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_btree_staging.h"
  19 #include "xfs_bmap_btree.h"
  20 #include "xfs_bmap.h"
  21 #include "xfs_error.h"
  22 #include "xfs_quota.h"
  23 #include "xfs_trace.h"
  24 #include "xfs_rmap.h"
  25 #include "xfs_ag.h"
  26
  27 static struct kmem_cache        *xfs_bmbt_cur_cache;
  28
  29 void
  30 xfs_bmbt_init_block(
  31         struct xfs_inode                *ip,
  32         struct xfs_btree_block          *buf,
  33         struct xfs_buf                  *bp,
  34         __u16                           level,
  35         __u16                           numrecs)
  36 {
  37         if (bp)
  38                 xfs_btree_init_buf(ip->i_mount, bp, &xfs_bmbt_ops, level,
  39                                 numrecs, ip->i_ino);
  40         else
  41                 xfs_btree_init_block(ip->i_mount, buf, &xfs_bmbt_ops, level,
  42                                 numrecs, ip->i_ino);
  43 }
  44
  45 /*
  46  * Convert on-disk form of btree root to in-memory form.
  47  */
  48 void
  49 xfs_bmdr_to_bmbt(
  50         struct xfs_inode        *ip,
  51         xfs_bmdr_block_t        *dblock,
  52         int                     dblocklen,
  53         struct xfs_btree_block  *rblock,
  54         int                     rblocklen)
  55 {
  56         struct xfs_mount        *mp = ip->i_mount;
  57         int                     dmxr;
  58         xfs_bmbt_key_t          *fkp;
  59         __be64                  *fpp;
  60         xfs_bmbt_key_t          *tkp;
  61         __be64                  *tpp;
  62
  63         xfs_bmbt_init_block(ip, rblock, NULL, 0, 0);
  64         rblock->bb_level = dblock->bb_level;
  65         ASSERT(be16_to_cpu(rblock->bb_level) > 0);
  66         rblock->bb_numrecs = dblock->bb_numrecs;
  67         dmxr = xfs_bmdr_maxrecs(dblocklen, 0);
  68         fkp = xfs_bmdr_key_addr(dblock, 1);
  69         tkp = xfs_bmbt_key_addr(mp, rblock, 1);
  70         fpp = xfs_bmdr_ptr_addr(dblock, 1, dmxr);
  71         tpp = xfs_bmap_broot_ptr_addr(mp, rblock, 1, rblocklen);
  72         dmxr = be16_to_cpu(dblock->bb_numrecs);
  73         memcpy(tkp, fkp, sizeof(*fkp) * dmxr);
  74         memcpy(tpp, fpp, sizeof(*fpp) * dmxr);
  75 }
  76
  77 void
  78 xfs_bmbt_disk_get_all(
  79         const struct xfs_bmbt_rec *rec,
  80         struct xfs_bmbt_irec    *irec)
  81 {
  82         uint64_t                l0 = get_unaligned_be64(&rec->l0);
  83         uint64_t                l1 = get_unaligned_be64(&rec->l1);
  84
  85         irec->br_startoff = (l0 & xfs_mask64lo(64 - BMBT_EXNTFLAG_BITLEN)) >> 9;
  86         irec->br_startblock = ((l0 & xfs_mask64lo(9)) << 43) | (l1 >> 21);
  87         irec->br_blockcount = l1 & xfs_mask64lo(21);
  88         if (l0 >> (64 - BMBT_EXNTFLAG_BITLEN))
  89                 irec->br_state = XFS_EXT_UNWRITTEN;
  90         else
  91                 irec->br_state = XFS_EXT_NORM;
  92 }
  93
  94 /*
  95  * Extract the blockcount field from an on disk bmap extent record.
  96  */
  97 xfs_filblks_t
  98 xfs_bmbt_disk_get_blockcount(
  99         const struct xfs_bmbt_rec       *r)
 100 {
 101         return (xfs_filblks_t)(be64_to_cpu(r->l1) & xfs_mask64lo(21));
 102 }
 103
 104 /*
 105  * Extract the startoff field from a disk format bmap extent record.
 106  */
 107 xfs_fileoff_t
 108 xfs_bmbt_disk_get_startoff(
 109         const struct xfs_bmbt_rec       *r)
 110 {
 111         return ((xfs_fileoff_t)be64_to_cpu(r->l0) &
 112                  xfs_mask64lo(64 - BMBT_EXNTFLAG_BITLEN)) >> 9;
 113 }
 114
 115 /*
 116  * Set all the fields in a bmap extent record from the uncompressed form.
 117  */
 118 void
 119 xfs_bmbt_disk_set_all(
 120         struct xfs_bmbt_rec     *r,
 121         struct xfs_bmbt_irec    *s)
 122 {
 123         int                     extent_flag = (s->br_state != XFS_EXT_NORM);
 124
 125         ASSERT(s->br_state == XFS_EXT_NORM || s->br_state == XFS_EXT_UNWRITTEN);
 126         ASSERT(!(s->br_startoff & xfs_mask64hi(64-BMBT_STARTOFF_BITLEN)));
 127         ASSERT(!(s->br_blockcount & xfs_mask64hi(64-BMBT_BLOCKCOUNT_BITLEN)));
 128         ASSERT(!(s->br_startblock & xfs_mask64hi(64-BMBT_STARTBLOCK_BITLEN)));
 129
 130         put_unaligned_be64(
 131                 ((xfs_bmbt_rec_base_t)extent_flag << 63) |
 132                  ((xfs_bmbt_rec_base_t)s->br_startoff << 9) |
 133                  ((xfs_bmbt_rec_base_t)s->br_startblock >> 43), &r->l0);
 134         put_unaligned_be64(
 135                 ((xfs_bmbt_rec_base_t)s->br_startblock << 21) |
 136                  ((xfs_bmbt_rec_base_t)s->br_blockcount &
 137                   (xfs_bmbt_rec_base_t)xfs_mask64lo(21)), &r->l1);
 138 }
 139
 140 /*
 141  * Convert in-memory form of btree root to on-disk form.
 142  */
 143 void
 144 xfs_bmbt_to_bmdr(
 145         struct xfs_mount        *mp,
 146         struct xfs_btree_block  *rblock,
 147         int                     rblocklen,
 148         xfs_bmdr_block_t        *dblock,
 149         int                     dblocklen)
 150 {
 151         int                     dmxr;
 152         xfs_bmbt_key_t          *fkp;
 153         __be64                  *fpp;
 154         xfs_bmbt_key_t          *tkp;
 155         __be64                  *tpp;
 156
 157         if (xfs_has_crc(mp)) {
 158                 ASSERT(rblock->bb_magic == cpu_to_be32(XFS_BMAP_CRC_MAGIC));
 159                 ASSERT(uuid_equal(&rblock->bb_u.l.bb_uuid,
 160                        &mp->m_sb.sb_meta_uuid));
 161                 ASSERT(rblock->bb_u.l.bb_blkno ==
 162                        cpu_to_be64(XFS_BUF_DADDR_NULL));
 163         } else
 164                 ASSERT(rblock->bb_magic == cpu_to_be32(XFS_BMAP_MAGIC));
 165         ASSERT(rblock->bb_u.l.bb_leftsib == cpu_to_be64(NULLFSBLOCK));
 166         ASSERT(rblock->bb_u.l.bb_rightsib == cpu_to_be64(NULLFSBLOCK));
 167         ASSERT(rblock->bb_level != 0);
 168         dblock->bb_level = rblock->bb_level;
 169         dblock->bb_numrecs = rblock->bb_numrecs;
 170         dmxr = xfs_bmdr_maxrecs(dblocklen, 0);
 171         fkp = xfs_bmbt_key_addr(mp, rblock, 1);
 172         tkp = xfs_bmdr_key_addr(dblock, 1);
 173         fpp = xfs_bmap_broot_ptr_addr(mp, rblock, 1, rblocklen);
 174         tpp = xfs_bmdr_ptr_addr(dblock, 1, dmxr);
 175         dmxr = be16_to_cpu(dblock->bb_numrecs);
 176         memcpy(tkp, fkp, sizeof(*fkp) * dmxr);
 177         memcpy(tpp, fpp, sizeof(*fpp) * dmxr);
 178 }
 179
 180 STATIC struct xfs_btree_cur *
 181 xfs_bmbt_dup_cursor(
 182         struct xfs_btree_cur    *cur)
 183 {
 184         struct xfs_btree_cur    *new;
 185
 186         new = xfs_bmbt_init_cursor(cur->bc_mp, cur->bc_tp,
 187                         cur->bc_ino.ip, cur->bc_ino.whichfork);
 188         new->bc_flags |= (cur->bc_flags &
 189                 (XFS_BTREE_BMBT_INVALID_OWNER | XFS_BTREE_BMBT_WASDEL));
 190         return new;
 191 }
 192
 193 STATIC void
 194 xfs_bmbt_update_cursor(
 195         struct xfs_btree_cur    *src,
 196         struct xfs_btree_cur    *dst)
 197 {
 198         ASSERT((dst->bc_tp->t_highest_agno != NULLAGNUMBER) ||
 199                (dst->bc_ino.ip->i_diflags & XFS_DIFLAG_REALTIME));
 200
 201         dst->bc_bmap.allocated += src->bc_bmap.allocated;
 202         dst->bc_tp->t_highest_agno = src->bc_tp->t_highest_agno;
 203
 204         src->bc_bmap.allocated = 0;
 205 }
 206
 207 STATIC int
 208 xfs_bmbt_alloc_block(
 209         struct xfs_btree_cur            *cur,
 210         const union xfs_btree_ptr       *start,
 211         union xfs_btree_ptr             *new,
 212         int                             *stat)
 213 {
 214         struct xfs_alloc_arg    args;
 215         int                     error;
 216
 217         memset(&args, 0, sizeof(args));
 218         args.tp = cur->bc_tp;
 219         args.mp = cur->bc_mp;
 220         xfs_rmap_ino_bmbt_owner(&args.oinfo, cur->bc_ino.ip->i_ino,
 221                         cur->bc_ino.whichfork);
 222         args.minlen = args.maxlen = args.prod = 1;
 223         args.wasdel = cur->bc_flags & XFS_BTREE_BMBT_WASDEL;
 224         if (!args.wasdel && args.tp->t_blk_res == 0)
 225                 return -ENOSPC;
 226
 227         /*
 228          * If we are coming here from something like unwritten extent
 229          * conversion, there has been no data extent allocation already done, so
 230          * we have to ensure that we attempt to locate the entire set of bmbt
 231          * allocations in the same AG, as xfs_bmapi_write() would have reserved.
 232          */
 233         if (cur->bc_tp->t_highest_agno == NULLAGNUMBER)
 234                 args.minleft = xfs_bmapi_minleft(cur->bc_tp, cur->bc_ino.ip,
 235                                         cur->bc_ino.whichfork);
 236
 237         error = xfs_alloc_vextent_start_ag(&args, be64_to_cpu(start->l));
 238         if (error)
 239                 return error;
 240
 241         if (args.fsbno == NULLFSBLOCK && args.minleft) {
 242                 /*
 243                  * Could not find an AG with enough free space to satisfy
 244                  * a full btree split.  Try again and if
 245                  * successful activate the lowspace algorithm.
 246                  */
 247                 args.minleft = 0;
 248                 error = xfs_alloc_vextent_start_ag(&args, 0);
 249                 if (error)
 250                         return error;
 251                 cur->bc_tp->t_flags |= XFS_TRANS_LOWMODE;
 252         }
 253         if (WARN_ON_ONCE(args.fsbno == NULLFSBLOCK)) {
 254                 *stat = 0;
 255                 return 0;
 256         }
 257
 258         ASSERT(args.len == 1);
 259         cur->bc_bmap.allocated++;
 260         cur->bc_ino.ip->i_nblocks++;
 261         xfs_trans_log_inode(args.tp, cur->bc_ino.ip, XFS_ILOG_CORE);
 262         xfs_trans_mod_dquot_byino(args.tp, cur->bc_ino.ip,
 263                         XFS_TRANS_DQ_BCOUNT, 1L);
 264
 265         new->l = cpu_to_be64(args.fsbno);
 266
 267         *stat = 1;
 268         return 0;
 269 }
 270
 271 STATIC int
 272 xfs_bmbt_free_block(
 273         struct xfs_btree_cur    *cur,
 274         struct xfs_buf          *bp)
 275 {
 276         struct xfs_mount        *mp = cur->bc_mp;
 277         struct xfs_inode        *ip = cur->bc_ino.ip;
 278         struct xfs_trans        *tp = cur->bc_tp;
 279         xfs_fsblock_t           fsbno = XFS_DADDR_TO_FSB(mp, xfs_buf_daddr(bp));
 280         struct xfs_owner_info   oinfo;
 281         int                     error;
 282
 283         xfs_rmap_ino_bmbt_owner(&oinfo, ip->i_ino, cur->bc_ino.whichfork);
 284         error = xfs_free_extent_later(cur->bc_tp, fsbno, 1, &oinfo,
 285                         XFS_AG_RESV_NONE, 0);
 286         if (error)
 287                 return error;
 288
 289         ip->i_nblocks--;
 290         xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
 291         xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, -1L);
 292         return 0;
 293 }
 294
 295 STATIC int
 296 xfs_bmbt_get_minrecs(
 297         struct xfs_btree_cur    *cur,
 298         int                     level)
 299 {
 300         if (level == cur->bc_nlevels - 1) {
 301                 struct xfs_ifork        *ifp = xfs_btree_ifork_ptr(cur);
 302
 303                 return xfs_bmbt_maxrecs(cur->bc_mp,
 304                                         ifp->if_broot_bytes, level == 0) / 2;
 305         }
 306
 307         return cur->bc_mp->m_bmap_dmnr[level != 0];
 308 }
 309
 310 int
 311 xfs_bmbt_get_maxrecs(
 312         struct xfs_btree_cur    *cur,
 313         int                     level)
 314 {
 315         if (level == cur->bc_nlevels - 1) {
 316                 struct xfs_ifork        *ifp = xfs_btree_ifork_ptr(cur);
 317
 318                 return xfs_bmbt_maxrecs(cur->bc_mp,
 319                                         ifp->if_broot_bytes, level == 0);
 320         }
 321
 322         return cur->bc_mp->m_bmap_dmxr[level != 0];
 323
 324 }
 325
 326 /*
 327  * Get the maximum records we could store in the on-disk format.
 328  *
 329  * For non-root nodes this is equivalent to xfs_bmbt_get_maxrecs, but
 330  * for the root node this checks the available space in the dinode fork
 331  * so that we can resize the in-memory buffer to match it.  After a
 332  * resize to the maximum size this function returns the same value
 333  * as xfs_bmbt_get_maxrecs for the root node, too.
 334  */
 335 STATIC int
 336 xfs_bmbt_get_dmaxrecs(
 337         struct xfs_btree_cur    *cur,
 338         int                     level)
 339 {
 340         if (level != cur->bc_nlevels - 1)
 341                 return cur->bc_mp->m_bmap_dmxr[level != 0];
 342         return xfs_bmdr_maxrecs(cur->bc_ino.forksize, level == 0);
 343 }
 344
 345 STATIC void
 346 xfs_bmbt_init_key_from_rec(
 347         union xfs_btree_key             *key,
 348         const union xfs_btree_rec       *rec)
 349 {
 350         key->bmbt.br_startoff =
 351                 cpu_to_be64(xfs_bmbt_disk_get_startoff(&rec->bmbt));
 352 }
 353
 354 STATIC void
 355 xfs_bmbt_init_high_key_from_rec(
 356         union xfs_btree_key             *key,
 357         const union xfs_btree_rec       *rec)
 358 {
 359         key->bmbt.br_startoff = cpu_to_be64(
 360                         xfs_bmbt_disk_get_startoff(&rec->bmbt) +
 361                         xfs_bmbt_disk_get_blockcount(&rec->bmbt) - 1);
 362 }
 363
 364 STATIC void
 365 xfs_bmbt_init_rec_from_cur(
 366         struct xfs_btree_cur    *cur,
 367         union xfs_btree_rec     *rec)
 368 {
 369         xfs_bmbt_disk_set_all(&rec->bmbt, &cur->bc_rec.b);
 370 }
 371
 372 STATIC int64_t
 373 xfs_bmbt_key_diff(
 374         struct xfs_btree_cur            *cur,
 375         const union xfs_btree_key       *key)
 376 {
 377         return (int64_t)be64_to_cpu(key->bmbt.br_startoff) -
 378                                       cur->bc_rec.b.br_startoff;
 379 }
 380
 381 STATIC int64_t
 382 xfs_bmbt_diff_two_keys(
 383         struct xfs_btree_cur            *cur,
 384         const union xfs_btree_key       *k1,
 385         const union xfs_btree_key       *k2,
 386         const union xfs_btree_key       *mask)
 387 {
 388         uint64_t                        a = be64_to_cpu(k1->bmbt.br_startoff);
 389         uint64_t                        b = be64_to_cpu(k2->bmbt.br_startoff);
 390
 391         ASSERT(!mask || mask->bmbt.br_startoff);
 392
 393         /*
 394          * Note: This routine previously casted a and b to int64 and subtracted
 395          * them to generate a result.  This lead to problems if b was the
 396          * "maximum" key value (all ones) being signed incorrectly, hence this
 397          * somewhat less efficient version.
 398          */
 399         if (a > b)
 400                 return 1;
 401         if (b > a)
 402                 return -1;
 403         return 0;
 404 }
 405
 406 static xfs_failaddr_t
 407 xfs_bmbt_verify(
 408         struct xfs_buf          *bp)
 409 {
 410         struct xfs_mount        *mp = bp->b_mount;
 411         struct xfs_btree_block  *block = XFS_BUF_TO_BLOCK(bp);
 412         xfs_failaddr_t          fa;
 413         unsigned int            level;
 414
 415         if (!xfs_verify_magic(bp, block->bb_magic))
 416                 return __this_address;
 417
 418         if (xfs_has_crc(mp)) {
 419                 /*
 420                  * XXX: need a better way of verifying the owner here. Right now
 421                  * just make sure there has been one set.
 422                  */
 423                 fa = xfs_btree_fsblock_v5hdr_verify(bp, XFS_RMAP_OWN_UNKNOWN);
 424                 if (fa)
 425                         return fa;
 426         }
 427
 428         /*
 429          * numrecs and level verification.
 430          *
 431          * We don't know what fork we belong to, so just verify that the level
 432          * is less than the maximum of the two. Later checks will be more
 433          * precise.
 434          */
 435         level = be16_to_cpu(block->bb_level);
 436         if (level > max(mp->m_bm_maxlevels[0], mp->m_bm_maxlevels[1]))
 437                 return __this_address;
 438
 439         return xfs_btree_fsblock_verify(bp, mp->m_bmap_dmxr[level != 0]);
 440 }
 441
 442 static void
 443 xfs_bmbt_read_verify(
 444         struct xfs_buf  *bp)
 445 {
 446         xfs_failaddr_t  fa;
 447
 448         if (!xfs_btree_fsblock_verify_crc(bp))
 449                 xfs_verifier_error(bp, -EFSBADCRC, __this_address);
 450         else {
 451                 fa = xfs_bmbt_verify(bp);
 452                 if (fa)
 453                         xfs_verifier_error(bp, -EFSCORRUPTED, fa);
 454         }
 455
 456         if (bp->b_error)
 457                 trace_xfs_btree_corrupt(bp, _RET_IP_);
 458 }
 459
 460 static void
 461 xfs_bmbt_write_verify(
 462         struct xfs_buf  *bp)
 463 {
 464         xfs_failaddr_t  fa;
 465
 466         fa = xfs_bmbt_verify(bp);
 467         if (fa) {
 468                 trace_xfs_btree_corrupt(bp, _RET_IP_);
 469                 xfs_verifier_error(bp, -EFSCORRUPTED, fa);
 470                 return;
 471         }
 472         xfs_btree_fsblock_calc_crc(bp);
 473 }
 474
 475 const struct xfs_buf_ops xfs_bmbt_buf_ops = {
 476         .name = "xfs_bmbt",
 477         .magic = { cpu_to_be32(XFS_BMAP_MAGIC),
 478                    cpu_to_be32(XFS_BMAP_CRC_MAGIC) },
 479         .verify_read = xfs_bmbt_read_verify,
 480         .verify_write = xfs_bmbt_write_verify,
 481         .verify_struct = xfs_bmbt_verify,
 482 };
 483
 484
 485 STATIC int
 486 xfs_bmbt_keys_inorder(
 487         struct xfs_btree_cur            *cur,
 488         const union xfs_btree_key       *k1,
 489         const union xfs_btree_key       *k2)
 490 {
 491         return be64_to_cpu(k1->bmbt.br_startoff) <
 492                 be64_to_cpu(k2->bmbt.br_startoff);
 493 }
 494
 495 STATIC int
 496 xfs_bmbt_recs_inorder(
 497         struct xfs_btree_cur            *cur,
 498         const union xfs_btree_rec       *r1,
 499         const union xfs_btree_rec       *r2)
 500 {
 501         return xfs_bmbt_disk_get_startoff(&r1->bmbt) +
 502                 xfs_bmbt_disk_get_blockcount(&r1->bmbt) <=
 503                 xfs_bmbt_disk_get_startoff(&r2->bmbt);
 504 }
 505
 506 STATIC enum xbtree_key_contig
 507 xfs_bmbt_keys_contiguous(
 508         struct xfs_btree_cur            *cur,
 509         const union xfs_btree_key       *key1,
 510         const union xfs_btree_key       *key2,
 511         const union xfs_btree_key       *mask)
 512 {
 513         ASSERT(!mask || mask->bmbt.br_startoff);
 514
 515         return xbtree_key_contig(be64_to_cpu(key1->bmbt.br_startoff),
 516                                  be64_to_cpu(key2->bmbt.br_startoff));
 517 }
 518
 519 const struct xfs_btree_ops xfs_bmbt_ops = {
 520         .name                   = "bmap",
 521         .type                   = XFS_BTREE_TYPE_INODE,
 522
 523         .rec_len                = sizeof(xfs_bmbt_rec_t),
 524         .key_len                = sizeof(xfs_bmbt_key_t),
 525         .ptr_len                = XFS_BTREE_LONG_PTR_LEN,
 526
 527         .lru_refs               = XFS_BMAP_BTREE_REF,
 528         .statoff                = XFS_STATS_CALC_INDEX(xs_bmbt_2),
 529
 530         .dup_cursor             = xfs_bmbt_dup_cursor,
 531         .update_cursor          = xfs_bmbt_update_cursor,
 532         .alloc_block            = xfs_bmbt_alloc_block,
 533         .free_block             = xfs_bmbt_free_block,
 534         .get_maxrecs            = xfs_bmbt_get_maxrecs,
 535         .get_minrecs            = xfs_bmbt_get_minrecs,
 536         .get_dmaxrecs           = xfs_bmbt_get_dmaxrecs,
 537         .init_key_from_rec      = xfs_bmbt_init_key_from_rec,
 538         .init_high_key_from_rec = xfs_bmbt_init_high_key_from_rec,
 539         .init_rec_from_cur      = xfs_bmbt_init_rec_from_cur,
 540         .key_diff               = xfs_bmbt_key_diff,
 541         .diff_two_keys          = xfs_bmbt_diff_two_keys,
 542         .buf_ops                = &xfs_bmbt_buf_ops,
 543         .keys_inorder           = xfs_bmbt_keys_inorder,
 544         .recs_inorder           = xfs_bmbt_recs_inorder,
 545         .keys_contiguous        = xfs_bmbt_keys_contiguous,
 546 };
 547
 548 /*
 549  * Create a new bmap btree cursor.
 550  *
 551  * For staging cursors -1 in passed in whichfork.
 552  */
 553 struct xfs_btree_cur *
 554 xfs_bmbt_init_cursor(
 555         struct xfs_mount        *mp,
 556         struct xfs_trans        *tp,
 557         struct xfs_inode        *ip,
 558         int                     whichfork)
 559 {
 560         struct xfs_btree_cur    *cur;
 561         unsigned int            maxlevels;
 562
 563         ASSERT(whichfork != XFS_COW_FORK);
 564
 565         /*
 566          * The Data fork always has larger maxlevel, so use that for staging
 567          * cursors.
 568          */
 569         switch (whichfork) {
 570         case XFS_STAGING_FORK:
 571                 maxlevels = mp->m_bm_maxlevels[XFS_DATA_FORK];
 572                 break;
 573         default:
 574                 maxlevels = mp->m_bm_maxlevels[whichfork];
 575                 break;
 576         }
 577         cur = xfs_btree_alloc_cursor(mp, tp, &xfs_bmbt_ops, maxlevels,
 578                         xfs_bmbt_cur_cache);
 579         cur->bc_ino.ip = ip;
 580         cur->bc_ino.whichfork = whichfork;
 581         cur->bc_bmap.allocated = 0;
 582         if (whichfork != XFS_STAGING_FORK) {
 583                 struct xfs_ifork        *ifp = xfs_ifork_ptr(ip, whichfork);
 584
 585                 cur->bc_nlevels = be16_to_cpu(ifp->if_broot->bb_level) + 1;
 586                 cur->bc_ino.forksize = xfs_inode_fork_size(ip, whichfork);
 587         }
 588         return cur;
 589 }
 590
 591 /* Calculate number of records in a block mapping btree block. */
 592 static inline unsigned int
 593 xfs_bmbt_block_maxrecs(
 594         unsigned int            blocklen,
 595         bool                    leaf)
 596 {
 597         if (leaf)
 598                 return blocklen / sizeof(xfs_bmbt_rec_t);
 599         return blocklen / (sizeof(xfs_bmbt_key_t) + sizeof(xfs_bmbt_ptr_t));
 600 }
 601
 602 /*
 603  * Swap in the new inode fork root.  Once we pass this point the newly rebuilt
 604  * mappings are in place and we have to kill off any old btree blocks.
 605  */
 606 void
 607 xfs_bmbt_commit_staged_btree(
 608         struct xfs_btree_cur    *cur,
 609         struct xfs_trans        *tp,
 610         int                     whichfork)
 611 {
 612         struct xbtree_ifakeroot *ifake = cur->bc_ino.ifake;
 613         struct xfs_ifork        *ifp;
 614         static const short      brootflag[2] = {XFS_ILOG_DBROOT, XFS_ILOG_ABROOT};
 615         static const short      extflag[2] = {XFS_ILOG_DEXT, XFS_ILOG_AEXT};
 616         int                     flags = XFS_ILOG_CORE;
 617
 618         ASSERT(cur->bc_flags & XFS_BTREE_STAGING);
 619         ASSERT(whichfork != XFS_COW_FORK);
 620
 621         /*
 622          * Free any resources hanging off the real fork, then shallow-copy the
 623          * staging fork's contents into the real fork to transfer everything
 624          * we just built.
 625          */
 626         ifp = xfs_ifork_ptr(cur->bc_ino.ip, whichfork);
 627         xfs_idestroy_fork(ifp);
 628         memcpy(ifp, ifake->if_fork, sizeof(struct xfs_ifork));
 629
 630         switch (ifp->if_format) {
 631         case XFS_DINODE_FMT_EXTENTS:
 632                 flags |= extflag[whichfork];
 633                 break;
 634         case XFS_DINODE_FMT_BTREE:
 635                 flags |= brootflag[whichfork];
 636                 break;
 637         default:
 638                 ASSERT(0);
 639                 break;
 640         }
 641         xfs_trans_log_inode(tp, cur->bc_ino.ip, flags);
 642         xfs_btree_commit_ifakeroot(cur, tp, whichfork);
 643 }
 644
 645 /*
 646  * Calculate number of records in a bmap btree block.
 647  */
 648 unsigned int
 649 xfs_bmbt_maxrecs(
 650         struct xfs_mount        *mp,
 651         unsigned int            blocklen,
 652         bool                    leaf)
 653 {
 654         blocklen -= xfs_bmbt_block_len(mp);
 655         return xfs_bmbt_block_maxrecs(blocklen, leaf);
 656 }
 657
 658 /*
 659  * Calculate the maximum possible height of the btree that the on-disk format
 660  * supports. This is used for sizing structures large enough to support every
 661  * possible configuration of a filesystem that might get mounted.
 662  */
 663 unsigned int
 664 xfs_bmbt_maxlevels_ondisk(void)
 665 {
 666         unsigned int            minrecs[2];
 667         unsigned int            blocklen;
 668
 669         blocklen = min(XFS_MIN_BLOCKSIZE - XFS_BTREE_SBLOCK_LEN,
 670                        XFS_MIN_CRC_BLOCKSIZE - XFS_BTREE_SBLOCK_CRC_LEN);
 671
 672         minrecs[0] = xfs_bmbt_block_maxrecs(blocklen, true) / 2;
 673         minrecs[1] = xfs_bmbt_block_maxrecs(blocklen, false) / 2;
 674
 675         /* One extra level for the inode root. */
 676         return xfs_btree_compute_maxlevels(minrecs,
 677                         XFS_MAX_EXTCNT_DATA_FORK_LARGE) + 1;
 678 }
 679
 680 /*
 681  * Calculate number of records in a bmap btree inode root.
 682  */
 683 int
 684 xfs_bmdr_maxrecs(
 685         int                     blocklen,
 686         int                     leaf)
 687 {
 688         blocklen -= sizeof(xfs_bmdr_block_t);
 689
 690         if (leaf)
 691                 return blocklen / sizeof(xfs_bmdr_rec_t);
 692         return blocklen / (sizeof(xfs_bmdr_key_t) + sizeof(xfs_bmdr_ptr_t));
 693 }
 694
 695 /*
 696  * Change the owner of a btree format fork fo the inode passed in. Change it to
 697  * the owner of that is passed in so that we can change owners before or after
 698  * we switch forks between inodes. The operation that the caller is doing will
 699  * determine whether is needs to change owner before or after the switch.
 700  *
 701  * For demand paged transactional modification, the fork switch should be done
 702  * after reading in all the blocks, modifying them and pinning them in the
 703  * transaction. For modification when the buffers are already pinned in memory,
 704  * the fork switch can be done before changing the owner as we won't need to
 705  * validate the owner until the btree buffers are unpinned and writes can occur
 706  * again.
 707  *
 708  * For recovery based ownership change, there is no transactional context and
 709  * so a buffer list must be supplied so that we can record the buffers that we
 710  * modified for the caller to issue IO on.
 711  */
 712 int
 713 xfs_bmbt_change_owner(
 714         struct xfs_trans        *tp,
 715         struct xfs_inode        *ip,
 716         int                     whichfork,
 717         xfs_ino_t               new_owner,
 718         struct list_head        *buffer_list)
 719 {
 720         struct xfs_btree_cur    *cur;
 721         int                     error;
 722
 723         ASSERT(tp || buffer_list);
 724         ASSERT(!(tp && buffer_list));
 725         ASSERT(xfs_ifork_ptr(ip, whichfork)->if_format == XFS_DINODE_FMT_BTREE);
 726
 727         cur = xfs_bmbt_init_cursor(ip->i_mount, tp, ip, whichfork);
 728         cur->bc_flags |= XFS_BTREE_BMBT_INVALID_OWNER;
 729
 730         error = xfs_btree_change_owner(cur, new_owner, buffer_list);
 731         xfs_btree_del_cursor(cur, error);
 732         return error;
 733 }
 734
 735 /* Calculate the bmap btree size for some records. */
 736 unsigned long long
 737 xfs_bmbt_calc_size(
 738         struct xfs_mount        *mp,
 739         unsigned long long      len)
 740 {
 741         return xfs_btree_calc_size(mp->m_bmap_dmnr, len);
 742 }
 743
 744 int __init
 745 xfs_bmbt_init_cur_cache(void)
 746 {
 747         xfs_bmbt_cur_cache = kmem_cache_create("xfs_bmbt_cur",
 748                         xfs_btree_cur_sizeof(xfs_bmbt_maxlevels_ondisk()),
 749                         0, 0, NULL);
 750
 751         if (!xfs_bmbt_cur_cache)
 752                 return -ENOMEM;
 753         return 0;
 754 }
 755
 756 void
 757 xfs_bmbt_destroy_cur_cache(void)
 758 {
 759         kmem_cache_destroy(xfs_bmbt_cur_cache);
 760         xfs_bmbt_cur_cache = NULL;
 761 }