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 #include <linux/six.h>
   8
   9 #include "bkey_methods.h"
  10 #include "buckets_types.h"
  11 #include "journal_types.h"
  12
  13 struct open_bucket;
  14 struct btree_update;
  15 struct btree_trans;
  16
  17 #define MAX_BSETS               3U
  18
  19 struct btree_nr_keys {
  20
  21         /*
  22          * Amount of live metadata (i.e. size of node after a compaction) in
  23          * units of u64s
  24          */
  25         u16                     live_u64s;
  26         u16                     bset_u64s[MAX_BSETS];
  27
  28         /* live keys only: */
  29         u16                     packed_keys;
  30         u16                     unpacked_keys;
  31 };
  32
  33 struct bset_tree {
  34         /*
  35          * We construct a binary tree in an array as if the array
  36          * started at 1, so that things line up on the same cachelines
  37          * better: see comments in bset.c at cacheline_to_bkey() for
  38          * details
  39          */
  40
  41         /* size of the binary tree and prev array */
  42         u16                     size;
  43
  44         /* function of size - precalculated for to_inorder() */
  45         u16                     extra;
  46
  47         u16                     data_offset;
  48         u16                     aux_data_offset;
  49         u16                     end_offset;
  50 };
  51
  52 struct btree_write {
  53         struct journal_entry_pin        journal;
  54 };
  55
  56 struct btree_alloc {
  57         struct open_buckets     ob;
  58         __BKEY_PADDED(k, BKEY_BTREE_PTR_VAL_U64s_MAX);
  59 };
  60
  61 struct btree_bkey_cached_common {
  62         struct six_lock         lock;
  63         u8                      level;
  64         u8                      btree_id;
  65 };
  66
  67 struct btree {
  68         struct btree_bkey_cached_common c;
  69
  70         struct rhash_head       hash;
  71         u64                     hash_val;
  72
  73         unsigned long           flags;
  74         u16                     written;
  75         u8                      nsets;
  76         u8                      nr_key_bits;
  77         u16                     version_ondisk;
  78
  79         struct bkey_format      format;
  80
  81         struct btree_node       *data;
  82         void                    *aux_data;
  83
  84         /*
  85          * Sets of sorted keys - the real btree node - plus a binary search tree
  86          *
  87          * set[0] is special; set[0]->tree, set[0]->prev and set[0]->data point
  88          * to the memory we have allocated for this btree node. Additionally,
  89          * set[0]->data points to the entire btree node as it exists on disk.
  90          */
  91         struct bset_tree        set[MAX_BSETS];
  92
  93         struct btree_nr_keys    nr;
  94         u16                     sib_u64s[2];
  95         u16                     whiteout_u64s;
  96         u8                      byte_order;
  97         u8                      unpack_fn_len;
  98
  99         struct btree_write      writes[2];
 100
 101         /* Key/pointer for this btree node */
 102         __BKEY_PADDED(key, BKEY_BTREE_PTR_VAL_U64s_MAX);
 103
 104         /*
 105          * XXX: add a delete sequence number, so when bch2_btree_node_relock()
 106          * fails because the lock sequence number has changed - i.e. the
 107          * contents were modified - we can still relock the node if it's still
 108          * the one we want, without redoing the traversal
 109          */
 110
 111         /*
 112          * For asynchronous splits/interior node updates:
 113          * When we do a split, we allocate new child nodes and update the parent
 114          * node to point to them: we update the parent in memory immediately,
 115          * but then we must wait until the children have been written out before
 116          * the update to the parent can be written - this is a list of the
 117          * btree_updates that are blocking this node from being
 118          * written:
 119          */
 120         struct list_head        write_blocked;
 121
 122         /*
 123          * Also for asynchronous splits/interior node updates:
 124          * If a btree node isn't reachable yet, we don't want to kick off
 125          * another write - because that write also won't yet be reachable and
 126          * marking it as completed before it's reachable would be incorrect:
 127          */
 128         unsigned long           will_make_reachable;
 129
 130         struct open_buckets     ob;
 131
 132         /* lru list */
 133         struct list_head        list;
 134 };
 135
 136 struct btree_cache {
 137         struct rhashtable       table;
 138         bool                    table_init_done;
 139         /*
 140          * We never free a struct btree, except on shutdown - we just put it on
 141          * the btree_cache_freed list and reuse it later. This simplifies the
 142          * code, and it doesn't cost us much memory as the memory usage is
 143          * dominated by buffers that hold the actual btree node data and those
 144          * can be freed - and the number of struct btrees allocated is
 145          * effectively bounded.
 146          *
 147          * btree_cache_freeable effectively is a small cache - we use it because
 148          * high order page allocations can be rather expensive, and it's quite
 149          * common to delete and allocate btree nodes in quick succession. It
 150          * should never grow past ~2-3 nodes in practice.
 151          */
 152         struct mutex            lock;
 153         struct list_head        live;
 154         struct list_head        freeable;
 155         struct list_head        freed_pcpu;
 156         struct list_head        freed_nonpcpu;
 157
 158         /* Number of elements in live + freeable lists */
 159         unsigned                used;
 160         unsigned                reserve;
 161         atomic_t                dirty;
 162         struct shrinker         shrink;
 163
 164         /*
 165          * If we need to allocate memory for a new btree node and that
 166          * allocation fails, we can cannibalize another node in the btree cache
 167          * to satisfy the allocation - lock to guarantee only one thread does
 168          * this at a time:
 169          */
 170         struct task_struct      *alloc_lock;
 171         struct closure_waitlist alloc_wait;
 172 };
 173
 174 struct btree_node_iter {
 175         struct btree_node_iter_set {
 176                 u16     k, end;
 177         } data[MAX_BSETS];
 178 };
 179
 180 /*
 181  * Iterate over all possible positions, synthesizing deleted keys for holes:
 182  */
 183 #define BTREE_ITER_SLOTS                (1 << 0)
 184 /*
 185  * Indicates that intent locks should be taken on leaf nodes, because we expect
 186  * to be doing updates:
 187  */
 188 #define BTREE_ITER_INTENT               (1 << 1)
 189 /*
 190  * Causes the btree iterator code to prefetch additional btree nodes from disk:
 191  */
 192 #define BTREE_ITER_PREFETCH             (1 << 2)
 193 /*
 194  * Indicates that this iterator should not be reused until transaction commit,
 195  * either because a pending update references it or because the update depends
 196  * on that particular key being locked (e.g. by the str_hash code, for hash
 197  * table consistency)
 198  */
 199 #define BTREE_ITER_KEEP_UNTIL_COMMIT    (1 << 3)
 200 /*
 201  * Used in bch2_btree_iter_traverse(), to indicate whether we're searching for
 202  * @pos or the first key strictly greater than @pos
 203  */
 204 #define BTREE_ITER_IS_EXTENTS           (1 << 4)
 205 #define BTREE_ITER_NOT_EXTENTS          (1 << 5)
 206 #define BTREE_ITER_CACHED               (1 << 6)
 207 #define BTREE_ITER_CACHED_NOFILL        (1 << 7)
 208 #define BTREE_ITER_CACHED_NOCREATE      (1 << 8)
 209 #define BTREE_ITER_WITH_KEY_CACHE       (1 << 9)
 210 #define BTREE_ITER_WITH_UPDATES         (1 << 10)
 211 #define BTREE_ITER_WITH_JOURNAL         (1 << 11)
 212 #define __BTREE_ITER_ALL_SNAPSHOTS      (1 << 12)
 213 #define BTREE_ITER_ALL_SNAPSHOTS        (1 << 13)
 214 #define BTREE_ITER_FILTER_SNAPSHOTS     (1 << 14)
 215 #define BTREE_ITER_NOPRESERVE           (1 << 15)
 216
 217 enum btree_path_uptodate {
 218         BTREE_ITER_UPTODATE             = 0,
 219         BTREE_ITER_NEED_RELOCK          = 1,
 220         BTREE_ITER_NEED_TRAVERSE        = 2,
 221 };
 222
 223 #define BTREE_ITER_NO_NODE_GET_LOCKS    ((struct btree *) 1)
 224 #define BTREE_ITER_NO_NODE_DROP         ((struct btree *) 2)
 225 #define BTREE_ITER_NO_NODE_LOCK_ROOT    ((struct btree *) 3)
 226 #define BTREE_ITER_NO_NODE_UP           ((struct btree *) 4)
 227 #define BTREE_ITER_NO_NODE_DOWN         ((struct btree *) 5)
 228 #define BTREE_ITER_NO_NODE_INIT         ((struct btree *) 6)
 229 #define BTREE_ITER_NO_NODE_ERROR        ((struct btree *) 7)
 230 #define BTREE_ITER_NO_NODE_CACHED       ((struct btree *) 8)
 231
 232 struct btree_path {
 233         u8                      idx;
 234         u8                      sorted_idx;
 235         u8                      ref;
 236         u8                      intent_ref;
 237
 238         /* btree_iter_copy starts here: */
 239         struct bpos             pos;
 240
 241         enum btree_id           btree_id:4;
 242         bool                    cached:1;
 243         bool                    preserve:1;
 244         enum btree_path_uptodate uptodate:2;
 245         /*
 246          * When true, failing to relock this path will cause the transaction to
 247          * restart:
 248          */
 249         bool                    should_be_locked:1;
 250         unsigned                level:3,
 251                                 locks_want:4,
 252                                 nodes_locked:4,
 253                                 nodes_intent_locked:4;
 254
 255         struct btree_path_level {
 256                 struct btree    *b;
 257                 struct btree_node_iter iter;
 258                 u32             lock_seq;
 259         }                       l[BTREE_MAX_DEPTH];
 260 #ifdef CONFIG_BCACHEFS_DEBUG
 261         unsigned long           ip_allocated;
 262 #endif
 263 };
 264
 265 static inline struct btree_path_level *path_l(struct btree_path *path)
 266 {
 267         return path->l + path->level;
 268 }
 269
 270 /*
 271  * @pos                 - iterator's current position
 272  * @level               - current btree depth
 273  * @locks_want          - btree level below which we start taking intent locks
 274  * @nodes_locked        - bitmask indicating which nodes in @nodes are locked
 275  * @nodes_intent_locked - bitmask indicating which locks are intent locks
 276  */
 277 struct btree_iter {
 278         struct btree_trans      *trans;
 279         struct btree_path       *path;
 280         struct btree_path       *update_path;
 281         struct btree_path       *key_cache_path;
 282
 283         enum btree_id           btree_id:4;
 284         unsigned                min_depth:4;
 285
 286         /* btree_iter_copy starts here: */
 287         u16                     flags;
 288
 289         /* When we're filtering by snapshot, the snapshot ID we're looking for: */
 290         unsigned                snapshot;
 291
 292         struct bpos             pos;
 293         struct bpos             pos_after_commit;
 294         /*
 295          * Current unpacked key - so that bch2_btree_iter_next()/
 296          * bch2_btree_iter_next_slot() can correctly advance pos.
 297          */
 298         struct bkey             k;
 299 #ifdef CONFIG_BCACHEFS_DEBUG
 300         unsigned long           ip_allocated;
 301 #endif
 302 };
 303
 304 struct btree_key_cache {
 305         struct mutex            lock;
 306         struct rhashtable       table;
 307         bool                    table_init_done;
 308         struct list_head        freed;
 309         struct shrinker         shrink;
 310         unsigned                shrink_iter;
 311
 312         size_t                  nr_freed;
 313         atomic_long_t           nr_keys;
 314         atomic_long_t           nr_dirty;
 315 };
 316
 317 struct bkey_cached_key {
 318         u32                     btree_id;
 319         struct bpos             pos;
 320 } __attribute__((packed, aligned(4)));
 321
 322 #define BKEY_CACHED_ACCESSED            0
 323 #define BKEY_CACHED_DIRTY               1
 324
 325 struct bkey_cached {
 326         struct btree_bkey_cached_common c;
 327
 328         unsigned long           flags;
 329         u8                      u64s;
 330         bool                    valid;
 331         u32                     btree_trans_barrier_seq;
 332         struct bkey_cached_key  key;
 333
 334         struct rhash_head       hash;
 335         struct list_head        list;
 336
 337         struct journal_preres   res;
 338         struct journal_entry_pin journal;
 339
 340         struct bkey_i           *k;
 341 };
 342
 343 struct btree_insert_entry {
 344         unsigned                flags;
 345         u8                      bkey_type;
 346         enum btree_id           btree_id:8;
 347         u8                      level:4;
 348         bool                    cached:1;
 349         bool                    insert_trigger_run:1;
 350         bool                    overwrite_trigger_run:1;
 351         /*
 352          * @old_k may be a key from the journal; @old_btree_u64s always refers
 353          * to the size of the key being overwritten in the btree:
 354          */
 355         u8                      old_btree_u64s;
 356         struct bkey_i           *k;
 357         struct btree_path       *path;
 358         /* key being overwritten: */
 359         struct bkey             old_k;
 360         const struct bch_val    *old_v;
 361         unsigned long           ip_allocated;
 362 };
 363
 364 #ifndef CONFIG_LOCKDEP
 365 #define BTREE_ITER_MAX          64
 366 #else
 367 #define BTREE_ITER_MAX          32
 368 #endif
 369
 370 struct btree_trans_commit_hook;
 371 typedef int (btree_trans_commit_hook_fn)(struct btree_trans *, struct btree_trans_commit_hook *);
 372
 373 struct btree_trans_commit_hook {
 374         btree_trans_commit_hook_fn      *fn;
 375         struct btree_trans_commit_hook  *next;
 376 };
 377
 378 #define BTREE_TRANS_MEM_MAX     (1U << 14)
 379
 380 struct btree_trans {
 381         struct bch_fs           *c;
 382         const char              *fn;
 383         struct list_head        list;
 384         struct btree            *locking;
 385         unsigned                locking_path_idx;
 386         struct bpos             locking_pos;
 387         u8                      locking_btree_id;
 388         u8                      locking_level;
 389         u8                      locking_lock_type;
 390         pid_t                   pid;
 391         int                     srcu_idx;
 392
 393         u8                      nr_sorted;
 394         u8                      nr_updates;
 395         u8                      traverse_all_idx;
 396         bool                    used_mempool:1;
 397         bool                    in_traverse_all:1;
 398         bool                    restarted:1;
 399         bool                    memory_allocation_failure:1;
 400         bool                    journal_transaction_names:1;
 401         bool                    is_initial_gc:1;
 402         /*
 403          * For when bch2_trans_update notices we'll be splitting a compressed
 404          * extent:
 405          */
 406         unsigned                extra_journal_res;
 407
 408         u64                     paths_allocated;
 409
 410         unsigned                mem_top;
 411         unsigned                mem_bytes;
 412         void                    *mem;
 413
 414         u8                      sorted[BTREE_ITER_MAX];
 415         struct btree_path       *paths;
 416         struct btree_insert_entry *updates;
 417
 418         /* update path: */
 419         struct btree_trans_commit_hook *hooks;
 420         struct jset_entry       *extra_journal_entries;
 421         unsigned                extra_journal_entry_u64s;
 422         struct journal_entry_pin *journal_pin;
 423
 424         struct journal_res      journal_res;
 425         struct journal_preres   journal_preres;
 426         u64                     *journal_seq;
 427         struct disk_reservation *disk_res;
 428         unsigned                flags;
 429         unsigned                journal_u64s;
 430         unsigned                journal_preres_u64s;
 431         struct replicas_delta_list *fs_usage_deltas;
 432 };
 433
 434 #define BTREE_FLAGS()                                                   \
 435         x(read_in_flight)                                               \
 436         x(read_error)                                                   \
 437         x(dirty)                                                        \
 438         x(need_write)                                                   \
 439         x(write_blocked)                                                \
 440         x(will_make_reachable)                                          \
 441         x(noevict)                                                      \
 442         x(write_idx)                                                    \
 443         x(accessed)                                                     \
 444         x(write_in_flight)                                              \
 445         x(write_in_flight_inner)                                        \
 446         x(just_written)                                                 \
 447         x(dying)                                                        \
 448         x(fake)                                                         \
 449         x(need_rewrite)                                                 \
 450         x(never_write)
 451
 452 enum btree_flags {
 453 #define x(flag) BTREE_NODE_##flag,
 454         BTREE_FLAGS()
 455 #undef x
 456 };
 457
 458 #define x(flag)                                                         \
 459 static inline bool btree_node_ ## flag(struct btree *b)                 \
 460 {       return test_bit(BTREE_NODE_ ## flag, &b->flags); }              \
 461                                                                         \
 462 static inline void set_btree_node_ ## flag(struct btree *b)             \
 463 {       set_bit(BTREE_NODE_ ## flag, &b->flags); }                      \
 464                                                                         \
 465 static inline void clear_btree_node_ ## flag(struct btree *b)           \
 466 {       clear_bit(BTREE_NODE_ ## flag, &b->flags); }
 467
 468 BTREE_FLAGS()
 469 #undef x
 470
 471 static inline struct btree_write *btree_current_write(struct btree *b)
 472 {
 473         return b->writes + btree_node_write_idx(b);
 474 }
 475
 476 static inline struct btree_write *btree_prev_write(struct btree *b)
 477 {
 478         return b->writes + (btree_node_write_idx(b) ^ 1);
 479 }
 480
 481 static inline struct bset_tree *bset_tree_last(struct btree *b)
 482 {
 483         EBUG_ON(!b->nsets);
 484         return b->set + b->nsets - 1;
 485 }
 486
 487 static inline void *
 488 __btree_node_offset_to_ptr(const struct btree *b, u16 offset)
 489 {
 490         return (void *) ((u64 *) b->data + 1 + offset);
 491 }
 492
 493 static inline u16
 494 __btree_node_ptr_to_offset(const struct btree *b, const void *p)
 495 {
 496         u16 ret = (u64 *) p - 1 - (u64 *) b->data;
 497
 498         EBUG_ON(__btree_node_offset_to_ptr(b, ret) != p);
 499         return ret;
 500 }
 501
 502 static inline struct bset *bset(const struct btree *b,
 503                                 const struct bset_tree *t)
 504 {
 505         return __btree_node_offset_to_ptr(b, t->data_offset);
 506 }
 507
 508 static inline void set_btree_bset_end(struct btree *b, struct bset_tree *t)
 509 {
 510         t->end_offset =
 511                 __btree_node_ptr_to_offset(b, vstruct_last(bset(b, t)));
 512 }
 513
 514 static inline void set_btree_bset(struct btree *b, struct bset_tree *t,
 515                                   const struct bset *i)
 516 {
 517         t->data_offset = __btree_node_ptr_to_offset(b, i);
 518         set_btree_bset_end(b, t);
 519 }
 520
 521 static inline struct bset *btree_bset_first(struct btree *b)
 522 {
 523         return bset(b, b->set);
 524 }
 525
 526 static inline struct bset *btree_bset_last(struct btree *b)
 527 {
 528         return bset(b, bset_tree_last(b));
 529 }
 530
 531 static inline u16
 532 __btree_node_key_to_offset(const struct btree *b, const struct bkey_packed *k)
 533 {
 534         return __btree_node_ptr_to_offset(b, k);
 535 }
 536
 537 static inline struct bkey_packed *
 538 __btree_node_offset_to_key(const struct btree *b, u16 k)
 539 {
 540         return __btree_node_offset_to_ptr(b, k);
 541 }
 542
 543 static inline unsigned btree_bkey_first_offset(const struct bset_tree *t)
 544 {
 545         return t->data_offset + offsetof(struct bset, _data) / sizeof(u64);
 546 }
 547
 548 #define btree_bkey_first(_b, _t)                                        \
 549 ({                                                                      \
 550         EBUG_ON(bset(_b, _t)->start !=                                  \
 551                 __btree_node_offset_to_key(_b, btree_bkey_first_offset(_t)));\
 552                                                                         \
 553         bset(_b, _t)->start;                                            \
 554 })
 555
 556 #define btree_bkey_last(_b, _t)                                         \
 557 ({                                                                      \
 558         EBUG_ON(__btree_node_offset_to_key(_b, (_t)->end_offset) !=     \
 559                 vstruct_last(bset(_b, _t)));                            \
 560                                                                         \
 561         __btree_node_offset_to_key(_b, (_t)->end_offset);               \
 562 })
 563
 564 static inline unsigned bset_u64s(struct bset_tree *t)
 565 {
 566         return t->end_offset - t->data_offset -
 567                 sizeof(struct bset) / sizeof(u64);
 568 }
 569
 570 static inline unsigned bset_dead_u64s(struct btree *b, struct bset_tree *t)
 571 {
 572         return bset_u64s(t) - b->nr.bset_u64s[t - b->set];
 573 }
 574
 575 static inline unsigned bset_byte_offset(struct btree *b, void *i)
 576 {
 577         return i - (void *) b->data;
 578 }
 579
 580 enum btree_node_type {
 581 #define x(kwd, val) BKEY_TYPE_##kwd = val,
 582         BCH_BTREE_IDS()
 583 #undef x
 584         BKEY_TYPE_btree,
 585 };
 586
 587 /* Type of a key in btree @id at level @level: */
 588 static inline enum btree_node_type __btree_node_type(unsigned level, enum btree_id id)
 589 {
 590         return level ? BKEY_TYPE_btree : (enum btree_node_type) id;
 591 }
 592
 593 /* Type of keys @b contains: */
 594 static inline enum btree_node_type btree_node_type(struct btree *b)
 595 {
 596         return __btree_node_type(b->c.level, b->c.btree_id);
 597 }
 598
 599 static inline bool btree_node_type_is_extents(enum btree_node_type type)
 600 {
 601         switch (type) {
 602         case BKEY_TYPE_extents:
 603         case BKEY_TYPE_reflink:
 604                 return true;
 605         default:
 606                 return false;
 607         }
 608 }
 609
 610 static inline bool btree_node_is_extents(struct btree *b)
 611 {
 612         return btree_node_type_is_extents(btree_node_type(b));
 613 }
 614
 615 #define BTREE_NODE_TYPE_HAS_TRANS_TRIGGERS              \
 616         ((1U << BKEY_TYPE_extents)|                     \
 617          (1U << BKEY_TYPE_inodes)|                      \
 618          (1U << BKEY_TYPE_stripes)|                     \
 619          (1U << BKEY_TYPE_reflink)|                     \
 620          (1U << BKEY_TYPE_btree))
 621
 622 #define BTREE_NODE_TYPE_HAS_MEM_TRIGGERS                \
 623         ((1U << BKEY_TYPE_alloc)|                       \
 624          (1U << BKEY_TYPE_inodes)|                      \
 625          (1U << BKEY_TYPE_stripes)|                     \
 626          (1U << BKEY_TYPE_snapshots))
 627
 628 #define BTREE_NODE_TYPE_HAS_TRIGGERS                    \
 629         (BTREE_NODE_TYPE_HAS_TRANS_TRIGGERS|            \
 630          BTREE_NODE_TYPE_HAS_MEM_TRIGGERS)
 631
 632 #define BTREE_ID_HAS_SNAPSHOTS                          \
 633         ((1U << BTREE_ID_extents)|                      \
 634          (1U << BTREE_ID_inodes)|                       \
 635          (1U << BTREE_ID_dirents)|                      \
 636          (1U << BTREE_ID_xattrs))
 637
 638 #define BTREE_ID_HAS_PTRS                               \
 639         ((1U << BTREE_ID_extents)|                      \
 640          (1U << BTREE_ID_reflink))
 641
 642 static inline bool btree_type_has_snapshots(enum btree_id id)
 643 {
 644         return (1 << id) & BTREE_ID_HAS_SNAPSHOTS;
 645 }
 646
 647 enum btree_update_flags {
 648         __BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE,
 649         __BTREE_UPDATE_KEY_CACHE_RECLAIM,
 650
 651         __BTREE_TRIGGER_NORUN,          /* Don't run triggers at all */
 652
 653         __BTREE_TRIGGER_INSERT,
 654         __BTREE_TRIGGER_OVERWRITE,
 655
 656         __BTREE_TRIGGER_GC,
 657         __BTREE_TRIGGER_BUCKET_INVALIDATE,
 658         __BTREE_TRIGGER_NOATOMIC,
 659 };
 660
 661 #define BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE (1U << __BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE)
 662 #define BTREE_UPDATE_KEY_CACHE_RECLAIM  (1U << __BTREE_UPDATE_KEY_CACHE_RECLAIM)
 663
 664 #define BTREE_TRIGGER_NORUN             (1U << __BTREE_TRIGGER_NORUN)
 665
 666 #define BTREE_TRIGGER_INSERT            (1U << __BTREE_TRIGGER_INSERT)
 667 #define BTREE_TRIGGER_OVERWRITE         (1U << __BTREE_TRIGGER_OVERWRITE)
 668
 669 #define BTREE_TRIGGER_GC                (1U << __BTREE_TRIGGER_GC)
 670 #define BTREE_TRIGGER_BUCKET_INVALIDATE (1U << __BTREE_TRIGGER_BUCKET_INVALIDATE)
 671 #define BTREE_TRIGGER_NOATOMIC          (1U << __BTREE_TRIGGER_NOATOMIC)
 672
 673 #define BTREE_TRIGGER_WANTS_OLD_AND_NEW         \
 674         ((1U << KEY_TYPE_alloc)|                \
 675          (1U << KEY_TYPE_alloc_v2)|             \
 676          (1U << KEY_TYPE_alloc_v3)|             \
 677          (1U << KEY_TYPE_stripe)|               \
 678          (1U << KEY_TYPE_inode)|                \
 679          (1U << KEY_TYPE_inode_v2)|             \
 680          (1U << KEY_TYPE_snapshot))
 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 struct btree_root {
 688         struct btree            *b;
 689
 690         /* On disk root - see async splits: */
 691         __BKEY_PADDED(key, BKEY_BTREE_PTR_VAL_U64s_MAX);
 692         u8                      level;
 693         u8                      alive;
 694         s8                      error;
 695 };
 696
 697 enum btree_insert_ret {
 698         BTREE_INSERT_OK,
 699         /* leaf node needs to be split */
 700         BTREE_INSERT_BTREE_NODE_FULL,
 701         BTREE_INSERT_NEED_MARK_REPLICAS,
 702         BTREE_INSERT_NEED_JOURNAL_RES,
 703         BTREE_INSERT_NEED_JOURNAL_RECLAIM,
 704 };
 705
 706 enum btree_gc_coalesce_fail_reason {
 707         BTREE_GC_COALESCE_FAIL_RESERVE_GET,
 708         BTREE_GC_COALESCE_FAIL_KEYLIST_REALLOC,
 709         BTREE_GC_COALESCE_FAIL_FORMAT_FITS,
 710 };
 711
 712 enum btree_node_sibling {
 713         btree_prev_sib,
 714         btree_next_sib,
 715 };
 716
 717 #endif /* _BCACHEFS_BTREE_TYPES_H */