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         u32                     alloc_seq;
 242         u32                     downgrade_seq;
 243
 244         /* btree_iter_copy starts here: */
 245         struct bpos             pos;
 246
 247         enum btree_id           btree_id:5;
 248         bool                    cached:1;
 249         bool                    preserve:1;
 250         enum btree_path_uptodate uptodate:2;
 251         /*
 252          * When true, failing to relock this path will cause the transaction to
 253          * restart:
 254          */
 255         bool                    should_be_locked:1;
 256         unsigned                level:3,
 257                                 locks_want:3;
 258         u8                      nodes_locked;
 259
 260         struct btree_path_level {
 261                 struct btree    *b;
 262                 struct btree_node_iter iter;
 263                 u32             lock_seq;
 264 #ifdef CONFIG_BCACHEFS_LOCK_TIME_STATS
 265                 u64             lock_taken_time;
 266 #endif
 267         }                       l[BTREE_MAX_DEPTH];
 268 #ifdef TRACK_PATH_ALLOCATED
 269         unsigned long           ip_allocated;
 270 #endif
 271 };
 272
 273 static inline struct btree_path_level *path_l(struct btree_path *path)
 274 {
 275         return path->l + path->level;
 276 }
 277
 278 static inline unsigned long btree_path_ip_allocated(struct btree_path *path)
 279 {
 280 #ifdef TRACK_PATH_ALLOCATED
 281         return path->ip_allocated;
 282 #else
 283         return _THIS_IP_;
 284 #endif
 285 }
 286
 287 /*
 288  * @pos                 - iterator's current position
 289  * @level               - current btree depth
 290  * @locks_want          - btree level below which we start taking intent locks
 291  * @nodes_locked        - bitmask indicating which nodes in @nodes are locked
 292  * @nodes_intent_locked - bitmask indicating which locks are intent locks
 293  */
 294 struct btree_iter {
 295         struct btree_trans      *trans;
 296         struct btree_path       *path;
 297         struct btree_path       *update_path;
 298         struct btree_path       *key_cache_path;
 299
 300         enum btree_id           btree_id:8;
 301         unsigned                min_depth:3;
 302         unsigned                advanced:1;
 303
 304         /* btree_iter_copy starts here: */
 305         u16                     flags;
 306
 307         /* When we're filtering by snapshot, the snapshot ID we're looking for: */
 308         unsigned                snapshot;
 309
 310         struct bpos             pos;
 311         /*
 312          * Current unpacked key - so that bch2_btree_iter_next()/
 313          * bch2_btree_iter_next_slot() can correctly advance pos.
 314          */
 315         struct bkey             k;
 316
 317         /* BTREE_ITER_WITH_JOURNAL: */
 318         size_t                  journal_idx;
 319         struct bpos             journal_pos;
 320 #ifdef TRACK_PATH_ALLOCATED
 321         unsigned long           ip_allocated;
 322 #endif
 323 };
 324
 325 struct btree_key_cache_freelist {
 326         struct bkey_cached      *objs[16];
 327         unsigned                nr;
 328 };
 329
 330 struct btree_key_cache {
 331         struct mutex            lock;
 332         struct rhashtable       table;
 333         bool                    table_init_done;
 334         struct list_head        freed_pcpu;
 335         struct list_head        freed_nonpcpu;
 336         struct shrinker         shrink;
 337         unsigned                shrink_iter;
 338         struct btree_key_cache_freelist __percpu *pcpu_freed;
 339
 340         atomic_long_t           nr_freed;
 341         atomic_long_t           nr_keys;
 342         atomic_long_t           nr_dirty;
 343 };
 344
 345 struct bkey_cached_key {
 346         u32                     btree_id;
 347         struct bpos             pos;
 348 } __packed __aligned(4);
 349
 350 #define BKEY_CACHED_ACCESSED            0
 351 #define BKEY_CACHED_DIRTY               1
 352
 353 struct bkey_cached {
 354         struct btree_bkey_cached_common c;
 355
 356         unsigned long           flags;
 357         u16                     u64s;
 358         bool                    valid;
 359         u32                     btree_trans_barrier_seq;
 360         struct bkey_cached_key  key;
 361
 362         struct rhash_head       hash;
 363         struct list_head        list;
 364
 365         struct journal_preres   res;
 366         struct journal_entry_pin journal;
 367         u64                     seq;
 368
 369         struct bkey_i           *k;
 370 };
 371
 372 static inline struct bpos btree_node_pos(struct btree_bkey_cached_common *b)
 373 {
 374         return !b->cached
 375                 ? container_of(b, struct btree, c)->key.k.p
 376                 : container_of(b, struct bkey_cached, c)->key.pos;
 377 }
 378
 379 struct btree_insert_entry {
 380         unsigned                flags;
 381         u8                      bkey_type;
 382         enum btree_id           btree_id:8;
 383         u8                      level:4;
 384         bool                    cached:1;
 385         bool                    insert_trigger_run:1;
 386         bool                    overwrite_trigger_run:1;
 387         bool                    key_cache_already_flushed:1;
 388         /*
 389          * @old_k may be a key from the journal; @old_btree_u64s always refers
 390          * to the size of the key being overwritten in the btree:
 391          */
 392         u8                      old_btree_u64s;
 393         struct bkey_i           *k;
 394         struct btree_path       *path;
 395         u64                     seq;
 396         /* key being overwritten: */
 397         struct bkey             old_k;
 398         const struct bch_val    *old_v;
 399         unsigned long           ip_allocated;
 400 };
 401
 402 #define BTREE_ITER_MAX          64
 403
 404 struct btree_trans_commit_hook;
 405 typedef int (btree_trans_commit_hook_fn)(struct btree_trans *, struct btree_trans_commit_hook *);
 406
 407 struct btree_trans_commit_hook {
 408         btree_trans_commit_hook_fn      *fn;
 409         struct btree_trans_commit_hook  *next;
 410 };
 411
 412 #define BTREE_TRANS_MEM_MAX     (1U << 16)
 413
 414 #define BTREE_TRANS_MAX_LOCK_HOLD_TIME_NS       10000
 415
 416 struct btree_trans {
 417         struct bch_fs           *c;
 418         const char              *fn;
 419         struct closure          ref;
 420         struct list_head        list;
 421         u64                     last_begin_time;
 422
 423         u8                      lock_may_not_fail;
 424         u8                      lock_must_abort;
 425         struct btree_bkey_cached_common *locking;
 426         struct six_lock_waiter  locking_wait;
 427
 428         int                     srcu_idx;
 429
 430         u8                      fn_idx;
 431         u8                      nr_sorted;
 432         u8                      nr_updates;
 433         u8                      nr_wb_updates;
 434         u8                      wb_updates_size;
 435         bool                    srcu_held:1;
 436         bool                    used_mempool:1;
 437         bool                    in_traverse_all:1;
 438         bool                    paths_sorted:1;
 439         bool                    memory_allocation_failure:1;
 440         bool                    journal_transaction_names:1;
 441         bool                    journal_replay_not_finished:1;
 442         bool                    is_initial_gc:1;
 443         bool                    notrace_relock_fail:1;
 444         enum bch_errcode        restarted:16;
 445         u32                     restart_count;
 446         unsigned long           last_begin_ip;
 447         unsigned long           last_restarted_ip;
 448         unsigned long           srcu_lock_time;
 449
 450         /*
 451          * For when bch2_trans_update notices we'll be splitting a compressed
 452          * extent:
 453          */
 454         unsigned                extra_journal_res;
 455         unsigned                nr_max_paths;
 456
 457         u64                     paths_allocated;
 458
 459         unsigned                mem_top;
 460         unsigned                mem_max;
 461         unsigned                mem_bytes;
 462         void                    *mem;
 463
 464         u8                      sorted[BTREE_ITER_MAX + 8];
 465         struct btree_path       paths[BTREE_ITER_MAX];
 466         struct btree_insert_entry updates[BTREE_ITER_MAX];
 467         struct btree_write_buffered_key *wb_updates;
 468
 469         /* update path: */
 470         struct btree_trans_commit_hook *hooks;
 471         darray_u64              extra_journal_entries;
 472         struct journal_entry_pin *journal_pin;
 473
 474         struct journal_res      journal_res;
 475         struct journal_preres   journal_preres;
 476         u64                     *journal_seq;
 477         struct disk_reservation *disk_res;
 478         unsigned                journal_u64s;
 479         unsigned                journal_preres_u64s;
 480         struct replicas_delta_list *fs_usage_deltas;
 481 };
 482
 483 #define BCH_BTREE_WRITE_TYPES()                                         \
 484         x(initial,              0)                                      \
 485         x(init_next_bset,       1)                                      \
 486         x(cache_reclaim,        2)                                      \
 487         x(journal_reclaim,      3)                                      \
 488         x(interior,             4)
 489
 490 enum btree_write_type {
 491 #define x(t, n) BTREE_WRITE_##t,
 492         BCH_BTREE_WRITE_TYPES()
 493 #undef x
 494         BTREE_WRITE_TYPE_NR,
 495 };
 496
 497 #define BTREE_WRITE_TYPE_MASK   (roundup_pow_of_two(BTREE_WRITE_TYPE_NR) - 1)
 498 #define BTREE_WRITE_TYPE_BITS   ilog2(roundup_pow_of_two(BTREE_WRITE_TYPE_NR))
 499
 500 #define BTREE_FLAGS()                                                   \
 501         x(read_in_flight)                                               \
 502         x(read_error)                                                   \
 503         x(dirty)                                                        \
 504         x(need_write)                                                   \
 505         x(write_blocked)                                                \
 506         x(will_make_reachable)                                          \
 507         x(noevict)                                                      \
 508         x(write_idx)                                                    \
 509         x(accessed)                                                     \
 510         x(write_in_flight)                                              \
 511         x(write_in_flight_inner)                                        \
 512         x(just_written)                                                 \
 513         x(dying)                                                        \
 514         x(fake)                                                         \
 515         x(need_rewrite)                                                 \
 516         x(never_write)
 517
 518 enum btree_flags {
 519         /* First bits for btree node write type */
 520         BTREE_NODE_FLAGS_START = BTREE_WRITE_TYPE_BITS - 1,
 521 #define x(flag) BTREE_NODE_##flag,
 522         BTREE_FLAGS()
 523 #undef x
 524 };
 525
 526 #define x(flag)                                                         \
 527 static inline bool btree_node_ ## flag(struct btree *b)                 \
 528 {       return test_bit(BTREE_NODE_ ## flag, &b->flags); }              \
 529                                                                         \
 530 static inline void set_btree_node_ ## flag(struct btree *b)             \
 531 {       set_bit(BTREE_NODE_ ## flag, &b->flags); }                      \
 532                                                                         \
 533 static inline void clear_btree_node_ ## flag(struct btree *b)           \
 534 {       clear_bit(BTREE_NODE_ ## flag, &b->flags); }
 535
 536 BTREE_FLAGS()
 537 #undef x
 538
 539 static inline struct btree_write *btree_current_write(struct btree *b)
 540 {
 541         return b->writes + btree_node_write_idx(b);
 542 }
 543
 544 static inline struct btree_write *btree_prev_write(struct btree *b)
 545 {
 546         return b->writes + (btree_node_write_idx(b) ^ 1);
 547 }
 548
 549 static inline struct bset_tree *bset_tree_last(struct btree *b)
 550 {
 551         EBUG_ON(!b->nsets);
 552         return b->set + b->nsets - 1;
 553 }
 554
 555 static inline void *
 556 __btree_node_offset_to_ptr(const struct btree *b, u16 offset)
 557 {
 558         return (void *) ((u64 *) b->data + 1 + offset);
 559 }
 560
 561 static inline u16
 562 __btree_node_ptr_to_offset(const struct btree *b, const void *p)
 563 {
 564         u16 ret = (u64 *) p - 1 - (u64 *) b->data;
 565
 566         EBUG_ON(__btree_node_offset_to_ptr(b, ret) != p);
 567         return ret;
 568 }
 569
 570 static inline struct bset *bset(const struct btree *b,
 571                                 const struct bset_tree *t)
 572 {
 573         return __btree_node_offset_to_ptr(b, t->data_offset);
 574 }
 575
 576 static inline void set_btree_bset_end(struct btree *b, struct bset_tree *t)
 577 {
 578         t->end_offset =
 579                 __btree_node_ptr_to_offset(b, vstruct_last(bset(b, t)));
 580 }
 581
 582 static inline void set_btree_bset(struct btree *b, struct bset_tree *t,
 583                                   const struct bset *i)
 584 {
 585         t->data_offset = __btree_node_ptr_to_offset(b, i);
 586         set_btree_bset_end(b, t);
 587 }
 588
 589 static inline struct bset *btree_bset_first(struct btree *b)
 590 {
 591         return bset(b, b->set);
 592 }
 593
 594 static inline struct bset *btree_bset_last(struct btree *b)
 595 {
 596         return bset(b, bset_tree_last(b));
 597 }
 598
 599 static inline u16
 600 __btree_node_key_to_offset(const struct btree *b, const struct bkey_packed *k)
 601 {
 602         return __btree_node_ptr_to_offset(b, k);
 603 }
 604
 605 static inline struct bkey_packed *
 606 __btree_node_offset_to_key(const struct btree *b, u16 k)
 607 {
 608         return __btree_node_offset_to_ptr(b, k);
 609 }
 610
 611 static inline unsigned btree_bkey_first_offset(const struct bset_tree *t)
 612 {
 613         return t->data_offset + offsetof(struct bset, _data) / sizeof(u64);
 614 }
 615
 616 #define btree_bkey_first(_b, _t)                                        \
 617 ({                                                                      \
 618         EBUG_ON(bset(_b, _t)->start !=                                  \
 619                 __btree_node_offset_to_key(_b, btree_bkey_first_offset(_t)));\
 620                                                                         \
 621         bset(_b, _t)->start;                                            \
 622 })
 623
 624 #define btree_bkey_last(_b, _t)                                         \
 625 ({                                                                      \
 626         EBUG_ON(__btree_node_offset_to_key(_b, (_t)->end_offset) !=     \
 627                 vstruct_last(bset(_b, _t)));                            \
 628                                                                         \
 629         __btree_node_offset_to_key(_b, (_t)->end_offset);               \
 630 })
 631
 632 static inline unsigned bset_u64s(struct bset_tree *t)
 633 {
 634         return t->end_offset - t->data_offset -
 635                 sizeof(struct bset) / sizeof(u64);
 636 }
 637
 638 static inline unsigned bset_dead_u64s(struct btree *b, struct bset_tree *t)
 639 {
 640         return bset_u64s(t) - b->nr.bset_u64s[t - b->set];
 641 }
 642
 643 static inline unsigned bset_byte_offset(struct btree *b, void *i)
 644 {
 645         return i - (void *) b->data;
 646 }
 647
 648 enum btree_node_type {
 649         BKEY_TYPE_btree,
 650 #define x(kwd, val, ...) BKEY_TYPE_##kwd = val + 1,
 651         BCH_BTREE_IDS()
 652 #undef x
 653         BKEY_TYPE_NR
 654 };
 655
 656 /* Type of a key in btree @id at level @level: */
 657 static inline enum btree_node_type __btree_node_type(unsigned level, enum btree_id id)
 658 {
 659         return level ? BKEY_TYPE_btree : (unsigned) id + 1;
 660 }
 661
 662 /* Type of keys @b contains: */
 663 static inline enum btree_node_type btree_node_type(struct btree *b)
 664 {
 665         return __btree_node_type(b->c.level, b->c.btree_id);
 666 }
 667
 668 const char *bch2_btree_node_type_str(enum btree_node_type);
 669
 670 #define BTREE_NODE_TYPE_HAS_TRANS_TRIGGERS              \
 671         (BIT_ULL(BKEY_TYPE_extents)|                    \
 672          BIT_ULL(BKEY_TYPE_alloc)|                      \
 673          BIT_ULL(BKEY_TYPE_inodes)|                     \
 674          BIT_ULL(BKEY_TYPE_stripes)|                    \
 675          BIT_ULL(BKEY_TYPE_reflink)|                    \
 676          BIT_ULL(BKEY_TYPE_btree))
 677
 678 #define BTREE_NODE_TYPE_HAS_MEM_TRIGGERS                \
 679         (BIT_ULL(BKEY_TYPE_alloc)|                      \
 680          BIT_ULL(BKEY_TYPE_inodes)|                     \
 681          BIT_ULL(BKEY_TYPE_stripes)|                    \
 682          BIT_ULL(BKEY_TYPE_snapshots))
 683
 684 #define BTREE_NODE_TYPE_HAS_TRIGGERS                    \
 685         (BTREE_NODE_TYPE_HAS_TRANS_TRIGGERS|            \
 686          BTREE_NODE_TYPE_HAS_MEM_TRIGGERS)
 687
 688 static inline bool btree_node_type_needs_gc(enum btree_node_type type)
 689 {
 690         return BTREE_NODE_TYPE_HAS_TRIGGERS & BIT_ULL(type);
 691 }
 692
 693 static inline bool btree_node_type_is_extents(enum btree_node_type type)
 694 {
 695         const unsigned mask = 0
 696 #define x(name, nr, flags, ...) |((!!((flags) & BTREE_ID_EXTENTS)) << (nr + 1))
 697         BCH_BTREE_IDS()
 698 #undef x
 699         ;
 700
 701         return (1U << type) & mask;
 702 }
 703
 704 static inline bool btree_id_is_extents(enum btree_id btree)
 705 {
 706         return btree_node_type_is_extents(__btree_node_type(0, btree));
 707 }
 708
 709 static inline bool btree_type_has_snapshots(enum btree_id id)
 710 {
 711         const unsigned mask = 0
 712 #define x(name, nr, flags, ...) |((!!((flags) & BTREE_ID_SNAPSHOTS)) << nr)
 713         BCH_BTREE_IDS()
 714 #undef x
 715         ;
 716
 717         return (1U << id) & mask;
 718 }
 719
 720 static inline bool btree_type_has_ptrs(enum btree_id id)
 721 {
 722         const unsigned mask = 0
 723 #define x(name, nr, flags, ...) |((!!((flags) & BTREE_ID_DATA)) << nr)
 724         BCH_BTREE_IDS()
 725 #undef x
 726         ;
 727
 728         return (1U << id) & mask;
 729 }
 730
 731 struct btree_root {
 732         struct btree            *b;
 733
 734         /* On disk root - see async splits: */
 735         __BKEY_PADDED(key, BKEY_BTREE_PTR_VAL_U64s_MAX);
 736         u8                      level;
 737         u8                      alive;
 738         s8                      error;
 739 };
 740
 741 enum btree_gc_coalesce_fail_reason {
 742         BTREE_GC_COALESCE_FAIL_RESERVE_GET,
 743         BTREE_GC_COALESCE_FAIL_KEYLIST_REALLOC,
 744         BTREE_GC_COALESCE_FAIL_FORMAT_FITS,
 745 };
 746
 747 enum btree_node_sibling {
 748         btree_prev_sib,
 749         btree_next_sib,
 750 };
 751
 752 #endif /* _BCACHEFS_BTREE_TYPES_H */