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;
 156
 157         /* Number of elements in live + freeable lists */
 158         unsigned                used;
 159         unsigned                reserve;
 160         atomic_t                dirty;
 161         struct shrinker         shrink;
 162
 163         /*
 164          * If we need to allocate memory for a new btree node and that
 165          * allocation fails, we can cannibalize another node in the btree cache
 166          * to satisfy the allocation - lock to guarantee only one thread does
 167          * this at a time:
 168          */
 169         struct task_struct      *alloc_lock;
 170         struct closure_waitlist alloc_wait;
 171 };
 172
 173 struct btree_node_iter {
 174         struct btree_node_iter_set {
 175                 u16     k, end;
 176         } data[MAX_BSETS];
 177 };
 178
 179 enum btree_iter_type {
 180         BTREE_ITER_KEYS,
 181         BTREE_ITER_NODES,
 182         BTREE_ITER_CACHED,
 183 };
 184
 185 #define BTREE_ITER_TYPE                 ((1 << 2) - 1)
 186
 187 /*
 188  * Iterate over all possible positions, synthesizing deleted keys for holes:
 189  */
 190 #define BTREE_ITER_SLOTS                (1 << 2)
 191 /*
 192  * Indicates that intent locks should be taken on leaf nodes, because we expect
 193  * to be doing updates:
 194  */
 195 #define BTREE_ITER_INTENT               (1 << 3)
 196 /*
 197  * Causes the btree iterator code to prefetch additional btree nodes from disk:
 198  */
 199 #define BTREE_ITER_PREFETCH             (1 << 4)
 200 /*
 201  * Indicates that this iterator should not be reused until transaction commit,
 202  * either because a pending update references it or because the update depends
 203  * on that particular key being locked (e.g. by the str_hash code, for hash
 204  * table consistency)
 205  */
 206 #define BTREE_ITER_KEEP_UNTIL_COMMIT    (1 << 5)
 207 /*
 208  * Used in bch2_btree_iter_traverse(), to indicate whether we're searching for
 209  * @pos or the first key strictly greater than @pos
 210  */
 211 #define BTREE_ITER_IS_EXTENTS           (1 << 6)
 212 #define BTREE_ITER_NOT_EXTENTS          (1 << 7)
 213 #define BTREE_ITER_ERROR                (1 << 8)
 214 #define BTREE_ITER_SET_POS_AFTER_COMMIT (1 << 9)
 215 #define BTREE_ITER_CACHED_NOFILL        (1 << 10)
 216 #define BTREE_ITER_CACHED_NOCREATE      (1 << 11)
 217 #define BTREE_ITER_WITH_UPDATES         (1 << 12)
 218 #define BTREE_ITER_ALL_SNAPSHOTS        (1 << 13)
 219
 220 enum btree_iter_uptodate {
 221         BTREE_ITER_UPTODATE             = 0,
 222         BTREE_ITER_NEED_PEEK            = 1,
 223         BTREE_ITER_NEED_RELOCK          = 2,
 224         BTREE_ITER_NEED_TRAVERSE        = 3,
 225 };
 226
 227 #define BTREE_ITER_NO_NODE_GET_LOCKS    ((struct btree *) 1)
 228 #define BTREE_ITER_NO_NODE_DROP         ((struct btree *) 2)
 229 #define BTREE_ITER_NO_NODE_LOCK_ROOT    ((struct btree *) 3)
 230 #define BTREE_ITER_NO_NODE_UP           ((struct btree *) 4)
 231 #define BTREE_ITER_NO_NODE_DOWN         ((struct btree *) 5)
 232 #define BTREE_ITER_NO_NODE_INIT         ((struct btree *) 6)
 233 #define BTREE_ITER_NO_NODE_ERROR        ((struct btree *) 7)
 234 #define BTREE_ITER_NO_NODE_CACHED       ((struct btree *) 8)
 235
 236 /*
 237  * @pos                 - iterator's current position
 238  * @level               - current btree depth
 239  * @locks_want          - btree level below which we start taking intent locks
 240  * @nodes_locked        - bitmask indicating which nodes in @nodes are locked
 241  * @nodes_intent_locked - bitmask indicating which locks are intent locks
 242  */
 243 struct btree_iter {
 244         struct btree_trans      *trans;
 245         unsigned long           ip_allocated;
 246
 247         u8                      idx;
 248         u8                      child_idx;
 249
 250         /* btree_iter_copy starts here: */
 251         u16                     flags;
 252
 253         /* When we're filtering by snapshot, the snapshot ID we're looking for: */
 254         unsigned                snapshot;
 255
 256         struct bpos             pos;
 257         struct bpos             real_pos;
 258         struct bpos             pos_after_commit;
 259
 260         enum btree_id           btree_id:4;
 261         enum btree_iter_uptodate uptodate:3;
 262         /*
 263          * True if we've returned a key (and thus are expected to keep it
 264          * locked), false after set_pos - for avoiding spurious transaction
 265          * restarts in bch2_trans_relock():
 266          */
 267         bool                    should_be_locked:1;
 268         unsigned                level:4,
 269                                 min_depth:4,
 270                                 locks_want:4,
 271                                 nodes_locked:4,
 272                                 nodes_intent_locked:4;
 273
 274         struct btree_iter_level {
 275                 struct btree    *b;
 276                 struct btree_node_iter iter;
 277                 u32             lock_seq;
 278         }                       l[BTREE_MAX_DEPTH];
 279
 280         /*
 281          * Current unpacked key - so that bch2_btree_iter_next()/
 282          * bch2_btree_iter_next_slot() can correctly advance pos.
 283          */
 284         struct bkey             k;
 285 };
 286
 287 static inline enum btree_iter_type
 288 btree_iter_type(const struct btree_iter *iter)
 289 {
 290         return iter->flags & BTREE_ITER_TYPE;
 291 }
 292
 293 static inline bool btree_iter_is_cached(const struct btree_iter *iter)
 294 {
 295         return btree_iter_type(iter) == BTREE_ITER_CACHED;
 296 }
 297
 298 static inline struct btree_iter_level *iter_l(struct btree_iter *iter)
 299 {
 300         return iter->l + iter->level;
 301 }
 302
 303 struct btree_key_cache {
 304         struct mutex            lock;
 305         struct rhashtable       table;
 306         bool                    table_init_done;
 307         struct list_head        freed;
 308         struct shrinker         shrink;
 309         unsigned                shrink_iter;
 310
 311         size_t                  nr_freed;
 312         atomic_long_t           nr_keys;
 313         atomic_long_t           nr_dirty;
 314 };
 315
 316 struct bkey_cached_key {
 317         u32                     btree_id;
 318         struct bpos             pos;
 319 } __attribute__((packed, aligned(4)));
 320
 321 #define BKEY_CACHED_ACCESSED            0
 322 #define BKEY_CACHED_DIRTY               1
 323
 324 struct bkey_cached {
 325         struct btree_bkey_cached_common c;
 326
 327         unsigned long           flags;
 328         u8                      u64s;
 329         bool                    valid;
 330         u32                     btree_trans_barrier_seq;
 331         struct bkey_cached_key  key;
 332
 333         struct rhash_head       hash;
 334         struct list_head        list;
 335
 336         struct journal_preres   res;
 337         struct journal_entry_pin journal;
 338
 339         struct bkey_i           *k;
 340 };
 341
 342 struct btree_insert_entry {
 343         unsigned                flags;
 344         u8                      bkey_type;
 345         enum btree_id           btree_id:8;
 346         u8                      level;
 347         unsigned                trans_triggers_run:1;
 348         struct bkey_i           *k;
 349         struct btree_iter       *iter;
 350 };
 351
 352 #ifndef CONFIG_LOCKDEP
 353 #define BTREE_ITER_MAX          64
 354 #else
 355 #define BTREE_ITER_MAX          32
 356 #endif
 357
 358 struct btree_trans_commit_hook;
 359 typedef int (btree_trans_commit_hook_fn)(struct btree_trans *, struct btree_trans_commit_hook *);
 360
 361 struct btree_trans_commit_hook {
 362         btree_trans_commit_hook_fn      *fn;
 363         struct btree_trans_commit_hook  *next;
 364 };
 365
 366 #define BTREE_TRANS_MEM_MAX     (1U << 14)
 367
 368 struct btree_trans {
 369         struct bch_fs           *c;
 370 #ifdef CONFIG_BCACHEFS_DEBUG
 371         struct list_head        list;
 372         struct btree            *locking;
 373         unsigned                locking_iter_idx;
 374         struct bpos             locking_pos;
 375         u8                      locking_btree_id;
 376         u8                      locking_level;
 377         pid_t                   pid;
 378 #endif
 379         unsigned long           ip;
 380         int                     srcu_idx;
 381
 382         u8                      nr_updates;
 383         unsigned                used_mempool:1;
 384         unsigned                error:1;
 385         unsigned                in_traverse_all:1;
 386         /*
 387          * For when bch2_trans_update notices we'll be splitting a compressed
 388          * extent:
 389          */
 390         unsigned                extra_journal_res;
 391
 392         u64                     iters_linked;
 393         u64                     iters_live;
 394         u64                     iters_touched;
 395
 396         unsigned                mem_top;
 397         unsigned                mem_bytes;
 398         void                    *mem;
 399
 400         struct btree_iter       *iters;
 401         struct btree_insert_entry *updates;
 402
 403         /* update path: */
 404         struct btree_trans_commit_hook *hooks;
 405         struct jset_entry       *extra_journal_entries;
 406         unsigned                extra_journal_entry_u64s;
 407         struct journal_entry_pin *journal_pin;
 408
 409         struct journal_res      journal_res;
 410         struct journal_preres   journal_preres;
 411         u64                     *journal_seq;
 412         struct disk_reservation *disk_res;
 413         unsigned                flags;
 414         unsigned                journal_u64s;
 415         unsigned                journal_preres_u64s;
 416         struct replicas_delta_list *fs_usage_deltas;
 417 };
 418
 419 #define BTREE_FLAG(flag)                                                \
 420 static inline bool btree_node_ ## flag(struct btree *b)                 \
 421 {       return test_bit(BTREE_NODE_ ## flag, &b->flags); }              \
 422                                                                         \
 423 static inline void set_btree_node_ ## flag(struct btree *b)             \
 424 {       set_bit(BTREE_NODE_ ## flag, &b->flags); }                      \
 425                                                                         \
 426 static inline void clear_btree_node_ ## flag(struct btree *b)           \
 427 {       clear_bit(BTREE_NODE_ ## flag, &b->flags); }
 428
 429 enum btree_flags {
 430         BTREE_NODE_read_in_flight,
 431         BTREE_NODE_read_error,
 432         BTREE_NODE_dirty,
 433         BTREE_NODE_need_write,
 434         BTREE_NODE_noevict,
 435         BTREE_NODE_write_idx,
 436         BTREE_NODE_accessed,
 437         BTREE_NODE_write_in_flight,
 438         BTREE_NODE_write_in_flight_inner,
 439         BTREE_NODE_just_written,
 440         BTREE_NODE_dying,
 441         BTREE_NODE_fake,
 442         BTREE_NODE_need_rewrite,
 443         BTREE_NODE_never_write,
 444 };
 445
 446 BTREE_FLAG(read_in_flight);
 447 BTREE_FLAG(read_error);
 448 BTREE_FLAG(need_write);
 449 BTREE_FLAG(noevict);
 450 BTREE_FLAG(write_idx);
 451 BTREE_FLAG(accessed);
 452 BTREE_FLAG(write_in_flight);
 453 BTREE_FLAG(write_in_flight_inner);
 454 BTREE_FLAG(just_written);
 455 BTREE_FLAG(dying);
 456 BTREE_FLAG(fake);
 457 BTREE_FLAG(need_rewrite);
 458 BTREE_FLAG(never_write);
 459
 460 static inline struct btree_write *btree_current_write(struct btree *b)
 461 {
 462         return b->writes + btree_node_write_idx(b);
 463 }
 464
 465 static inline struct btree_write *btree_prev_write(struct btree *b)
 466 {
 467         return b->writes + (btree_node_write_idx(b) ^ 1);
 468 }
 469
 470 static inline struct bset_tree *bset_tree_last(struct btree *b)
 471 {
 472         EBUG_ON(!b->nsets);
 473         return b->set + b->nsets - 1;
 474 }
 475
 476 static inline void *
 477 __btree_node_offset_to_ptr(const struct btree *b, u16 offset)
 478 {
 479         return (void *) ((u64 *) b->data + 1 + offset);
 480 }
 481
 482 static inline u16
 483 __btree_node_ptr_to_offset(const struct btree *b, const void *p)
 484 {
 485         u16 ret = (u64 *) p - 1 - (u64 *) b->data;
 486
 487         EBUG_ON(__btree_node_offset_to_ptr(b, ret) != p);
 488         return ret;
 489 }
 490
 491 static inline struct bset *bset(const struct btree *b,
 492                                 const struct bset_tree *t)
 493 {
 494         return __btree_node_offset_to_ptr(b, t->data_offset);
 495 }
 496
 497 static inline void set_btree_bset_end(struct btree *b, struct bset_tree *t)
 498 {
 499         t->end_offset =
 500                 __btree_node_ptr_to_offset(b, vstruct_last(bset(b, t)));
 501 }
 502
 503 static inline void set_btree_bset(struct btree *b, struct bset_tree *t,
 504                                   const struct bset *i)
 505 {
 506         t->data_offset = __btree_node_ptr_to_offset(b, i);
 507         set_btree_bset_end(b, t);
 508 }
 509
 510 static inline struct bset *btree_bset_first(struct btree *b)
 511 {
 512         return bset(b, b->set);
 513 }
 514
 515 static inline struct bset *btree_bset_last(struct btree *b)
 516 {
 517         return bset(b, bset_tree_last(b));
 518 }
 519
 520 static inline u16
 521 __btree_node_key_to_offset(const struct btree *b, const struct bkey_packed *k)
 522 {
 523         return __btree_node_ptr_to_offset(b, k);
 524 }
 525
 526 static inline struct bkey_packed *
 527 __btree_node_offset_to_key(const struct btree *b, u16 k)
 528 {
 529         return __btree_node_offset_to_ptr(b, k);
 530 }
 531
 532 static inline unsigned btree_bkey_first_offset(const struct bset_tree *t)
 533 {
 534         return t->data_offset + offsetof(struct bset, _data) / sizeof(u64);
 535 }
 536
 537 #define btree_bkey_first(_b, _t)                                        \
 538 ({                                                                      \
 539         EBUG_ON(bset(_b, _t)->start !=                                  \
 540                 __btree_node_offset_to_key(_b, btree_bkey_first_offset(_t)));\
 541                                                                         \
 542         bset(_b, _t)->start;                                            \
 543 })
 544
 545 #define btree_bkey_last(_b, _t)                                         \
 546 ({                                                                      \
 547         EBUG_ON(__btree_node_offset_to_key(_b, (_t)->end_offset) !=     \
 548                 vstruct_last(bset(_b, _t)));                            \
 549                                                                         \
 550         __btree_node_offset_to_key(_b, (_t)->end_offset);               \
 551 })
 552
 553 static inline unsigned bset_u64s(struct bset_tree *t)
 554 {
 555         return t->end_offset - t->data_offset -
 556                 sizeof(struct bset) / sizeof(u64);
 557 }
 558
 559 static inline unsigned bset_dead_u64s(struct btree *b, struct bset_tree *t)
 560 {
 561         return bset_u64s(t) - b->nr.bset_u64s[t - b->set];
 562 }
 563
 564 static inline unsigned bset_byte_offset(struct btree *b, void *i)
 565 {
 566         return i - (void *) b->data;
 567 }
 568
 569 enum btree_node_type {
 570 #define x(kwd, val) BKEY_TYPE_##kwd = val,
 571         BCH_BTREE_IDS()
 572 #undef x
 573         BKEY_TYPE_btree,
 574 };
 575
 576 /* Type of a key in btree @id at level @level: */
 577 static inline enum btree_node_type __btree_node_type(unsigned level, enum btree_id id)
 578 {
 579         return level ? BKEY_TYPE_btree : (enum btree_node_type) id;
 580 }
 581
 582 /* Type of keys @b contains: */
 583 static inline enum btree_node_type btree_node_type(struct btree *b)
 584 {
 585         return __btree_node_type(b->c.level, b->c.btree_id);
 586 }
 587
 588 static inline bool btree_node_type_is_extents(enum btree_node_type type)
 589 {
 590         switch (type) {
 591         case BKEY_TYPE_extents:
 592         case BKEY_TYPE_reflink:
 593                 return true;
 594         default:
 595                 return false;
 596         }
 597 }
 598
 599 static inline bool btree_node_is_extents(struct btree *b)
 600 {
 601         return btree_node_type_is_extents(btree_node_type(b));
 602 }
 603
 604 static inline enum btree_node_type btree_iter_key_type(struct btree_iter *iter)
 605 {
 606         return __btree_node_type(iter->level, iter->btree_id);
 607 }
 608
 609 static inline bool btree_iter_is_extents(struct btree_iter *iter)
 610 {
 611         return btree_node_type_is_extents(btree_iter_key_type(iter));
 612 }
 613
 614 #define BTREE_NODE_TYPE_HAS_TRANS_TRIGGERS              \
 615         ((1U << BKEY_TYPE_extents)|                     \
 616          (1U << BKEY_TYPE_inodes)|                      \
 617          (1U << BKEY_TYPE_stripes)|                     \
 618          (1U << BKEY_TYPE_reflink)|                     \
 619          (1U << BKEY_TYPE_btree))
 620
 621 #define BTREE_NODE_TYPE_HAS_MEM_TRIGGERS                \
 622         ((1U << BKEY_TYPE_alloc)|                       \
 623          (1U << BKEY_TYPE_stripes))
 624
 625 #define BTREE_NODE_TYPE_HAS_TRIGGERS                    \
 626         (BTREE_NODE_TYPE_HAS_TRANS_TRIGGERS|            \
 627          BTREE_NODE_TYPE_HAS_MEM_TRIGGERS)
 628
 629 #define BTREE_ID_HAS_SNAPSHOTS                          \
 630         ((1U << BTREE_ID_extents)|                      \
 631          (1U << BTREE_ID_inodes)|                       \
 632          (1U << BTREE_ID_dirents)|                      \
 633          (1U << BTREE_ID_xattrs))
 634
 635 #define BTREE_ID_HAS_PTRS                               \
 636         ((1U << BTREE_ID_extents)|                      \
 637          (1U << BTREE_ID_reflink))
 638
 639 static inline bool btree_type_has_snapshots(enum btree_id id)
 640 {
 641         return (1 << id) & BTREE_ID_HAS_SNAPSHOTS;
 642 }
 643
 644 enum btree_update_flags {
 645         __BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE,
 646
 647         __BTREE_TRIGGER_NORUN,          /* Don't run triggers at all */
 648
 649         __BTREE_TRIGGER_INSERT,
 650         __BTREE_TRIGGER_OVERWRITE,
 651
 652         __BTREE_TRIGGER_GC,
 653         __BTREE_TRIGGER_BUCKET_INVALIDATE,
 654         __BTREE_TRIGGER_NOATOMIC,
 655 };
 656
 657 #define BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE (1U << __BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE)
 658
 659 #define BTREE_TRIGGER_NORUN             (1U << __BTREE_TRIGGER_NORUN)
 660
 661 #define BTREE_TRIGGER_INSERT            (1U << __BTREE_TRIGGER_INSERT)
 662 #define BTREE_TRIGGER_OVERWRITE         (1U << __BTREE_TRIGGER_OVERWRITE)
 663
 664 #define BTREE_TRIGGER_GC                (1U << __BTREE_TRIGGER_GC)
 665 #define BTREE_TRIGGER_BUCKET_INVALIDATE (1U << __BTREE_TRIGGER_BUCKET_INVALIDATE)
 666 #define BTREE_TRIGGER_NOATOMIC          (1U << __BTREE_TRIGGER_NOATOMIC)
 667
 668 #define BTREE_TRIGGER_WANTS_OLD_AND_NEW         \
 669         ((1U << KEY_TYPE_stripe)|               \
 670          (1U << KEY_TYPE_inode))
 671
 672 static inline bool btree_node_type_needs_gc(enum btree_node_type type)
 673 {
 674         return BTREE_NODE_TYPE_HAS_TRIGGERS & (1U << type);
 675 }
 676
 677 struct btree_root {
 678         struct btree            *b;
 679
 680         /* On disk root - see async splits: */
 681         __BKEY_PADDED(key, BKEY_BTREE_PTR_VAL_U64s_MAX);
 682         u8                      level;
 683         u8                      alive;
 684         s8                      error;
 685 };
 686
 687 /*
 688  * Optional hook that will be called just prior to a btree node update, when
 689  * we're holding the write lock and we know what key is about to be overwritten:
 690  */
 691
 692 enum btree_insert_ret {
 693         BTREE_INSERT_OK,
 694         /* leaf node needs to be split */
 695         BTREE_INSERT_BTREE_NODE_FULL,
 696         BTREE_INSERT_NEED_MARK_REPLICAS,
 697         BTREE_INSERT_NEED_JOURNAL_RES,
 698         BTREE_INSERT_NEED_JOURNAL_RECLAIM,
 699 };
 700
 701 enum btree_gc_coalesce_fail_reason {
 702         BTREE_GC_COALESCE_FAIL_RESERVE_GET,
 703         BTREE_GC_COALESCE_FAIL_KEYLIST_REALLOC,
 704         BTREE_GC_COALESCE_FAIL_FORMAT_FITS,
 705 };
 706
 707 enum btree_node_sibling {
 708         btree_prev_sib,
 709         btree_next_sib,
 710 };
 711
 712 typedef struct btree_nr_keys (*sort_fix_overlapping_fn)(struct bset *,
 713                                                         struct btree *,
 714                                                         struct btree_node_iter *);
 715
 716 #endif /* _BCACHEFS_BTREE_TYPES_H */