fs/xfs/xfs_bmap.c

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