git.sesse.net Git - bcachefs-tools-debian/blob - libbcachefs/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 "bkey_methods.h"
   9 #include "buckets_types.h"
  10 #include "darray.h"
  11 #include "errcode.h"
  12 #include "journal_types.h"
  13 #include "replicas_types.h"
  14 #include "six.h"
  15
  16 struct open_bucket;
  17 struct btree_update;
  18 struct btree_trans;
  19
  20 #define MAX_BSETS               3U
  21
  22 struct btree_nr_keys {
  23
  24         /*
  25          * Amount of live metadata (i.e. size of node after a compaction) in
  26          * units of u64s
  27          */
  28         u16                     live_u64s;
  29         u16                     bset_u64s[MAX_BSETS];
  30
  31         /* live keys only: */
  32         u16                     packed_keys;
  33         u16                     unpacked_keys;
  34 };
  35
  36 struct bset_tree {
  37         /*
  38          * We construct a binary tree in an array as if the array
  39          * started at 1, so that things line up on the same cachelines
  40          * better: see comments in bset.c at cacheline_to_bkey() for
  41          * details
  42          */
  43
  44         /* size of the binary tree and prev array */
  45         u16                     size;
  46
  47         /* function of size - precalculated for to_inorder() */
  48         u16                     extra;
  49
  50         u16                     data_offset;
  51         u16                     aux_data_offset;
  52         u16                     end_offset;
  53 };
  54
  55 struct btree_write {
  56         struct journal_entry_pin        journal;
  57 };
  58
  59 struct btree_alloc {
  60         struct open_buckets     ob;
  61         __BKEY_PADDED(k, BKEY_BTREE_PTR_VAL_U64s_MAX);
  62 };
  63
  64 struct btree_bkey_cached_common {
  65         struct six_lock         lock;
  66         u8                      level;
  67         u8                      btree_id;
  68         bool                    cached;
  69 };
  70
  71 struct btree {
  72         struct btree_bkey_cached_common c;
  73
  74         struct rhash_head       hash;
  75         u64                     hash_val;
  76
  77         unsigned long           flags;
  78         u16                     written;
  79         u8                      nsets;
  80         u8                      nr_key_bits;
  81         u16                     version_ondisk;
  82
  83         struct bkey_format      format;
  84
  85         struct btree_node       *data;
  86         void                    *aux_data;
  87
  88         /*
  89          * Sets of sorted keys - the real btree node - plus a binary search tree
  90          *
  91          * set[0] is special; set[0]->tree, set[0]->prev and set[0]->data point
  92          * to the memory we have allocated for this btree node. Additionally,
  93          * set[0]->data points to the entire btree node as it exists on disk.
  94          */
  95         struct bset_tree        set[MAX_BSETS];
  96
  97         struct btree_nr_keys    nr;
  98         u16                     sib_u64s[2];
  99         u16                     whiteout_u64s;
 100         u8                      byte_order;
 101         u8                      unpack_fn_len;
 102
 103         struct btree_write      writes[2];
 104
 105         /* Key/pointer for this btree node */
 106         __BKEY_PADDED(key, BKEY_BTREE_PTR_VAL_U64s_MAX);
 107
 108         /*
 109          * XXX: add a delete sequence number, so when bch2_btree_node_relock()
 110          * fails because the lock sequence number has changed - i.e. the
 111          * contents were modified - we can still relock the node if it's still
 112          * the one we want, without redoing the traversal
 113          */
 114
 115         /*
 116          * For asynchronous splits/interior node updates:
 117          * When we do a split, we allocate new child nodes and update the parent
 118          * node to point to them: we update the parent in memory immediately,
 119          * but then we must wait until the children have been written out before
 120          * the update to the parent can be written - this is a list of the
 121          * btree_updates that are blocking this node from being
 122          * written:
 123          */
 124         struct list_head        write_blocked;
 125
 126         /*
 127          * Also for asynchronous splits/interior node updates:
 128          * If a btree node isn't reachable yet, we don't want to kick off
 129          * another write - because that write also won't yet be reachable and
 130          * marking it as completed before it's reachable would be incorrect:
 131          */
 132         unsigned long           will_make_reachable;
 133
 134         struct open_buckets     ob;
 135
 136         /* lru list */
 137         struct list_head        list;
 138 };
 139
 140 struct btree_cache {
 141         struct rhashtable       table;
 142         bool                    table_init_done;
 143         /*
 144          * We never free a struct btree, except on shutdown - we just put it on
 145          * the btree_cache_freed list and reuse it later. This simplifies the
 146          * code, and it doesn't cost us much memory as the memory usage is
 147          * dominated by buffers that hold the actual btree node data and those
 148          * can be freed - and the number of struct btrees allocated is
 149          * effectively bounded.
 150          *
 151          * btree_cache_freeable effectively is a small cache - we use it because
 152          * high order page allocations can be rather expensive, and it's quite
 153          * common to delete and allocate btree nodes in quick succession. It
 154          * should never grow past ~2-3 nodes in practice.
 155          */
 156         struct mutex            lock;
 157         struct list_head        live;
 158         struct list_head        freeable;
 159         struct list_head        freed_pcpu;
 160         struct list_head        freed_nonpcpu;
 161
 162         /* Number of elements in live + freeable lists */
 163         unsigned                used;
 164         unsigned                reserve;
 165         unsigned                freed;
 166         unsigned                not_freed_lock_intent;
 167         unsigned                not_freed_lock_write;
 168         unsigned                not_freed_dirty;
 169         unsigned                not_freed_read_in_flight;
 170         unsigned                not_freed_write_in_flight;
 171         unsigned                not_freed_noevict;
 172         unsigned                not_freed_write_blocked;
 173         unsigned                not_freed_will_make_reachable;
 174         unsigned                not_freed_access_bit;
 175         atomic_t                dirty;
 176         struct shrinker         shrink;
 177
 178         /*
 179          * If we need to allocate memory for a new btree node and that
 180          * allocation fails, we can cannibalize another node in the btree cache
 181          * to satisfy the allocation - lock to guarantee only one thread does
 182          * this at a time:
 183          */
 184         struct task_struct      *alloc_lock;
 185         struct closure_waitlist alloc_wait;
 186 };
 187
 188 struct btree_node_iter {
 189         struct btree_node_iter_set {
 190                 u16     k, end;
 191         } data[MAX_BSETS];
 192 };
 193
 194 /*
 195  * Iterate over all possible positions, synthesizing deleted keys for holes:
 196  */
 197 static const __maybe_unused u16 BTREE_ITER_SLOTS                = 1 << 0;
 198 static const __maybe_unused u16 BTREE_ITER_ALL_LEVELS           = 1 << 1;
 199 /*
 200  * Indicates that intent locks should be taken on leaf nodes, because we expect
 201  * to be doing updates:
 202  */
 203 static const __maybe_unused u16 BTREE_ITER_INTENT               = 1 << 2;
 204 /*
 205  * Causes the btree iterator code to prefetch additional btree nodes from disk:
 206  */
 207 static const __maybe_unused u16 BTREE_ITER_PREFETCH             = 1 << 3;
 208 /*
 209  * Used in bch2_btree_iter_traverse(), to indicate whether we're searching for
 210  * @pos or the first key strictly greater than @pos
 211  */
 212 static const __maybe_unused u16 BTREE_ITER_IS_EXTENTS           = 1 << 4;
 213 static const __maybe_unused u16 BTREE_ITER_NOT_EXTENTS          = 1 << 5;
 214 static const __maybe_unused u16 BTREE_ITER_CACHED               = 1 << 6;
 215 static const __maybe_unused u16 BTREE_ITER_WITH_KEY_CACHE       = 1 << 7;
 216 static const __maybe_unused u16 BTREE_ITER_WITH_UPDATES         = 1 << 8;
 217 static const __maybe_unused u16 BTREE_ITER_WITH_JOURNAL         = 1 << 9;
 218 static const __maybe_unused u16 __BTREE_ITER_ALL_SNAPSHOTS      = 1 << 10;
 219 static const __maybe_unused u16 BTREE_ITER_ALL_SNAPSHOTS        = 1 << 11;
 220 static const __maybe_unused u16 BTREE_ITER_FILTER_SNAPSHOTS     = 1 << 12;
 221 static const __maybe_unused u16 BTREE_ITER_NOPRESERVE           = 1 << 13;
 222 static const __maybe_unused u16 BTREE_ITER_CACHED_NOFILL        = 1 << 14;
 223 static const __maybe_unused u16 BTREE_ITER_KEY_CACHE_FILL       = 1 << 15;
 224 #define __BTREE_ITER_FLAGS_END                                         16
 225
 226 enum btree_path_uptodate {
 227         BTREE_ITER_UPTODATE             = 0,
 228         BTREE_ITER_NEED_RELOCK          = 1,
 229         BTREE_ITER_NEED_TRAVERSE        = 2,
 230 };
 231
 232 #if defined(CONFIG_BCACHEFS_LOCK_TIME_STATS) || defined(CONFIG_BCACHEFS_DEBUG)
 233 #define TRACK_PATH_ALLOCATED
 234 #endif
 235
 236 struct btree_path {
 237         u8                      idx;
 238         u8                      sorted_idx;
 239         u8                      ref;
 240         u8                      intent_ref;
 241
 242         /* btree_iter_copy starts here: */
 243         struct bpos             pos;
 244
 245         enum btree_id           btree_id:5;
 246         bool                    cached:1;
 247         bool                    preserve:1;
 248         enum btree_path_uptodate uptodate:2;
 249         /*
 250          * When true, failing to relock this path will cause the transaction to
 251          * restart:
 252          */
 253         bool                    should_be_locked:1;
 254         unsigned                level:3,
 255                                 locks_want:3;
 256         u8                      nodes_locked;
 257
 258         struct btree_path_level {
 259                 struct btree    *b;
 260                 struct btree_node_iter iter;
 261                 u32             lock_seq;
 262 #ifdef CONFIG_BCACHEFS_LOCK_TIME_STATS
 263                 u64             lock_taken_time;
 264 #endif
 265         }                       l[BTREE_MAX_DEPTH];
 266 #ifdef TRACK_PATH_ALLOCATED
 267         unsigned long           ip_allocated;
 268 #endif
 269 };
 270
 271 static inline struct btree_path_level *path_l(struct btree_path *path)
 272 {
 273         return path->l + path->level;
 274 }
 275
 276 static inline unsigned long btree_path_ip_allocated(struct btree_path *path)
 277 {
 278 #ifdef TRACK_PATH_ALLOCATED
 279         return path->ip_allocated;
 280 #else
 281         return _THIS_IP_;
 282 #endif
 283 }
 284
 285 /*
 286  * @pos                 - iterator's current position
 287  * @level               - current btree depth
 288  * @locks_want          - btree level below which we start taking intent locks
 289  * @nodes_locked        - bitmask indicating which nodes in @nodes are locked
 290  * @nodes_intent_locked - bitmask indicating which locks are intent locks
 291  */
 292 struct btree_iter {
 293         struct btree_trans      *trans;
 294         struct btree_path       *path;
 295         struct btree_path       *update_path;
 296         struct btree_path       *key_cache_path;
 297
 298         enum btree_id           btree_id:8;
 299         unsigned                min_depth:3;
 300         unsigned                advanced:1;
 301
 302         /* btree_iter_copy starts here: */
 303         u16                     flags;
 304
 305         /* When we're filtering by snapshot, the snapshot ID we're looking for: */
 306         unsigned                snapshot;
 307
 308         struct bpos             pos;
 309         /*
 310          * Current unpacked key - so that bch2_btree_iter_next()/
 311          * bch2_btree_iter_next_slot() can correctly advance pos.
 312          */
 313         struct bkey             k;
 314
 315         /* BTREE_ITER_WITH_JOURNAL: */
 316         size_t                  journal_idx;
 317         struct bpos             journal_pos;
 318 #ifdef TRACK_PATH_ALLOCATED
 319         unsigned long           ip_allocated;
 320 #endif
 321 };
 322
 323 struct btree_key_cache_freelist {
 324         struct bkey_cached      *objs[16];
 325         unsigned                nr;
 326 };
 327
 328 struct btree_key_cache {
 329         struct mutex            lock;
 330         struct rhashtable       table;
 331         bool                    table_init_done;
 332         struct list_head        freed_pcpu;
 333         struct list_head        freed_nonpcpu;
 334         struct shrinker         shrink;
 335         unsigned                shrink_iter;
 336         struct btree_key_cache_freelist __percpu *pcpu_freed;
 337
 338         atomic_long_t           nr_freed;
 339         atomic_long_t           nr_keys;
 340         atomic_long_t           nr_dirty;
 341 };
 342
 343 struct bkey_cached_key {
 344         u32                     btree_id;
 345         struct bpos             pos;
 346 } __packed __aligned(4);
 347
 348 #define BKEY_CACHED_ACCESSED            0
 349 #define BKEY_CACHED_DIRTY               1
 350
 351 struct bkey_cached {
 352         struct btree_bkey_cached_common c;
 353
 354         unsigned long           flags;
 355         u16                     u64s;
 356         bool                    valid;
 357         u32                     btree_trans_barrier_seq;
 358         struct bkey_cached_key  key;
 359
 360         struct rhash_head       hash;
 361         struct list_head        list;
 362
 363         struct journal_preres   res;
 364         struct journal_entry_pin journal;
 365         u64                     seq;
 366
 367         struct bkey_i           *k;
 368 };
 369
 370 static inline struct bpos btree_node_pos(struct btree_bkey_cached_common *b)
 371 {
 372         return !b->cached
 373                 ? container_of(b, struct btree, c)->key.k.p
 374                 : container_of(b, struct bkey_cached, c)->key.pos;
 375 }
 376
 377 struct btree_insert_entry {
 378         unsigned                flags;
 379         u8                      bkey_type;
 380         enum btree_id           btree_id:8;
 381         u8                      level:4;
 382         bool                    cached:1;
 383         bool                    insert_trigger_run:1;
 384         bool                    overwrite_trigger_run:1;
 385         bool                    key_cache_already_flushed:1;
 386         /*
 387          * @old_k may be a key from the journal; @old_btree_u64s always refers
 388          * to the size of the key being overwritten in the btree:
 389          */
 390         u8                      old_btree_u64s;
 391         struct bkey_i           *k;
 392         struct btree_path       *path;
 393         u64                     seq;
 394         /* key being overwritten: */
 395         struct bkey             old_k;
 396         const struct bch_val    *old_v;
 397         unsigned long           ip_allocated;
 398 };
 399
 400 #define BTREE_ITER_MAX          64
 401
 402 struct btree_trans_commit_hook;
 403 typedef int (btree_trans_commit_hook_fn)(struct btree_trans *, struct btree_trans_commit_hook *);
 404
 405 struct btree_trans_commit_hook {
 406         btree_trans_commit_hook_fn      *fn;
 407         struct btree_trans_commit_hook  *next;
 408 };
 409
 410 #define BTREE_TRANS_MEM_MAX     (1U << 16)
 411
 412 #define BTREE_TRANS_MAX_LOCK_HOLD_TIME_NS       10000
 413
 414 struct btree_trans {
 415         struct bch_fs           *c;
 416         const char              *fn;
 417         struct closure          ref;
 418         struct list_head        list;
 419         u64                     last_begin_time;
 420
 421         u8                      lock_may_not_fail;
 422         u8                      lock_must_abort;
 423         struct btree_bkey_cached_common *locking;
 424         struct six_lock_waiter  locking_wait;
 425
 426         int                     srcu_idx;
 427
 428         u8                      fn_idx;
 429         u8                      nr_sorted;
 430         u8                      nr_updates;
 431         u8                      nr_wb_updates;
 432         u8                      wb_updates_size;
 433         bool                    used_mempool:1;
 434         bool                    in_traverse_all:1;
 435         bool                    paths_sorted:1;
 436         bool                    memory_allocation_failure:1;
 437         bool                    journal_transaction_names:1;
 438         bool                    journal_replay_not_finished:1;
 439         bool                    is_initial_gc:1;
 440         bool                    notrace_relock_fail:1;
 441         enum bch_errcode        restarted:16;
 442         u32                     restart_count;
 443         unsigned long           last_begin_ip;
 444         unsigned long           last_restarted_ip;
 445         unsigned long           srcu_lock_time;
 446
 447         /*
 448          * For when bch2_trans_update notices we'll be splitting a compressed
 449          * extent:
 450          */
 451         unsigned                extra_journal_res;
 452         unsigned                nr_max_paths;
 453
 454         u64                     paths_allocated;
 455
 456         unsigned                mem_top;
 457         unsigned                mem_max;
 458         unsigned                mem_bytes;
 459         void                    *mem;
 460
 461         u8                      sorted[BTREE_ITER_MAX + 8];
 462         struct btree_path       paths[BTREE_ITER_MAX];
 463         struct btree_insert_entry updates[BTREE_ITER_MAX];
 464         struct btree_write_buffered_key *wb_updates;
 465
 466         /* update path: */
 467         struct btree_trans_commit_hook *hooks;
 468         darray_u64              extra_journal_entries;
 469         struct journal_entry_pin *journal_pin;
 470
 471         struct journal_res      journal_res;
 472         struct journal_preres   journal_preres;
 473         u64                     *journal_seq;
 474         struct disk_reservation *disk_res;
 475         unsigned                journal_u64s;
 476         unsigned                journal_preres_u64s;
 477         struct replicas_delta_list *fs_usage_deltas;
 478 };
 479
 480 #define BCH_BTREE_WRITE_TYPES()                                         \
 481         x(initial,              0)                                      \
 482         x(init_next_bset,       1)                                      \
 483         x(cache_reclaim,        2)                                      \
 484         x(journal_reclaim,      3)                                      \
 485         x(interior,             4)
 486
 487 enum btree_write_type {
 488 #define x(t, n) BTREE_WRITE_##t,
 489         BCH_BTREE_WRITE_TYPES()
 490 #undef x
 491         BTREE_WRITE_TYPE_NR,
 492 };
 493
 494 #define BTREE_WRITE_TYPE_MASK   (roundup_pow_of_two(BTREE_WRITE_TYPE_NR) - 1)
 495 #define BTREE_WRITE_TYPE_BITS   ilog2(roundup_pow_of_two(BTREE_WRITE_TYPE_NR))
 496
 497 #define BTREE_FLAGS()                                                   \
 498         x(read_in_flight)                                               \
 499         x(read_error)                                                   \
 500         x(dirty)                                                        \
 501         x(need_write)                                                   \
 502         x(write_blocked)                                                \
 503         x(will_make_reachable)                                          \
 504         x(noevict)                                                      \
 505         x(write_idx)                                                    \
 506         x(accessed)                                                     \
 507         x(write_in_flight)                                              \
 508         x(write_in_flight_inner)                                        \
 509         x(just_written)                                                 \
 510         x(dying)                                                        \
 511         x(fake)                                                         \
 512         x(need_rewrite)                                                 \
 513         x(never_write)
 514
 515 enum btree_flags {
 516         /* First bits for btree node write type */
 517         BTREE_NODE_FLAGS_START = BTREE_WRITE_TYPE_BITS - 1,
 518 #define x(flag) BTREE_NODE_##flag,
 519         BTREE_FLAGS()
 520 #undef x
 521 };
 522
 523 #define x(flag)                                                         \
 524 static inline bool btree_node_ ## flag(struct btree *b)                 \
 525 {       return test_bit(BTREE_NODE_ ## flag, &b->flags); }              \
 526                                                                         \
 527 static inline void set_btree_node_ ## flag(struct btree *b)             \
 528 {       set_bit(BTREE_NODE_ ## flag, &b->flags); }                      \
 529                                                                         \
 530 static inline void clear_btree_node_ ## flag(struct btree *b)           \
 531 {       clear_bit(BTREE_NODE_ ## flag, &b->flags); }
 532
 533 BTREE_FLAGS()
 534 #undef x
 535
 536 static inline struct btree_write *btree_current_write(struct btree *b)
 537 {
 538         return b->writes + btree_node_write_idx(b);
 539 }
 540
 541 static inline struct btree_write *btree_prev_write(struct btree *b)
 542 {
 543         return b->writes + (btree_node_write_idx(b) ^ 1);
 544 }
 545
 546 static inline struct bset_tree *bset_tree_last(struct btree *b)
 547 {
 548         EBUG_ON(!b->nsets);
 549         return b->set + b->nsets - 1;
 550 }
 551
 552 static inline void *
 553 __btree_node_offset_to_ptr(const struct btree *b, u16 offset)
 554 {
 555         return (void *) ((u64 *) b->data + 1 + offset);
 556 }
 557
 558 static inline u16
 559 __btree_node_ptr_to_offset(const struct btree *b, const void *p)
 560 {
 561         u16 ret = (u64 *) p - 1 - (u64 *) b->data;
 562
 563         EBUG_ON(__btree_node_offset_to_ptr(b, ret) != p);
 564         return ret;
 565 }
 566
 567 static inline struct bset *bset(const struct btree *b,
 568                                 const struct bset_tree *t)
 569 {
 570         return __btree_node_offset_to_ptr(b, t->data_offset);
 571 }
 572
 573 static inline void set_btree_bset_end(struct btree *b, struct bset_tree *t)
 574 {
 575         t->end_offset =
 576                 __btree_node_ptr_to_offset(b, vstruct_last(bset(b, t)));
 577 }
 578
 579 static inline void set_btree_bset(struct btree *b, struct bset_tree *t,
 580                                   const struct bset *i)
 581 {
 582         t->data_offset = __btree_node_ptr_to_offset(b, i);
 583         set_btree_bset_end(b, t);
 584 }
 585
 586 static inline struct bset *btree_bset_first(struct btree *b)
 587 {
 588         return bset(b, b->set);
 589 }
 590
 591 static inline struct bset *btree_bset_last(struct btree *b)
 592 {
 593         return bset(b, bset_tree_last(b));
 594 }
 595
 596 static inline u16
 597 __btree_node_key_to_offset(const struct btree *b, const struct bkey_packed *k)
 598 {
 599         return __btree_node_ptr_to_offset(b, k);
 600 }
 601
 602 static inline struct bkey_packed *
 603 __btree_node_offset_to_key(const struct btree *b, u16 k)
 604 {
 605         return __btree_node_offset_to_ptr(b, k);
 606 }
 607
 608 static inline unsigned btree_bkey_first_offset(const struct bset_tree *t)
 609 {
 610         return t->data_offset + offsetof(struct bset, _data) / sizeof(u64);
 611 }
 612
 613 #define btree_bkey_first(_b, _t)                                        \
 614 ({                                                                      \
 615         EBUG_ON(bset(_b, _t)->start !=                                  \
 616                 __btree_node_offset_to_key(_b, btree_bkey_first_offset(_t)));\
 617                                                                         \
 618         bset(_b, _t)->start;                                            \
 619 })
 620
 621 #define btree_bkey_last(_b, _t)                                         \
 622 ({                                                                      \
 623         EBUG_ON(__btree_node_offset_to_key(_b, (_t)->end_offset) !=     \
 624                 vstruct_last(bset(_b, _t)));                            \
 625                                                                         \
 626         __btree_node_offset_to_key(_b, (_t)->end_offset);               \
 627 })
 628
 629 static inline unsigned bset_u64s(struct bset_tree *t)
 630 {
 631         return t->end_offset - t->data_offset -
 632                 sizeof(struct bset) / sizeof(u64);
 633 }
 634
 635 static inline unsigned bset_dead_u64s(struct btree *b, struct bset_tree *t)
 636 {
 637         return bset_u64s(t) - b->nr.bset_u64s[t - b->set];
 638 }
 639
 640 static inline unsigned bset_byte_offset(struct btree *b, void *i)
 641 {
 642         return i - (void *) b->data;
 643 }
 644
 645 enum btree_node_type {
 646 #define x(kwd, val, ...) BKEY_TYPE_##kwd = val,
 647         BCH_BTREE_IDS()
 648 #undef x
 649         BKEY_TYPE_btree,
 650 };
 651
 652 /* Type of a key in btree @id at level @level: */
 653 static inline enum btree_node_type __btree_node_type(unsigned level, enum btree_id id)
 654 {
 655         return level ? BKEY_TYPE_btree : (enum btree_node_type) id;
 656 }
 657
 658 /* Type of keys @b contains: */
 659 static inline enum btree_node_type btree_node_type(struct btree *b)
 660 {
 661         return __btree_node_type(b->c.level, b->c.btree_id);
 662 }
 663
 664 #define BTREE_NODE_TYPE_HAS_TRANS_TRIGGERS              \
 665         (BIT(BKEY_TYPE_extents)|                        \
 666          BIT(BKEY_TYPE_alloc)|                          \
 667          BIT(BKEY_TYPE_inodes)|                         \
 668          BIT(BKEY_TYPE_stripes)|                        \
 669          BIT(BKEY_TYPE_reflink)|                        \
 670          BIT(BKEY_TYPE_btree))
 671
 672 #define BTREE_NODE_TYPE_HAS_MEM_TRIGGERS                \
 673         (BIT(BKEY_TYPE_alloc)|                          \
 674          BIT(BKEY_TYPE_inodes)|                         \
 675          BIT(BKEY_TYPE_stripes)|                        \
 676          BIT(BKEY_TYPE_snapshots))
 677
 678 #define BTREE_NODE_TYPE_HAS_TRIGGERS                    \
 679         (BTREE_NODE_TYPE_HAS_TRANS_TRIGGERS|            \
 680          BTREE_NODE_TYPE_HAS_MEM_TRIGGERS)
 681
 682 static inline bool btree_node_type_needs_gc(enum btree_node_type type)
 683 {
 684         return BTREE_NODE_TYPE_HAS_TRIGGERS & (1U << type);
 685 }
 686
 687 static inline bool btree_node_type_is_extents(enum btree_node_type type)
 688 {
 689         const unsigned mask = 0
 690 #define x(name, nr, flags, ...) |((!!((flags) & BTREE_ID_EXTENTS)) << nr)
 691         BCH_BTREE_IDS()
 692 #undef x
 693         ;
 694
 695         return (1U << type) & mask;
 696 }
 697
 698 static inline bool btree_id_is_extents(enum btree_id btree)
 699 {
 700         return btree_node_type_is_extents((enum btree_node_type) btree);
 701 }
 702
 703 static inline bool btree_type_has_snapshots(enum btree_id id)
 704 {
 705         const unsigned mask = 0
 706 #define x(name, nr, flags, ...) |((!!((flags) & BTREE_ID_SNAPSHOTS)) << nr)
 707         BCH_BTREE_IDS()
 708 #undef x
 709         ;
 710
 711         return (1U << id) & mask;
 712 }
 713
 714 static inline bool btree_type_has_ptrs(enum btree_id id)
 715 {
 716         const unsigned mask = 0
 717 #define x(name, nr, flags, ...) |((!!((flags) & BTREE_ID_DATA)) << nr)
 718         BCH_BTREE_IDS()
 719 #undef x
 720         ;
 721
 722         return (1U << id) & mask;
 723 }
 724
 725 struct btree_root {
 726         struct btree            *b;
 727
 728         /* On disk root - see async splits: */
 729         __BKEY_PADDED(key, BKEY_BTREE_PTR_VAL_U64s_MAX);
 730         u8                      level;
 731         u8                      alive;
 732         s8                      error;
 733 };
 734
 735 enum btree_gc_coalesce_fail_reason {
 736         BTREE_GC_COALESCE_FAIL_RESERVE_GET,
 737         BTREE_GC_COALESCE_FAIL_KEYLIST_REALLOC,
 738         BTREE_GC_COALESCE_FAIL_FORMAT_FITS,
 739 };
 740
 741 enum btree_node_sibling {
 742         btree_prev_sib,
 743         btree_next_sib,
 744 };
 745
 746 #endif /* _BCACHEFS_BTREE_TYPES_H */