fs/bcachefs/btree_types.h

   1 /* SPDX-License-Identifier: GPL-2.0 */
   2 #ifndef _BCACHEFS_BTREE_TYPES_H
   3 #define _BCACHEFS_BTREE_TYPES_H
   4
   5 #include <linux/list.h>
   6 #include <linux/rhashtable.h>
   7
   8 #include "bbpos_types.h"
   9 #include "btree_key_cache_types.h"
  10 #include "buckets_types.h"
  11 #include "darray.h"
  12 #include "errcode.h"
  13 #include "journal_types.h"
  14 #include "replicas_types.h"
  15 #include "six.h"
  16
  17 struct open_bucket;
  18 struct btree_update;
  19 struct btree_trans;
  20
  21 #define MAX_BSETS               3U
  22
  23 struct btree_nr_keys {
  24
  25         /*
  26          * Amount of live metadata (i.e. size of node after a compaction) in
  27          * units of u64s
  28          */
  29         u16                     live_u64s;
  30         u16                     bset_u64s[MAX_BSETS];
  31
  32         /* live keys only: */
  33         u16                     packed_keys;
  34         u16                     unpacked_keys;
  35 };
  36
  37 struct bset_tree {
  38         /*
  39          * We construct a binary tree in an array as if the array
  40          * started at 1, so that things line up on the same cachelines
  41          * better: see comments in bset.c at cacheline_to_bkey() for
  42          * details
  43          */
  44
  45         /* size of the binary tree and prev array */
  46         u16                     size;
  47
  48         /* function of size - precalculated for to_inorder() */
  49         u16                     extra;
  50
  51         u16                     data_offset;
  52         u16                     aux_data_offset;
  53         u16                     end_offset;
  54 };
  55
  56 struct btree_write {
  57         struct journal_entry_pin        journal;
  58 };
  59
  60 struct btree_alloc {
  61         struct open_buckets     ob;
  62         __BKEY_PADDED(k, BKEY_BTREE_PTR_VAL_U64s_MAX);
  63 };
  64
  65 struct btree_bkey_cached_common {
  66         struct six_lock         lock;
  67         u8                      level;
  68         u8                      btree_id;
  69         bool                    cached;
  70 };
  71
  72 struct btree {
  73         struct btree_bkey_cached_common c;
  74
  75         struct rhash_head       hash;
  76         u64                     hash_val;
  77
  78         unsigned long           flags;
  79         u16                     written;
  80         u8                      nsets;
  81         u8                      nr_key_bits;
  82         u16                     version_ondisk;
  83
  84         struct bkey_format      format;
  85
  86         struct btree_node       *data;
  87         void                    *aux_data;
  88
  89         /*
  90          * Sets of sorted keys - the real btree node - plus a binary search tree
  91          *
  92          * set[0] is special; set[0]->tree, set[0]->prev and set[0]->data point
  93          * to the memory we have allocated for this btree node. Additionally,
  94          * set[0]->data points to the entire btree node as it exists on disk.
  95          */
  96         struct bset_tree        set[MAX_BSETS];
  97
  98         struct btree_nr_keys    nr;
  99         u16                     sib_u64s[2];
 100         u16                     whiteout_u64s;
 101         u8                      byte_order;
 102         u8                      unpack_fn_len;
 103
 104         struct btree_write      writes[2];
 105
 106         /* Key/pointer for this btree node */
 107         __BKEY_PADDED(key, BKEY_BTREE_PTR_VAL_U64s_MAX);
 108
 109         /*
 110          * XXX: add a delete sequence number, so when bch2_btree_node_relock()
 111          * fails because the lock sequence number has changed - i.e. the
 112          * contents were modified - we can still relock the node if it's still
 113          * the one we want, without redoing the traversal
 114          */
 115
 116         /*
 117          * For asynchronous splits/interior node updates:
 118          * When we do a split, we allocate new child nodes and update the parent
 119          * node to point to them: we update the parent in memory immediately,
 120          * but then we must wait until the children have been written out before
 121          * the update to the parent can be written - this is a list of the
 122          * btree_updates that are blocking this node from being
 123          * written:
 124          */
 125         struct list_head        write_blocked;
 126
 127         /*
 128          * Also for asynchronous splits/interior node updates:
 129          * If a btree node isn't reachable yet, we don't want to kick off
 130          * another write - because that write also won't yet be reachable and
 131          * marking it as completed before it's reachable would be incorrect:
 132          */
 133         unsigned long           will_make_reachable;
 134
 135         struct open_buckets     ob;
 136
 137         /* lru list */
 138         struct list_head        list;
 139 };
 140
 141 struct btree_cache {
 142         struct rhashtable       table;
 143         bool                    table_init_done;
 144         /*
 145          * We never free a struct btree, except on shutdown - we just put it on
 146          * the btree_cache_freed list and reuse it later. This simplifies the
 147          * code, and it doesn't cost us much memory as the memory usage is
 148          * dominated by buffers that hold the actual btree node data and those
 149          * can be freed - and the number of struct btrees allocated is
 150          * effectively bounded.
 151          *
 152          * btree_cache_freeable effectively is a small cache - we use it because
 153          * high order page allocations can be rather expensive, and it's quite
 154          * common to delete and allocate btree nodes in quick succession. It
 155          * should never grow past ~2-3 nodes in practice.
 156          */
 157         struct mutex            lock;
 158         struct list_head        live;
 159         struct list_head        freeable;
 160         struct list_head        freed_pcpu;
 161         struct list_head        freed_nonpcpu;
 162
 163         /* Number of elements in live + freeable lists */
 164         unsigned                used;
 165         unsigned                reserve;
 166         unsigned                freed;
 167         unsigned                not_freed_lock_intent;
 168         unsigned                not_freed_lock_write;
 169         unsigned                not_freed_dirty;
 170         unsigned                not_freed_read_in_flight;
 171         unsigned                not_freed_write_in_flight;
 172         unsigned                not_freed_noevict;
 173         unsigned                not_freed_write_blocked;
 174         unsigned                not_freed_will_make_reachable;
 175         unsigned                not_freed_access_bit;
 176         atomic_t                dirty;
 177         struct shrinker         *shrink;
 178
 179         unsigned                used_by_btree[BTREE_ID_NR];
 180
 181         /*
 182          * If we need to allocate memory for a new btree node and that
 183          * allocation fails, we can cannibalize another node in the btree cache
 184          * to satisfy the allocation - lock to guarantee only one thread does
 185          * this at a time:
 186          */
 187         struct task_struct      *alloc_lock;
 188         struct closure_waitlist alloc_wait;
 189
 190         struct bbpos            pinned_nodes_start;
 191         struct bbpos            pinned_nodes_end;
 192         u64                     pinned_nodes_leaf_mask;
 193         u64                     pinned_nodes_interior_mask;
 194 };
 195
 196 struct btree_node_iter {
 197         struct btree_node_iter_set {
 198                 u16     k, end;
 199         } data[MAX_BSETS];
 200 };
 201
 202 #define BTREE_ITER_FLAGS()                      \
 203         x(slots)                                \
 204         x(intent)                               \
 205         x(prefetch)                             \
 206         x(is_extents)                           \
 207         x(not_extents)                          \
 208         x(cached)                               \
 209         x(with_key_cache)                       \
 210         x(with_updates)                         \
 211         x(with_journal)                         \
 212         x(snapshot_field)                       \
 213         x(all_snapshots)                        \
 214         x(filter_snapshots)                     \
 215         x(nopreserve)                           \
 216         x(cached_nofill)                        \
 217         x(key_cache_fill)                       \
 218
 219 #define STR_HASH_FLAGS()                        \
 220         x(must_create)                          \
 221         x(must_replace)
 222
 223 #define BTREE_UPDATE_FLAGS()                    \
 224         x(internal_snapshot_node)               \
 225         x(nojournal)                            \
 226         x(key_cache_reclaim)
 227
 228
 229 /*
 230  * BTREE_TRIGGER_norun - don't run triggers at all
 231  *
 232  * BTREE_TRIGGER_transactional - we're running transactional triggers as part of
 233  * a transaction commit: triggers may generate new updates
 234  *
 235  * BTREE_TRIGGER_atomic - we're running atomic triggers during a transaction
 236  * commit: we have our journal reservation, we're holding btree node write
 237  * locks, and we know the transaction is going to commit (returning an error
 238  * here is a fatal error, causing us to go emergency read-only)
 239  *
 240  * BTREE_TRIGGER_gc - we're in gc/fsck: running triggers to recalculate e.g. disk usage
 241  *
 242  * BTREE_TRIGGER_insert - @new is entering the btree
 243  * BTREE_TRIGGER_overwrite - @old is leaving the btree
 244  *
 245  * BTREE_TRIGGER_bucket_invalidate - signal from bucket invalidate path to alloc
 246  * trigger
 247  */
 248 #define BTREE_TRIGGER_FLAGS()                   \
 249         x(norun)                                \
 250         x(transactional)                        \
 251         x(atomic)                               \
 252         x(check_repair)                         \
 253         x(gc)                                   \
 254         x(insert)                               \
 255         x(overwrite)                            \
 256         x(is_root)                              \
 257         x(bucket_invalidate)
 258
 259 enum {
 260 #define x(n) BTREE_ITER_FLAG_BIT_##n,
 261         BTREE_ITER_FLAGS()
 262         STR_HASH_FLAGS()
 263         BTREE_UPDATE_FLAGS()
 264         BTREE_TRIGGER_FLAGS()
 265 #undef x
 266 };
 267
 268 /* iter flags must fit in a u16: */
 269 //BUILD_BUG_ON(BTREE_ITER_FLAG_BIT_key_cache_fill > 15);
 270
 271 enum btree_iter_update_trigger_flags {
 272 #define x(n) BTREE_ITER_##n     = 1U << BTREE_ITER_FLAG_BIT_##n,
 273         BTREE_ITER_FLAGS()
 274 #undef x
 275 #define x(n) STR_HASH_##n       = 1U << BTREE_ITER_FLAG_BIT_##n,
 276         STR_HASH_FLAGS()
 277 #undef x
 278 #define x(n) BTREE_UPDATE_##n   = 1U << BTREE_ITER_FLAG_BIT_##n,
 279         BTREE_UPDATE_FLAGS()
 280 #undef x
 281 #define x(n) BTREE_TRIGGER_##n  = 1U << BTREE_ITER_FLAG_BIT_##n,
 282         BTREE_TRIGGER_FLAGS()
 283 #undef x
 284 };
 285
 286 enum btree_path_uptodate {
 287         BTREE_ITER_UPTODATE             = 0,
 288         BTREE_ITER_NEED_RELOCK          = 1,
 289         BTREE_ITER_NEED_TRAVERSE        = 2,
 290 };
 291
 292 #if defined(CONFIG_BCACHEFS_LOCK_TIME_STATS) || defined(CONFIG_BCACHEFS_DEBUG)
 293 #define TRACK_PATH_ALLOCATED
 294 #endif
 295
 296 typedef u16 btree_path_idx_t;
 297
 298 struct btree_path {
 299         btree_path_idx_t        sorted_idx;
 300         u8                      ref;
 301         u8                      intent_ref;
 302
 303         /* btree_iter_copy starts here: */
 304         struct bpos             pos;
 305
 306         enum btree_id           btree_id:5;
 307         bool                    cached:1;
 308         bool                    preserve:1;
 309         enum btree_path_uptodate uptodate:2;
 310         /*
 311          * When true, failing to relock this path will cause the transaction to
 312          * restart:
 313          */
 314         bool                    should_be_locked:1;
 315         unsigned                level:3,
 316                                 locks_want:3;
 317         u8                      nodes_locked;
 318
 319         struct btree_path_level {
 320                 struct btree    *b;
 321                 struct btree_node_iter iter;
 322                 u32             lock_seq;
 323 #ifdef CONFIG_BCACHEFS_LOCK_TIME_STATS
 324                 u64             lock_taken_time;
 325 #endif
 326         }                       l[BTREE_MAX_DEPTH];
 327 #ifdef TRACK_PATH_ALLOCATED
 328         unsigned long           ip_allocated;
 329 #endif
 330 };
 331
 332 static inline struct btree_path_level *path_l(struct btree_path *path)
 333 {
 334         return path->l + path->level;
 335 }
 336
 337 static inline unsigned long btree_path_ip_allocated(struct btree_path *path)
 338 {
 339 #ifdef TRACK_PATH_ALLOCATED
 340         return path->ip_allocated;
 341 #else
 342         return _THIS_IP_;
 343 #endif
 344 }
 345
 346 /*
 347  * @pos                 - iterator's current position
 348  * @level               - current btree depth
 349  * @locks_want          - btree level below which we start taking intent locks
 350  * @nodes_locked        - bitmask indicating which nodes in @nodes are locked
 351  * @nodes_intent_locked - bitmask indicating which locks are intent locks
 352  */
 353 struct btree_iter {
 354         struct btree_trans      *trans;
 355         btree_path_idx_t        path;
 356         btree_path_idx_t        update_path;
 357         btree_path_idx_t        key_cache_path;
 358
 359         enum btree_id           btree_id:8;
 360         u8                      min_depth;
 361
 362         /* btree_iter_copy starts here: */
 363         u16                     flags;
 364
 365         /* When we're filtering by snapshot, the snapshot ID we're looking for: */
 366         unsigned                snapshot;
 367
 368         struct bpos             pos;
 369         /*
 370          * Current unpacked key - so that bch2_btree_iter_next()/
 371          * bch2_btree_iter_next_slot() can correctly advance pos.
 372          */
 373         struct bkey             k;
 374
 375         /* BTREE_ITER_with_journal: */
 376         size_t                  journal_idx;
 377 #ifdef TRACK_PATH_ALLOCATED
 378         unsigned long           ip_allocated;
 379 #endif
 380 };
 381
 382 #define BKEY_CACHED_ACCESSED            0
 383 #define BKEY_CACHED_DIRTY               1
 384
 385 struct bkey_cached {
 386         struct btree_bkey_cached_common c;
 387
 388         unsigned long           flags;
 389         unsigned long           btree_trans_barrier_seq;
 390         u16                     u64s;
 391         bool                    valid;
 392         struct bkey_cached_key  key;
 393
 394         struct rhash_head       hash;
 395         struct list_head        list;
 396
 397         struct journal_entry_pin journal;
 398         u64                     seq;
 399
 400         struct bkey_i           *k;
 401 };
 402
 403 static inline struct bpos btree_node_pos(struct btree_bkey_cached_common *b)
 404 {
 405         return !b->cached
 406                 ? container_of(b, struct btree, c)->key.k.p
 407                 : container_of(b, struct bkey_cached, c)->key.pos;
 408 }
 409
 410 struct btree_insert_entry {
 411         unsigned                flags;
 412         u8                      bkey_type;
 413         enum btree_id           btree_id:8;
 414         u8                      level:4;
 415         bool                    cached:1;
 416         bool                    insert_trigger_run:1;
 417         bool                    overwrite_trigger_run:1;
 418         bool                    key_cache_already_flushed:1;
 419         /*
 420          * @old_k may be a key from the journal; @old_btree_u64s always refers
 421          * to the size of the key being overwritten in the btree:
 422          */
 423         u8                      old_btree_u64s;
 424         btree_path_idx_t        path;
 425         struct bkey_i           *k;
 426         /* key being overwritten: */
 427         struct bkey             old_k;
 428         const struct bch_val    *old_v;
 429         unsigned long           ip_allocated;
 430 };
 431
 432 /* Number of btree paths we preallocate, usually enough */
 433 #define BTREE_ITER_INITIAL              64
 434 /*
 435  * Lmiit for btree_trans_too_many_iters(); this is enough that almost all code
 436  * paths should run inside this limit, and if they don't it usually indicates a
 437  * bug (leaking/duplicated btree paths).
 438  *
 439  * exception: some fsck paths
 440  *
 441  * bugs with excessive path usage seem to have possibly been eliminated now, so
 442  * we might consider eliminating this (and btree_trans_too_many_iter()) at some
 443  * point.
 444  */
 445 #define BTREE_ITER_NORMAL_LIMIT         256
 446 /* never exceed limit */
 447 #define BTREE_ITER_MAX                  (1U << 10)
 448
 449 struct btree_trans_commit_hook;
 450 typedef int (btree_trans_commit_hook_fn)(struct btree_trans *, struct btree_trans_commit_hook *);
 451
 452 struct btree_trans_commit_hook {
 453         btree_trans_commit_hook_fn      *fn;
 454         struct btree_trans_commit_hook  *next;
 455 };
 456
 457 #define BTREE_TRANS_MEM_MAX     (1U << 16)
 458
 459 #define BTREE_TRANS_MAX_LOCK_HOLD_TIME_NS       10000
 460
 461 struct btree_trans_paths {
 462         unsigned long           nr_paths;
 463         struct btree_path       paths[];
 464 };
 465
 466 struct btree_trans {
 467         struct bch_fs           *c;
 468
 469         unsigned long           *paths_allocated;
 470         struct btree_path       *paths;
 471         btree_path_idx_t        *sorted;
 472         struct btree_insert_entry *updates;
 473
 474         void                    *mem;
 475         unsigned                mem_top;
 476         unsigned                mem_bytes;
 477
 478         btree_path_idx_t        nr_sorted;
 479         btree_path_idx_t        nr_paths;
 480         btree_path_idx_t        nr_paths_max;
 481         u8                      fn_idx;
 482         u8                      nr_updates;
 483         u8                      lock_must_abort;
 484         bool                    lock_may_not_fail:1;
 485         bool                    srcu_held:1;
 486         bool                    locked:1;
 487         bool                    write_locked:1;
 488         bool                    used_mempool:1;
 489         bool                    in_traverse_all:1;
 490         bool                    paths_sorted:1;
 491         bool                    memory_allocation_failure:1;
 492         bool                    journal_transaction_names:1;
 493         bool                    journal_replay_not_finished:1;
 494         bool                    notrace_relock_fail:1;
 495         enum bch_errcode        restarted:16;
 496         u32                     restart_count;
 497
 498         u64                     last_begin_time;
 499         unsigned long           last_begin_ip;
 500         unsigned long           last_restarted_ip;
 501         unsigned long           last_unlock_ip;
 502         unsigned long           srcu_lock_time;
 503
 504         const char              *fn;
 505         struct btree_bkey_cached_common *locking;
 506         struct six_lock_waiter  locking_wait;
 507         int                     srcu_idx;
 508
 509         /* update path: */
 510         u16                     journal_entries_u64s;
 511         u16                     journal_entries_size;
 512         struct jset_entry       *journal_entries;
 513
 514         struct btree_trans_commit_hook *hooks;
 515         struct journal_entry_pin *journal_pin;
 516
 517         struct journal_res      journal_res;
 518         u64                     *journal_seq;
 519         struct disk_reservation *disk_res;
 520
 521         struct bch_fs_usage_base fs_usage_delta;
 522
 523         unsigned                journal_u64s;
 524         unsigned                extra_disk_res; /* XXX kill */
 525         struct replicas_delta_list *fs_usage_deltas;
 526
 527         /* Entries before this are zeroed out on every bch2_trans_get() call */
 528
 529         struct list_head        list;
 530         struct closure          ref;
 531
 532         unsigned long           _paths_allocated[BITS_TO_LONGS(BTREE_ITER_INITIAL)];
 533         struct btree_trans_paths trans_paths;
 534         struct btree_path       _paths[BTREE_ITER_INITIAL];
 535         btree_path_idx_t        _sorted[BTREE_ITER_INITIAL + 4];
 536         struct btree_insert_entry _updates[BTREE_ITER_INITIAL];
 537 };
 538
 539 static inline struct btree_path *btree_iter_path(struct btree_trans *trans, struct btree_iter *iter)
 540 {
 541         return trans->paths + iter->path;
 542 }
 543
 544 static inline struct btree_path *btree_iter_key_cache_path(struct btree_trans *trans, struct btree_iter *iter)
 545 {
 546         return iter->key_cache_path
 547                 ? trans->paths + iter->key_cache_path
 548                 : NULL;
 549 }
 550
 551 #define BCH_BTREE_WRITE_TYPES()                                         \
 552         x(initial,              0)                                      \
 553         x(init_next_bset,       1)                                      \
 554         x(cache_reclaim,        2)                                      \
 555         x(journal_reclaim,      3)                                      \
 556         x(interior,             4)
 557
 558 enum btree_write_type {
 559 #define x(t, n) BTREE_WRITE_##t,
 560         BCH_BTREE_WRITE_TYPES()
 561 #undef x
 562         BTREE_WRITE_TYPE_NR,
 563 };
 564
 565 #define BTREE_WRITE_TYPE_MASK   (roundup_pow_of_two(BTREE_WRITE_TYPE_NR) - 1)
 566 #define BTREE_WRITE_TYPE_BITS   ilog2(roundup_pow_of_two(BTREE_WRITE_TYPE_NR))
 567
 568 #define BTREE_FLAGS()                                                   \
 569         x(read_in_flight)                                               \
 570         x(read_error)                                                   \
 571         x(dirty)                                                        \
 572         x(need_write)                                                   \
 573         x(write_blocked)                                                \
 574         x(will_make_reachable)                                          \
 575         x(noevict)                                                      \
 576         x(write_idx)                                                    \
 577         x(accessed)                                                     \
 578         x(write_in_flight)                                              \
 579         x(write_in_flight_inner)                                        \
 580         x(just_written)                                                 \
 581         x(dying)                                                        \
 582         x(fake)                                                         \
 583         x(need_rewrite)                                                 \
 584         x(never_write)
 585
 586 enum btree_flags {
 587         /* First bits for btree node write type */
 588         BTREE_NODE_FLAGS_START = BTREE_WRITE_TYPE_BITS - 1,
 589 #define x(flag) BTREE_NODE_##flag,
 590         BTREE_FLAGS()
 591 #undef x
 592 };
 593
 594 #define x(flag)                                                         \
 595 static inline bool btree_node_ ## flag(struct btree *b)                 \
 596 {       return test_bit(BTREE_NODE_ ## flag, &b->flags); }              \
 597                                                                         \
 598 static inline void set_btree_node_ ## flag(struct btree *b)             \
 599 {       set_bit(BTREE_NODE_ ## flag, &b->flags); }                      \
 600                                                                         \
 601 static inline void clear_btree_node_ ## flag(struct btree *b)           \
 602 {       clear_bit(BTREE_NODE_ ## flag, &b->flags); }
 603
 604 BTREE_FLAGS()
 605 #undef x
 606
 607 static inline struct btree_write *btree_current_write(struct btree *b)
 608 {
 609         return b->writes + btree_node_write_idx(b);
 610 }
 611
 612 static inline struct btree_write *btree_prev_write(struct btree *b)
 613 {
 614         return b->writes + (btree_node_write_idx(b) ^ 1);
 615 }
 616
 617 static inline struct bset_tree *bset_tree_last(struct btree *b)
 618 {
 619         EBUG_ON(!b->nsets);
 620         return b->set + b->nsets - 1;
 621 }
 622
 623 static inline void *
 624 __btree_node_offset_to_ptr(const struct btree *b, u16 offset)
 625 {
 626         return (void *) ((u64 *) b->data + 1 + offset);
 627 }
 628
 629 static inline u16
 630 __btree_node_ptr_to_offset(const struct btree *b, const void *p)
 631 {
 632         u16 ret = (u64 *) p - 1 - (u64 *) b->data;
 633
 634         EBUG_ON(__btree_node_offset_to_ptr(b, ret) != p);
 635         return ret;
 636 }
 637
 638 static inline struct bset *bset(const struct btree *b,
 639                                 const struct bset_tree *t)
 640 {
 641         return __btree_node_offset_to_ptr(b, t->data_offset);
 642 }
 643
 644 static inline void set_btree_bset_end(struct btree *b, struct bset_tree *t)
 645 {
 646         t->end_offset =
 647                 __btree_node_ptr_to_offset(b, vstruct_last(bset(b, t)));
 648 }
 649
 650 static inline void set_btree_bset(struct btree *b, struct bset_tree *t,
 651                                   const struct bset *i)
 652 {
 653         t->data_offset = __btree_node_ptr_to_offset(b, i);
 654         set_btree_bset_end(b, t);
 655 }
 656
 657 static inline struct bset *btree_bset_first(struct btree *b)
 658 {
 659         return bset(b, b->set);
 660 }
 661
 662 static inline struct bset *btree_bset_last(struct btree *b)
 663 {
 664         return bset(b, bset_tree_last(b));
 665 }
 666
 667 static inline u16
 668 __btree_node_key_to_offset(const struct btree *b, const struct bkey_packed *k)
 669 {
 670         return __btree_node_ptr_to_offset(b, k);
 671 }
 672
 673 static inline struct bkey_packed *
 674 __btree_node_offset_to_key(const struct btree *b, u16 k)
 675 {
 676         return __btree_node_offset_to_ptr(b, k);
 677 }
 678
 679 static inline unsigned btree_bkey_first_offset(const struct bset_tree *t)
 680 {
 681         return t->data_offset + offsetof(struct bset, _data) / sizeof(u64);
 682 }
 683
 684 #define btree_bkey_first(_b, _t)                                        \
 685 ({                                                                      \
 686         EBUG_ON(bset(_b, _t)->start !=                                  \
 687                 __btree_node_offset_to_key(_b, btree_bkey_first_offset(_t)));\
 688                                                                         \
 689         bset(_b, _t)->start;                                            \
 690 })
 691
 692 #define btree_bkey_last(_b, _t)                                         \
 693 ({                                                                      \
 694         EBUG_ON(__btree_node_offset_to_key(_b, (_t)->end_offset) !=     \
 695                 vstruct_last(bset(_b, _t)));                            \
 696                                                                         \
 697         __btree_node_offset_to_key(_b, (_t)->end_offset);               \
 698 })
 699
 700 static inline unsigned bset_u64s(struct bset_tree *t)
 701 {
 702         return t->end_offset - t->data_offset -
 703                 sizeof(struct bset) / sizeof(u64);
 704 }
 705
 706 static inline unsigned bset_dead_u64s(struct btree *b, struct bset_tree *t)
 707 {
 708         return bset_u64s(t) - b->nr.bset_u64s[t - b->set];
 709 }
 710
 711 static inline unsigned bset_byte_offset(struct btree *b, void *i)
 712 {
 713         return i - (void *) b->data;
 714 }
 715
 716 enum btree_node_type {
 717         BKEY_TYPE_btree,
 718 #define x(kwd, val, ...) BKEY_TYPE_##kwd = val + 1,
 719         BCH_BTREE_IDS()
 720 #undef x
 721         BKEY_TYPE_NR
 722 };
 723
 724 /* Type of a key in btree @id at level @level: */
 725 static inline enum btree_node_type __btree_node_type(unsigned level, enum btree_id id)
 726 {
 727         return level ? BKEY_TYPE_btree : (unsigned) id + 1;
 728 }
 729
 730 /* Type of keys @b contains: */
 731 static inline enum btree_node_type btree_node_type(struct btree *b)
 732 {
 733         return __btree_node_type(b->c.level, b->c.btree_id);
 734 }
 735
 736 const char *bch2_btree_node_type_str(enum btree_node_type);
 737
 738 #define BTREE_NODE_TYPE_HAS_TRANS_TRIGGERS              \
 739         (BIT_ULL(BKEY_TYPE_extents)|                    \
 740          BIT_ULL(BKEY_TYPE_alloc)|                      \
 741          BIT_ULL(BKEY_TYPE_inodes)|                     \
 742          BIT_ULL(BKEY_TYPE_stripes)|                    \
 743          BIT_ULL(BKEY_TYPE_reflink)|                    \
 744          BIT_ULL(BKEY_TYPE_subvolumes)|                 \
 745          BIT_ULL(BKEY_TYPE_btree))
 746
 747 #define BTREE_NODE_TYPE_HAS_ATOMIC_TRIGGERS             \
 748         (BIT_ULL(BKEY_TYPE_alloc)|                      \
 749          BIT_ULL(BKEY_TYPE_inodes)|                     \
 750          BIT_ULL(BKEY_TYPE_stripes)|                    \
 751          BIT_ULL(BKEY_TYPE_snapshots))
 752
 753 #define BTREE_NODE_TYPE_HAS_TRIGGERS                    \
 754         (BTREE_NODE_TYPE_HAS_TRANS_TRIGGERS|            \
 755          BTREE_NODE_TYPE_HAS_ATOMIC_TRIGGERS)
 756
 757 static inline bool btree_node_type_needs_gc(enum btree_node_type type)
 758 {
 759         return BTREE_NODE_TYPE_HAS_TRIGGERS & BIT_ULL(type);
 760 }
 761
 762 static inline bool btree_node_type_is_extents(enum btree_node_type type)
 763 {
 764         const unsigned mask = 0
 765 #define x(name, nr, flags, ...) |((!!((flags) & BTREE_ID_EXTENTS)) << (nr + 1))
 766         BCH_BTREE_IDS()
 767 #undef x
 768         ;
 769
 770         return (1U << type) & mask;
 771 }
 772
 773 static inline bool btree_id_is_extents(enum btree_id btree)
 774 {
 775         return btree_node_type_is_extents(__btree_node_type(0, btree));
 776 }
 777
 778 static inline bool btree_type_has_snapshots(enum btree_id id)
 779 {
 780         const unsigned mask = 0
 781 #define x(name, nr, flags, ...) |((!!((flags) & BTREE_ID_SNAPSHOTS)) << nr)
 782         BCH_BTREE_IDS()
 783 #undef x
 784         ;
 785
 786         return (1U << id) & mask;
 787 }
 788
 789 static inline bool btree_type_has_snapshot_field(enum btree_id id)
 790 {
 791         const unsigned mask = 0
 792 #define x(name, nr, flags, ...) |((!!((flags) & (BTREE_ID_SNAPSHOT_FIELD|BTREE_ID_SNAPSHOTS))) << nr)
 793         BCH_BTREE_IDS()
 794 #undef x
 795         ;
 796
 797         return (1U << id) & mask;
 798 }
 799
 800 static inline bool btree_type_has_ptrs(enum btree_id id)
 801 {
 802         const unsigned mask = 0
 803 #define x(name, nr, flags, ...) |((!!((flags) & BTREE_ID_DATA)) << nr)
 804         BCH_BTREE_IDS()
 805 #undef x
 806         ;
 807
 808         return (1U << id) & mask;
 809 }
 810
 811 struct btree_root {
 812         struct btree            *b;
 813
 814         /* On disk root - see async splits: */
 815         __BKEY_PADDED(key, BKEY_BTREE_PTR_VAL_U64s_MAX);
 816         u8                      level;
 817         u8                      alive;
 818         s16                     error;
 819 };
 820
 821 enum btree_gc_coalesce_fail_reason {
 822         BTREE_GC_COALESCE_FAIL_RESERVE_GET,
 823         BTREE_GC_COALESCE_FAIL_KEYLIST_REALLOC,
 824         BTREE_GC_COALESCE_FAIL_FORMAT_FITS,
 825 };
 826
 827 enum btree_node_sibling {
 828         btree_prev_sib,
 829         btree_next_sib,
 830 };
 831
 832 struct get_locks_fail {
 833         unsigned        l;
 834         struct btree    *b;
 835 };
 836
 837 #endif /* _BCACHEFS_BTREE_TYPES_H */