fs/xfs/libxfs/xfs_refcount_btree.c

   1 // SPDX-License-Identifier: GPL-2.0+
   2 /*
   3  * Copyright (C) 2016 Oracle.  All Rights Reserved.
   4  * Author: Darrick J. Wong <darrick.wong@oracle.com>
   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_mount.h"
  13 #include "xfs_btree.h"
  14 #include "xfs_btree_staging.h"
  15 #include "xfs_refcount_btree.h"
  16 #include "xfs_refcount.h"
  17 #include "xfs_alloc.h"
  18 #include "xfs_error.h"
  19 #include "xfs_health.h"
  20 #include "xfs_trace.h"
  21 #include "xfs_trans.h"
  22 #include "xfs_bit.h"
  23 #include "xfs_rmap.h"
  24 #include "xfs_ag.h"
  25
  26 static struct kmem_cache        *xfs_refcountbt_cur_cache;
  27
  28 static struct xfs_btree_cur *
  29 xfs_refcountbt_dup_cursor(
  30         struct xfs_btree_cur    *cur)
  31 {
  32         return xfs_refcountbt_init_cursor(cur->bc_mp, cur->bc_tp,
  33                         cur->bc_ag.agbp, to_perag(cur->bc_group));
  34 }
  35
  36 STATIC void
  37 xfs_refcountbt_set_root(
  38         struct xfs_btree_cur            *cur,
  39         const union xfs_btree_ptr       *ptr,
  40         int                             inc)
  41 {
  42         struct xfs_buf          *agbp = cur->bc_ag.agbp;
  43         struct xfs_agf          *agf = agbp->b_addr;
  44         struct xfs_perag        *pag = agbp->b_pag;
  45
  46         ASSERT(ptr->s != 0);
  47
  48         agf->agf_refcount_root = ptr->s;
  49         be32_add_cpu(&agf->agf_refcount_level, inc);
  50         pag->pagf_refcount_level += inc;
  51
  52         xfs_alloc_log_agf(cur->bc_tp, agbp,
  53                         XFS_AGF_REFCOUNT_ROOT | XFS_AGF_REFCOUNT_LEVEL);
  54 }
  55
  56 STATIC int
  57 xfs_refcountbt_alloc_block(
  58         struct xfs_btree_cur            *cur,
  59         const union xfs_btree_ptr       *start,
  60         union xfs_btree_ptr             *new,
  61         int                             *stat)
  62 {
  63         struct xfs_buf          *agbp = cur->bc_ag.agbp;
  64         struct xfs_agf          *agf = agbp->b_addr;
  65         struct xfs_alloc_arg    args;           /* block allocation args */
  66         int                     error;          /* error return value */
  67
  68         memset(&args, 0, sizeof(args));
  69         args.tp = cur->bc_tp;
  70         args.mp = cur->bc_mp;
  71         args.pag = to_perag(cur->bc_group);
  72         args.oinfo = XFS_RMAP_OINFO_REFC;
  73         args.minlen = args.maxlen = args.prod = 1;
  74         args.resv = XFS_AG_RESV_METADATA;
  75
  76         error = xfs_alloc_vextent_near_bno(&args,
  77                         xfs_agbno_to_fsb(args.pag, xfs_refc_block(args.mp)));
  78         if (error)
  79                 goto out_error;
  80         if (args.fsbno == NULLFSBLOCK) {
  81                 *stat = 0;
  82                 return 0;
  83         }
  84         ASSERT(args.agno == cur->bc_group->xg_gno);
  85         ASSERT(args.len == 1);
  86
  87         new->s = cpu_to_be32(args.agbno);
  88         be32_add_cpu(&agf->agf_refcount_blocks, 1);
  89         xfs_alloc_log_agf(cur->bc_tp, agbp, XFS_AGF_REFCOUNT_BLOCKS);
  90
  91         *stat = 1;
  92         return 0;
  93
  94 out_error:
  95         return error;
  96 }
  97
  98 STATIC int
  99 xfs_refcountbt_free_block(
 100         struct xfs_btree_cur    *cur,
 101         struct xfs_buf          *bp)
 102 {
 103         struct xfs_mount        *mp = cur->bc_mp;
 104         struct xfs_buf          *agbp = cur->bc_ag.agbp;
 105         struct xfs_agf          *agf = agbp->b_addr;
 106         xfs_fsblock_t           fsbno = XFS_DADDR_TO_FSB(mp, xfs_buf_daddr(bp));
 107
 108         be32_add_cpu(&agf->agf_refcount_blocks, -1);
 109         xfs_alloc_log_agf(cur->bc_tp, agbp, XFS_AGF_REFCOUNT_BLOCKS);
 110         return xfs_free_extent_later(cur->bc_tp, fsbno, 1,
 111                         &XFS_RMAP_OINFO_REFC, XFS_AG_RESV_METADATA, 0);
 112 }
 113
 114 STATIC int
 115 xfs_refcountbt_get_minrecs(
 116         struct xfs_btree_cur    *cur,
 117         int                     level)
 118 {
 119         return cur->bc_mp->m_refc_mnr[level != 0];
 120 }
 121
 122 STATIC int
 123 xfs_refcountbt_get_maxrecs(
 124         struct xfs_btree_cur    *cur,
 125         int                     level)
 126 {
 127         return cur->bc_mp->m_refc_mxr[level != 0];
 128 }
 129
 130 STATIC void
 131 xfs_refcountbt_init_key_from_rec(
 132         union xfs_btree_key             *key,
 133         const union xfs_btree_rec       *rec)
 134 {
 135         key->refc.rc_startblock = rec->refc.rc_startblock;
 136 }
 137
 138 STATIC void
 139 xfs_refcountbt_init_high_key_from_rec(
 140         union xfs_btree_key             *key,
 141         const union xfs_btree_rec       *rec)
 142 {
 143         __u32                           x;
 144
 145         x = be32_to_cpu(rec->refc.rc_startblock);
 146         x += be32_to_cpu(rec->refc.rc_blockcount) - 1;
 147         key->refc.rc_startblock = cpu_to_be32(x);
 148 }
 149
 150 STATIC void
 151 xfs_refcountbt_init_rec_from_cur(
 152         struct xfs_btree_cur    *cur,
 153         union xfs_btree_rec     *rec)
 154 {
 155         const struct xfs_refcount_irec *irec = &cur->bc_rec.rc;
 156         uint32_t                start;
 157
 158         start = xfs_refcount_encode_startblock(irec->rc_startblock,
 159                         irec->rc_domain);
 160         rec->refc.rc_startblock = cpu_to_be32(start);
 161         rec->refc.rc_blockcount = cpu_to_be32(cur->bc_rec.rc.rc_blockcount);
 162         rec->refc.rc_refcount = cpu_to_be32(cur->bc_rec.rc.rc_refcount);
 163 }
 164
 165 STATIC void
 166 xfs_refcountbt_init_ptr_from_cur(
 167         struct xfs_btree_cur    *cur,
 168         union xfs_btree_ptr     *ptr)
 169 {
 170         struct xfs_agf          *agf = cur->bc_ag.agbp->b_addr;
 171
 172         ASSERT(cur->bc_group->xg_gno == be32_to_cpu(agf->agf_seqno));
 173
 174         ptr->s = agf->agf_refcount_root;
 175 }
 176
 177 STATIC int64_t
 178 xfs_refcountbt_key_diff(
 179         struct xfs_btree_cur            *cur,
 180         const union xfs_btree_key       *key)
 181 {
 182         const struct xfs_refcount_key   *kp = &key->refc;
 183         const struct xfs_refcount_irec  *irec = &cur->bc_rec.rc;
 184         uint32_t                        start;
 185
 186         start = xfs_refcount_encode_startblock(irec->rc_startblock,
 187                         irec->rc_domain);
 188         return (int64_t)be32_to_cpu(kp->rc_startblock) - start;
 189 }
 190
 191 STATIC int64_t
 192 xfs_refcountbt_diff_two_keys(
 193         struct xfs_btree_cur            *cur,
 194         const union xfs_btree_key       *k1,
 195         const union xfs_btree_key       *k2,
 196         const union xfs_btree_key       *mask)
 197 {
 198         ASSERT(!mask || mask->refc.rc_startblock);
 199
 200         return (int64_t)be32_to_cpu(k1->refc.rc_startblock) -
 201                         be32_to_cpu(k2->refc.rc_startblock);
 202 }
 203
 204 STATIC xfs_failaddr_t
 205 xfs_refcountbt_verify(
 206         struct xfs_buf          *bp)
 207 {
 208         struct xfs_mount        *mp = bp->b_mount;
 209         struct xfs_btree_block  *block = XFS_BUF_TO_BLOCK(bp);
 210         struct xfs_perag        *pag = bp->b_pag;
 211         xfs_failaddr_t          fa;
 212         unsigned int            level;
 213
 214         if (!xfs_verify_magic(bp, block->bb_magic))
 215                 return __this_address;
 216
 217         if (!xfs_has_reflink(mp))
 218                 return __this_address;
 219         fa = xfs_btree_agblock_v5hdr_verify(bp);
 220         if (fa)
 221                 return fa;
 222
 223         level = be16_to_cpu(block->bb_level);
 224         if (pag && xfs_perag_initialised_agf(pag)) {
 225                 unsigned int    maxlevel = pag->pagf_refcount_level;
 226
 227 #ifdef CONFIG_XFS_ONLINE_REPAIR
 228                 /*
 229                  * Online repair could be rewriting the refcount btree, so
 230                  * we'll validate against the larger of either tree while this
 231                  * is going on.
 232                  */
 233                 maxlevel = max_t(unsigned int, maxlevel,
 234                                 pag->pagf_repair_refcount_level);
 235 #endif
 236                 if (level >= maxlevel)
 237                         return __this_address;
 238         } else if (level >= mp->m_refc_maxlevels)
 239                 return __this_address;
 240
 241         return xfs_btree_agblock_verify(bp, mp->m_refc_mxr[level != 0]);
 242 }
 243
 244 STATIC void
 245 xfs_refcountbt_read_verify(
 246         struct xfs_buf  *bp)
 247 {
 248         xfs_failaddr_t  fa;
 249
 250         if (!xfs_btree_agblock_verify_crc(bp))
 251                 xfs_verifier_error(bp, -EFSBADCRC, __this_address);
 252         else {
 253                 fa = xfs_refcountbt_verify(bp);
 254                 if (fa)
 255                         xfs_verifier_error(bp, -EFSCORRUPTED, fa);
 256         }
 257
 258         if (bp->b_error)
 259                 trace_xfs_btree_corrupt(bp, _RET_IP_);
 260 }
 261
 262 STATIC void
 263 xfs_refcountbt_write_verify(
 264         struct xfs_buf  *bp)
 265 {
 266         xfs_failaddr_t  fa;
 267
 268         fa = xfs_refcountbt_verify(bp);
 269         if (fa) {
 270                 trace_xfs_btree_corrupt(bp, _RET_IP_);
 271                 xfs_verifier_error(bp, -EFSCORRUPTED, fa);
 272                 return;
 273         }
 274         xfs_btree_agblock_calc_crc(bp);
 275
 276 }
 277
 278 const struct xfs_buf_ops xfs_refcountbt_buf_ops = {
 279         .name                   = "xfs_refcountbt",
 280         .magic                  = { 0, cpu_to_be32(XFS_REFC_CRC_MAGIC) },
 281         .verify_read            = xfs_refcountbt_read_verify,
 282         .verify_write           = xfs_refcountbt_write_verify,
 283         .verify_struct          = xfs_refcountbt_verify,
 284 };
 285
 286 STATIC int
 287 xfs_refcountbt_keys_inorder(
 288         struct xfs_btree_cur            *cur,
 289         const union xfs_btree_key       *k1,
 290         const union xfs_btree_key       *k2)
 291 {
 292         return be32_to_cpu(k1->refc.rc_startblock) <
 293                be32_to_cpu(k2->refc.rc_startblock);
 294 }
 295
 296 STATIC int
 297 xfs_refcountbt_recs_inorder(
 298         struct xfs_btree_cur            *cur,
 299         const union xfs_btree_rec       *r1,
 300         const union xfs_btree_rec       *r2)
 301 {
 302         return  be32_to_cpu(r1->refc.rc_startblock) +
 303                 be32_to_cpu(r1->refc.rc_blockcount) <=
 304                 be32_to_cpu(r2->refc.rc_startblock);
 305 }
 306
 307 STATIC enum xbtree_key_contig
 308 xfs_refcountbt_keys_contiguous(
 309         struct xfs_btree_cur            *cur,
 310         const union xfs_btree_key       *key1,
 311         const union xfs_btree_key       *key2,
 312         const union xfs_btree_key       *mask)
 313 {
 314         ASSERT(!mask || mask->refc.rc_startblock);
 315
 316         return xbtree_key_contig(be32_to_cpu(key1->refc.rc_startblock),
 317                                  be32_to_cpu(key2->refc.rc_startblock));
 318 }
 319
 320 const struct xfs_btree_ops xfs_refcountbt_ops = {
 321         .name                   = "refcount",
 322         .type                   = XFS_BTREE_TYPE_AG,
 323
 324         .rec_len                = sizeof(struct xfs_refcount_rec),
 325         .key_len                = sizeof(struct xfs_refcount_key),
 326         .ptr_len                = XFS_BTREE_SHORT_PTR_LEN,
 327
 328         .lru_refs               = XFS_REFC_BTREE_REF,
 329         .statoff                = XFS_STATS_CALC_INDEX(xs_refcbt_2),
 330         .sick_mask              = XFS_SICK_AG_REFCNTBT,
 331
 332         .dup_cursor             = xfs_refcountbt_dup_cursor,
 333         .set_root               = xfs_refcountbt_set_root,
 334         .alloc_block            = xfs_refcountbt_alloc_block,
 335         .free_block             = xfs_refcountbt_free_block,
 336         .get_minrecs            = xfs_refcountbt_get_minrecs,
 337         .get_maxrecs            = xfs_refcountbt_get_maxrecs,
 338         .init_key_from_rec      = xfs_refcountbt_init_key_from_rec,
 339         .init_high_key_from_rec = xfs_refcountbt_init_high_key_from_rec,
 340         .init_rec_from_cur      = xfs_refcountbt_init_rec_from_cur,
 341         .init_ptr_from_cur      = xfs_refcountbt_init_ptr_from_cur,
 342         .key_diff               = xfs_refcountbt_key_diff,
 343         .buf_ops                = &xfs_refcountbt_buf_ops,
 344         .diff_two_keys          = xfs_refcountbt_diff_two_keys,
 345         .keys_inorder           = xfs_refcountbt_keys_inorder,
 346         .recs_inorder           = xfs_refcountbt_recs_inorder,
 347         .keys_contiguous        = xfs_refcountbt_keys_contiguous,
 348 };
 349
 350 /*
 351  * Create a new refcount btree cursor.
 352  *
 353  * For staging cursors tp and agbp are NULL.
 354  */
 355 struct xfs_btree_cur *
 356 xfs_refcountbt_init_cursor(
 357         struct xfs_mount        *mp,
 358         struct xfs_trans        *tp,
 359         struct xfs_buf          *agbp,
 360         struct xfs_perag        *pag)
 361 {
 362         struct xfs_btree_cur    *cur;
 363
 364         ASSERT(pag_agno(pag) < mp->m_sb.sb_agcount);
 365
 366         cur = xfs_btree_alloc_cursor(mp, tp, &xfs_refcountbt_ops,
 367                         mp->m_refc_maxlevels, xfs_refcountbt_cur_cache);
 368         cur->bc_group = xfs_group_hold(pag_group(pag));
 369         cur->bc_refc.nr_ops = 0;
 370         cur->bc_refc.shape_changes = 0;
 371         cur->bc_ag.agbp = agbp;
 372         if (agbp) {
 373                 struct xfs_agf          *agf = agbp->b_addr;
 374
 375                 cur->bc_nlevels = be32_to_cpu(agf->agf_refcount_level);
 376         }
 377         return cur;
 378 }
 379
 380 /*
 381  * Swap in the new btree root.  Once we pass this point the newly rebuilt btree
 382  * is in place and we have to kill off all the old btree blocks.
 383  */
 384 void
 385 xfs_refcountbt_commit_staged_btree(
 386         struct xfs_btree_cur    *cur,
 387         struct xfs_trans        *tp,
 388         struct xfs_buf          *agbp)
 389 {
 390         struct xfs_agf          *agf = agbp->b_addr;
 391         struct xbtree_afakeroot *afake = cur->bc_ag.afake;
 392
 393         ASSERT(cur->bc_flags & XFS_BTREE_STAGING);
 394
 395         agf->agf_refcount_root = cpu_to_be32(afake->af_root);
 396         agf->agf_refcount_level = cpu_to_be32(afake->af_levels);
 397         agf->agf_refcount_blocks = cpu_to_be32(afake->af_blocks);
 398         xfs_alloc_log_agf(tp, agbp, XFS_AGF_REFCOUNT_BLOCKS |
 399                                     XFS_AGF_REFCOUNT_ROOT |
 400                                     XFS_AGF_REFCOUNT_LEVEL);
 401         xfs_btree_commit_afakeroot(cur, tp, agbp);
 402 }
 403
 404 /* Calculate number of records in a refcount btree block. */
 405 static inline unsigned int
 406 xfs_refcountbt_block_maxrecs(
 407         unsigned int            blocklen,
 408         bool                    leaf)
 409 {
 410         if (leaf)
 411                 return blocklen / sizeof(struct xfs_refcount_rec);
 412         return blocklen / (sizeof(struct xfs_refcount_key) +
 413                            sizeof(xfs_refcount_ptr_t));
 414 }
 415
 416 /*
 417  * Calculate the number of records in a refcount btree block.
 418  */
 419 unsigned int
 420 xfs_refcountbt_maxrecs(
 421         struct xfs_mount        *mp,
 422         unsigned int            blocklen,
 423         bool                    leaf)
 424 {
 425         blocklen -= XFS_REFCOUNT_BLOCK_LEN;
 426         return xfs_refcountbt_block_maxrecs(blocklen, leaf);
 427 }
 428
 429 /* Compute the max possible height of the maximally sized refcount btree. */
 430 unsigned int
 431 xfs_refcountbt_maxlevels_ondisk(void)
 432 {
 433         unsigned int            minrecs[2];
 434         unsigned int            blocklen;
 435
 436         blocklen = XFS_MIN_CRC_BLOCKSIZE - XFS_BTREE_SBLOCK_CRC_LEN;
 437
 438         minrecs[0] = xfs_refcountbt_block_maxrecs(blocklen, true) / 2;
 439         minrecs[1] = xfs_refcountbt_block_maxrecs(blocklen, false) / 2;
 440
 441         return xfs_btree_compute_maxlevels(minrecs, XFS_MAX_CRC_AG_BLOCKS);
 442 }
 443
 444 /* Compute the maximum height of a refcount btree. */
 445 void
 446 xfs_refcountbt_compute_maxlevels(
 447         struct xfs_mount                *mp)
 448 {
 449         if (!xfs_has_reflink(mp)) {
 450                 mp->m_refc_maxlevels = 0;
 451                 return;
 452         }
 453
 454         mp->m_refc_maxlevels = xfs_btree_compute_maxlevels(
 455                         mp->m_refc_mnr, mp->m_sb.sb_agblocks);
 456         ASSERT(mp->m_refc_maxlevels <= xfs_refcountbt_maxlevels_ondisk());
 457 }
 458
 459 /* Calculate the refcount btree size for some records. */
 460 xfs_extlen_t
 461 xfs_refcountbt_calc_size(
 462         struct xfs_mount        *mp,
 463         unsigned long long      len)
 464 {
 465         return xfs_btree_calc_size(mp->m_refc_mnr, len);
 466 }
 467
 468 /*
 469  * Calculate the maximum refcount btree size.
 470  */
 471 xfs_extlen_t
 472 xfs_refcountbt_max_size(
 473         struct xfs_mount        *mp,
 474         xfs_agblock_t           agblocks)
 475 {
 476         /* Bail out if we're uninitialized, which can happen in mkfs. */
 477         if (mp->m_refc_mxr[0] == 0)
 478                 return 0;
 479
 480         return xfs_refcountbt_calc_size(mp, agblocks);
 481 }
 482
 483 /*
 484  * Figure out how many blocks to reserve and how many are used by this btree.
 485  */
 486 int
 487 xfs_refcountbt_calc_reserves(
 488         struct xfs_mount        *mp,
 489         struct xfs_trans        *tp,
 490         struct xfs_perag        *pag,
 491         xfs_extlen_t            *ask,
 492         xfs_extlen_t            *used)
 493 {
 494         struct xfs_buf          *agbp;
 495         struct xfs_agf          *agf;
 496         xfs_agblock_t           agblocks;
 497         xfs_extlen_t            tree_len;
 498         int                     error;
 499
 500         if (!xfs_has_reflink(mp))
 501                 return 0;
 502
 503         error = xfs_alloc_read_agf(pag, tp, 0, &agbp);
 504         if (error)
 505                 return error;
 506
 507         agf = agbp->b_addr;
 508         agblocks = be32_to_cpu(agf->agf_length);
 509         tree_len = be32_to_cpu(agf->agf_refcount_blocks);
 510         xfs_trans_brelse(tp, agbp);
 511
 512         /*
 513          * The log is permanently allocated, so the space it occupies will
 514          * never be available for the kinds of things that would require btree
 515          * expansion.  We therefore can pretend the space isn't there.
 516          */
 517         if (xfs_ag_contains_log(mp, pag_agno(pag)))
 518                 agblocks -= mp->m_sb.sb_logblocks;
 519
 520         *ask += xfs_refcountbt_max_size(mp, agblocks);
 521         *used += tree_len;
 522
 523         return error;
 524 }
 525
 526 int __init
 527 xfs_refcountbt_init_cur_cache(void)
 528 {
 529         xfs_refcountbt_cur_cache = kmem_cache_create("xfs_refcbt_cur",
 530                         xfs_btree_cur_sizeof(xfs_refcountbt_maxlevels_ondisk()),
 531                         0, 0, NULL);
 532
 533         if (!xfs_refcountbt_cur_cache)
 534                 return -ENOMEM;
 535         return 0;
 536 }
 537
 538 void
 539 xfs_refcountbt_destroy_cur_cache(void)
 540 {
 541         kmem_cache_destroy(xfs_refcountbt_cur_cache);
 542         xfs_refcountbt_cur_cache = NULL;
 543 }