fs/xfs/libxfs/xfs_bmap_btree.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * Copyright (c) 2000-2003,2005 Silicon Graphics, Inc.
   4  * All Rights Reserved.
   5  */
   6 #include "xfs.h"
   7 #include "xfs_fs.h"
   8 #include "xfs_shared.h"
   9 #include "xfs_format.h"
  10 #include "xfs_log_format.h"
  11 #include "xfs_trans_resv.h"
  12 #include "xfs_bit.h"
  13 #include "xfs_mount.h"
  14 #include "xfs_inode.h"
  15 #include "xfs_trans.h"
  16 #include "xfs_alloc.h"
  17 #include "xfs_btree.h"
  18 #include "xfs_bmap_btree.h"
  19 #include "xfs_bmap.h"
  20 #include "xfs_error.h"
  21 #include "xfs_quota.h"
  22 #include "xfs_trace.h"
  23 #include "xfs_rmap.h"
  24 #include "xfs_ag.h"
  25
  26 static struct kmem_cache        *xfs_bmbt_cur_cache;
  27
  28 /*
  29  * Convert on-disk form of btree root to in-memory form.
  30  */
  31 void
  32 xfs_bmdr_to_bmbt(
  33         struct xfs_inode        *ip,
  34         xfs_bmdr_block_t        *dblock,
  35         int                     dblocklen,
  36         struct xfs_btree_block  *rblock,
  37         int                     rblocklen)
  38 {
  39         struct xfs_mount        *mp = ip->i_mount;
  40         int                     dmxr;
  41         xfs_bmbt_key_t          *fkp;
  42         __be64                  *fpp;
  43         xfs_bmbt_key_t          *tkp;
  44         __be64                  *tpp;
  45
  46         xfs_btree_init_block_int(mp, rblock, XFS_BUF_DADDR_NULL,
  47                                  XFS_BTNUM_BMAP, 0, 0, ip->i_ino,
  48                                  XFS_BTREE_LONG_PTRS);
  49         rblock->bb_level = dblock->bb_level;
  50         ASSERT(be16_to_cpu(rblock->bb_level) > 0);
  51         rblock->bb_numrecs = dblock->bb_numrecs;
  52         dmxr = xfs_bmdr_maxrecs(dblocklen, 0);
  53         fkp = XFS_BMDR_KEY_ADDR(dblock, 1);
  54         tkp = XFS_BMBT_KEY_ADDR(mp, rblock, 1);
  55         fpp = XFS_BMDR_PTR_ADDR(dblock, 1, dmxr);
  56         tpp = XFS_BMAP_BROOT_PTR_ADDR(mp, rblock, 1, rblocklen);
  57         dmxr = be16_to_cpu(dblock->bb_numrecs);
  58         memcpy(tkp, fkp, sizeof(*fkp) * dmxr);
  59         memcpy(tpp, fpp, sizeof(*fpp) * dmxr);
  60 }
  61
  62 void
  63 xfs_bmbt_disk_get_all(
  64         const struct xfs_bmbt_rec *rec,
  65         struct xfs_bmbt_irec    *irec)
  66 {
  67         uint64_t                l0 = get_unaligned_be64(&rec->l0);
  68         uint64_t                l1 = get_unaligned_be64(&rec->l1);
  69
  70         irec->br_startoff = (l0 & xfs_mask64lo(64 - BMBT_EXNTFLAG_BITLEN)) >> 9;
  71         irec->br_startblock = ((l0 & xfs_mask64lo(9)) << 43) | (l1 >> 21);
  72         irec->br_blockcount = l1 & xfs_mask64lo(21);
  73         if (l0 >> (64 - BMBT_EXNTFLAG_BITLEN))
  74                 irec->br_state = XFS_EXT_UNWRITTEN;
  75         else
  76                 irec->br_state = XFS_EXT_NORM;
  77 }
  78
  79 /*
  80  * Extract the blockcount field from an on disk bmap extent record.
  81  */
  82 xfs_filblks_t
  83 xfs_bmbt_disk_get_blockcount(
  84         const struct xfs_bmbt_rec       *r)
  85 {
  86         return (xfs_filblks_t)(be64_to_cpu(r->l1) & xfs_mask64lo(21));
  87 }
  88
  89 /*
  90  * Extract the startoff field from a disk format bmap extent record.
  91  */
  92 xfs_fileoff_t
  93 xfs_bmbt_disk_get_startoff(
  94         const struct xfs_bmbt_rec       *r)
  95 {
  96         return ((xfs_fileoff_t)be64_to_cpu(r->l0) &
  97                  xfs_mask64lo(64 - BMBT_EXNTFLAG_BITLEN)) >> 9;
  98 }
  99
 100 /*
 101  * Set all the fields in a bmap extent record from the uncompressed form.
 102  */
 103 void
 104 xfs_bmbt_disk_set_all(
 105         struct xfs_bmbt_rec     *r,
 106         struct xfs_bmbt_irec    *s)
 107 {
 108         int                     extent_flag = (s->br_state != XFS_EXT_NORM);
 109
 110         ASSERT(s->br_state == XFS_EXT_NORM || s->br_state == XFS_EXT_UNWRITTEN);
 111         ASSERT(!(s->br_startoff & xfs_mask64hi(64-BMBT_STARTOFF_BITLEN)));
 112         ASSERT(!(s->br_blockcount & xfs_mask64hi(64-BMBT_BLOCKCOUNT_BITLEN)));
 113         ASSERT(!(s->br_startblock & xfs_mask64hi(64-BMBT_STARTBLOCK_BITLEN)));
 114
 115         put_unaligned_be64(
 116                 ((xfs_bmbt_rec_base_t)extent_flag << 63) |
 117                  ((xfs_bmbt_rec_base_t)s->br_startoff << 9) |
 118                  ((xfs_bmbt_rec_base_t)s->br_startblock >> 43), &r->l0);
 119         put_unaligned_be64(
 120                 ((xfs_bmbt_rec_base_t)s->br_startblock << 21) |
 121                  ((xfs_bmbt_rec_base_t)s->br_blockcount &
 122                   (xfs_bmbt_rec_base_t)xfs_mask64lo(21)), &r->l1);
 123 }
 124
 125 /*
 126  * Convert in-memory form of btree root to on-disk form.
 127  */
 128 void
 129 xfs_bmbt_to_bmdr(
 130         struct xfs_mount        *mp,
 131         struct xfs_btree_block  *rblock,
 132         int                     rblocklen,
 133         xfs_bmdr_block_t        *dblock,
 134         int                     dblocklen)
 135 {
 136         int                     dmxr;
 137         xfs_bmbt_key_t          *fkp;
 138         __be64                  *fpp;
 139         xfs_bmbt_key_t          *tkp;
 140         __be64                  *tpp;
 141
 142         if (xfs_has_crc(mp)) {
 143                 ASSERT(rblock->bb_magic == cpu_to_be32(XFS_BMAP_CRC_MAGIC));
 144                 ASSERT(uuid_equal(&rblock->bb_u.l.bb_uuid,
 145                        &mp->m_sb.sb_meta_uuid));
 146                 ASSERT(rblock->bb_u.l.bb_blkno ==
 147                        cpu_to_be64(XFS_BUF_DADDR_NULL));
 148         } else
 149                 ASSERT(rblock->bb_magic == cpu_to_be32(XFS_BMAP_MAGIC));
 150         ASSERT(rblock->bb_u.l.bb_leftsib == cpu_to_be64(NULLFSBLOCK));
 151         ASSERT(rblock->bb_u.l.bb_rightsib == cpu_to_be64(NULLFSBLOCK));
 152         ASSERT(rblock->bb_level != 0);
 153         dblock->bb_level = rblock->bb_level;
 154         dblock->bb_numrecs = rblock->bb_numrecs;
 155         dmxr = xfs_bmdr_maxrecs(dblocklen, 0);
 156         fkp = XFS_BMBT_KEY_ADDR(mp, rblock, 1);
 157         tkp = XFS_BMDR_KEY_ADDR(dblock, 1);
 158         fpp = XFS_BMAP_BROOT_PTR_ADDR(mp, rblock, 1, rblocklen);
 159         tpp = XFS_BMDR_PTR_ADDR(dblock, 1, dmxr);
 160         dmxr = be16_to_cpu(dblock->bb_numrecs);
 161         memcpy(tkp, fkp, sizeof(*fkp) * dmxr);
 162         memcpy(tpp, fpp, sizeof(*fpp) * dmxr);
 163 }
 164
 165 STATIC struct xfs_btree_cur *
 166 xfs_bmbt_dup_cursor(
 167         struct xfs_btree_cur    *cur)
 168 {
 169         struct xfs_btree_cur    *new;
 170
 171         new = xfs_bmbt_init_cursor(cur->bc_mp, cur->bc_tp,
 172                         cur->bc_ino.ip, cur->bc_ino.whichfork);
 173
 174         /*
 175          * Copy the firstblock, dfops, and flags values,
 176          * since init cursor doesn't get them.
 177          */
 178         new->bc_ino.flags = cur->bc_ino.flags;
 179
 180         return new;
 181 }
 182
 183 STATIC void
 184 xfs_bmbt_update_cursor(
 185         struct xfs_btree_cur    *src,
 186         struct xfs_btree_cur    *dst)
 187 {
 188         ASSERT((dst->bc_tp->t_highest_agno != NULLAGNUMBER) ||
 189                (dst->bc_ino.ip->i_diflags & XFS_DIFLAG_REALTIME));
 190
 191         dst->bc_ino.allocated += src->bc_ino.allocated;
 192         dst->bc_tp->t_highest_agno = src->bc_tp->t_highest_agno;
 193
 194         src->bc_ino.allocated = 0;
 195 }
 196
 197 STATIC int
 198 xfs_bmbt_alloc_block(
 199         struct xfs_btree_cur            *cur,
 200         const union xfs_btree_ptr       *start,
 201         union xfs_btree_ptr             *new,
 202         int                             *stat)
 203 {
 204         struct xfs_alloc_arg    args;
 205         int                     error;
 206
 207         memset(&args, 0, sizeof(args));
 208         args.tp = cur->bc_tp;
 209         args.mp = cur->bc_mp;
 210         xfs_rmap_ino_bmbt_owner(&args.oinfo, cur->bc_ino.ip->i_ino,
 211                         cur->bc_ino.whichfork);
 212         args.minlen = args.maxlen = args.prod = 1;
 213         args.wasdel = cur->bc_ino.flags & XFS_BTCUR_BMBT_WASDEL;
 214         if (!args.wasdel && args.tp->t_blk_res == 0)
 215                 return -ENOSPC;
 216
 217         /*
 218          * If we are coming here from something like unwritten extent
 219          * conversion, there has been no data extent allocation already done, so
 220          * we have to ensure that we attempt to locate the entire set of bmbt
 221          * allocations in the same AG, as xfs_bmapi_write() would have reserved.
 222          */
 223         if (cur->bc_tp->t_highest_agno == NULLAGNUMBER)
 224                 args.minleft = xfs_bmapi_minleft(cur->bc_tp, cur->bc_ino.ip,
 225                                         cur->bc_ino.whichfork);
 226
 227         error = xfs_alloc_vextent_start_ag(&args, be64_to_cpu(start->l));
 228         if (error)
 229                 return error;
 230
 231         if (args.fsbno == NULLFSBLOCK && args.minleft) {
 232                 /*
 233                  * Could not find an AG with enough free space to satisfy
 234                  * a full btree split.  Try again and if
 235                  * successful activate the lowspace algorithm.
 236                  */
 237                 args.minleft = 0;
 238                 error = xfs_alloc_vextent_start_ag(&args, 0);
 239                 if (error)
 240                         return error;
 241                 cur->bc_tp->t_flags |= XFS_TRANS_LOWMODE;
 242         }
 243         if (WARN_ON_ONCE(args.fsbno == NULLFSBLOCK)) {
 244                 *stat = 0;
 245                 return 0;
 246         }
 247
 248         ASSERT(args.len == 1);
 249         cur->bc_ino.allocated++;
 250         cur->bc_ino.ip->i_nblocks++;
 251         xfs_trans_log_inode(args.tp, cur->bc_ino.ip, XFS_ILOG_CORE);
 252         xfs_trans_mod_dquot_byino(args.tp, cur->bc_ino.ip,
 253                         XFS_TRANS_DQ_BCOUNT, 1L);
 254
 255         new->l = cpu_to_be64(args.fsbno);
 256
 257         *stat = 1;
 258         return 0;
 259 }
 260
 261 STATIC int
 262 xfs_bmbt_free_block(
 263         struct xfs_btree_cur    *cur,
 264         struct xfs_buf          *bp)
 265 {
 266         struct xfs_mount        *mp = cur->bc_mp;
 267         struct xfs_inode        *ip = cur->bc_ino.ip;
 268         struct xfs_trans        *tp = cur->bc_tp;
 269         xfs_fsblock_t           fsbno = XFS_DADDR_TO_FSB(mp, xfs_buf_daddr(bp));
 270         struct xfs_owner_info   oinfo;
 271         int                     error;
 272
 273         xfs_rmap_ino_bmbt_owner(&oinfo, ip->i_ino, cur->bc_ino.whichfork);
 274         error = xfs_free_extent_later(cur->bc_tp, fsbno, 1, &oinfo);
 275         if (error)
 276                 return error;
 277
 278         ip->i_nblocks--;
 279         xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
 280         xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, -1L);
 281         return 0;
 282 }
 283
 284 STATIC int
 285 xfs_bmbt_get_minrecs(
 286         struct xfs_btree_cur    *cur,
 287         int                     level)
 288 {
 289         if (level == cur->bc_nlevels - 1) {
 290                 struct xfs_ifork        *ifp;
 291
 292                 ifp = xfs_ifork_ptr(cur->bc_ino.ip,
 293                                     cur->bc_ino.whichfork);
 294
 295                 return xfs_bmbt_maxrecs(cur->bc_mp,
 296                                         ifp->if_broot_bytes, level == 0) / 2;
 297         }
 298
 299         return cur->bc_mp->m_bmap_dmnr[level != 0];
 300 }
 301
 302 int
 303 xfs_bmbt_get_maxrecs(
 304         struct xfs_btree_cur    *cur,
 305         int                     level)
 306 {
 307         if (level == cur->bc_nlevels - 1) {
 308                 struct xfs_ifork        *ifp;
 309
 310                 ifp = xfs_ifork_ptr(cur->bc_ino.ip,
 311                                     cur->bc_ino.whichfork);
 312
 313                 return xfs_bmbt_maxrecs(cur->bc_mp,
 314                                         ifp->if_broot_bytes, level == 0);
 315         }
 316
 317         return cur->bc_mp->m_bmap_dmxr[level != 0];
 318
 319 }
 320
 321 /*
 322  * Get the maximum records we could store in the on-disk format.
 323  *
 324  * For non-root nodes this is equivalent to xfs_bmbt_get_maxrecs, but
 325  * for the root node this checks the available space in the dinode fork
 326  * so that we can resize the in-memory buffer to match it.  After a
 327  * resize to the maximum size this function returns the same value
 328  * as xfs_bmbt_get_maxrecs for the root node, too.
 329  */
 330 STATIC int
 331 xfs_bmbt_get_dmaxrecs(
 332         struct xfs_btree_cur    *cur,
 333         int                     level)
 334 {
 335         if (level != cur->bc_nlevels - 1)
 336                 return cur->bc_mp->m_bmap_dmxr[level != 0];
 337         return xfs_bmdr_maxrecs(cur->bc_ino.forksize, level == 0);
 338 }
 339
 340 STATIC void
 341 xfs_bmbt_init_key_from_rec(
 342         union xfs_btree_key             *key,
 343         const union xfs_btree_rec       *rec)
 344 {
 345         key->bmbt.br_startoff =
 346                 cpu_to_be64(xfs_bmbt_disk_get_startoff(&rec->bmbt));
 347 }
 348
 349 STATIC void
 350 xfs_bmbt_init_high_key_from_rec(
 351         union xfs_btree_key             *key,
 352         const union xfs_btree_rec       *rec)
 353 {
 354         key->bmbt.br_startoff = cpu_to_be64(
 355                         xfs_bmbt_disk_get_startoff(&rec->bmbt) +
 356                         xfs_bmbt_disk_get_blockcount(&rec->bmbt) - 1);
 357 }
 358
 359 STATIC void
 360 xfs_bmbt_init_rec_from_cur(
 361         struct xfs_btree_cur    *cur,
 362         union xfs_btree_rec     *rec)
 363 {
 364         xfs_bmbt_disk_set_all(&rec->bmbt, &cur->bc_rec.b);
 365 }
 366
 367 STATIC void
 368 xfs_bmbt_init_ptr_from_cur(
 369         struct xfs_btree_cur    *cur,
 370         union xfs_btree_ptr     *ptr)
 371 {
 372         ptr->l = 0;
 373 }
 374
 375 STATIC int64_t
 376 xfs_bmbt_key_diff(
 377         struct xfs_btree_cur            *cur,
 378         const union xfs_btree_key       *key)
 379 {
 380         return (int64_t)be64_to_cpu(key->bmbt.br_startoff) -
 381                                       cur->bc_rec.b.br_startoff;
 382 }
 383
 384 STATIC int64_t
 385 xfs_bmbt_diff_two_keys(
 386         struct xfs_btree_cur            *cur,
 387         const union xfs_btree_key       *k1,
 388         const union xfs_btree_key       *k2,
 389         const union xfs_btree_key       *mask)
 390 {
 391         uint64_t                        a = be64_to_cpu(k1->bmbt.br_startoff);
 392         uint64_t                        b = be64_to_cpu(k2->bmbt.br_startoff);
 393
 394         ASSERT(!mask || mask->bmbt.br_startoff);
 395
 396         /*
 397          * Note: This routine previously casted a and b to int64 and subtracted
 398          * them to generate a result.  This lead to problems if b was the
 399          * "maximum" key value (all ones) being signed incorrectly, hence this
 400          * somewhat less efficient version.
 401          */
 402         if (a > b)
 403                 return 1;
 404         if (b > a)
 405                 return -1;
 406         return 0;
 407 }
 408
 409 static xfs_failaddr_t
 410 xfs_bmbt_verify(
 411         struct xfs_buf          *bp)
 412 {
 413         struct xfs_mount        *mp = bp->b_mount;
 414         struct xfs_btree_block  *block = XFS_BUF_TO_BLOCK(bp);
 415         xfs_failaddr_t          fa;
 416         unsigned int            level;
 417
 418         if (!xfs_verify_magic(bp, block->bb_magic))
 419                 return __this_address;
 420
 421         if (xfs_has_crc(mp)) {
 422                 /*
 423                  * XXX: need a better way of verifying the owner here. Right now
 424                  * just make sure there has been one set.
 425                  */
 426                 fa = xfs_btree_lblock_v5hdr_verify(bp, XFS_RMAP_OWN_UNKNOWN);
 427                 if (fa)
 428                         return fa;
 429         }
 430
 431         /*
 432          * numrecs and level verification.
 433          *
 434          * We don't know what fork we belong to, so just verify that the level
 435          * is less than the maximum of the two. Later checks will be more
 436          * precise.
 437          */
 438         level = be16_to_cpu(block->bb_level);
 439         if (level > max(mp->m_bm_maxlevels[0], mp->m_bm_maxlevels[1]))
 440                 return __this_address;
 441
 442         return xfs_btree_lblock_verify(bp, mp->m_bmap_dmxr[level != 0]);
 443 }
 444
 445 static void
 446 xfs_bmbt_read_verify(
 447         struct xfs_buf  *bp)
 448 {
 449         xfs_failaddr_t  fa;
 450
 451         if (!xfs_btree_lblock_verify_crc(bp))
 452                 xfs_verifier_error(bp, -EFSBADCRC, __this_address);
 453         else {
 454                 fa = xfs_bmbt_verify(bp);
 455                 if (fa)
 456                         xfs_verifier_error(bp, -EFSCORRUPTED, fa);
 457         }
 458
 459         if (bp->b_error)
 460                 trace_xfs_btree_corrupt(bp, _RET_IP_);
 461 }
 462
 463 static void
 464 xfs_bmbt_write_verify(
 465         struct xfs_buf  *bp)
 466 {
 467         xfs_failaddr_t  fa;
 468
 469         fa = xfs_bmbt_verify(bp);
 470         if (fa) {
 471                 trace_xfs_btree_corrupt(bp, _RET_IP_);
 472                 xfs_verifier_error(bp, -EFSCORRUPTED, fa);
 473                 return;
 474         }
 475         xfs_btree_lblock_calc_crc(bp);
 476 }
 477
 478 const struct xfs_buf_ops xfs_bmbt_buf_ops = {
 479         .name = "xfs_bmbt",
 480         .magic = { cpu_to_be32(XFS_BMAP_MAGIC),
 481                    cpu_to_be32(XFS_BMAP_CRC_MAGIC) },
 482         .verify_read = xfs_bmbt_read_verify,
 483         .verify_write = xfs_bmbt_write_verify,
 484         .verify_struct = xfs_bmbt_verify,
 485 };
 486
 487
 488 STATIC int
 489 xfs_bmbt_keys_inorder(
 490         struct xfs_btree_cur            *cur,
 491         const union xfs_btree_key       *k1,
 492         const union xfs_btree_key       *k2)
 493 {
 494         return be64_to_cpu(k1->bmbt.br_startoff) <
 495                 be64_to_cpu(k2->bmbt.br_startoff);
 496 }
 497
 498 STATIC int
 499 xfs_bmbt_recs_inorder(
 500         struct xfs_btree_cur            *cur,
 501         const union xfs_btree_rec       *r1,
 502         const union xfs_btree_rec       *r2)
 503 {
 504         return xfs_bmbt_disk_get_startoff(&r1->bmbt) +
 505                 xfs_bmbt_disk_get_blockcount(&r1->bmbt) <=
 506                 xfs_bmbt_disk_get_startoff(&r2->bmbt);
 507 }
 508
 509 STATIC enum xbtree_key_contig
 510 xfs_bmbt_keys_contiguous(
 511         struct xfs_btree_cur            *cur,
 512         const union xfs_btree_key       *key1,
 513         const union xfs_btree_key       *key2,
 514         const union xfs_btree_key       *mask)
 515 {
 516         ASSERT(!mask || mask->bmbt.br_startoff);
 517
 518         return xbtree_key_contig(be64_to_cpu(key1->bmbt.br_startoff),
 519                                  be64_to_cpu(key2->bmbt.br_startoff));
 520 }
 521
 522 static const struct xfs_btree_ops xfs_bmbt_ops = {
 523         .rec_len                = sizeof(xfs_bmbt_rec_t),
 524         .key_len                = sizeof(xfs_bmbt_key_t),
 525
 526         .dup_cursor             = xfs_bmbt_dup_cursor,
 527         .update_cursor          = xfs_bmbt_update_cursor,
 528         .alloc_block            = xfs_bmbt_alloc_block,
 529         .free_block             = xfs_bmbt_free_block,
 530         .get_maxrecs            = xfs_bmbt_get_maxrecs,
 531         .get_minrecs            = xfs_bmbt_get_minrecs,
 532         .get_dmaxrecs           = xfs_bmbt_get_dmaxrecs,
 533         .init_key_from_rec      = xfs_bmbt_init_key_from_rec,
 534         .init_high_key_from_rec = xfs_bmbt_init_high_key_from_rec,
 535         .init_rec_from_cur      = xfs_bmbt_init_rec_from_cur,
 536         .init_ptr_from_cur      = xfs_bmbt_init_ptr_from_cur,
 537         .key_diff               = xfs_bmbt_key_diff,
 538         .diff_two_keys          = xfs_bmbt_diff_two_keys,
 539         .buf_ops                = &xfs_bmbt_buf_ops,
 540         .keys_inorder           = xfs_bmbt_keys_inorder,
 541         .recs_inorder           = xfs_bmbt_recs_inorder,
 542         .keys_contiguous        = xfs_bmbt_keys_contiguous,
 543 };
 544
 545 /*
 546  * Allocate a new bmap btree cursor.
 547  */
 548 struct xfs_btree_cur *                          /* new bmap btree cursor */
 549 xfs_bmbt_init_cursor(
 550         struct xfs_mount        *mp,            /* file system mount point */
 551         struct xfs_trans        *tp,            /* transaction pointer */
 552         struct xfs_inode        *ip,            /* inode owning the btree */
 553         int                     whichfork)      /* data or attr fork */
 554 {
 555         struct xfs_ifork        *ifp = xfs_ifork_ptr(ip, whichfork);
 556         struct xfs_btree_cur    *cur;
 557         ASSERT(whichfork != XFS_COW_FORK);
 558
 559         cur = xfs_btree_alloc_cursor(mp, tp, XFS_BTNUM_BMAP,
 560                         mp->m_bm_maxlevels[whichfork], xfs_bmbt_cur_cache);
 561         cur->bc_nlevels = be16_to_cpu(ifp->if_broot->bb_level) + 1;
 562         cur->bc_statoff = XFS_STATS_CALC_INDEX(xs_bmbt_2);
 563
 564         cur->bc_ops = &xfs_bmbt_ops;
 565         cur->bc_flags = XFS_BTREE_LONG_PTRS | XFS_BTREE_ROOT_IN_INODE;
 566         if (xfs_has_crc(mp))
 567                 cur->bc_flags |= XFS_BTREE_CRC_BLOCKS;
 568
 569         cur->bc_ino.forksize = xfs_inode_fork_size(ip, whichfork);
 570         cur->bc_ino.ip = ip;
 571         cur->bc_ino.allocated = 0;
 572         cur->bc_ino.flags = 0;
 573         cur->bc_ino.whichfork = whichfork;
 574
 575         return cur;
 576 }
 577
 578 /* Calculate number of records in a block mapping btree block. */
 579 static inline unsigned int
 580 xfs_bmbt_block_maxrecs(
 581         unsigned int            blocklen,
 582         bool                    leaf)
 583 {
 584         if (leaf)
 585                 return blocklen / sizeof(xfs_bmbt_rec_t);
 586         return blocklen / (sizeof(xfs_bmbt_key_t) + sizeof(xfs_bmbt_ptr_t));
 587 }
 588
 589 /*
 590  * Calculate number of records in a bmap btree block.
 591  */
 592 int
 593 xfs_bmbt_maxrecs(
 594         struct xfs_mount        *mp,
 595         int                     blocklen,
 596         int                     leaf)
 597 {
 598         blocklen -= XFS_BMBT_BLOCK_LEN(mp);
 599         return xfs_bmbt_block_maxrecs(blocklen, leaf);
 600 }
 601
 602 /*
 603  * Calculate the maximum possible height of the btree that the on-disk format
 604  * supports. This is used for sizing structures large enough to support every
 605  * possible configuration of a filesystem that might get mounted.
 606  */
 607 unsigned int
 608 xfs_bmbt_maxlevels_ondisk(void)
 609 {
 610         unsigned int            minrecs[2];
 611         unsigned int            blocklen;
 612
 613         blocklen = min(XFS_MIN_BLOCKSIZE - XFS_BTREE_SBLOCK_LEN,
 614                        XFS_MIN_CRC_BLOCKSIZE - XFS_BTREE_SBLOCK_CRC_LEN);
 615
 616         minrecs[0] = xfs_bmbt_block_maxrecs(blocklen, true) / 2;
 617         minrecs[1] = xfs_bmbt_block_maxrecs(blocklen, false) / 2;
 618
 619         /* One extra level for the inode root. */
 620         return xfs_btree_compute_maxlevels(minrecs,
 621                         XFS_MAX_EXTCNT_DATA_FORK_LARGE) + 1;
 622 }
 623
 624 /*
 625  * Calculate number of records in a bmap btree inode root.
 626  */
 627 int
 628 xfs_bmdr_maxrecs(
 629         int                     blocklen,
 630         int                     leaf)
 631 {
 632         blocklen -= sizeof(xfs_bmdr_block_t);
 633
 634         if (leaf)
 635                 return blocklen / sizeof(xfs_bmdr_rec_t);
 636         return blocklen / (sizeof(xfs_bmdr_key_t) + sizeof(xfs_bmdr_ptr_t));
 637 }
 638
 639 /*
 640  * Change the owner of a btree format fork fo the inode passed in. Change it to
 641  * the owner of that is passed in so that we can change owners before or after
 642  * we switch forks between inodes. The operation that the caller is doing will
 643  * determine whether is needs to change owner before or after the switch.
 644  *
 645  * For demand paged transactional modification, the fork switch should be done
 646  * after reading in all the blocks, modifying them and pinning them in the
 647  * transaction. For modification when the buffers are already pinned in memory,
 648  * the fork switch can be done before changing the owner as we won't need to
 649  * validate the owner until the btree buffers are unpinned and writes can occur
 650  * again.
 651  *
 652  * For recovery based ownership change, there is no transactional context and
 653  * so a buffer list must be supplied so that we can record the buffers that we
 654  * modified for the caller to issue IO on.
 655  */
 656 int
 657 xfs_bmbt_change_owner(
 658         struct xfs_trans        *tp,
 659         struct xfs_inode        *ip,
 660         int                     whichfork,
 661         xfs_ino_t               new_owner,
 662         struct list_head        *buffer_list)
 663 {
 664         struct xfs_btree_cur    *cur;
 665         int                     error;
 666
 667         ASSERT(tp || buffer_list);
 668         ASSERT(!(tp && buffer_list));
 669         ASSERT(xfs_ifork_ptr(ip, whichfork)->if_format == XFS_DINODE_FMT_BTREE);
 670
 671         cur = xfs_bmbt_init_cursor(ip->i_mount, tp, ip, whichfork);
 672         cur->bc_ino.flags |= XFS_BTCUR_BMBT_INVALID_OWNER;
 673
 674         error = xfs_btree_change_owner(cur, new_owner, buffer_list);
 675         xfs_btree_del_cursor(cur, error);
 676         return error;
 677 }
 678
 679 /* Calculate the bmap btree size for some records. */
 680 unsigned long long
 681 xfs_bmbt_calc_size(
 682         struct xfs_mount        *mp,
 683         unsigned long long      len)
 684 {
 685         return xfs_btree_calc_size(mp->m_bmap_dmnr, len);
 686 }
 687
 688 int __init
 689 xfs_bmbt_init_cur_cache(void)
 690 {
 691         xfs_bmbt_cur_cache = kmem_cache_create("xfs_bmbt_cur",
 692                         xfs_btree_cur_sizeof(xfs_bmbt_maxlevels_ondisk()),
 693                         0, 0, NULL);
 694
 695         if (!xfs_bmbt_cur_cache)
 696                 return -ENOMEM;
 697         return 0;
 698 }
 699
 700 void
 701 xfs_bmbt_destroy_cur_cache(void)
 702 {
 703         kmem_cache_destroy(xfs_bmbt_cur_cache);
 704         xfs_bmbt_cur_cache = NULL;
 705 }