
3289                 flags |= xfs_ilog_fext(whichfork);
3290         } else {
3291                 ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) == 0);
3292                 xfs_bmap_forkoff_reset(ip->i_mount, ip, whichfork);
3293         }
3294         ifp->if_flags &= ~XFS_IFINLINE;
3295         ifp->if_flags |= XFS_IFEXTENTS;
3296         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
3297         flags |= XFS_ILOG_CORE;
3298 done:
3299         *logflagsp = flags;
3300         return error;
3301 }
3302
3303 /*
3304  * Search the extent records for the entry containing block bno.
3305  * If bno lies in a hole, point to the next entry.  If bno lies
3306  * past eof, *eofp will be set, and *prevp will contain the last
3307  * entry (null if none).  Else, *lastxp will be set to the index
3308  * of the found entry; *gotp will contain the entry.
3309  */
3310 STATIC xfs_bmbt_rec_host_t *            /* pointer to found extent entry */
3311 xfs_bmap_search_multi_extents(
3312         xfs_ifork_t     *ifp,           /* inode fork pointer */
3313         xfs_fileoff_t   bno,            /* block number searched for */
3314         int             *eofp,          /* out: end of file found */
3315         xfs_extnum_t    *lastxp,        /* out: last extent index */
3316         xfs_bmbt_irec_t *gotp,          /* out: extent entry found */
3317         xfs_bmbt_irec_t *prevp)         /* out: previous extent entry found */
3318 {
3319         xfs_bmbt_rec_host_t *ep;                /* extent record pointer */
3320         xfs_extnum_t    lastx;          /* last extent index */
3321
3322         /*
3323          * Initialize the extent entry structure to catch access to
3324          * uninitialized br_startblock field.
3325          */
3326         gotp->br_startoff = 0xffa5a5a5a5a5a5a5LL;
3327         gotp->br_blockcount = 0xa55a5a5a5a5a5a5aLL;
3328         gotp->br_state = XFS_EXT_INVALID;
3329 #if XFS_BIG_BLKNOS
3330         gotp->br_startblock = 0xffffa5a5a5a5a5a5LL;
3331 #else
3332         gotp->br_startblock = 0xffffa5a5;
3333 #endif
3334         prevp->br_startoff = NULLFILEOFF;
3335
3336         ep = xfs_iext_bno_to_ext(ifp, bno, &lastx);
3337         if (lastx > 0) {
3338                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx - 1), prevp);
3339         }
3340         if (lastx < (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t))) {
3341                 xfs_bmbt_get_all(ep, gotp);
3342                 *eofp = 0;
3343         } else {
3344                 if (lastx > 0) {
3345                         *gotp = *prevp;
3346                 }
3347                 *eofp = 1;
3348                 ep = NULL;
3349         }
3350         *lastxp = lastx;
3351         return ep;
3352 }
3353
3354 /*
3355  * Search the extents list for the inode, for the extent containing bno.
3356  * If bno lies in a hole, point to the next entry.  If bno lies past eof,
3357  * *eofp will be set, and *prevp will contain the last entry (null if none).
3358  * Else, *lastxp will be set to the index of the found
3359  * entry; *gotp will contain the entry.
3360  */
3361 STATIC xfs_bmbt_rec_host_t *                 /* pointer to found extent entry */
3362 xfs_bmap_search_extents(
3363         xfs_inode_t     *ip,            /* incore inode pointer */
3364         xfs_fileoff_t   bno,            /* block number searched for */
3365         int             fork,           /* data or attr fork */
3366         int             *eofp,          /* out: end of file found */
3367         xfs_extnum_t    *lastxp,        /* out: last extent index */
3368         xfs_bmbt_irec_t *gotp,          /* out: extent entry found */
3369         xfs_bmbt_irec_t *prevp)         /* out: previous extent entry found */
3370 {
3371         xfs_ifork_t     *ifp;           /* inode fork pointer */
3372         xfs_bmbt_rec_host_t  *ep;            /* extent record pointer */
3373
3374         XFS_STATS_INC(xs_look_exlist);
3375         ifp = XFS_IFORK_PTR(ip, fork);
3376
3377         ep = xfs_bmap_search_multi_extents(ifp, bno, eofp, lastxp, gotp, prevp);
3378
3379         if (unlikely(!(gotp->br_startblock) && (*lastxp != NULLEXTNUM) &&
3380                      !(XFS_IS_REALTIME_INODE(ip) && fork == XFS_DATA_FORK))) {
3381                 xfs_alert_tag(ip->i_mount, XFS_PTAG_FSBLOCK_ZERO,
3382                                 "Access to block zero in inode %llu "
3383                                 "start_block: %llx start_off: %llx "
3384                                 "blkcnt: %llx extent-state: %x lastx: %x\n",
3385                         (unsigned long long)ip->i_ino,
3386                         (unsigned long long)gotp->br_startblock,
3387                         (unsigned long long)gotp->br_startoff,
3388                         (unsigned long long)gotp->br_blockcount,
3389                         gotp->br_state, *lastxp);
3390                 *lastxp = NULLEXTNUM;
3391                 *eofp = 1;
3392                 return NULL;
3393         }
3394         return ep;
3395 }
3396
3397 /*
3398  * Compute the worst-case number of indirect blocks that will be used
3399  * for ip's delayed extent of length "len".
3400  */
3401 STATIC xfs_filblks_t
3402 xfs_bmap_worst_indlen(
3403         xfs_inode_t     *ip,            /* incore inode pointer */
3404         xfs_filblks_t   len)            /* delayed extent length */
3405 {
3406         int             level;          /* btree level number */
3407         int             maxrecs;        /* maximum record count at this level */
3408         xfs_mount_t     *mp;            /* mount structure */
3409         xfs_filblks_t   rval;           /* return value */
3410
3411         mp = ip->i_mount;
3412         maxrecs = mp->m_bmap_dmxr[0];
3413         for (level = 0, rval = 0;
3414              level < XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK);
3415              level++) {
3416                 len += maxrecs - 1;
3417                 do_div(len, maxrecs);
3418                 rval += len;
3419                 if (len == 1)
3420                         return rval + XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK) -
3421                                 level - 1;
3422                 if (level == 0)
3423                         maxrecs = mp->m_bmap_dmxr[1];
3424         }
3425         return rval;
3426 }
3427
3428 /*
3429  * Convert inode from non-attributed to attributed.
3430  * Must not be in a transaction, ip must not be locked.
3431  */
3432 int                                             /* error code */
3433 xfs_bmap_add_attrfork(
3434         xfs_inode_t             *ip,            /* incore inode pointer */
3435         int                     size,           /* space new attribute needs */
3436         int                     rsvd)           /* xact may use reserved blks */
3437 {
3438         xfs_fsblock_t           firstblock;     /* 1st block/ag allocated */
3439         xfs_bmap_free_t         flist;          /* freed extent records */
3440         xfs_mount_t             *mp;            /* mount structure */
3441         xfs_trans_t             *tp;            /* transaction pointer */
3442         int                     blks;           /* space reservation */
3443         int                     version = 1;    /* superblock attr version */
3444         int                     committed;      /* xaction was committed */
3445         int                     logflags;       /* logging flags */
3446         int                     error;          /* error return value */
3447
3448         ASSERT(XFS_IFORK_Q(ip) == 0);
3449
3450         mp = ip->i_mount;
3451         ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
3452         tp = xfs_trans_alloc(mp, XFS_TRANS_ADDAFORK);
3453         blks = XFS_ADDAFORK_SPACE_RES(mp);
3454         if (rsvd)
3455                 tp->t_flags |= XFS_TRANS_RESERVE;
3456         if ((error = xfs_trans_reserve(tp, blks, XFS_ADDAFORK_LOG_RES(mp), 0,
3457                         XFS_TRANS_PERM_LOG_RES, XFS_ADDAFORK_LOG_COUNT)))
3458                 goto error0;
3459         xfs_ilock(ip, XFS_ILOCK_EXCL);
3460         error = xfs_trans_reserve_quota_nblks(tp, ip, blks, 0, rsvd ?
3461                         XFS_QMOPT_RES_REGBLKS | XFS_QMOPT_FORCE_RES :
3462                         XFS_QMOPT_RES_REGBLKS);
3463         if (error) {
3464                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3465                 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES);
3466                 return error;
3467         }
3468         if (XFS_IFORK_Q(ip))
3469                 goto error1;
3470         if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS) {
3471                 /*
3472                  * For inodes coming from pre-6.2 filesystems.
3473                  */
3474                 ASSERT(ip->i_d.di_aformat == 0);
3475                 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
3476         }
3477         ASSERT(ip->i_d.di_anextents == 0);
3478
3479         xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
3480         xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
3481
3482         switch (ip->i_d.di_format) {
3483         case XFS_DINODE_FMT_DEV:
3484                 ip->i_d.di_forkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
3485                 break;
3486         case XFS_DINODE_FMT_UUID:
3487                 ip->i_d.di_forkoff = roundup(sizeof(uuid_t), 8) >> 3;
3488                 break;
3489         case XFS_DINODE_FMT_LOCAL:
3490         case XFS_DINODE_FMT_EXTENTS:
3491         case XFS_DINODE_FMT_BTREE:
3492                 ip->i_d.di_forkoff = xfs_attr_shortform_bytesfit(ip, size);
3493                 if (!ip->i_d.di_forkoff)
3494                         ip->i_d.di_forkoff = xfs_default_attroffset(ip) >> 3;
3495                 else if (mp->m_flags & XFS_MOUNT_ATTR2)
3496                         version = 2;
3497                 break;
3498         default:
3499                 ASSERT(0);
3500                 error = XFS_ERROR(EINVAL);
3501                 goto error1;
3502         }
3503
3504         ASSERT(ip->i_afp == NULL);
3505         ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP);
3506         ip->i_afp->if_flags = XFS_IFEXTENTS;
3507         logflags = 0;
3508         xfs_bmap_init(&flist, &firstblock);
3509         switch (ip->i_d.di_format) {
3510         case XFS_DINODE_FMT_LOCAL:
3511                 error = xfs_bmap_add_attrfork_local(tp, ip, &firstblock, &flist,
3512                         &logflags);
3513                 break;
3514         case XFS_DINODE_FMT_EXTENTS:
3515                 error = xfs_bmap_add_attrfork_extents(tp, ip, &firstblock,
3516                         &flist, &logflags);
3517                 break;
3518         case XFS_DINODE_FMT_BTREE:
3519                 error = xfs_bmap_add_attrfork_btree(tp, ip, &firstblock, &flist,
3520                         &logflags);
3521                 break;
3522         default:
3523                 error = 0;
3524                 break;
3525         }
3526         if (logflags)
3527                 xfs_trans_log_inode(tp, ip, logflags);
3528         if (error)
3529                 goto error2;
3530         if (!xfs_sb_version_hasattr(&mp->m_sb) ||
3531            (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2)) {
3532                 __int64_t sbfields = 0;
3533
3534                 spin_lock(&mp->m_sb_lock);
3535                 if (!xfs_sb_version_hasattr(&mp->m_sb)) {
3536                         xfs_sb_version_addattr(&mp->m_sb);
3537                         sbfields |= XFS_SB_VERSIONNUM;
3538                 }
3539                 if (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2) {
3540                         xfs_sb_version_addattr2(&mp->m_sb);
3541                         sbfields |= (XFS_SB_VERSIONNUM | XFS_SB_FEATURES2);
3542                 }
3543                 if (sbfields) {
3544                         spin_unlock(&mp->m_sb_lock);
3545                         xfs_mod_sb(tp, sbfields);
3546                 } else
3547                         spin_unlock(&mp->m_sb_lock);
3548         }
3549
3550         error = xfs_bmap_finish(&tp, &flist, &committed);
3551         if (error)
3552                 goto error2;
3553         return xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
3554 error2:
3555         xfs_bmap_cancel(&flist);
3556 error1:
3557         xfs_iunlock(ip, XFS_ILOCK_EXCL);
3558 error0:
3559         xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
3560         return error;
3561 }
3562
3563 /*
3564  * Add the extent to the list of extents to be free at transaction end.
3565  * The list is maintained sorted (by block number).
3566  */
3567 /* ARGSUSED */
3568 void
3569 xfs_bmap_add_free(
3570         xfs_fsblock_t           bno,            /* fs block number of extent */
3571         xfs_filblks_t           len,            /* length of extent */
3572         xfs_bmap_free_t         *flist,         /* list of extents */
3573         xfs_mount_t             *mp)            /* mount point structure */
3574 {
3575         xfs_bmap_free_item_t    *cur;           /* current (next) element */
3576         xfs_bmap_free_item_t    *new;           /* new element */
3577         xfs_bmap_free_item_t    *prev;          /* previous element */
3578 #ifdef DEBUG
3579         xfs_agnumber_t          agno;
3580         xfs_agblock_t           agbno;
3581
3582         ASSERT(bno != NULLFSBLOCK);
3583         ASSERT(len > 0);
3584         ASSERT(len <= MAXEXTLEN);
3585         ASSERT(!isnullstartblock(bno));
3586         agno = XFS_FSB_TO_AGNO(mp, bno);
3587         agbno = XFS_FSB_TO_AGBNO(mp, bno);
3588         ASSERT(agno < mp->m_sb.sb_agcount);
3589         ASSERT(agbno < mp->m_sb.sb_agblocks);
3590         ASSERT(len < mp->m_sb.sb_agblocks);
3591         ASSERT(agbno + len <= mp->m_sb.sb_agblocks);
3592 #endif
3593         ASSERT(xfs_bmap_free_item_zone != NULL);
3594         new = kmem_zone_alloc(xfs_bmap_free_item_zone, KM_SLEEP);
3595         new->xbfi_startblock = bno;
3596         new->xbfi_blockcount = (xfs_extlen_t)len;
3597         for (prev = NULL, cur = flist->xbf_first;
3598              cur != NULL;
3599              prev = cur, cur = cur->xbfi_next) {
3600                 if (cur->xbfi_startblock >= bno)
3601                         break;
3602         }
3603         if (prev)
3604                 prev->xbfi_next = new;
3605         else
3606                 flist->xbf_first = new;
3607         new->xbfi_next = cur;
3608         flist->xbf_count++;
3609 }
3610
3611 /*
3612  * Compute and fill in the value of the maximum depth of a bmap btree
3613  * in this filesystem.  Done once, during mount.
3614  */
3615 void
3616 xfs_bmap_compute_maxlevels(
3617         xfs_mount_t     *mp,            /* file system mount structure */
3618         int             whichfork)      /* data or attr fork */
3619 {
3620         int             level;          /* btree level */
3621         uint            maxblocks;      /* max blocks at this level */
3622         uint            maxleafents;    /* max leaf entries possible */
3623         int             maxrootrecs;    /* max records in root block */
3624         int             minleafrecs;    /* min records in leaf block */
3625         int             minnoderecs;    /* min records in node block */
3626         int             sz;             /* root block size */
3627
3628         /*
3629          * The maximum number of extents in a file, hence the maximum
3630          * number of leaf entries, is controlled by the type of di_nextents
3631          * (a signed 32-bit number, xfs_extnum_t), or by di_anextents
3632          * (a signed 16-bit number, xfs_aextnum_t).
3633          *
3634          * Note that we can no longer assume that if we are in ATTR1 that
3635          * the fork offset of all the inodes will be
3636          * (xfs_default_attroffset(ip) >> 3) because we could have mounted
3637          * with ATTR2 and then mounted back with ATTR1, keeping the
3638          * di_forkoff's fixed but probably at various positions. Therefore,
3639          * for both ATTR1 and ATTR2 we have to assume the worst case scenario
3640          * of a minimum size available.
3641          */
3642         if (whichfork == XFS_DATA_FORK) {
3643                 maxleafents = MAXEXTNUM;
3644                 sz = XFS_BMDR_SPACE_CALC(MINDBTPTRS);
3645         } else {
3646                 maxleafents = MAXAEXTNUM;
3647                 sz = XFS_BMDR_SPACE_CALC(MINABTPTRS);
3648         }
3649         maxrootrecs = xfs_bmdr_maxrecs(mp, sz, 0);
3650         minleafrecs = mp->m_bmap_dmnr[0];
3651         minnoderecs = mp->m_bmap_dmnr[1];
3652         maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs;
3653         for (level = 1; maxblocks > 1; level++) {
3654                 if (maxblocks <= maxrootrecs)
3655                         maxblocks = 1;
3656                 else
3657                         maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs;
3658         }
3659         mp->m_bm_maxlevels[whichfork] = level;
3660 }
3661
3662 /*
3663  * Routine to be called at transaction's end by xfs_bmapi, xfs_bunmapi
3664  * caller.  Frees all the extents that need freeing, which must be done
3665  * last due to locking considerations.  We never free any extents in
3666  * the first transaction.
3667  *
3668  * Return 1 if the given transaction was committed and a new one
3669  * started, and 0 otherwise in the committed parameter.
3670  */
3671 int                                             /* error */
3672 xfs_bmap_finish(
3673         xfs_trans_t             **tp,           /* transaction pointer addr */
3674         xfs_bmap_free_t         *flist,         /* i/o: list extents to free */
3675         int                     *committed)     /* xact committed or not */
3676 {
3677         xfs_efd_log_item_t      *efd;           /* extent free data */
3678         xfs_efi_log_item_t      *efi;           /* extent free intention */
3679         int                     error;          /* error return value */
3680         xfs_bmap_free_item_t    *free;          /* free extent item */
3681         unsigned int            logres;         /* new log reservation */
3682         unsigned int            logcount;       /* new log count */
3683         xfs_mount_t             *mp;            /* filesystem mount structure */
3684         xfs_bmap_free_item_t    *next;          /* next item on free list */
3685         xfs_trans_t             *ntp;           /* new transaction pointer */
3686
3687         ASSERT((*tp)->t_flags & XFS_TRANS_PERM_LOG_RES);
3688         if (flist->xbf_count == 0) {
3689                 *committed = 0;
3690                 return 0;
3691         }
3692         ntp = *tp;
3693         efi = xfs_trans_get_efi(ntp, flist->xbf_count);
3694         for (free = flist->xbf_first; free; free = free->xbfi_next)
3695                 xfs_trans_log_efi_extent(ntp, efi, free->xbfi_startblock,
3696                         free->xbfi_blockcount);
3697         logres = ntp->t_log_res;
3698         logcount = ntp->t_log_count;
3699         ntp = xfs_trans_dup(*tp);
3700         error = xfs_trans_commit(*tp, 0);
3701         *tp = ntp;
3702         *committed = 1;
3703         /*
3704          * We have a new transaction, so we should return committed=1,
3705          * even though we're returning an error.
3706          */
3707         if (error)
3708                 return error;
3709
3710         /*
3711          * transaction commit worked ok so we can drop the extra ticket
3712          * reference that we gained in xfs_trans_dup()
3713          */
3714         xfs_log_ticket_put(ntp->t_ticket);
3715
3716         if ((error = xfs_trans_reserve(ntp, 0, logres, 0, XFS_TRANS_PERM_LOG_RES,
3717                         logcount)))
3718                 return error;
3719         efd = xfs_trans_get_efd(ntp, efi, flist->xbf_count);
3720         for (free = flist->xbf_first; free != NULL; free = next) {
3721                 next = free->xbfi_next;
3722                 if ((error = xfs_free_extent(ntp, free->xbfi_startblock,
3723                                 free->xbfi_blockcount))) {
3724                         /*
3725                          * The bmap free list will be cleaned up at a
3726                          * higher level.  The EFI will be canceled when
3727                          * this transaction is aborted.
3728                          * Need to force shutdown here to make sure it
3729                          * happens, since this transaction may not be
3730                          * dirty yet.
3731                          */
3732                         mp = ntp->t_mountp;
3733                         if (!XFS_FORCED_SHUTDOWN(mp))
3734                                 xfs_force_shutdown(mp,
3735                                                    (error == EFSCORRUPTED) ?
3736                                                    SHUTDOWN_CORRUPT_INCORE :
3737                                                    SHUTDOWN_META_IO_ERROR);
3738                         return error;
3739                 }
3740                 xfs_trans_log_efd_extent(ntp, efd, free->xbfi_startblock,
3741                         free->xbfi_blockcount);
3742                 xfs_bmap_del_free(flist, NULL, free);
3743         }
3744         return 0;
3745 }
3746
3747 /*
3748  * Free up any items left in the list.
3749  */
3750 void
3751 xfs_bmap_cancel(
3752         xfs_bmap_free_t         *flist) /* list of bmap_free_items */
3753 {
3754         xfs_bmap_free_item_t    *free;  /* free list item */
3755         xfs_bmap_free_item_t    *next;
3756
3757         if (flist->xbf_count == 0)
3758                 return;
3759         ASSERT(flist->xbf_first != NULL);
3760         for (free = flist->xbf_first; free; free = next) {
3761                 next = free->xbfi_next;
3762                 xfs_bmap_del_free(flist, NULL, free);
3763         }
3764         ASSERT(flist->xbf_count == 0);
3765 }
3766
3767 /*
3768  * Returns the file-relative block number of the first unused block(s)
3769  * in the file with at least "len" logically contiguous blocks free.
3770  * This is the lowest-address hole if the file has holes, else the first block
3771  * past the end of file.
3772  * Return 0 if the file is currently local (in-inode).
3773  */
3774 int                                             /* error */
3775 xfs_bmap_first_unused(
3776         xfs_trans_t     *tp,                    /* transaction pointer */
3777         xfs_inode_t     *ip,                    /* incore inode */
3778         xfs_extlen_t    len,                    /* size of hole to find */
3779         xfs_fileoff_t   *first_unused,          /* unused block */
3780         int             whichfork)              /* data or attr fork */
3781 {
3782         int             error;                  /* error return value */
3783         int             idx;                    /* extent record index */
3784         xfs_ifork_t     *ifp;                   /* inode fork pointer */
3785         xfs_fileoff_t   lastaddr;               /* last block number seen */
3786         xfs_fileoff_t   lowest;                 /* lowest useful block */
3787         xfs_fileoff_t   max;                    /* starting useful block */
3788         xfs_fileoff_t   off;                    /* offset for this block */
3789         xfs_extnum_t    nextents;               /* number of extent entries */
3790
3791         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE ||
3792                XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ||
3793                XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
3794         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
3795                 *first_unused = 0;
3796                 return 0;
3797         }
3798         ifp = XFS_IFORK_PTR(ip, whichfork);
3799         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
3800             (error = xfs_iread_extents(tp, ip, whichfork)))
3801                 return error;
3802         lowest = *first_unused;
3803         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3804         for (idx = 0, lastaddr = 0, max = lowest; idx < nextents; idx++) {
3805                 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, idx);
3806                 off = xfs_bmbt_get_startoff(ep);
3807                 /*
3808                  * See if the hole before this extent will work.
3809                  */
3810                 if (off >= lowest + len && off - max >= len) {
3811                         *first_unused = max;
3812                         return 0;
3813                 }
3814                 lastaddr = off + xfs_bmbt_get_blockcount(ep);
3815                 max = XFS_FILEOFF_MAX(lastaddr, lowest);
3816         }
3817         *first_unused = max;
3818         return 0;
3819 }
3820
3821 /*
3822  * Returns the file-relative block number of the last block + 1 before
3823  * last_block (input value) in the file.
3824  * This is not based on i_size, it is based on the extent records.
3825  * Returns 0 for local files, as they do not have extent records.
3826  */
3827 int                                             /* error */
3828 xfs_bmap_last_before(
3829         xfs_trans_t     *tp,                    /* transaction pointer */
3830         xfs_inode_t     *ip,                    /* incore inode */
3831         xfs_fileoff_t   *last_block,            /* last block */
3832         int             whichfork)              /* data or attr fork */
3833 {
3834         xfs_fileoff_t   bno;                    /* input file offset */
3835         int             eof;                    /* hit end of file */
3836         xfs_bmbt_rec_host_t *ep;                /* pointer to last extent */
3837         int             error;                  /* error return value */
3838         xfs_bmbt_irec_t got;                    /* current extent value */
3839         xfs_ifork_t     *ifp;                   /* inode fork pointer */
3840         xfs_extnum_t    lastx;                  /* last extent used */
3841         xfs_bmbt_irec_t prev;                   /* previous extent value */
3842
3843         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
3844             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
3845             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL)
3846                return XFS_ERROR(EIO);
3847         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
3848                 *last_block = 0;
3849                 return 0;
3850         }
3851         ifp = XFS_IFORK_PTR(ip, whichfork);
3852         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
3853             (error = xfs_iread_extents(tp, ip, whichfork)))
3854                 return error;
3855         bno = *last_block - 1;
3856         ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
3857                 &prev);
3858         if (eof || xfs_bmbt_get_startoff(ep) > bno) {
3859                 if (prev.br_startoff == NULLFILEOFF)
3860                         *last_block = 0;
3861                 else
3862                         *last_block = prev.br_startoff + prev.br_blockcount;
3863         }
3864         /*
3865          * Otherwise *last_block is already the right answer.
3866          */
3867         return 0;
3868 }
3869
3870 STATIC int
3871 xfs_bmap_last_extent(
3872         struct xfs_trans        *tp,
3873         struct xfs_inode        *ip,
3874         int                     whichfork,
3875         struct xfs_bmbt_irec    *rec,
3876         int                     *is_empty)
3877 {
3878         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, whichfork);
3879         int                     error;
3880         int                     nextents;
3881
3882         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
3883                 error = xfs_iread_extents(tp, ip, whichfork);
3884                 if (error)
3885                         return error;
3886         }
3887
3888         nextents = ifp->if_bytes / sizeof(xfs_bmbt_rec_t);
3889         if (nextents == 0) {
3890                 *is_empty = 1;
3891                 return 0;
3892         }
3893
3894         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, nextents - 1), rec);
3895         *is_empty = 0;
3896         return 0;
3897 }
3898
3899 /*
3900  * Check the last inode extent to determine whether this allocation will result
3901  * in blocks being allocated at the end of the file. When we allocate new data
3902  * blocks at the end of the file which do not start at the previous data block,
3903  * we will try to align the new blocks at stripe unit boundaries.
3904  *
3905  * Returns 0 in bma->aeof if the file (fork) is empty as any new write will be
3906  * at, or past the EOF.
3907  */
3908 STATIC int
3909 xfs_bmap_isaeof(
3910         struct xfs_bmalloca     *bma,
3911         int                     whichfork)
3912 {
3913         struct xfs_bmbt_irec    rec;
3914         int                     is_empty;
3915         int                     error;
3916
3917         bma->aeof = 0;
3918         error = xfs_bmap_last_extent(NULL, bma->ip, whichfork, &rec,
3919                                      &is_empty);
3920         if (error || is_empty)
3921                 return error;
3922
3923         /*
3924          * Check if we are allocation or past the last extent, or at least into
3925          * the last delayed allocated extent.
3926          */
3927         bma->aeof = bma->offset >= rec.br_startoff + rec.br_blockcount ||
3928                 (bma->offset >= rec.br_startoff &&
3929                  isnullstartblock(rec.br_startblock));
3930         return 0;
3931 }
3932
3933 /*
3934  * Check if the endoff is outside the last extent. If so the caller will grow
3935  * the allocation to a stripe unit boundary.  All offsets are considered outside
3936  * the end of file for an empty fork, so 1 is returned in *eof in that case.
3937  */
3938 int
3939 xfs_bmap_eof(
3940         struct xfs_inode        *ip,
3941         xfs_fileoff_t           endoff,
3942         int                     whichfork,
3943         int                     *eof)
3944 {
3945         struct xfs_bmbt_irec    rec;
3946         int                     error;
3947
3948         error = xfs_bmap_last_extent(NULL, ip, whichfork, &rec, eof);
3949         if (error || *eof)
3950                 return error;
3951
3952         *eof = endoff >= rec.br_startoff + rec.br_blockcount;
3953         return 0;
3954 }
3955
3956 /*
3957  * Returns the file-relative block number of the first block past eof in
3958  * the file.  This is not based on i_size, it is based on the extent records.
3959  * Returns 0 for local files, as they do not have extent records.
3960  */
3961 int
3962 xfs_bmap_last_offset(
3963         struct xfs_trans        *tp,
3964         struct xfs_inode        *ip,
3965         xfs_fileoff_t           *last_block,
3966         int                     whichfork)
3967 {
3968         struct xfs_bmbt_irec    rec;
3969         int                     is_empty;
3970         int                     error;
3971
3972         *last_block = 0;
3973
3974         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL)
3975                 return 0;
3976
3977         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
3978             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
3979                return XFS_ERROR(EIO);
3980
3981         error = xfs_bmap_last_extent(NULL, ip, whichfork, &rec, &is_empty);
3982         if (error || is_empty)
3983                 return error;
3984
3985         *last_block = rec.br_startoff + rec.br_blockcount;
3986         return 0;
3987 }
3988
3989 /*
3990  * Returns whether the selected fork of the inode has exactly one
3991  * block or not.  For the data fork we check this matches di_size,
3992  * implying the file's range is 0..bsize-1.
3993  */
3994 int                                     /* 1=>1 block, 0=>otherwise */
3995 xfs_bmap_one_block(
3996         xfs_inode_t     *ip,            /* incore inode */
3997         int             whichfork)      /* data or attr fork */
3998 {
3999         xfs_bmbt_rec_host_t *ep;        /* ptr to fork's extent */
4000         xfs_ifork_t     *ifp;           /* inode fork pointer */
4001         int             rval;           /* return value */
4002         xfs_bmbt_irec_t s;              /* internal version of extent */
4003
4004 #ifndef DEBUG
4005         if (whichfork == XFS_DATA_FORK)
4006                 return XFS_ISIZE(ip) == ip->i_mount->m_sb.sb_blocksize;
4007 #endif  /* !DEBUG */
4008         if (XFS_IFORK_NEXTENTS(ip, whichfork) != 1)
4009                 return 0;
4010         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
4011                 return 0;
4012         ifp = XFS_IFORK_PTR(ip, whichfork);
4013         ASSERT(ifp->if_flags & XFS_IFEXTENTS);
4014         ep = xfs_iext_get_ext(ifp, 0);
4015         xfs_bmbt_get_all(ep, &s);
4016         rval = s.br_startoff == 0 && s.br_blockcount == 1;
4017         if (rval && whichfork == XFS_DATA_FORK)
4018                 ASSERT(XFS_ISIZE(ip) == ip->i_mount->m_sb.sb_blocksize);
4019         return rval;
4020 }
4021
4022 STATIC int
4023 xfs_bmap_sanity_check(
4024         struct xfs_mount        *mp,
4025         struct xfs_buf          *bp,
4026         int                     level)
4027 {
4028         struct xfs_btree_block  *block = XFS_BUF_TO_BLOCK(bp);
4029
4030         if (block->bb_magic != cpu_to_be32(XFS_BMAP_MAGIC) ||
4031             be16_to_cpu(block->bb_level) != level ||
4032             be16_to_cpu(block->bb_numrecs) == 0 ||
4033             be16_to_cpu(block->bb_numrecs) > mp->m_bmap_dmxr[level != 0])
4034                 return 0;
4035         return 1;
4036 }
4037
4038 /*
4039  * Read in the extents to if_extents.
4040  * All inode fields are set up by caller, we just traverse the btree
4041  * and copy the records in. If the file system cannot contain unwritten
4042  * extents, the records are checked for no "state" flags.
4043  */
4044 int                                     /* error */
4045 xfs_bmap_read_extents(
4046         xfs_trans_t             *tp,    /* transaction pointer */
4047         xfs_inode_t             *ip,    /* incore inode */
4048         int                     whichfork) /* data or attr fork */
4049 {
4050         struct xfs_btree_block  *block; /* current btree block */
4051         xfs_fsblock_t           bno;    /* block # of "block" */
4052         xfs_buf_t               *bp;    /* buffer for "block" */
4053         int                     error;  /* error return value */
4054         xfs_exntfmt_t           exntf;  /* XFS_EXTFMT_NOSTATE, if checking */
4055         xfs_extnum_t            i, j;   /* index into the extents list */
4056         xfs_ifork_t             *ifp;   /* fork structure */
4057         int                     level;  /* btree level, for checking */
4058         xfs_mount_t             *mp;    /* file system mount structure */
4059         __be64                  *pp;    /* pointer to block address */
4060         /* REFERENCED */
4061         xfs_extnum_t            room;   /* number of entries there's room for */
4062
4063         bno = NULLFSBLOCK;
4064         mp = ip->i_mount;
4065         ifp = XFS_IFORK_PTR(ip, whichfork);
4066         exntf = (whichfork != XFS_DATA_FORK) ? XFS_EXTFMT_NOSTATE :
4067                                         XFS_EXTFMT_INODE(ip);
4068         block = ifp->if_broot;
4069         /*
4070          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
4071          */
4072         level = be16_to_cpu(block->bb_level);
4073         ASSERT(level > 0);
4074         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
4075         bno = be64_to_cpu(*pp);
4076         ASSERT(bno != NULLDFSBNO);
4077         ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
4078         ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
4079         /*
4080          * Go down the tree until leaf level is reached, following the first
4081          * pointer (leftmost) at each level.
4082          */
4083         while (level-- > 0) {
4084                 error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
4085                                 XFS_BMAP_BTREE_REF, &xfs_bmbt_buf_ops);
4086                 if (error)
4087                         return error;
4088                 block = XFS_BUF_TO_BLOCK(bp);
4089                 XFS_WANT_CORRUPTED_GOTO(
4090                         xfs_bmap_sanity_check(mp, bp, level),
4091                         error0);
4092                 if (level == 0)
4093                         break;
4094                 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
4095                 bno = be64_to_cpu(*pp);
4096                 XFS_WANT_CORRUPTED_GOTO(XFS_FSB_SANITY_CHECK(mp, bno), error0);
4097                 xfs_trans_brelse(tp, bp);
4098         }
4099         /*
4100          * Here with bp and block set to the leftmost leaf node in the tree.
4101          */
4102         room = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4103         i = 0;
4104         /*
4105          * Loop over all leaf nodes.  Copy information to the extent records.
4106          */
4107         for (;;) {
4108                 xfs_bmbt_rec_t  *frp;
4109                 xfs_fsblock_t   nextbno;
4110                 xfs_extnum_t    num_recs;
4111                 xfs_extnum_t    start;
4112
4113                 num_recs = xfs_btree_get_numrecs(block);
4114                 if (unlikely(i + num_recs > room)) {
4115                         ASSERT(i + num_recs <= room);
4116                         xfs_warn(ip->i_mount,
4117                                 "corrupt dinode %Lu, (btree extents).",
4118                                 (unsigned long long) ip->i_ino);
4119                         XFS_CORRUPTION_ERROR("xfs_bmap_read_extents(1)",
4120                                 XFS_ERRLEVEL_LOW, ip->i_mount, block);
4121                         goto error0;
4122                 }
4123                 XFS_WANT_CORRUPTED_GOTO(
4124                         xfs_bmap_sanity_check(mp, bp, 0),
4125                         error0);
4126                 /*
4127                  * Read-ahead the next leaf block, if any.
4128                  */
4129                 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
4130                 if (nextbno != NULLFSBLOCK)
4131                         xfs_btree_reada_bufl(mp, nextbno, 1,
4132                                              &xfs_bmbt_buf_ops);
4133                 /*
4134                  * Copy records into the extent records.
4135                  */
4136                 frp = XFS_BMBT_REC_ADDR(mp, block, 1);
4137                 start = i;
4138                 for (j = 0; j < num_recs; j++, i++, frp++) {
4139                         xfs_bmbt_rec_host_t *trp = xfs_iext_get_ext(ifp, i);
4140                         trp->l0 = be64_to_cpu(frp->l0);
4141                         trp->l1 = be64_to_cpu(frp->l1);
4142                 }
4143                 if (exntf == XFS_EXTFMT_NOSTATE) {
4144                         /*
4145                          * Check all attribute bmap btree records and
4146                          * any "older" data bmap btree records for a
4147                          * set bit in the "extent flag" position.
4148                          */
4149                         if (unlikely(xfs_check_nostate_extents(ifp,
4150                                         start, num_recs))) {
4151                                 XFS_ERROR_REPORT("xfs_bmap_read_extents(2)",
4152                                                  XFS_ERRLEVEL_LOW,
4153                                                  ip->i_mount);
4154                                 goto error0;
4155                         }
4156                 }
4157                 xfs_trans_brelse(tp, bp);
4158                 bno = nextbno;
4159                 /*
4160                  * If we've reached the end, stop.
4161                  */
4162                 if (bno == NULLFSBLOCK)
4163                         break;
4164                 error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
4165                                 XFS_BMAP_BTREE_REF, &xfs_bmbt_buf_ops);
4166                 if (error)
4167                         return error;
4168                 block = XFS_BUF_TO_BLOCK(bp);
4169         }
4170         ASSERT(i == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)));
4171         ASSERT(i == XFS_IFORK_NEXTENTS(ip, whichfork));
4172         XFS_BMAP_TRACE_EXLIST(ip, i, whichfork);
4173         return 0;
4174 error0:
4175         xfs_trans_brelse(tp, bp);
4176         return XFS_ERROR(EFSCORRUPTED);
4177 }
4178
4179 #ifdef DEBUG
4180 /*
4181  * Add bmap trace insert entries for all the contents of the extent records.
4182  */
4183 void
4184 xfs_bmap_trace_exlist(
4185         xfs_inode_t     *ip,            /* incore inode pointer */
4186         xfs_extnum_t    cnt,            /* count of entries in the list */
4187         int             whichfork,      /* data or attr fork */
4188         unsigned long   caller_ip)
4189 {
4190         xfs_extnum_t    idx;            /* extent record index */
4191         xfs_ifork_t     *ifp;           /* inode fork pointer */
4192         int             state = 0;
4193
4194         if (whichfork == XFS_ATTR_FORK)
4195                 state |= BMAP_ATTRFORK;
4196
4197         ifp = XFS_IFORK_PTR(ip, whichfork);
4198         ASSERT(cnt == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)));
4199         for (idx = 0; idx < cnt; idx++)
4200                 trace_xfs_extlist(ip, idx, whichfork, caller_ip);
4201 }
4202
4203 /*
4204  * Validate that the bmbt_irecs being returned from bmapi are valid
4205  * given the callers original parameters.  Specifically check the
4206  * ranges of the returned irecs to ensure that they only extent beyond
4207  * the given parameters if the XFS_BMAPI_ENTIRE flag was set.
4208  */
4209 STATIC void
4210 xfs_bmap_validate_ret(
4211         xfs_fileoff_t           bno,
4212         xfs_filblks_t           len,
4213         int                     flags,
4214         xfs_bmbt_irec_t         *mval,
4215         int                     nmap,
4216         int                     ret_nmap)
4217 {
4218         int                     i;              /* index to map values */
4219
4220         ASSERT(ret_nmap <= nmap);
4221
4222         for (i = 0; i < ret_nmap; i++) {
4223                 ASSERT(mval[i].br_blockcount > 0);
4224                 if (!(flags & XFS_BMAPI_ENTIRE)) {
4225                         ASSERT(mval[i].br_startoff >= bno);
4226                         ASSERT(mval[i].br_blockcount <= len);
4227                         ASSERT(mval[i].br_startoff + mval[i].br_blockcount <=
4228                                bno + len);
4229                 } else {
4230                         ASSERT(mval[i].br_startoff < bno + len);
4231                         ASSERT(mval[i].br_startoff + mval[i].br_blockcount >
4232                                bno);
4233                 }
4234                 ASSERT(i == 0 ||
4235                        mval[i - 1].br_startoff + mval[i - 1].br_blockcount ==
4236                        mval[i].br_startoff);
4237                 ASSERT(mval[i].br_startblock != DELAYSTARTBLOCK &&
4238                        mval[i].br_startblock != HOLESTARTBLOCK);
4239                 ASSERT(mval[i].br_state == XFS_EXT_NORM ||
4240                        mval[i].br_state == XFS_EXT_UNWRITTEN);
4241         }
4242 }
4243 #endif /* DEBUG */
4244
4245
4246 /*
4247  * Trim the returned map to the required bounds
4248  */
4249 STATIC void
4250 xfs_bmapi_trim_map(
4251         struct xfs_bmbt_irec    *mval,
4252         struct xfs_bmbt_irec    *got,
4253         xfs_fileoff_t           *bno,
4254         xfs_filblks_t           len,
4255         xfs_fileoff_t           obno,
4256         xfs_fileoff_t           end,
4257         int                     n,
4258         int                     flags)
4259 {
4260         if ((flags & XFS_BMAPI_ENTIRE) ||
4261             got->br_startoff + got->br_blockcount <= obno) {
4262                 *mval = *got;
4263                 if (isnullstartblock(got->br_startblock))
4264                         mval->br_startblock = DELAYSTARTBLOCK;
4265                 return;
4266         }
4267
4268         if (obno > *bno)
4269                 *bno = obno;
4270         ASSERT((*bno >= obno) || (n == 0));
4271         ASSERT(*bno < end);
4272         mval->br_startoff = *bno;
4273         if (isnullstartblock(got->br_startblock))
4274                 mval->br_startblock = DELAYSTARTBLOCK;
4275         else
4276                 mval->br_startblock = got->br_startblock +
4277                                         (*bno - got->br_startoff);
4278         /*
4279          * Return the minimum of what we got and what we asked for for
4280          * the length.  We can use the len variable here because it is
4281          * modified below and we could have been there before coming
4282          * here if the first part of the allocation didn't overlap what
4283          * was asked for.
4284          */
4285         mval->br_blockcount = XFS_FILBLKS_MIN(end - *bno,
4286                         got->br_blockcount - (*bno - got->br_startoff));
4287         mval->br_state = got->br_state;
4288         ASSERT(mval->br_blockcount <= len);
4289         return;
4290 }
4291
4292 /*
4293  * Update and validate the extent map to return
4294  */
4295 STATIC void
4296 xfs_bmapi_update_map(
4297         struct xfs_bmbt_irec    **map,
4298         xfs_fileoff_t           *bno,
4299         xfs_filblks_t           *len,
4300         xfs_fileoff_t           obno,
4301         xfs_fileoff_t           end,
4302         int                     *n,
4303         int                     flags)
4304 {
4305         xfs_bmbt_irec_t *mval = *map;
4306
4307         ASSERT((flags & XFS_BMAPI_ENTIRE) ||
4308                ((mval->br_startoff + mval->br_blockcount) <= end));
4309         ASSERT((flags & XFS_BMAPI_ENTIRE) || (mval->br_blockcount <= *len) ||
4310                (mval->br_startoff < obno));
4311
4312         *bno = mval->br_startoff + mval->br_blockcount;
4313         *len = end - *bno;
4314         if (*n > 0 && mval->br_startoff == mval[-1].br_startoff) {
4315                 /* update previous map with new information */
4316                 ASSERT(mval->br_startblock == mval[-1].br_startblock);
4317                 ASSERT(mval->br_blockcount > mval[-1].br_blockcount);
4318                 ASSERT(mval->br_state == mval[-1].br_state);
4319                 mval[-1].br_blockcount = mval->br_blockcount;
4320                 mval[-1].br_state = mval->br_state;
4321         } else if (*n > 0 && mval->br_startblock != DELAYSTARTBLOCK &&
4322                    mval[-1].br_startblock != DELAYSTARTBLOCK &&
4323                    mval[-1].br_startblock != HOLESTARTBLOCK &&
4324                    mval->br_startblock == mval[-1].br_startblock +
4325                                           mval[-1].br_blockcount &&
4326                    ((flags & XFS_BMAPI_IGSTATE) ||
4327                         mval[-1].br_state == mval->br_state)) {
4328                 ASSERT(mval->br_startoff ==
4329                        mval[-1].br_startoff + mval[-1].br_blockcount);
4330                 mval[-1].br_blockcount += mval->br_blockcount;
4331         } else if (*n > 0 &&
4332                    mval->br_startblock == DELAYSTARTBLOCK &&
4333                    mval[-1].br_startblock == DELAYSTARTBLOCK &&
4334                    mval->br_startoff ==
4335                    mval[-1].br_startoff + mval[-1].br_blockcount) {
4336                 mval[-1].br_blockcount += mval->br_blockcount;
4337                 mval[-1].br_state = mval->br_state;
4338         } else if (!((*n == 0) &&
4339                      ((mval->br_startoff + mval->br_blockcount) <=
4340                       obno))) {
4341                 mval++;
4342                 (*n)++;
4343         }
4344         *map = mval;
4345 }
4346
4347 /*
4348  * Map file blocks to filesystem blocks without allocation.
4349  */
4350 int
4351 xfs_bmapi_read(
4352         struct xfs_inode        *ip,
4353         xfs_fileoff_t           bno,
4354         xfs_filblks_t           len,
4355         struct xfs_bmbt_irec    *mval,
4356         int                     *nmap,
4357         int                     flags)
4358 {
4359         struct xfs_mount        *mp = ip->i_mount;
4360         struct xfs_ifork        *ifp;
4361         struct xfs_bmbt_irec    got;
4362         struct xfs_bmbt_irec    prev;
4363         xfs_fileoff_t           obno;
4364         xfs_fileoff_t           end;
4365         xfs_extnum_t            lastx;
4366         int                     error;
4367         int                     eof;
4368         int                     n = 0;
4369         int                     whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4370                                                 XFS_ATTR_FORK : XFS_DATA_FORK;
4371
4372         ASSERT(*nmap >= 1);
4373         ASSERT(!(flags & ~(XFS_BMAPI_ATTRFORK|XFS_BMAPI_ENTIRE|
4374                            XFS_BMAPI_IGSTATE)));
4375
4376         if (unlikely(XFS_TEST_ERROR(
4377             (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4378              XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
4379              mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4380                 XFS_ERROR_REPORT("xfs_bmapi_read", XFS_ERRLEVEL_LOW, mp);
4381                 return XFS_ERROR(EFSCORRUPTED);
4382         }
4383
4384         if (XFS_FORCED_SHUTDOWN(mp))
4385                 return XFS_ERROR(EIO);
4386
4387         XFS_STATS_INC(xs_blk_mapr);
4388
4389         ifp = XFS_IFORK_PTR(ip, whichfork);
4390
4391         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4392                 error = xfs_iread_extents(NULL, ip, whichfork);
4393                 if (error)
4394                         return error;
4395         }
4396
4397         xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got, &prev);
4398         end = bno + len;
4399         obno = bno;
4400
4401         while (bno < end && n < *nmap) {
4402                 /* Reading past eof, act as though there's a hole up to end. */
4403                 if (eof)
4404                         got.br_startoff = end;
4405                 if (got.br_startoff > bno) {
4406                         /* Reading in a hole.  */
4407                         mval->br_startoff = bno;
4408                         mval->br_startblock = HOLESTARTBLOCK;
4409                         mval->br_blockcount =
4410                                 XFS_FILBLKS_MIN(len, got.br_startoff - bno);
4411                         mval->br_state = XFS_EXT_NORM;
4412                         bno += mval->br_blockcount;
4413                         len -= mval->br_blockcount;
4414                         mval++;
4415                         n++;
4416                         continue;
4417                 }
4418
4419                 /* set up the extent map to return. */
4420                 xfs_bmapi_trim_map(mval, &got, &bno, len, obno, end, n, flags);
4421                 xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4422
4423                 /* If we're done, stop now. */
4424                 if (bno >= end || n >= *nmap)
4425                         break;
4426
4427                 /* Else go on to the next record. */
4428                 if (++lastx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t))
4429                         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx), &got);
4430                 else
4431                         eof = 1;
4432         }
4433         *nmap = n;
4434         return 0;
4435 }
4436
4437 STATIC int
4438 xfs_bmapi_reserve_delalloc(
4439         struct xfs_inode        *ip,
4440         xfs_fileoff_t           aoff,
4441         xfs_filblks_t           len,
4442         struct xfs_bmbt_irec    *got,
4443         struct xfs_bmbt_irec    *prev,
4444         xfs_extnum_t            *lastx,
4445         int                     eof)
4446 {
4447         struct xfs_mount        *mp = ip->i_mount;
4448         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
4449         xfs_extlen_t            alen;
4450         xfs_extlen_t            indlen;
4451         char                    rt = XFS_IS_REALTIME_INODE(ip);
4452         xfs_extlen_t            extsz;
4453         int                     error;
4454
4455         alen = XFS_FILBLKS_MIN(len, MAXEXTLEN);
4456         if (!eof)
4457                 alen = XFS_FILBLKS_MIN(alen, got->br_startoff - aoff);
4458
4459         /* Figure out the extent size, adjust alen */
4460         extsz = xfs_get_extsz_hint(ip);
4461         if (extsz) {
4462                 /*
4463                  * Make sure we don't exceed a single extent length when we
4464                  * align the extent by reducing length we are going to
4465                  * allocate by the maximum amount extent size aligment may
4466                  * require.
4467                  */
4468                 alen = XFS_FILBLKS_MIN(len, MAXEXTLEN - (2 * extsz - 1));
4469                 error = xfs_bmap_extsize_align(mp, got, prev, extsz, rt, eof,
4470                                                1, 0, &aoff, &alen);
4471                 ASSERT(!error);
4472         }
4473
4474         if (rt)
4475                 extsz = alen / mp->m_sb.sb_rextsize;
4476
4477         /*
4478          * Make a transaction-less quota reservation for delayed allocation
4479          * blocks.  This number gets adjusted later.  We return if we haven't
4480          * allocated blocks already inside this loop.
4481          */
4482         error = xfs_trans_reserve_quota_nblks(NULL, ip, (long)alen, 0,
4483                         rt ? XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4484         if (error)
4485                 return error;
4486
4487         /*
4488          * Split changing sb for alen and indlen since they could be coming
4489          * from different places.
4490          */
4491         indlen = (xfs_extlen_t)xfs_bmap_worst_indlen(ip, alen);
4492         ASSERT(indlen > 0);
4493
4494         if (rt) {
4495                 error = xfs_mod_incore_sb(mp, XFS_SBS_FREXTENTS,
4496                                           -((int64_t)extsz), 0);
4497         } else {
4498                 error = xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
4499                                                  -((int64_t)alen), 0);
4500         }
4501
4502         if (error)
4503                 goto out_unreserve_quota;
4504
4505         error = xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
4506                                          -((int64_t)indlen), 0);
4507         if (error)
4508                 goto out_unreserve_blocks;
4509
4510
4511         ip->i_delayed_blks += alen;
4512
4513         got->br_startoff = aoff;
4514         got->br_startblock = nullstartblock(indlen);
4515         got->br_blockcount = alen;
4516         got->br_state = XFS_EXT_NORM;
4517         xfs_bmap_add_extent_hole_delay(ip, lastx, got);
4518
4519         /*
4520          * Update our extent pointer, given that xfs_bmap_add_extent_hole_delay
4521          * might have merged it into one of the neighbouring ones.
4522          */
4523         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *lastx), got);
4524
4525         ASSERT(got->br_startoff <= aoff);
4526         ASSERT(got->br_startoff + got->br_blockcount >= aoff + alen);
4527         ASSERT(isnullstartblock(got->br_startblock));
4528         ASSERT(got->br_state == XFS_EXT_NORM);
4529         return 0;
4530
4531 out_unreserve_blocks:
4532         if (rt)
4533                 xfs_mod_incore_sb(mp, XFS_SBS_FREXTENTS, extsz, 0);
4534         else
4535                 xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS, alen, 0);
4536 out_unreserve_quota:
4537         if (XFS_IS_QUOTA_ON(mp))
4538                 xfs_trans_unreserve_quota_nblks(NULL, ip, (long)alen, 0, rt ?
4539                                 XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4540         return error;
4541 }
4542
4543 /*
4544  * Map file blocks to filesystem blocks, adding delayed allocations as needed.
4545  */
4546 int
4547 xfs_bmapi_delay(
4548         struct xfs_inode        *ip,    /* incore inode */
4549         xfs_fileoff_t           bno,    /* starting file offs. mapped */
4550         xfs_filblks_t           len,    /* length to map in file */
4551         struct xfs_bmbt_irec    *mval,  /* output: map values */
4552         int                     *nmap,  /* i/o: mval size/count */
4553         int                     flags)  /* XFS_BMAPI_... */
4554 {
4555         struct xfs_mount        *mp = ip->i_mount;
4556         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
4557         struct xfs_bmbt_irec    got;    /* current file extent record */
4558         struct xfs_bmbt_irec    prev;   /* previous file extent record */
4559         xfs_fileoff_t           obno;   /* old block number (offset) */
4560         xfs_fileoff_t           end;    /* end of mapped file region */
4561         xfs_extnum_t            lastx;  /* last useful extent number */
4562         int                     eof;    /* we've hit the end of extents */
4563         int                     n = 0;  /* current extent index */
4564         int                     error = 0;
4565
4566         ASSERT(*nmap >= 1);
4567         ASSERT(*nmap <= XFS_BMAP_MAX_NMAP);
4568         ASSERT(!(flags & ~XFS_BMAPI_ENTIRE));
4569
4570         if (unlikely(XFS_TEST_ERROR(
4571             (XFS_IFORK_FORMAT(ip, XFS_DATA_FORK) != XFS_DINODE_FMT_EXTENTS &&
4572              XFS_IFORK_FORMAT(ip, XFS_DATA_FORK) != XFS_DINODE_FMT_BTREE),
4573              mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4574                 XFS_ERROR_REPORT("xfs_bmapi_delay", XFS_ERRLEVEL_LOW, mp);
4575                 return XFS_ERROR(EFSCORRUPTED);
4576         }
4577
4578         if (XFS_FORCED_SHUTDOWN(mp))
4579                 return XFS_ERROR(EIO);
4580
4581         XFS_STATS_INC(xs_blk_mapw);
4582
4583         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4584                 error = xfs_iread_extents(NULL, ip, XFS_DATA_FORK);
4585                 if (error)
4586                         return error;
4587         }
4588
4589         xfs_bmap_search_extents(ip, bno, XFS_DATA_FORK, &eof, &lastx, &got, &prev);
4590         end = bno + len;
4591         obno = bno;
4592
4593         while (bno < end && n < *nmap) {
4594                 if (eof || got.br_startoff > bno) {
4595                         error = xfs_bmapi_reserve_delalloc(ip, bno, len, &got,
4596                                                            &prev, &lastx, eof);
4597                         if (error) {
4598                                 if (n == 0) {
4599                                         *nmap = 0;
4600                                         return error;
4601                                 }
4602                                 break;
4603                         }
4604                 }
4605
4606                 /* set up the extent map to return. */
4607                 xfs_bmapi_trim_map(mval, &got, &bno, len, obno, end, n, flags);
4608                 xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4609
4610                 /* If we're done, stop now. */
4611                 if (bno >= end || n >= *nmap)
4612                         break;
4613
4614                 /* Else go on to the next record. */
4615                 prev = got;
4616                 if (++lastx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t))
4617                         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx), &got);
4618                 else
4619                         eof = 1;
4620         }
4621
4622         *nmap = n;
4623         return 0;
4624 }
4625
4626
4627 STATIC int
4628 __xfs_bmapi_allocate(
4629         struct xfs_bmalloca     *bma)
4630 {
4631         struct xfs_mount        *mp = bma->ip->i_mount;
4632         int                     whichfork = (bma->flags & XFS_BMAPI_ATTRFORK) ?
4633                                                 XFS_ATTR_FORK : XFS_DATA_FORK;
4634         struct xfs_ifork        *ifp = XFS_IFORK_PTR(bma->ip, whichfork);
4635         int                     tmp_logflags = 0;
4636         int                     error;
4637         int                     rt;
4638
4639         ASSERT(bma->length > 0);
4640
4641         rt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(bma->ip);
4642
4643         /*
4644          * For the wasdelay case, we could also just allocate the stuff asked
4645          * for in this bmap call but that wouldn't be as good.
4646          */
4647         if (bma->wasdel) {
4648                 bma->length = (xfs_extlen_t)bma->got.br_blockcount;
4649                 bma->offset = bma->got.br_startoff;
4650                 if (bma->idx != NULLEXTNUM && bma->idx) {
4651                         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx - 1),
4652                                          &bma->prev);
4653                 }
4654         } else {
4655                 bma->length = XFS_FILBLKS_MIN(bma->length, MAXEXTLEN);
4656                 if (!bma->eof)
4657                         bma->length = XFS_FILBLKS_MIN(bma->length,
4658                                         bma->got.br_startoff - bma->offset);
4659         }
4660
4661         /*
4662          * Indicate if this is the first user data in the file, or just any
4663          * user data.
4664          */
4665         if (!(bma->flags & XFS_BMAPI_METADATA)) {
4666                 bma->userdata = (bma->offset == 0) ?
4667                         XFS_ALLOC_INITIAL_USER_DATA : XFS_ALLOC_USERDATA;
4668         }
4669
4670         bma->minlen = (bma->flags & XFS_BMAPI_CONTIG) ? bma->length : 1;
4671
4672         /*
4673          * Only want to do the alignment at the eof if it is userdata and
4674          * allocation length is larger than a stripe unit.
4675          */
4676         if (mp->m_dalign && bma->length >= mp->m_dalign &&
4677             !(bma->flags & XFS_BMAPI_METADATA) && whichfork == XFS_DATA_FORK) {
4678                 error = xfs_bmap_isaeof(bma, whichfork);
4679                 if (error)
4680                         return error;
4681         }
4682
4683         if (bma->flags & XFS_BMAPI_STACK_SWITCH)
4684                 bma->stack_switch = 1;
4685
4686         error = xfs_bmap_alloc(bma);
4687         if (error)
4688                 return error;
4689
4690         if (bma->flist->xbf_low)
4691                 bma->minleft = 0;
4692         if (bma->cur)
4693                 bma->cur->bc_private.b.firstblock = *bma->firstblock;
4694         if (bma->blkno == NULLFSBLOCK)
4695                 return 0;
4696         if ((ifp->if_flags & XFS_IFBROOT) && !bma->cur) {
4697                 bma->cur = xfs_bmbt_init_cursor(mp, bma->tp, bma->ip, whichfork);
4698                 bma->cur->bc_private.b.firstblock = *bma->firstblock;
4699                 bma->cur->bc_private.b.flist = bma->flist;
4700         }
4701         /*
4702          * Bump the number of extents we've allocated
4703          * in this call.
4704          */
4705         bma->nallocs++;
4706
4707         if (bma->cur)
4708                 bma->cur->bc_private.b.flags =
4709                         bma->wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
4710
4711         bma->got.br_startoff = bma->offset;
4712         bma->got.br_startblock = bma->blkno;
4713         bma->got.br_blockcount = bma->length;
4714         bma->got.br_state = XFS_EXT_NORM;
4715
4716         /*
4717          * A wasdelay extent has been initialized, so shouldn't be flagged
4718          * as unwritten.
4719          */
4720         if (!bma->wasdel && (bma->flags & XFS_BMAPI_PREALLOC) &&
4721             xfs_sb_version_hasextflgbit(&mp->m_sb))
4722                 bma->got.br_state = XFS_EXT_UNWRITTEN;
4723
4724         if (bma->wasdel)
4725                 error = xfs_bmap_add_extent_delay_real(bma);
4726         else
4727                 error = xfs_bmap_add_extent_hole_real(bma, whichfork);
4728
4729         bma->logflags |= tmp_logflags;
4730         if (error)
4731                 return error;
4732
4733         /*
4734          * Update our extent pointer, given that xfs_bmap_add_extent_delay_real
4735          * or xfs_bmap_add_extent_hole_real might have merged it into one of
4736          * the neighbouring ones.
4737          */
4738         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx), &bma->got);
4739
4740         ASSERT(bma->got.br_startoff <= bma->offset);
4741         ASSERT(bma->got.br_startoff + bma->got.br_blockcount >=
4742                bma->offset + bma->length);
4743         ASSERT(bma->got.br_state == XFS_EXT_NORM ||
4744                bma->got.br_state == XFS_EXT_UNWRITTEN);
4745         return 0;
4746 }
4747
4748 static void
4749 xfs_bmapi_allocate_worker(
4750         struct work_struct      *work)
4751 {
4752         struct xfs_bmalloca     *args = container_of(work,
4753                                                 struct xfs_bmalloca, work);
4754         unsigned long           pflags;
4755
4756         /* we are in a transaction context here */
4757         current_set_flags_nested(&pflags, PF_FSTRANS);
4758
4759         args->result = __xfs_bmapi_allocate(args);
4760         complete(args->done);
4761
4762         current_restore_flags_nested(&pflags, PF_FSTRANS);
4763 }
4764
4765 /*
4766  * Some allocation requests often come in with little stack to work on. Push
4767  * them off to a worker thread so there is lots of stack to use. Otherwise just
4768  * call directly to avoid the context switch overhead here.
4769  */
4770 int
4771 xfs_bmapi_allocate(
4772         struct xfs_bmalloca     *args)
4773 {
4774         DECLARE_COMPLETION_ONSTACK(done);
4775
4776         if (!args->stack_switch)
4777                 return __xfs_bmapi_allocate(args);
4778
4779
4780         args->done = &done;
4781         INIT_WORK_ONSTACK(&args->work, xfs_bmapi_allocate_worker);
4782         queue_work(xfs_alloc_wq, &args->work);
4783         wait_for_completion(&done);
4784         return args->result;
4785 }
4786
4787 STATIC int
4788 xfs_bmapi_convert_unwritten(
4789         struct xfs_bmalloca     *bma,
4790         struct xfs_bmbt_irec    *mval,
4791         xfs_filblks_t           len,
4792         int                     flags)
4793 {
4794         int                     whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4795                                                 XFS_ATTR_FORK : XFS_DATA_FORK;
4796         struct xfs_ifork        *ifp = XFS_IFORK_PTR(bma->ip, whichfork);
4797         int                     tmp_logflags = 0;
4798         int                     error;
4799
4800         /* check if we need to do unwritten->real conversion */
4801         if (mval->br_state == XFS_EXT_UNWRITTEN &&
4802             (flags & XFS_BMAPI_PREALLOC))
4803                 return 0;
4804
4805         /* check if we need to do real->unwritten conversion */
4806         if (mval->br_state == XFS_EXT_NORM &&
4807             (flags & (XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT)) !=
4808                         (XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT))
4809                 return 0;
4810
4811         /*
4812          * Modify (by adding) the state flag, if writing.
4813          */
4814         ASSERT(mval->br_blockcount <= len);
4815         if ((ifp->if_flags & XFS_IFBROOT) && !bma->cur) {
4816                 bma->cur = xfs_bmbt_init_cursor(bma->ip->i_mount, bma->tp,
4817                                         bma->ip, whichfork);
4818                 bma->cur->bc_private.b.firstblock = *bma->firstblock;
4819                 bma->cur->bc_private.b.flist = bma->flist;
4820         }
4821         mval->br_state = (mval->br_state == XFS_EXT_UNWRITTEN)
4822                                 ? XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
4823
4824         error = xfs_bmap_add_extent_unwritten_real(bma->tp, bma->ip, &bma->idx,
4825                         &bma->cur, mval, bma->firstblock, bma->flist,
4826                         &tmp_logflags);
4827         bma->logflags |= tmp_logflags;
4828         if (error)
4829                 return error;
4830
4831         /*
4832          * Update our extent pointer, given that
4833          * xfs_bmap_add_extent_unwritten_real might have merged it into one
4834          * of the neighbouring ones.
4835          */
4836         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx), &bma->got);
4837
4838         /*
4839          * We may have combined previously unwritten space with written space,
4840          * so generate another request.
4841          */
4842         if (mval->br_blockcount < len)
4843                 return EAGAIN;
4844         return 0;
4845 }
4846
4847 /*
4848  * Map file blocks to filesystem blocks, and allocate blocks or convert the
4849  * extent state if necessary.  Details behaviour is controlled by the flags
4850  * parameter.  Only allocates blocks from a single allocation group, to avoid
4851  * locking problems.
4852  *
4853  * The returned value in "firstblock" from the first call in a transaction
4854  * must be remembered and presented to subsequent calls in "firstblock".
4855  * An upper bound for the number of blocks to be allocated is supplied to
4856  * the first call in "total"; if no allocation group has that many free
4857  * blocks then the call will fail (return NULLFSBLOCK in "firstblock").
4858  */
4859 int
4860 xfs_bmapi_write(
4861         struct xfs_trans        *tp,            /* transaction pointer */
4862         struct xfs_inode        *ip,            /* incore inode */
4863         xfs_fileoff_t           bno,            /* starting file offs. mapped */
4864         xfs_filblks_t           len,            /* length to map in file */
4865         int                     flags,          /* XFS_BMAPI_... */
4866         xfs_fsblock_t           *firstblock,    /* first allocated block
4867                                                    controls a.g. for allocs */
4868         xfs_extlen_t            total,          /* total blocks needed */
4869         struct xfs_bmbt_irec    *mval,          /* output: map values */
4870         int                     *nmap,          /* i/o: mval size/count */
4871         struct xfs_bmap_free    *flist)         /* i/o: list extents to free */
4872 {
4873         struct xfs_mount        *mp = ip->i_mount;
4874         struct xfs_ifork        *ifp;
4875         struct xfs_bmalloca     bma = { 0 };    /* args for xfs_bmap_alloc */
4876         xfs_fileoff_t           end;            /* end of mapped file region */
4877         int                     eof;            /* after the end of extents */
4878         int                     error;          /* error return */
4879         int                     n;              /* current extent index */
4880         xfs_fileoff_t           obno;           /* old block number (offset) */
4881         int                     whichfork;      /* data or attr fork */
4882         char                    inhole;         /* current location is hole in file */
4883         char                    wasdelay;       /* old extent was delayed */
4884
4885 #ifdef DEBUG
4886         xfs_fileoff_t           orig_bno;       /* original block number value */
4887         int                     orig_flags;     /* original flags arg value */
4888         xfs_filblks_t           orig_len;       /* original value of len arg */
4889         struct xfs_bmbt_irec    *orig_mval;     /* original value of mval */
4890         int                     orig_nmap;      /* original value of *nmap */
4891
4892         orig_bno = bno;
4893         orig_len = len;
4894         orig_flags = flags;
4895         orig_mval = mval;
4896         orig_nmap = *nmap;
4897 #endif
4898
4899         ASSERT(*nmap >= 1);
4900         ASSERT(*nmap <= XFS_BMAP_MAX_NMAP);
4901         ASSERT(!(flags & XFS_BMAPI_IGSTATE));
4902         ASSERT(tp != NULL);
4903         ASSERT(len > 0);
4904
4905         whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4906                 XFS_ATTR_FORK : XFS_DATA_FORK;
4907
4908         if (unlikely(XFS_TEST_ERROR(
4909             (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4910              XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
4911              XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL),
4912              mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4913                 XFS_ERROR_REPORT("xfs_bmapi_write", XFS_ERRLEVEL_LOW, mp);
4914                 return XFS_ERROR(EFSCORRUPTED);
4915         }
4916
4917         if (XFS_FORCED_SHUTDOWN(mp))
4918                 return XFS_ERROR(EIO);
4919
4920         ifp = XFS_IFORK_PTR(ip, whichfork);
4921
4922         XFS_STATS_INC(xs_blk_mapw);
4923
4924         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
4925                 error = xfs_bmap_local_to_extents(tp, ip, firstblock, total,
4926                                                   &bma.logflags, whichfork);
4927                 if (error)
4928                         goto error0;
4929         }
4930
4931         if (*firstblock == NULLFSBLOCK) {
4932                 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE)
4933                         bma.minleft = be16_to_cpu(ifp->if_broot->bb_level) + 1;
4934                 else
4935                         bma.minleft = 1;
4936         } else {
4937                 bma.minleft = 0;
4938         }
4939
4940         if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4941                 error = xfs_iread_extents(tp, ip, whichfork);
4942                 if (error)
4943                         goto error0;
4944         }
4945
4946         xfs_bmap_search_extents(ip, bno, whichfork, &eof, &bma.idx, &bma.got,
4947                                 &bma.prev);
4948         n = 0;
4949         end = bno + len;
4950         obno = bno;
4951
4952         bma.tp = tp;
4953         bma.ip = ip;
4954         bma.total = total;
4955         bma.userdata = 0;
4956         bma.flist = flist;
4957         bma.firstblock = firstblock;
4958
4959         while (bno < end && n < *nmap) {
4960                 inhole = eof || bma.got.br_startoff > bno;
4961                 wasdelay = !inhole && isnullstartblock(bma.got.br_startblock);
4962
4963                 /*
4964                  * First, deal with the hole before the allocated space
4965                  * that we found, if any.
4966                  */
4967                 if (inhole || wasdelay) {
4968                         bma.eof = eof;
4969                         bma.conv = !!(flags & XFS_BMAPI_CONVERT);
4970                         bma.wasdel = wasdelay;
4971                         bma.offset = bno;
4972                         bma.flags = flags;
4973
4974                         /*
4975                          * There's a 32/64 bit type mismatch between the
4976                          * allocation length request (which can be 64 bits in
4977                          * length) and the bma length request, which is
4978                          * xfs_extlen_t and therefore 32 bits. Hence we have to
4979                          * check for 32-bit overflows and handle them here.
4980                          */
4981                         if (len > (xfs_filblks_t)MAXEXTLEN)
4982                                 bma.length = MAXEXTLEN;
4983                         else
4984                                 bma.length = len;
4985
4986                         ASSERT(len > 0);
4987                         ASSERT(bma.length > 0);
4988                         error = xfs_bmapi_allocate(&bma);
4989                         if (error)
4990                                 goto error0;
4991                         if (bma.blkno == NULLFSBLOCK)
4992                                 break;
4993                 }
4994
4995                 /* Deal with the allocated space we found.  */
4996                 xfs_bmapi_trim_map(mval, &bma.got, &bno, len, obno,
4997                                                         end, n, flags);
4998
4999                 /* Execute unwritten extent conversion if necessary */
5000                 error = xfs_bmapi_convert_unwritten(&bma, mval, len, flags);
5001                 if (error == EAGAIN)
5002                         continue;
5003                 if (error)
5004                         goto error0;
5005
5006                 /* update the extent map to return */
5007                 xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
5008
5009                 /*
5010                  * If we're done, stop now.  Stop when we've allocated
5011                  * XFS_BMAP_MAX_NMAP extents no matter what.  Otherwise
5012                  * the transaction may get too big.
5013                  */
5014                 if (bno >= end || n >= *nmap || bma.nallocs >= *nmap)
5015                         break;
5016
5017                 /* Else go on to the next record. */
5018                 bma.prev = bma.got;
5019                 if (++bma.idx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t)) {
5020                         xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma.idx),
5021                                          &bma.got);
5022                 } else
5023                         eof = 1;
5024         }
5025         *nmap = n;
5026
5027         /*
5028          * Transform from btree to extents, give it cur.
5029          */
5030         if (xfs_bmap_wants_extents(ip, whichfork)) {
5031                 int             tmp_logflags = 0;
5032
5033                 ASSERT(bma.cur);
5034                 error = xfs_bmap_btree_to_extents(tp, ip, bma.cur,
5035                         &tmp_logflags, whichfork);
5036                 bma.logflags |= tmp_logflags;
5037                 if (error)
5038                         goto error0;
5039         }
5040
5041         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE ||
5042                XFS_IFORK_NEXTENTS(ip, whichfork) >
5043                 XFS_IFORK_MAXEXT(ip, whichfork));
5044         error = 0;
5045 error0:
5046         /*
5047          * Log everything.  Do this after conversion, there's no point in
5048          * logging the extent records if we've converted to btree format.
5049          */
5050         if ((bma.logflags & xfs_ilog_fext(whichfork)) &&
5051             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
5052                 bma.logflags &= ~xfs_ilog_fext(whichfork);
5053         else if ((bma.logflags & xfs_ilog_fbroot(whichfork)) &&
5054                  XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
5055                 bma.logflags &= ~xfs_ilog_fbroot(whichfork);
5056         /*
5057          * Log whatever the flags say, even if error.  Otherwise we might miss
5058          * detecting a case where the data is changed, there's an error,
5059          * and it's not logged so we don't shutdown when we should.
5060          */
5061         if (bma.logflags)
5062                 xfs_trans_log_inode(tp, ip, bma.logflags);
5063
5064         if (bma.cur) {
5065                 if (!error) {
5066                         ASSERT(*firstblock == NULLFSBLOCK ||
5067                                XFS_FSB_TO_AGNO(mp, *firstblock) ==
5068                                XFS_FSB_TO_AGNO(mp,
5069                                        bma.cur->bc_private.b.firstblock) ||
5070                                (flist->xbf_low &&
5071                                 XFS_FSB_TO_AGNO(mp, *firstblock) <
5072                                 XFS_FSB_TO_AGNO(mp,
5073                                         bma.cur->bc_private.b.firstblock)));
5074                         *firstblock = bma.cur->bc_private.b.firstblock;
5075                 }
5076                 xfs_btree_del_cursor(bma.cur,
5077                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5078         }
5079         if (!error)
5080                 xfs_bmap_validate_ret(orig_bno, orig_len, orig_flags, orig_mval,
5081                         orig_nmap, *nmap);
5082         return error;
5083 }
5084
5085 /*
5086  * Unmap (remove) blocks from a file.
5087  * If nexts is nonzero then the number of extents to remove is limited to
5088  * that value.  If not all extents in the block range can be removed then
5089  * *done is set.
5090  */
5091 int                                             /* error */
5092 xfs_bunmapi(
5093         xfs_trans_t             *tp,            /* transaction pointer */
5094         struct xfs_inode        *ip,            /* incore inode */
5095         xfs_fileoff_t           bno,            /* starting offset to unmap */
5096         xfs_filblks_t           len,            /* length to unmap in file */
5097         int                     flags,          /* misc flags */
5098         xfs_extnum_t            nexts,          /* number of extents max */
5099         xfs_fsblock_t           *firstblock,    /* first allocated block
5100                                                    controls a.g. for allocs */
5101         xfs_bmap_free_t         *flist,         /* i/o: list extents to free */
5102         int                     *done)          /* set if not done yet */
5103 {
5104         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
5105         xfs_bmbt_irec_t         del;            /* extent being deleted */
5106         int                     eof;            /* is deleting at eof */
5107         xfs_bmbt_rec_host_t     *ep;            /* extent record pointer */
5108         int                     error;          /* error return value */
5109         xfs_extnum_t            extno;          /* extent number in list */
5110         xfs_bmbt_irec_t         got;            /* current extent record */
5111         xfs_ifork_t             *ifp;           /* inode fork pointer */
5112         int                     isrt;           /* freeing in rt area */
5113         xfs_extnum_t            lastx;          /* last extent index used */
5114         int                     logflags;       /* transaction logging flags */
5115         xfs_extlen_t            mod;            /* rt extent offset */
5116         xfs_mount_t             *mp;            /* mount structure */
5117         xfs_extnum_t            nextents;       /* number of file extents */
5118         xfs_bmbt_irec_t         prev;           /* previous extent record */
5119         xfs_fileoff_t           start;          /* first file offset deleted */
5120         int                     tmp_logflags;   /* partial logging flags */
5121         int                     wasdel;         /* was a delayed alloc extent */
5122         int                     whichfork;      /* data or attribute fork */
5123         xfs_fsblock_t           sum;
5124
5125         trace_xfs_bunmap(ip, bno, len, flags, _RET_IP_);
5126
5127         whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
5128                 XFS_ATTR_FORK : XFS_DATA_FORK;
5129         ifp = XFS_IFORK_PTR(ip, whichfork);
5130         if (unlikely(
5131             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
5132             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
5133                 XFS_ERROR_REPORT("xfs_bunmapi", XFS_ERRLEVEL_LOW,
5134                                  ip->i_mount);
5135                 return XFS_ERROR(EFSCORRUPTED);
5136         }
5137         mp = ip->i_mount;
5138         if (XFS_FORCED_SHUTDOWN(mp))
5139                 return XFS_ERROR(EIO);
5140
5141         ASSERT(len > 0);
5142         ASSERT(nexts >= 0);
5143
5144         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5145             (error = xfs_iread_extents(tp, ip, whichfork)))
5146                 return error;
5147         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5148         if (nextents == 0) {
5149                 *done = 1;
5150                 return 0;
5151         }
5152         XFS_STATS_INC(xs_blk_unmap);
5153         isrt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
5154         start = bno;
5155         bno = start + len - 1;
5156         ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
5157                 &prev);
5158
5159         /*
5160          * Check to see if the given block number is past the end of the
5161          * file, back up to the last block if so...
5162          */
5163         if (eof) {
5164                 ep = xfs_iext_get_ext(ifp, --lastx);
5165                 xfs_bmbt_get_all(ep, &got);
5166                 bno = got.br_startoff + got.br_blockcount - 1;
5167         }
5168         logflags = 0;
5169         if (ifp->if_flags & XFS_IFBROOT) {
5170                 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
5171                 cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5172                 cur->bc_private.b.firstblock = *firstblock;
5173                 cur->bc_private.b.flist = flist;
5174                 cur->bc_private.b.flags = 0;
5175         } else
5176                 cur = NULL;
5177
5178         if (isrt) {
5179                 /*
5180                  * Synchronize by locking the bitmap inode.
5181                  */
5182                 xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL);
5183                 xfs_trans_ijoin(tp, mp->m_rbmip, XFS_ILOCK_EXCL);
5184         }
5185
5186         extno = 0;
5187         while (bno != (xfs_fileoff_t)-1 && bno >= start && lastx >= 0 &&
5188                (nexts == 0 || extno < nexts)) {
5189                 /*
5190                  * Is the found extent after a hole in which bno lives?
5191                  * Just back up to the previous extent, if so.
5192                  */
5193                 if (got.br_startoff > bno) {
5194                         if (--lastx < 0)
5195                                 break;
5196                         ep = xfs_iext_get_ext(ifp, lastx);
5197                         xfs_bmbt_get_all(ep, &got);
5198                 }
5199                 /*
5200                  * Is the last block of this extent before the range
5201                  * we're supposed to delete?  If so, we're done.
5202                  */
5203                 bno = XFS_FILEOFF_MIN(bno,
5204                         got.br_startoff + got.br_blockcount - 1);
5205                 if (bno < start)
5206                         break;
5207                 /*
5208                  * Then deal with the (possibly delayed) allocated space
5209                  * we found.
5210                  */
5211                 ASSERT(ep != NULL);
5212                 del = got;
5213                 wasdel = isnullstartblock(del.br_startblock);
5214                 if (got.br_startoff < start) {
5215                         del.br_startoff = start;
5216                         del.br_blockcount -= start - got.br_startoff;
5217                         if (!wasdel)
5218                                 del.br_startblock += start - got.br_startoff;
5219                 }
5220                 if (del.br_startoff + del.br_blockcount > bno + 1)
5221                         del.br_blockcount = bno + 1 - del.br_startoff;
5222                 sum = del.br_startblock + del.br_blockcount;
5223                 if (isrt &&
5224                     (mod = do_mod(sum, mp->m_sb.sb_rextsize))) {
5225                         /*
5226                          * Realtime extent not lined up at the end.
5227                          * The extent could have been split into written
5228                          * and unwritten pieces, or we could just be
5229                          * unmapping part of it.  But we can't really
5230                          * get rid of part of a realtime extent.
5231                          */
5232                         if (del.br_state == XFS_EXT_UNWRITTEN ||
5233                             !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5234                                 /*
5235                                  * This piece is unwritten, or we're not
5236                                  * using unwritten extents.  Skip over it.
5237                                  */
5238                                 ASSERT(bno >= mod);
5239                                 bno -= mod > del.br_blockcount ?
5240                                         del.br_blockcount : mod;
5241                                 if (bno < got.br_startoff) {
5242                                         if (--lastx >= 0)
5243                                                 xfs_bmbt_get_all(xfs_iext_get_ext(
5244                                                         ifp, lastx), &got);
5245                                 }
5246                                 continue;
5247                         }
5248                         /*
5249                          * It's written, turn it unwritten.
5250                          * This is better than zeroing it.
5251                          */
5252                         ASSERT(del.br_state == XFS_EXT_NORM);
5253                         ASSERT(xfs_trans_get_block_res(tp) > 0);
5254                         /*
5255                          * If this spans a realtime extent boundary,
5256                          * chop it back to the start of the one we end at.
5257                          */
5258                         if (del.br_blockcount > mod) {
5259                                 del.br_startoff += del.br_blockcount - mod;
5260                                 del.br_startblock += del.br_blockcount - mod;
5261                                 del.br_blockcount = mod;
5262                         }
5263                         del.br_state = XFS_EXT_UNWRITTEN;
5264                         error = xfs_bmap_add_extent_unwritten_real(tp, ip,
5265                                         &lastx, &cur, &del, firstblock, flist,
5266                                         &logflags);
5267                         if (error)
5268                                 goto error0;
5269                         goto nodelete;
5270                 }
5271                 if (isrt && (mod = do_mod(del.br_startblock, mp->m_sb.sb_rextsize))) {
5272                         /*
5273                          * Realtime extent is lined up at the end but not
5274                          * at the front.  We'll get rid of full extents if
5275                          * we can.
5276                          */
5277                         mod = mp->m_sb.sb_rextsize - mod;
5278                         if (del.br_blockcount > mod) {
5279                                 del.br_blockcount -= mod;
5280                                 del.br_startoff += mod;
5281                                 del.br_startblock += mod;
5282                         } else if ((del.br_startoff == start &&
5283                                     (del.br_state == XFS_EXT_UNWRITTEN ||
5284                                      xfs_trans_get_block_res(tp) == 0)) ||
5285                                    !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5286                                 /*
5287                                  * Can't make it unwritten.  There isn't
5288                                  * a full extent here so just skip it.
5289                                  */
5290                                 ASSERT(bno >= del.br_blockcount);
5291                                 bno -= del.br_blockcount;
5292                                 if (got.br_startoff > bno) {
5293                                         if (--lastx >= 0) {
5294                                                 ep = xfs_iext_get_ext(ifp,
5295                                                                       lastx);
5296                                                 xfs_bmbt_get_all(ep, &got);
5297                                         }
5298                                 }
5299                                 continue;
5300                         } else if (del.br_state == XFS_EXT_UNWRITTEN) {
5301                                 /*
5302                                  * This one is already unwritten.
5303                                  * It must have a written left neighbor.
5304                                  * Unwrite the killed part of that one and
5305                                  * try again.
5306                                  */
5307                                 ASSERT(lastx > 0);
5308                                 xfs_bmbt_get_all(xfs_iext_get_ext(ifp,
5309                                                 lastx - 1), &prev);
5310                                 ASSERT(prev.br_state == XFS_EXT_NORM);
5311                                 ASSERT(!isnullstartblock(prev.br_startblock));
5312                                 ASSERT(del.br_startblock ==
5313                                        prev.br_startblock + prev.br_blockcount);
5314                                 if (prev.br_startoff < start) {
5315                                         mod = start - prev.br_startoff;
5316                                         prev.br_blockcount -= mod;
5317                                         prev.br_startblock += mod;
5318                                         prev.br_startoff = start;
5319                                 }
5320                                 prev.br_state = XFS_EXT_UNWRITTEN;
5321                                 lastx--;
5322                                 error = xfs_bmap_add_extent_unwritten_real(tp,
5323                                                 ip, &lastx, &cur, &prev,
5324                                                 firstblock, flist, &logflags);
5325                                 if (error)
5326                                         goto error0;
5327                                 goto nodelete;
5328                         } else {
5329                                 ASSERT(del.br_state == XFS_EXT_NORM);
5330                                 del.br_state = XFS_EXT_UNWRITTEN;
5331                                 error = xfs_bmap_add_extent_unwritten_real(tp,
5332                                                 ip, &lastx, &cur, &del,
5333                                                 firstblock, flist, &logflags);
5334                                 if (error)
5335                                         goto error0;
5336                                 goto nodelete;
5337                         }
5338                 }
5339                 if (wasdel) {
5340                         ASSERT(startblockval(del.br_startblock) > 0);
5341                         /* Update realtime/data freespace, unreserve quota */
5342                         if (isrt) {
5343                                 xfs_filblks_t rtexts;
5344
5345                                 rtexts = XFS_FSB_TO_B(mp, del.br_blockcount);
5346                                 do_div(rtexts, mp->m_sb.sb_rextsize);
5347                                 xfs_mod_incore_sb(mp, XFS_SBS_FREXTENTS,
5348                                                 (int64_t)rtexts, 0);
5349                                 (void)xfs_trans_reserve_quota_nblks(NULL,
5350                                         ip, -((long)del.br_blockcount), 0,
5351                                         XFS_QMOPT_RES_RTBLKS);
5352                         } else {
5353                                 xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
5354                                                 (int64_t)del.br_blockcount, 0);
5355                                 (void)xfs_trans_reserve_quota_nblks(NULL,
5356                                         ip, -((long)del.br_blockcount), 0,
5357                                         XFS_QMOPT_RES_REGBLKS);
5358                         }
5359                         ip->i_delayed_blks -= del.br_blockcount;
5360                         if (cur)
5361                                 cur->bc_private.b.flags |=
5362                                         XFS_BTCUR_BPRV_WASDEL;
5363                 } else if (cur)
5364                         cur->bc_private.b.flags &= ~XFS_BTCUR_BPRV_WASDEL;
5365                 /*
5366                  * If it's the case where the directory code is running
5367                  * with no block reservation, and the deleted block is in
5368                  * the middle of its extent, and the resulting insert
5369                  * of an extent would cause transformation to btree format,
5370                  * then reject it.  The calling code will then swap
5371                  * blocks around instead.
5372                  * We have to do this now, rather than waiting for the
5373                  * conversion to btree format, since the transaction
5374                  * will be dirty.
5375                  */
5376                 if (!wasdel && xfs_trans_get_block_res(tp) == 0 &&
5377                     XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
5378                     XFS_IFORK_NEXTENTS(ip, whichfork) >= /* Note the >= */
5379                         XFS_IFORK_MAXEXT(ip, whichfork) &&
5380                     del.br_startoff > got.br_startoff &&
5381                     del.br_startoff + del.br_blockcount <
5382                     got.br_startoff + got.br_blockcount) {
5383                         error = XFS_ERROR(ENOSPC);
5384                         goto error0;
5385                 }
5386                 error = xfs_bmap_del_extent(ip, tp, &lastx, flist, cur, &del,
5387                                 &tmp_logflags, whichfork);
5388                 logflags |= tmp_logflags;
5389                 if (error)
5390                         goto error0;
5391                 bno = del.br_startoff - 1;
5392 nodelete:
5393                 /*
5394                  * If not done go on to the next (previous) record.
5395                  */
5396                 if (bno != (xfs_fileoff_t)-1 && bno >= start) {
5397                         if (lastx >= 0) {
5398                                 ep = xfs_iext_get_ext(ifp, lastx);
5399                                 if (xfs_bmbt_get_startoff(ep) > bno) {
5400                                         if (--lastx >= 0)
5401                                                 ep = xfs_iext_get_ext(ifp,
5402                                                                       lastx);
5403                                 }
5404                                 xfs_bmbt_get_all(ep, &got);
5405                         }
5406                         extno++;
5407                 }
5408         }
5409         *done = bno == (xfs_fileoff_t)-1 || bno < start || lastx < 0;
5410
5411         /*
5412          * Convert to a btree if necessary.
5413          */
5414         if (xfs_bmap_needs_btree(ip, whichfork)) {
5415                 ASSERT(cur == NULL);
5416                 error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist,
5417                         &cur, 0, &tmp_logflags, whichfork);
5418                 logflags |= tmp_logflags;
5419                 if (error)
5420                         goto error0;
5421         }
5422         /*
5423          * transform from btree to extents, give it cur
5424          */
5425         else if (xfs_bmap_wants_extents(ip, whichfork)) {
5426                 ASSERT(cur != NULL);
5427                 error = xfs_bmap_btree_to_extents(tp, ip, cur, &tmp_logflags,
5428                         whichfork);
5429                 logflags |= tmp_logflags;
5430                 if (error)
5431                         goto error0;
5432         }
5433         /*
5434          * transform from extents to local?
5435          */
5436         error = 0;
5437 error0:
5438         /*
5439          * Log everything.  Do this after conversion, there's no point in
5440          * logging the extent records if we've converted to btree format.
5441          */
5442         if ((logflags & xfs_ilog_fext(whichfork)) &&
5443             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
5444                 logflags &= ~xfs_ilog_fext(whichfork);
5445         else if ((logflags & xfs_ilog_fbroot(whichfork)) &&
5446                  XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
5447                 logflags &= ~xfs_ilog_fbroot(whichfork);
5448         /*
5449          * Log inode even in the error case, if the transaction
5450          * is dirty we'll need to shut down the filesystem.
5451          */
5452         if (logflags)
5453                 xfs_trans_log_inode(tp, ip, logflags);
5454         if (cur) {
5455                 if (!error) {
5456                         *firstblock = cur->bc_private.b.firstblock;
5457                         cur->bc_private.b.allocated = 0;
5458                 }
5459                 xfs_btree_del_cursor(cur,
5460                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5461         }
5462         return error;
5463 }
5464
5465 /*
5466  * returns 1 for success, 0 if we failed to map the extent.
5467  */
5468 STATIC int
5469 xfs_getbmapx_fix_eof_hole(
5470         xfs_inode_t             *ip,            /* xfs incore inode pointer */
5471         struct getbmapx         *out,           /* output structure */
5472         int                     prealloced,     /* this is a file with
5473                                                  * preallocated data space */
5474         __int64_t               end,            /* last block requested */
5475         xfs_fsblock_t           startblock)
5476 {
5477         __int64_t               fixlen;
5478         xfs_mount_t             *mp;            /* file system mount point */
5479         xfs_ifork_t             *ifp;           /* inode fork pointer */
5480         xfs_extnum_t            lastx;          /* last extent pointer */
5481         xfs_fileoff_t           fileblock;
5482
5483         if (startblock == HOLESTARTBLOCK) {
5484                 mp = ip->i_mount;
5485                 out->bmv_block = -1;
5486                 fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, XFS_ISIZE(ip)));
5487                 fixlen -= out->bmv_offset;
5488                 if (prealloced && out->bmv_offset + out->bmv_length == end) {
5489                         /* Came to hole at EOF. Trim it. */
5490                         if (fixlen <= 0)
5491                                 return 0;
5492                         out->bmv_length = fixlen;
5493                 }
5494         } else {
5495                 if (startblock == DELAYSTARTBLOCK)
5496                         out->bmv_block = -2;
5497                 else
5498                         out->bmv_block = xfs_fsb_to_db(ip, startblock);
5499                 fileblock = XFS_BB_TO_FSB(ip->i_mount, out->bmv_offset);
5500                 ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
5501                 if (xfs_iext_bno_to_ext(ifp, fileblock, &lastx) &&
5502                    (lastx == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t))-1))
5503                         out->bmv_oflags |= BMV_OF_LAST;
5504         }
5505
5506         return 1;
5507 }
5508
5509 /*
5510  * Get inode's extents as described in bmv, and format for output.
5511  * Calls formatter to fill the user's buffer until all extents
5512  * are mapped, until the passed-in bmv->bmv_count slots have
5513  * been filled, or until the formatter short-circuits the loop,
5514  * if it is tracking filled-in extents on its own.
5515  */
5516 int                                             /* error code */
5517 xfs_getbmap(
5518         xfs_inode_t             *ip,
5519         struct getbmapx         *bmv,           /* user bmap structure */
5520         xfs_bmap_format_t       formatter,      /* format to user */
5521         void                    *arg)           /* formatter arg */
5522 {
5523         __int64_t               bmvend;         /* last block requested */
5524         int                     error = 0;      /* return value */
5525         __int64_t               fixlen;         /* length for -1 case */
5526         int                     i;              /* extent number */
5527         int                     lock;           /* lock state */
5528         xfs_bmbt_irec_t         *map;           /* buffer for user's data */
5529         xfs_mount_t             *mp;            /* file system mount point */
5530         int                     nex;            /* # of user extents can do */
5531         int                     nexleft;        /* # of user extents left */
5532         int                     subnex;         /* # of bmapi's can do */
5533         int                     nmap;           /* number of map entries */
5534         struct getbmapx         *out;           /* output structure */
5535         int                     whichfork;      /* data or attr fork */
5536         int                     prealloced;     /* this is a file with
5537                                                  * preallocated data space */
5538         int                     iflags;         /* interface flags */
5539         int                     bmapi_flags;    /* flags for xfs_bmapi */
5540         int                     cur_ext = 0;
5541
5542         mp = ip->i_mount;
5543         iflags = bmv->bmv_iflags;
5544         whichfork = iflags & BMV_IF_ATTRFORK ? XFS_ATTR_FORK : XFS_DATA_FORK;
5545
5546         if (whichfork == XFS_ATTR_FORK) {
5547                 if (XFS_IFORK_Q(ip)) {
5548                         if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS &&
5549                             ip->i_d.di_aformat != XFS_DINODE_FMT_BTREE &&
5550                             ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)
5551                                 return XFS_ERROR(EINVAL);
5552                 } else if (unlikely(
5553                            ip->i_d.di_aformat != 0 &&
5554                            ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS)) {
5555                         XFS_ERROR_REPORT("xfs_getbmap", XFS_ERRLEVEL_LOW,
5556                                          ip->i_mount);
5557                         return XFS_ERROR(EFSCORRUPTED);
5558                 }
5559
5560                 prealloced = 0;
5561                 fixlen = 1LL << 32;
5562         } else {
5563                 if (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS &&
5564                     ip->i_d.di_format != XFS_DINODE_FMT_BTREE &&
5565                     ip->i_d.di_format != XFS_DINODE_FMT_LOCAL)
5566                         return XFS_ERROR(EINVAL);
5567
5568                 if (xfs_get_extsz_hint(ip) ||
5569                     ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC|XFS_DIFLAG_APPEND)){
5570                         prealloced = 1;
5571                         fixlen = mp->m_super->s_maxbytes;
5572                 } else {
5573                         prealloced = 0;
5574                         fixlen = XFS_ISIZE(ip);
5575                 }
5576         }
5577
5578         if (bmv->bmv_length == -1) {
5579                 fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, fixlen));
5580                 bmv->bmv_length =
5581                         max_t(__int64_t, fixlen - bmv->bmv_offset, 0);
5582         } else if (bmv->bmv_length == 0) {
5583                 bmv->bmv_entries = 0;
5584                 return 0;
5585         } else if (bmv->bmv_length < 0) {
5586                 return XFS_ERROR(EINVAL);
5587         }
5588
5589         nex = bmv->bmv_count - 1;
5590         if (nex <= 0)
5591                 return XFS_ERROR(EINVAL);
5592         bmvend = bmv->bmv_offset + bmv->bmv_length;
5593
5594
5595         if (bmv->bmv_count > ULONG_MAX / sizeof(struct getbmapx))
5596                 return XFS_ERROR(ENOMEM);
5597         out = kmem_zalloc(bmv->bmv_count * sizeof(struct getbmapx), KM_MAYFAIL);
5598         if (!out) {
5599                 out = kmem_zalloc_large(bmv->bmv_count *
5600                                         sizeof(struct getbmapx));
5601                 if (!out)
5602                         return XFS_ERROR(ENOMEM);
5603         }
5604
5605         xfs_ilock(ip, XFS_IOLOCK_SHARED);
5606         if (whichfork == XFS_DATA_FORK && !(iflags & BMV_IF_DELALLOC)) {
5607                 if (ip->i_delayed_blks || XFS_ISIZE(ip) > ip->i_d.di_size) {
5608                         error = -filemap_write_and_wait(VFS_I(ip)->i_mapping);
5609                         if (error)
5610                                 goto out_unlock_iolock;
5611                 }
5612                 /*
5613                  * even after flushing the inode, there can still be delalloc
5614                  * blocks on the inode beyond EOF due to speculative
5615                  * preallocation. These are not removed until the release
5616                  * function is called or the inode is inactivated. Hence we
5617                  * cannot assert here that ip->i_delayed_blks == 0.
5618                  */
5619         }
5620
5621         lock = xfs_ilock_map_shared(ip);
5622
5623         /*
5624          * Don't let nex be bigger than the number of extents
5625          * we can have assuming alternating holes and real extents.
5626          */
5627         if (nex > XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1)
5628                 nex = XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1;
5629
5630         bmapi_flags = xfs_bmapi_aflag(whichfork);
5631         if (!(iflags & BMV_IF_PREALLOC))
5632                 bmapi_flags |= XFS_BMAPI_IGSTATE;
5633
5634         /*
5635          * Allocate enough space to handle "subnex" maps at a time.
5636          */
5637         error = ENOMEM;
5638         subnex = 16;
5639         map = kmem_alloc(subnex * sizeof(*map), KM_MAYFAIL | KM_NOFS);
5640         if (!map)
5641                 goto out_unlock_ilock;
5642
5643         bmv->bmv_entries = 0;
5644
5645         if (XFS_IFORK_NEXTENTS(ip, whichfork) == 0 &&
5646             (whichfork == XFS_ATTR_FORK || !(iflags & BMV_IF_DELALLOC))) {
5647                 error = 0;
5648                 goto out_free_map;
5649         }
5650
5651         nexleft = nex;
5652
5653         do {
5654                 nmap = (nexleft > subnex) ? subnex : nexleft;
5655                 error = xfs_bmapi_read(ip, XFS_BB_TO_FSBT(mp, bmv->bmv_offset),
5656                                        XFS_BB_TO_FSB(mp, bmv->bmv_length),
5657                                        map, &nmap, bmapi_flags);
5658                 if (error)
5659                         goto out_free_map;
5660                 ASSERT(nmap <= subnex);
5661
5662                 for (i = 0; i < nmap && nexleft && bmv->bmv_length; i++) {
5663                         out[cur_ext].bmv_oflags = 0;
5664                         if (map[i].br_state == XFS_EXT_UNWRITTEN)
5665                                 out[cur_ext].bmv_oflags |= BMV_OF_PREALLOC;
5666                         else if (map[i].br_startblock == DELAYSTARTBLOCK)
5667                                 out[cur_ext].bmv_oflags |= BMV_OF_DELALLOC;
5668                         out[cur_ext].bmv_offset =
5669                                 XFS_FSB_TO_BB(mp, map[i].br_startoff);
5670                         out[cur_ext].bmv_length =
5671                                 XFS_FSB_TO_BB(mp, map[i].br_blockcount);
5672                         out[cur_ext].bmv_unused1 = 0;
5673                         out[cur_ext].bmv_unused2 = 0;
5674
5675                         /*
5676                          * delayed allocation extents that start beyond EOF can
5677                          * occur due to speculative EOF allocation when the
5678                          * delalloc extent is larger than the largest freespace
5679                          * extent at conversion time. These extents cannot be
5680                          * converted by data writeback, so can exist here even
5681                          * if we are not supposed to be finding delalloc
5682                          * extents.
5683                          */
5684                         if (map[i].br_startblock == DELAYSTARTBLOCK &&
5685                             map[i].br_startoff <= XFS_B_TO_FSB(mp, XFS_ISIZE(ip)))
5686                                 ASSERT((iflags & BMV_IF_DELALLOC) != 0);
5687
5688                         if (map[i].br_startblock == HOLESTARTBLOCK &&
5689                             whichfork == XFS_ATTR_FORK) {
5690                                 /* came to the end of attribute fork */
5691                                 out[cur_ext].bmv_oflags |= BMV_OF_LAST;
5692                                 goto out_free_map;
5693                         }
5694
5695                         if (!xfs_getbmapx_fix_eof_hole(ip, &out[cur_ext],
5696                                         prealloced, bmvend,
5697                                         map[i].br_startblock))
5698                                 goto out_free_map;
5699
5700                         bmv->bmv_offset =
5701                                 out[cur_ext].bmv_offset +
5702                                 out[cur_ext].bmv_length;
5703                         bmv->bmv_length =
5704                                 max_t(__int64_t, 0, bmvend - bmv->bmv_offset);
5705
5706                         /*
5707                          * In case we don't want to return the hole,
5708                          * don't increase cur_ext so that we can reuse
5709                          * it in the next loop.
5710                          */
5711                         if ((iflags & BMV_IF_NO_HOLES) &&
5712                             map[i].br_startblock == HOLESTARTBLOCK) {
5713                                 memset(&out[cur_ext], 0, sizeof(out[cur_ext]));
5714                                 continue;
5715                         }
5716
5717                         nexleft--;
5718                         bmv->bmv_entries++;
5719                         cur_ext++;
5720                 }
5721         } while (nmap && nexleft && bmv->bmv_length);
5722
5723  out_free_map:
5724         kmem_free(map);
5725  out_unlock_ilock:
5726         xfs_iunlock_map_shared(ip, lock);
5727  out_unlock_iolock:
5728         xfs_iunlock(ip, XFS_IOLOCK_SHARED);
5729
5730         for (i = 0; i < cur_ext; i++) {
5731                 int full = 0;   /* user array is full */
5732
5733                 /* format results & advance arg */
5734                 error = formatter(&arg, &out[i], &full);
5735                 if (error || full)
5736                         break;
5737         }
5738
5739         if (is_vmalloc_addr(out))
5740                 kmem_free_large(out);
5741         else
5742                 kmem_free(out);
5743         return error;
5744 }
5745
5746 #ifdef DEBUG
5747 STATIC struct xfs_buf *
5748 xfs_bmap_get_bp(
5749         struct xfs_btree_cur    *cur,
5750         xfs_fsblock_t           bno)
5751 {
5752         struct xfs_log_item_desc *lidp;
5753         int                     i;
5754
5755         if (!cur)
5756                 return NULL;
5757
5758         for (i = 0; i < XFS_BTREE_MAXLEVELS; i++) {
5759                 if (!cur->bc_bufs[i])
5760                         break;
5761                 if (XFS_BUF_ADDR(cur->bc_bufs[i]) == bno)
5762                         return cur->bc_bufs[i];
5763         }
5764
5765         /* Chase down all the log items to see if the bp is there */
5766         list_for_each_entry(lidp, &cur->bc_tp->t_items, lid_trans) {
5767                 struct xfs_buf_log_item *bip;
5768                 bip = (struct xfs_buf_log_item *)lidp->lid_item;
5769                 if (bip->bli_item.li_type == XFS_LI_BUF &&
5770                     XFS_BUF_ADDR(bip->bli_buf) == bno)
5771                         return bip->bli_buf;
5772         }
5773
5774         return NULL;
5775 }
5776
5777 STATIC void
5778 xfs_check_block(
5779         struct xfs_btree_block  *block,
5780         xfs_mount_t             *mp,
5781         int                     root,
5782         short                   sz)
5783 {
5784         int                     i, j, dmxr;
5785         __be64                  *pp, *thispa;   /* pointer to block address */
5786         xfs_bmbt_key_t          *prevp, *keyp;
5787
5788         ASSERT(be16_to_cpu(block->bb_level) > 0);
5789
5790         prevp = NULL;
5791         for( i = 1; i <= xfs_btree_get_numrecs(block); i++) {
5792                 dmxr = mp->m_bmap_dmxr[0];
5793                 keyp = XFS_BMBT_KEY_ADDR(mp, block, i);
5794
5795                 if (prevp) {
5796                         ASSERT(be64_to_cpu(prevp->br_startoff) <
5797                                be64_to_cpu(keyp->br_startoff));
5798                 }
5799                 prevp = keyp;
5800
5801                 /*
5802                  * Compare the block numbers to see if there are dups.
5803                  */
5804                 if (root)
5805                         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, i, sz);
5806                 else
5807                         pp = XFS_BMBT_PTR_ADDR(mp, block, i, dmxr);
5808
5809                 for (j = i+1; j <= be16_to_cpu(block->bb_numrecs); j++) {
5810                         if (root)
5811                                 thispa = XFS_BMAP_BROOT_PTR_ADDR(mp, block, j, sz);
5812                         else
5813                                 thispa = XFS_BMBT_PTR_ADDR(mp, block, j, dmxr);
5814                         if (*thispa == *pp) {
5815                                 xfs_warn(mp, "%s: thispa(%d) == pp(%d) %Ld",
5816                                         __func__, j, i,
5817                                         (unsigned long long)be64_to_cpu(*thispa));
5818                                 panic("%s: ptrs are equal in node\n",
5819                                         __func__);
5820                         }
5821                 }
5822         }
5823 }
5824
5825 /*
5826  * Check that the extents for the inode ip are in the right order in all
5827  * btree leaves.
5828  */
5829
5830 STATIC void
5831 xfs_bmap_check_leaf_extents(
5832         xfs_btree_cur_t         *cur,   /* btree cursor or null */
5833         xfs_inode_t             *ip,            /* incore inode pointer */
5834         int                     whichfork)      /* data or attr fork */
5835 {
5836         struct xfs_btree_block  *block; /* current btree block */
5837         xfs_fsblock_t           bno;    /* block # of "block" */
5838         xfs_buf_t               *bp;    /* buffer for "block" */
5839         int                     error;  /* error return value */
5840         xfs_extnum_t            i=0, j; /* index into the extents list */
5841         xfs_ifork_t             *ifp;   /* fork structure */
5842         int                     level;  /* btree level, for checking */
5843         xfs_mount_t             *mp;    /* file system mount structure */
5844         __be64                  *pp;    /* pointer to block address */
5845         xfs_bmbt_rec_t          *ep;    /* pointer to current extent */
5846         xfs_bmbt_rec_t          last = {0, 0}; /* last extent in prev block */
5847         xfs_bmbt_rec_t          *nextp; /* pointer to next extent */
5848         int                     bp_release = 0;
5849
5850         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE) {
5851                 return;
5852         }
5853
5854         bno = NULLFSBLOCK;
5855         mp = ip->i_mount;
5856         ifp = XFS_IFORK_PTR(ip, whichfork);
5857         block = ifp->if_broot;
5858         /*
5859          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
5860          */
5861         level = be16_to_cpu(block->bb_level);
5862         ASSERT(level > 0);
5863         xfs_check_block(block, mp, 1, ifp->if_broot_bytes);
5864         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
5865         bno = be64_to_cpu(*pp);
5866
5867         ASSERT(bno != NULLDFSBNO);
5868         ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
5869         ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
5870
5871         /*
5872          * Go down the tree until leaf level is reached, following the first
5873          * pointer (leftmost) at each level.
5874          */
5875         while (level-- > 0) {
5876                 /* See if buf is in cur first */
5877                 bp_release = 0;
5878                 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
5879                 if (!bp) {
5880                         bp_release = 1;
5881                         error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
5882                                                 XFS_BMAP_BTREE_REF,
5883                                                 &xfs_bmbt_buf_ops);
5884                         if (error)
5885                                 goto error_norelse;
5886                 }
5887                 block = XFS_BUF_TO_BLOCK(bp);
5888                 XFS_WANT_CORRUPTED_GOTO(
5889                         xfs_bmap_sanity_check(mp, bp, level),
5890                         error0);
5891                 if (level == 0)
5892                         break;
5893
5894                 /*
5895                  * Check this block for basic sanity (increasing keys and
5896                  * no duplicate blocks).
5897                  */
5898
5899                 xfs_check_block(block, mp, 0, 0);
5900                 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
5901                 bno = be64_to_cpu(*pp);
5902                 XFS_WANT_CORRUPTED_GOTO(XFS_FSB_SANITY_CHECK(mp, bno), error0);
5903                 if (bp_release) {
5904                         bp_release = 0;
5905                         xfs_trans_brelse(NULL, bp);
5906                 }
5907         }
5908
5909         /*
5910          * Here with bp and block set to the leftmost leaf node in the tree.
5911          */
5912         i = 0;
5913
5914         /*
5915          * Loop over all leaf nodes checking that all extents are in the right order.
5916          */
5917         for (;;) {
5918                 xfs_fsblock_t   nextbno;
5919                 xfs_extnum_t    num_recs;
5920
5921
5922                 num_recs = xfs_btree_get_numrecs(block);
5923
5924                 /*
5925                  * Read-ahead the next leaf block, if any.
5926                  */
5927
5928                 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
5929
5930                 /*
5931                  * Check all the extents to make sure they are OK.
5932                  * If we had a previous block, the last entry should
5933                  * conform with the first entry in this one.
5934                  */
5935
5936                 ep = XFS_BMBT_REC_ADDR(mp, block, 1);
5937                 if (i) {
5938                         ASSERT(xfs_bmbt_disk_get_startoff(&last) +
5939                                xfs_bmbt_disk_get_blockcount(&last) <=
5940                                xfs_bmbt_disk_get_startoff(ep));
5941                 }
5942                 for (j = 1; j < num_recs; j++) {
5943                         nextp = XFS_BMBT_REC_ADDR(mp, block, j + 1);
5944                         ASSERT(xfs_bmbt_disk_get_startoff(ep) +
5945                                xfs_bmbt_disk_get_blockcount(ep) <=
5946                                xfs_bmbt_disk_get_startoff(nextp));
5947                         ep = nextp;
5948                 }
5949
5950                 last = *ep;
5951                 i += num_recs;
5952                 if (bp_release) {
5953                         bp_release = 0;
5954                         xfs_trans_brelse(NULL, bp);
5955                 }
5956                 bno = nextbno;
5957                 /*
5958                  * If we've reached the end, stop.
5959                  */
5960                 if (bno == NULLFSBLOCK)
5961                         break;
5962
5963                 bp_release = 0;
5964                 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
5965                 if (!bp) {
5966                         bp_release = 1;
5967                         error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
5968                                                 XFS_BMAP_BTREE_REF,
5969                                                 &xfs_bmbt_buf_ops);
5970                         if (error)
5971                                 goto error_norelse;
5972                 }
5973                 block = XFS_BUF_TO_BLOCK(bp);
5974         }
5975         if (bp_release) {
5976                 bp_release = 0;
5977                 xfs_trans_brelse(NULL, bp);
5978         }
5979         return;
5980
5981 error0:
5982         xfs_warn(mp, "%s: at error0", __func__);
5983         if (bp_release)
5984                 xfs_trans_brelse(NULL, bp);
5985 error_norelse:
5986         xfs_warn(mp, "%s: BAD after btree leaves for %d extents",
5987                 __func__, i);
5988         panic("%s: CORRUPTED BTREE OR SOMETHING", __func__);
5989         return;
5990 }
5991 #endif
5992
5993 /*
5994  * Count fsblocks of the given fork.
5995  */
5996 int                                             /* error */
5997 xfs_bmap_count_blocks(
5998         xfs_trans_t             *tp,            /* transaction pointer */
5999         xfs_inode_t             *ip,            /* incore inode */
6000         int                     whichfork,      /* data or attr fork */
6001         int                     *count)         /* out: count of blocks */
6002 {
6003         struct xfs_btree_block  *block; /* current btree block */
6004         xfs_fsblock_t           bno;    /* block # of "block" */
6005         xfs_ifork_t             *ifp;   /* fork structure */
6006         int                     level;  /* btree level, for checking */
6007         xfs_mount_t             *mp;    /* file system mount structure */
6008         __be64                  *pp;    /* pointer to block address */
6009
6010         bno = NULLFSBLOCK;
6011         mp = ip->i_mount;
6012         ifp = XFS_IFORK_PTR(ip, whichfork);
6013         if ( XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ) {
6014                 xfs_bmap_count_leaves(ifp, 0,
6015                         ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t),
6016                         count);
6017                 return 0;
6018         }
6019
6020         /*
6021          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
6022          */
6023         block = ifp->if_broot;
6024         level = be16_to_cpu(block->bb_level);
6025         ASSERT(level > 0);
6026         pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
6027         bno = be64_to_cpu(*pp);
6028         ASSERT(bno != NULLDFSBNO);
6029         ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
6030         ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
6031
6032         if (unlikely(xfs_bmap_count_tree(mp, tp, ifp, bno, level, count) < 0)) {
6033                 XFS_ERROR_REPORT("xfs_bmap_count_blocks(2)", XFS_ERRLEVEL_LOW,
6034                                  mp);
6035                 return XFS_ERROR(EFSCORRUPTED);
6036         }
6037
6038         return 0;
6039 }
6040
6041 /*
6042  * Recursively walks each level of a btree
6043  * to count total fsblocks is use.
6044  */
6045 STATIC int                                     /* error */
6046 xfs_bmap_count_tree(
6047         xfs_mount_t     *mp,            /* file system mount point */
6048         xfs_trans_t     *tp,            /* transaction pointer */
6049         xfs_ifork_t     *ifp,           /* inode fork pointer */
6050         xfs_fsblock_t   blockno,        /* file system block number */
6051         int             levelin,        /* level in btree */
6052         int             *count)         /* Count of blocks */
6053 {
6054         int                     error;
6055         xfs_buf_t               *bp, *nbp;
6056         int                     level = levelin;
6057         __be64                  *pp;
6058         xfs_fsblock_t           bno = blockno;
6059         xfs_fsblock_t           nextbno;
6060         struct xfs_btree_block  *block, *nextblock;
6061         int                     numrecs;
6062
6063         error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp, XFS_BMAP_BTREE_REF,
6064                                                 &xfs_bmbt_buf_ops);
6065         if (error)
6066                 return error;
6067         *count += 1;
6068         block = XFS_BUF_TO_BLOCK(bp);
6069
6070         if (--level) {
6071                 /* Not at node above leaves, count this level of nodes */
6072                 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
6073                 while (nextbno != NULLFSBLOCK) {
6074                         error = xfs_btree_read_bufl(mp, tp, nextbno, 0, &nbp,
6075                                                 XFS_BMAP_BTREE_REF,
6076                                                 &xfs_bmbt_buf_ops);
6077                         if (error)
6078                                 return error;
6079                         *count += 1;
6080                         nextblock = XFS_BUF_TO_BLOCK(nbp);
6081                         nextbno = be64_to_cpu(nextblock->bb_u.l.bb_rightsib);
6082                         xfs_trans_brelse(tp, nbp);
6083                 }
6084
6085                 /* Dive to the next level */
6086                 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
6087                 bno = be64_to_cpu(*pp);
6088                 if (unlikely((error =
6089                      xfs_bmap_count_tree(mp, tp, ifp, bno, level, count)) < 0)) {
6090                         xfs_trans_brelse(tp, bp);
6091                         XFS_ERROR_REPORT("xfs_bmap_count_tree(1)",
6092                                          XFS_ERRLEVEL_LOW, mp);
6093                         return XFS_ERROR(EFSCORRUPTED);
6094                 }
6095                 xfs_trans_brelse(tp, bp);
6096         } else {
6097                 /* count all level 1 nodes and their leaves */
6098                 for (;;) {
6099                         nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
6100                         numrecs = be16_to_cpu(block->bb_numrecs);
6101                         xfs_bmap_disk_count_leaves(mp, block, numrecs, count);
6102                         xfs_trans_brelse(tp, bp);
6103                         if (nextbno == NULLFSBLOCK)
6104                                 break;
6105                         bno = nextbno;
6106                         error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
6107                                                 XFS_BMAP_BTREE_REF,
6108                                                 &xfs_bmbt_buf_ops);
6109                         if (error)
6110                                 return error;
6111                         *count += 1;
6112                         block = XFS_BUF_TO_BLOCK(bp);
6113                 }
6114         }
6115         return 0;
6116 }
6117
6118 /*
6119  * Count leaf blocks given a range of extent records.
6120  */
6121 STATIC void
6122 xfs_bmap_count_leaves(
6123         xfs_ifork_t             *ifp,
6124         xfs_extnum_t            idx,
6125         int                     numrecs,
6126         int                     *count)
6127 {
6128         int             b;
6129
6130         for (b = 0; b < numrecs; b++) {
6131                 xfs_bmbt_rec_host_t *frp = xfs_iext_get_ext(ifp, idx + b);
6132                 *count += xfs_bmbt_get_blockcount(frp);
6133         }
6134 }
6135
6136 /*
6137  * Count leaf blocks given a range of extent records originally
6138  * in btree format.
6139  */
6140 STATIC void
6141 xfs_bmap_disk_count_leaves(
6142         struct xfs_mount        *mp,
6143         struct xfs_btree_block  *block,
6144         int                     numrecs,
6145         int                     *count)
6146 {
6147         int             b;
6148         xfs_bmbt_rec_t  *frp;
6149
6150         for (b = 1; b <= numrecs; b++) {
6151                 frp = XFS_BMBT_REC_ADDR(mp, block, b);
6152                 *count += xfs_bmbt_disk_get_blockcount(frp);
6153         }
6154 }
6155
6156 /*
6157  * dead simple method of punching delalyed allocation blocks from a range in
6158  * the inode. Walks a block at a time so will be slow, but is only executed in
6159  * rare error cases so the overhead is not critical. This will alays punch out
6160  * both the start and end blocks, even if the ranges only partially overlap
6161  * them, so it is up to the caller to ensure that partial blocks are not
6162  * passed in.
6163  */
6164 int
6165 xfs_bmap_punch_delalloc_range(
6166         struct xfs_inode        *ip,
6167         xfs_fileoff_t           start_fsb,
6168         xfs_fileoff_t           length)
6169 {
6170         xfs_fileoff_t           remaining = length;
6171         int                     error = 0;
6172
6173         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
6174
6175         do {
6176                 int             done;
6177                 xfs_bmbt_irec_t imap;
6178                 int             nimaps = 1;
6179                 xfs_fsblock_t   firstblock;
6180                 xfs_bmap_free_t flist;
6181
6182                 /*
6183                  * Map the range first and check that it is a delalloc extent
6184                  * before trying to unmap the range. Otherwise we will be
6185                  * trying to remove a real extent (which requires a
6186                  * transaction) or a hole, which is probably a bad idea...
6187                  */
6188                 error = xfs_bmapi_read(ip, start_fsb, 1, &imap, &nimaps,
6189                                        XFS_BMAPI_ENTIRE);
6190
6191                 if (error) {
6192                         /* something screwed, just bail */
6193                         if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
6194                                 xfs_alert(ip->i_mount,
6195                         "Failed delalloc mapping lookup ino %lld fsb %lld.",
6196                                                 ip->i_ino, start_fsb);
6197                         }
6198                         break;
6199                 }
6200                 if (!nimaps) {
6201                         /* nothing there */
6202                         goto next_block;
6203                 }
6204                 if (imap.br_startblock != DELAYSTARTBLOCK) {
6205                         /* been converted, ignore */
6206                         goto next_block;
6207                 }
6208                 WARN_ON(imap.br_blockcount == 0);
6209
6210                 /*
6211                  * Note: while we initialise the firstblock/flist pair, they
6212                  * should never be used because blocks should never be
6213                  * allocated or freed for a delalloc extent and hence we need
6214                  * don't cancel or finish them after the xfs_bunmapi() call.
6215                  */
6216                 xfs_bmap_init(&flist, &firstblock);
6217                 error = xfs_bunmapi(NULL, ip, start_fsb, 1, 0, 1, &firstblock,
6218                                         &flist, &done);
6219                 if (error)
6220                         break;
6221
6222                 ASSERT(!flist.xbf_count && !flist.xbf_first);
6223 next_block:
6224                 start_fsb++;
6225                 remaining--;
6226         } while(remaining > 0);
6227
6228         return error;
6229 }
6230
6231 /*
6232  * Convert the given file system block to a disk block.  We have to treat it
6233  * differently based on whether the file is a real time file or not, because the
6234  * bmap code does.
6235  */
6236 xfs_daddr_t
6237 xfs_fsb_to_db(struct xfs_inode *ip, xfs_fsblock_t fsb)
6238 {
6239         return (XFS_IS_REALTIME_INODE(ip) ? \
6240                  (xfs_daddr_t)XFS_FSB_TO_BB((ip)->i_mount, (fsb)) : \
6241                  XFS_FSB_TO_DADDR((ip)->i_mount, (fsb)));
6242 }