libxfs/xfs_rmap_btree.c

   1 /*
   2  * Copyright (c) 2014 Red Hat, 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 "libxfs_priv.h"
  19 #include "xfs_fs.h"
  20 #include "xfs_shared.h"
  21 #include "xfs_format.h"
  22 #include "xfs_log_format.h"
  23 #include "xfs_trans_resv.h"
  24 #include "xfs_bit.h"
  25 #include "xfs_sb.h"
  26 #include "xfs_mount.h"
  27 #include "xfs_defer.h"
  28 #include "xfs_inode.h"
  29 #include "xfs_trans.h"
  30 #include "xfs_alloc.h"
  31 #include "xfs_btree.h"
  32 #include "xfs_rmap.h"
  33 #include "xfs_rmap_btree.h"
  34 #include "xfs_trace.h"
  35 #include "xfs_cksum.h"
  36
  37 /*
  38  * Reverse map btree.
  39  *
  40  * This is a per-ag tree used to track the owner(s) of a given extent. With
  41  * reflink it is possible for there to be multiple owners, which is a departure
  42  * from classic XFS. Owner records for data extents are inserted when the
  43  * extent is mapped and removed when an extent is unmapped.  Owner records for
  44  * all other block types (i.e. metadata) are inserted when an extent is
  45  * allocated and removed when an extent is freed. There can only be one owner
  46  * of a metadata extent, usually an inode or some other metadata structure like
  47  * an AG btree.
  48  *
  49  * The rmap btree is part of the free space management, so blocks for the tree
  50  * are sourced from the agfl. Hence we need transaction reservation support for
  51  * this tree so that the freelist is always large enough. This also impacts on
  52  * the minimum space we need to leave free in the AG.
  53  *
  54  * The tree is ordered by [ag block, owner, offset]. This is a large key size,
  55  * but it is the only way to enforce unique keys when a block can be owned by
  56  * multiple files at any offset. There's no need to order/search by extent
  57  * size for online updating/management of the tree. It is intended that most
  58  * reverse lookups will be to find the owner(s) of a particular block, or to
  59  * try to recover tree and file data from corrupt primary metadata.
  60  */
  61
  62 static struct xfs_btree_cur *
  63 xfs_rmapbt_dup_cursor(
  64         struct xfs_btree_cur    *cur)
  65 {
  66         return xfs_rmapbt_init_cursor(cur->bc_mp, cur->bc_tp,
  67                         cur->bc_private.a.agbp, cur->bc_private.a.agno);
  68 }
  69
  70 STATIC void
  71 xfs_rmapbt_set_root(
  72         struct xfs_btree_cur    *cur,
  73         union xfs_btree_ptr     *ptr,
  74         int                     inc)
  75 {
  76         struct xfs_buf          *agbp = cur->bc_private.a.agbp;
  77         struct xfs_agf          *agf = XFS_BUF_TO_AGF(agbp);
  78         xfs_agnumber_t          seqno = be32_to_cpu(agf->agf_seqno);
  79         int                     btnum = cur->bc_btnum;
  80         struct xfs_perag        *pag = xfs_perag_get(cur->bc_mp, seqno);
  81
  82         ASSERT(ptr->s != 0);
  83
  84         agf->agf_roots[btnum] = ptr->s;
  85         be32_add_cpu(&agf->agf_levels[btnum], inc);
  86         pag->pagf_levels[btnum] += inc;
  87         xfs_perag_put(pag);
  88
  89         xfs_alloc_log_agf(cur->bc_tp, agbp, XFS_AGF_ROOTS | XFS_AGF_LEVELS);
  90 }
  91
  92 STATIC int
  93 xfs_rmapbt_alloc_block(
  94         struct xfs_btree_cur    *cur,
  95         union xfs_btree_ptr     *start,
  96         union xfs_btree_ptr     *new,
  97         int                     *stat)
  98 {
  99         int                     error;
 100         xfs_agblock_t           bno;
 101
 102         XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
 103
 104         /* Allocate the new block from the freelist. If we can't, give up.  */
 105         error = xfs_alloc_get_freelist(cur->bc_tp, cur->bc_private.a.agbp,
 106                                        &bno, 1);
 107         if (error) {
 108                 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
 109                 return error;
 110         }
 111
 112         trace_xfs_rmapbt_alloc_block(cur->bc_mp, cur->bc_private.a.agno,
 113                         bno, 1);
 114         if (bno == NULLAGBLOCK) {
 115                 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
 116                 *stat = 0;
 117                 return 0;
 118         }
 119
 120         xfs_extent_busy_reuse(cur->bc_mp, cur->bc_private.a.agno, bno, 1,
 121                         false);
 122
 123         xfs_trans_agbtree_delta(cur->bc_tp, 1);
 124         new->s = cpu_to_be32(bno);
 125
 126         XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
 127         *stat = 1;
 128         return 0;
 129 }
 130
 131 STATIC int
 132 xfs_rmapbt_free_block(
 133         struct xfs_btree_cur    *cur,
 134         struct xfs_buf          *bp)
 135 {
 136         struct xfs_buf          *agbp = cur->bc_private.a.agbp;
 137         struct xfs_agf          *agf = XFS_BUF_TO_AGF(agbp);
 138         xfs_agblock_t           bno;
 139         int                     error;
 140
 141         bno = xfs_daddr_to_agbno(cur->bc_mp, XFS_BUF_ADDR(bp));
 142         trace_xfs_rmapbt_free_block(cur->bc_mp, cur->bc_private.a.agno,
 143                         bno, 1);
 144         error = xfs_alloc_put_freelist(cur->bc_tp, agbp, NULL, bno, 1);
 145         if (error)
 146                 return error;
 147
 148         xfs_extent_busy_insert(cur->bc_tp, be32_to_cpu(agf->agf_seqno), bno, 1,
 149                               XFS_EXTENT_BUSY_SKIP_DISCARD);
 150         xfs_trans_agbtree_delta(cur->bc_tp, -1);
 151
 152         return 0;
 153 }
 154
 155 STATIC int
 156 xfs_rmapbt_get_minrecs(
 157         struct xfs_btree_cur    *cur,
 158         int                     level)
 159 {
 160         return cur->bc_mp->m_rmap_mnr[level != 0];
 161 }
 162
 163 STATIC int
 164 xfs_rmapbt_get_maxrecs(
 165         struct xfs_btree_cur    *cur,
 166         int                     level)
 167 {
 168         return cur->bc_mp->m_rmap_mxr[level != 0];
 169 }
 170
 171 STATIC void
 172 xfs_rmapbt_init_key_from_rec(
 173         union xfs_btree_key     *key,
 174         union xfs_btree_rec     *rec)
 175 {
 176         key->rmap.rm_startblock = rec->rmap.rm_startblock;
 177         key->rmap.rm_owner = rec->rmap.rm_owner;
 178         key->rmap.rm_offset = rec->rmap.rm_offset;
 179 }
 180
 181 /*
 182  * The high key for a reverse mapping record can be computed by shifting
 183  * the startblock and offset to the highest value that would still map
 184  * to that record.  In practice this means that we add blockcount-1 to
 185  * the startblock for all records, and if the record is for a data/attr
 186  * fork mapping, we add blockcount-1 to the offset too.
 187  */
 188 STATIC void
 189 xfs_rmapbt_init_high_key_from_rec(
 190         union xfs_btree_key     *key,
 191         union xfs_btree_rec     *rec)
 192 {
 193         __uint64_t              off;
 194         int                     adj;
 195
 196         adj = be32_to_cpu(rec->rmap.rm_blockcount) - 1;
 197
 198         key->rmap.rm_startblock = rec->rmap.rm_startblock;
 199         be32_add_cpu(&key->rmap.rm_startblock, adj);
 200         key->rmap.rm_owner = rec->rmap.rm_owner;
 201         key->rmap.rm_offset = rec->rmap.rm_offset;
 202         if (XFS_RMAP_NON_INODE_OWNER(be64_to_cpu(rec->rmap.rm_owner)) ||
 203             XFS_RMAP_IS_BMBT_BLOCK(be64_to_cpu(rec->rmap.rm_offset)))
 204                 return;
 205         off = be64_to_cpu(key->rmap.rm_offset);
 206         off = (XFS_RMAP_OFF(off) + adj) | (off & ~XFS_RMAP_OFF_MASK);
 207         key->rmap.rm_offset = cpu_to_be64(off);
 208 }
 209
 210 STATIC void
 211 xfs_rmapbt_init_rec_from_cur(
 212         struct xfs_btree_cur    *cur,
 213         union xfs_btree_rec     *rec)
 214 {
 215         rec->rmap.rm_startblock = cpu_to_be32(cur->bc_rec.r.rm_startblock);
 216         rec->rmap.rm_blockcount = cpu_to_be32(cur->bc_rec.r.rm_blockcount);
 217         rec->rmap.rm_owner = cpu_to_be64(cur->bc_rec.r.rm_owner);
 218         rec->rmap.rm_offset = cpu_to_be64(
 219                         xfs_rmap_irec_offset_pack(&cur->bc_rec.r));
 220 }
 221
 222 STATIC void
 223 xfs_rmapbt_init_ptr_from_cur(
 224         struct xfs_btree_cur    *cur,
 225         union xfs_btree_ptr     *ptr)
 226 {
 227         struct xfs_agf          *agf = XFS_BUF_TO_AGF(cur->bc_private.a.agbp);
 228
 229         ASSERT(cur->bc_private.a.agno == be32_to_cpu(agf->agf_seqno));
 230         ASSERT(agf->agf_roots[cur->bc_btnum] != 0);
 231
 232         ptr->s = agf->agf_roots[cur->bc_btnum];
 233 }
 234
 235 STATIC __int64_t
 236 xfs_rmapbt_key_diff(
 237         struct xfs_btree_cur    *cur,
 238         union xfs_btree_key     *key)
 239 {
 240         struct xfs_rmap_irec    *rec = &cur->bc_rec.r;
 241         struct xfs_rmap_key     *kp = &key->rmap;
 242         __u64                   x, y;
 243         __int64_t               d;
 244
 245         d = (__int64_t)be32_to_cpu(kp->rm_startblock) - rec->rm_startblock;
 246         if (d)
 247                 return d;
 248
 249         x = be64_to_cpu(kp->rm_owner);
 250         y = rec->rm_owner;
 251         if (x > y)
 252                 return 1;
 253         else if (y > x)
 254                 return -1;
 255
 256         x = XFS_RMAP_OFF(be64_to_cpu(kp->rm_offset));
 257         y = rec->rm_offset;
 258         if (x > y)
 259                 return 1;
 260         else if (y > x)
 261                 return -1;
 262         return 0;
 263 }
 264
 265 STATIC __int64_t
 266 xfs_rmapbt_diff_two_keys(
 267         struct xfs_btree_cur    *cur,
 268         union xfs_btree_key     *k1,
 269         union xfs_btree_key     *k2)
 270 {
 271         struct xfs_rmap_key     *kp1 = &k1->rmap;
 272         struct xfs_rmap_key     *kp2 = &k2->rmap;
 273         __int64_t               d;
 274         __u64                   x, y;
 275
 276         d = (__int64_t)be32_to_cpu(kp1->rm_startblock) -
 277                        be32_to_cpu(kp2->rm_startblock);
 278         if (d)
 279                 return d;
 280
 281         x = be64_to_cpu(kp1->rm_owner);
 282         y = be64_to_cpu(kp2->rm_owner);
 283         if (x > y)
 284                 return 1;
 285         else if (y > x)
 286                 return -1;
 287
 288         x = XFS_RMAP_OFF(be64_to_cpu(kp1->rm_offset));
 289         y = XFS_RMAP_OFF(be64_to_cpu(kp2->rm_offset));
 290         if (x > y)
 291                 return 1;
 292         else if (y > x)
 293                 return -1;
 294         return 0;
 295 }
 296
 297 static bool
 298 xfs_rmapbt_verify(
 299         struct xfs_buf          *bp)
 300 {
 301         struct xfs_mount        *mp = bp->b_target->bt_mount;
 302         struct xfs_btree_block  *block = XFS_BUF_TO_BLOCK(bp);
 303         struct xfs_perag        *pag = bp->b_pag;
 304         unsigned int            level;
 305
 306         /*
 307          * magic number and level verification
 308          *
 309          * During growfs operations, we can't verify the exact level or owner as
 310          * the perag is not fully initialised and hence not attached to the
 311          * buffer.  In this case, check against the maximum tree depth.
 312          *
 313          * Similarly, during log recovery we will have a perag structure
 314          * attached, but the agf information will not yet have been initialised
 315          * from the on disk AGF. Again, we can only check against maximum limits
 316          * in this case.
 317          */
 318         if (block->bb_magic != cpu_to_be32(XFS_RMAP_CRC_MAGIC))
 319                 return false;
 320
 321         if (!xfs_sb_version_hasrmapbt(&mp->m_sb))
 322                 return false;
 323         if (!xfs_btree_sblock_v5hdr_verify(bp))
 324                 return false;
 325
 326         level = be16_to_cpu(block->bb_level);
 327         if (pag && pag->pagf_init) {
 328                 if (level >= pag->pagf_levels[XFS_BTNUM_RMAPi])
 329                         return false;
 330         } else if (level >= mp->m_rmap_maxlevels)
 331                 return false;
 332
 333         return xfs_btree_sblock_verify(bp, mp->m_rmap_mxr[level != 0]);
 334 }
 335
 336 static void
 337 xfs_rmapbt_read_verify(
 338         struct xfs_buf  *bp)
 339 {
 340         if (!xfs_btree_sblock_verify_crc(bp))
 341                 xfs_buf_ioerror(bp, -EFSBADCRC);
 342         else if (!xfs_rmapbt_verify(bp))
 343                 xfs_buf_ioerror(bp, -EFSCORRUPTED);
 344
 345         if (bp->b_error) {
 346                 trace_xfs_btree_corrupt(bp, _RET_IP_);
 347                 xfs_verifier_error(bp);
 348         }
 349 }
 350
 351 static void
 352 xfs_rmapbt_write_verify(
 353         struct xfs_buf  *bp)
 354 {
 355         if (!xfs_rmapbt_verify(bp)) {
 356                 trace_xfs_btree_corrupt(bp, _RET_IP_);
 357                 xfs_buf_ioerror(bp, -EFSCORRUPTED);
 358                 xfs_verifier_error(bp);
 359                 return;
 360         }
 361         xfs_btree_sblock_calc_crc(bp);
 362
 363 }
 364
 365 const struct xfs_buf_ops xfs_rmapbt_buf_ops = {
 366         .name                   = "xfs_rmapbt",
 367         .verify_read            = xfs_rmapbt_read_verify,
 368         .verify_write           = xfs_rmapbt_write_verify,
 369 };
 370
 371 #if defined(DEBUG) || defined(XFS_WARN)
 372 STATIC int
 373 xfs_rmapbt_keys_inorder(
 374         struct xfs_btree_cur    *cur,
 375         union xfs_btree_key     *k1,
 376         union xfs_btree_key     *k2)
 377 {
 378         __uint32_t              x;
 379         __uint32_t              y;
 380         __uint64_t              a;
 381         __uint64_t              b;
 382
 383         x = be32_to_cpu(k1->rmap.rm_startblock);
 384         y = be32_to_cpu(k2->rmap.rm_startblock);
 385         if (x < y)
 386                 return 1;
 387         else if (x > y)
 388                 return 0;
 389         a = be64_to_cpu(k1->rmap.rm_owner);
 390         b = be64_to_cpu(k2->rmap.rm_owner);
 391         if (a < b)
 392                 return 1;
 393         else if (a > b)
 394                 return 0;
 395         a = XFS_RMAP_OFF(be64_to_cpu(k1->rmap.rm_offset));
 396         b = XFS_RMAP_OFF(be64_to_cpu(k2->rmap.rm_offset));
 397         if (a <= b)
 398                 return 1;
 399         return 0;
 400 }
 401
 402 STATIC int
 403 xfs_rmapbt_recs_inorder(
 404         struct xfs_btree_cur    *cur,
 405         union xfs_btree_rec     *r1,
 406         union xfs_btree_rec     *r2)
 407 {
 408         __uint32_t              x;
 409         __uint32_t              y;
 410         __uint64_t              a;
 411         __uint64_t              b;
 412
 413         x = be32_to_cpu(r1->rmap.rm_startblock);
 414         y = be32_to_cpu(r2->rmap.rm_startblock);
 415         if (x < y)
 416                 return 1;
 417         else if (x > y)
 418                 return 0;
 419         a = be64_to_cpu(r1->rmap.rm_owner);
 420         b = be64_to_cpu(r2->rmap.rm_owner);
 421         if (a < b)
 422                 return 1;
 423         else if (a > b)
 424                 return 0;
 425         a = XFS_RMAP_OFF(be64_to_cpu(r1->rmap.rm_offset));
 426         b = XFS_RMAP_OFF(be64_to_cpu(r2->rmap.rm_offset));
 427         if (a <= b)
 428                 return 1;
 429         return 0;
 430 }
 431 #endif  /* DEBUG */
 432
 433 static const struct xfs_btree_ops xfs_rmapbt_ops = {
 434         .rec_len                = sizeof(struct xfs_rmap_rec),
 435         .key_len                = 2 * sizeof(struct xfs_rmap_key),
 436
 437         .dup_cursor             = xfs_rmapbt_dup_cursor,
 438         .set_root               = xfs_rmapbt_set_root,
 439         .alloc_block            = xfs_rmapbt_alloc_block,
 440         .free_block             = xfs_rmapbt_free_block,
 441         .get_minrecs            = xfs_rmapbt_get_minrecs,
 442         .get_maxrecs            = xfs_rmapbt_get_maxrecs,
 443         .init_key_from_rec      = xfs_rmapbt_init_key_from_rec,
 444         .init_high_key_from_rec = xfs_rmapbt_init_high_key_from_rec,
 445         .init_rec_from_cur      = xfs_rmapbt_init_rec_from_cur,
 446         .init_ptr_from_cur      = xfs_rmapbt_init_ptr_from_cur,
 447         .key_diff               = xfs_rmapbt_key_diff,
 448         .buf_ops                = &xfs_rmapbt_buf_ops,
 449         .diff_two_keys          = xfs_rmapbt_diff_two_keys,
 450 #if defined(DEBUG) || defined(XFS_WARN)
 451         .keys_inorder           = xfs_rmapbt_keys_inorder,
 452         .recs_inorder           = xfs_rmapbt_recs_inorder,
 453 #endif
 454 };
 455
 456 /*
 457  * Allocate a new allocation btree cursor.
 458  */
 459 struct xfs_btree_cur *
 460 xfs_rmapbt_init_cursor(
 461         struct xfs_mount        *mp,
 462         struct xfs_trans        *tp,
 463         struct xfs_buf          *agbp,
 464         xfs_agnumber_t          agno)
 465 {
 466         struct xfs_agf          *agf = XFS_BUF_TO_AGF(agbp);
 467         struct xfs_btree_cur    *cur;
 468
 469         cur = kmem_zone_zalloc(xfs_btree_cur_zone, KM_NOFS);
 470         cur->bc_tp = tp;
 471         cur->bc_mp = mp;
 472         /* Overlapping btree; 2 keys per pointer. */
 473         cur->bc_btnum = XFS_BTNUM_RMAP;
 474         cur->bc_flags = XFS_BTREE_CRC_BLOCKS | XFS_BTREE_OVERLAPPING;
 475         cur->bc_blocklog = mp->m_sb.sb_blocklog;
 476         cur->bc_ops = &xfs_rmapbt_ops;
 477         cur->bc_nlevels = be32_to_cpu(agf->agf_levels[XFS_BTNUM_RMAP]);
 478
 479         cur->bc_private.a.agbp = agbp;
 480         cur->bc_private.a.agno = agno;
 481
 482         return cur;
 483 }
 484
 485 /*
 486  * Calculate number of records in an rmap btree block.
 487  */
 488 int
 489 xfs_rmapbt_maxrecs(
 490         struct xfs_mount        *mp,
 491         int                     blocklen,
 492         int                     leaf)
 493 {
 494         blocklen -= XFS_RMAP_BLOCK_LEN;
 495
 496         if (leaf)
 497                 return blocklen / sizeof(struct xfs_rmap_rec);
 498         return blocklen /
 499                 (2 * sizeof(struct xfs_rmap_key) + sizeof(xfs_rmap_ptr_t));
 500 }
 501
 502 /* Compute the maximum height of an rmap btree. */
 503 void
 504 xfs_rmapbt_compute_maxlevels(
 505         struct xfs_mount                *mp)
 506 {
 507         mp->m_rmap_maxlevels = xfs_btree_compute_maxlevels(mp,
 508                         mp->m_rmap_mnr, mp->m_sb.sb_agblocks);
 509 }