1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _BCACHEFS_BTREE_TYPES_H
3 #define _BCACHEFS_BTREE_TYPES_H
5 #include <linux/list.h>
6 #include <linux/rhashtable.h>
8 #include "btree_key_cache_types.h"
9 #include "buckets_types.h"
12 #include "journal_types.h"
13 #include "replicas_types.h"
22 struct btree_nr_keys {
25 * Amount of live metadata (i.e. size of node after a compaction) in
29 u16 bset_u64s[MAX_BSETS];
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
44 /* size of the binary tree and prev array */
47 /* function of size - precalculated for to_inorder() */
56 struct journal_entry_pin journal;
60 struct open_buckets ob;
61 __BKEY_PADDED(k, BKEY_BTREE_PTR_VAL_U64s_MAX);
64 struct btree_bkey_cached_common {
72 struct btree_bkey_cached_common c;
74 struct rhash_head hash;
83 struct bkey_format format;
85 struct btree_node *data;
89 * Sets of sorted keys - the real btree node - plus a binary search tree
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.
95 struct bset_tree set[MAX_BSETS];
97 struct btree_nr_keys nr;
103 struct btree_write writes[2];
105 /* Key/pointer for this btree node */
106 __BKEY_PADDED(key, BKEY_BTREE_PTR_VAL_U64s_MAX);
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
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
124 struct list_head write_blocked;
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:
132 unsigned long will_make_reachable;
134 struct open_buckets ob;
137 struct list_head list;
141 struct rhashtable table;
142 bool table_init_done;
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.
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.
157 struct list_head live;
158 struct list_head freeable;
159 struct list_head freed_pcpu;
160 struct list_head freed_nonpcpu;
162 /* Number of elements in live + freeable lists */
166 struct shrinker *shrink;
169 * If we need to allocate memory for a new btree node and that
170 * allocation fails, we can cannibalize another node in the btree cache
171 * to satisfy the allocation - lock to guarantee only one thread does
174 struct task_struct *alloc_lock;
175 struct closure_waitlist alloc_wait;
178 struct btree_node_iter {
179 struct btree_node_iter_set {
185 * Iterate over all possible positions, synthesizing deleted keys for holes:
187 static const __maybe_unused u16 BTREE_ITER_SLOTS = 1 << 0;
189 * Indicates that intent locks should be taken on leaf nodes, because we expect
190 * to be doing updates:
192 static const __maybe_unused u16 BTREE_ITER_INTENT = 1 << 1;
194 * Causes the btree iterator code to prefetch additional btree nodes from disk:
196 static const __maybe_unused u16 BTREE_ITER_PREFETCH = 1 << 2;
198 * Used in bch2_btree_iter_traverse(), to indicate whether we're searching for
199 * @pos or the first key strictly greater than @pos
201 static const __maybe_unused u16 BTREE_ITER_IS_EXTENTS = 1 << 3;
202 static const __maybe_unused u16 BTREE_ITER_NOT_EXTENTS = 1 << 4;
203 static const __maybe_unused u16 BTREE_ITER_CACHED = 1 << 5;
204 static const __maybe_unused u16 BTREE_ITER_WITH_KEY_CACHE = 1 << 6;
205 static const __maybe_unused u16 BTREE_ITER_WITH_UPDATES = 1 << 7;
206 static const __maybe_unused u16 BTREE_ITER_WITH_JOURNAL = 1 << 8;
207 static const __maybe_unused u16 __BTREE_ITER_ALL_SNAPSHOTS = 1 << 9;
208 static const __maybe_unused u16 BTREE_ITER_ALL_SNAPSHOTS = 1 << 10;
209 static const __maybe_unused u16 BTREE_ITER_FILTER_SNAPSHOTS = 1 << 11;
210 static const __maybe_unused u16 BTREE_ITER_NOPRESERVE = 1 << 12;
211 static const __maybe_unused u16 BTREE_ITER_CACHED_NOFILL = 1 << 13;
212 static const __maybe_unused u16 BTREE_ITER_KEY_CACHE_FILL = 1 << 14;
213 #define __BTREE_ITER_FLAGS_END 15
215 enum btree_path_uptodate {
216 BTREE_ITER_UPTODATE = 0,
217 BTREE_ITER_NEED_RELOCK = 1,
218 BTREE_ITER_NEED_TRAVERSE = 2,
221 #if defined(CONFIG_BCACHEFS_LOCK_TIME_STATS) || defined(CONFIG_BCACHEFS_DEBUG)
222 #define TRACK_PATH_ALLOCATED
233 /* btree_iter_copy starts here: */
236 enum btree_id btree_id:5;
239 enum btree_path_uptodate uptodate:2;
241 * When true, failing to relock this path will cause the transaction to
244 bool should_be_locked:1;
249 struct btree_path_level {
251 struct btree_node_iter iter;
253 #ifdef CONFIG_BCACHEFS_LOCK_TIME_STATS
256 } l[BTREE_MAX_DEPTH];
257 #ifdef TRACK_PATH_ALLOCATED
258 unsigned long ip_allocated;
262 static inline struct btree_path_level *path_l(struct btree_path *path)
264 return path->l + path->level;
267 static inline unsigned long btree_path_ip_allocated(struct btree_path *path)
269 #ifdef TRACK_PATH_ALLOCATED
270 return path->ip_allocated;
277 * @pos - iterator's current position
278 * @level - current btree depth
279 * @locks_want - btree level below which we start taking intent locks
280 * @nodes_locked - bitmask indicating which nodes in @nodes are locked
281 * @nodes_intent_locked - bitmask indicating which locks are intent locks
284 struct btree_trans *trans;
285 struct btree_path *path;
286 struct btree_path *update_path;
287 struct btree_path *key_cache_path;
289 enum btree_id btree_id:8;
292 /* btree_iter_copy starts here: */
295 /* When we're filtering by snapshot, the snapshot ID we're looking for: */
300 * Current unpacked key - so that bch2_btree_iter_next()/
301 * bch2_btree_iter_next_slot() can correctly advance pos.
305 /* BTREE_ITER_WITH_JOURNAL: */
307 #ifdef TRACK_PATH_ALLOCATED
308 unsigned long ip_allocated;
312 #define BKEY_CACHED_ACCESSED 0
313 #define BKEY_CACHED_DIRTY 1
316 struct btree_bkey_cached_common c;
321 u32 btree_trans_barrier_seq;
322 struct bkey_cached_key key;
324 struct rhash_head hash;
325 struct list_head list;
327 struct journal_entry_pin journal;
333 static inline struct bpos btree_node_pos(struct btree_bkey_cached_common *b)
336 ? container_of(b, struct btree, c)->key.k.p
337 : container_of(b, struct bkey_cached, c)->key.pos;
340 struct btree_insert_entry {
343 enum btree_id btree_id:8;
346 bool insert_trigger_run:1;
347 bool overwrite_trigger_run:1;
348 bool key_cache_already_flushed:1;
350 * @old_k may be a key from the journal; @old_btree_u64s always refers
351 * to the size of the key being overwritten in the btree:
355 struct btree_path *path;
356 /* key being overwritten: */
358 const struct bch_val *old_v;
359 unsigned long ip_allocated;
362 #define BTREE_ITER_MAX 64
364 struct btree_trans_commit_hook;
365 typedef int (btree_trans_commit_hook_fn)(struct btree_trans *, struct btree_trans_commit_hook *);
367 struct btree_trans_commit_hook {
368 btree_trans_commit_hook_fn *fn;
369 struct btree_trans_commit_hook *next;
372 #define BTREE_TRANS_MEM_MAX (1U << 16)
374 #define BTREE_TRANS_MAX_LOCK_HOLD_TIME_NS 10000
380 struct list_head list;
383 u8 lock_may_not_fail;
385 struct btree_bkey_cached_common *locking;
386 struct six_lock_waiter locking_wait;
397 bool in_traverse_all:1;
399 bool memory_allocation_failure:1;
400 bool journal_transaction_names:1;
401 bool journal_replay_not_finished:1;
402 bool notrace_relock_fail:1;
404 enum bch_errcode restarted:16;
406 unsigned long last_begin_ip;
407 unsigned long last_restarted_ip;
408 unsigned long srcu_lock_time;
411 * For when bch2_trans_update notices we'll be splitting a compressed
414 unsigned extra_journal_res;
415 unsigned nr_max_paths;
424 u8 sorted[BTREE_ITER_MAX + 8];
425 struct btree_path paths[BTREE_ITER_MAX];
426 struct btree_insert_entry updates[BTREE_ITER_MAX];
427 struct btree_write_buffered_key *wb_updates;
430 struct btree_trans_commit_hook *hooks;
431 darray_u64 extra_journal_entries;
432 struct journal_entry_pin *journal_pin;
434 struct journal_res journal_res;
436 struct disk_reservation *disk_res;
437 unsigned journal_u64s;
438 struct replicas_delta_list *fs_usage_deltas;
441 #define BCH_BTREE_WRITE_TYPES() \
443 x(init_next_bset, 1) \
444 x(cache_reclaim, 2) \
445 x(journal_reclaim, 3) \
448 enum btree_write_type {
449 #define x(t, n) BTREE_WRITE_##t,
450 BCH_BTREE_WRITE_TYPES()
455 #define BTREE_WRITE_TYPE_MASK (roundup_pow_of_two(BTREE_WRITE_TYPE_NR) - 1)
456 #define BTREE_WRITE_TYPE_BITS ilog2(roundup_pow_of_two(BTREE_WRITE_TYPE_NR))
458 #define BTREE_FLAGS() \
464 x(will_make_reachable) \
469 x(write_in_flight_inner) \
477 /* First bits for btree node write type */
478 BTREE_NODE_FLAGS_START = BTREE_WRITE_TYPE_BITS - 1,
479 #define x(flag) BTREE_NODE_##flag,
485 static inline bool btree_node_ ## flag(struct btree *b) \
486 { return test_bit(BTREE_NODE_ ## flag, &b->flags); } \
488 static inline void set_btree_node_ ## flag(struct btree *b) \
489 { set_bit(BTREE_NODE_ ## flag, &b->flags); } \
491 static inline void clear_btree_node_ ## flag(struct btree *b) \
492 { clear_bit(BTREE_NODE_ ## flag, &b->flags); }
497 static inline struct btree_write *btree_current_write(struct btree *b)
499 return b->writes + btree_node_write_idx(b);
502 static inline struct btree_write *btree_prev_write(struct btree *b)
504 return b->writes + (btree_node_write_idx(b) ^ 1);
507 static inline struct bset_tree *bset_tree_last(struct btree *b)
510 return b->set + b->nsets - 1;
514 __btree_node_offset_to_ptr(const struct btree *b, u16 offset)
516 return (void *) ((u64 *) b->data + 1 + offset);
520 __btree_node_ptr_to_offset(const struct btree *b, const void *p)
522 u16 ret = (u64 *) p - 1 - (u64 *) b->data;
524 EBUG_ON(__btree_node_offset_to_ptr(b, ret) != p);
528 static inline struct bset *bset(const struct btree *b,
529 const struct bset_tree *t)
531 return __btree_node_offset_to_ptr(b, t->data_offset);
534 static inline void set_btree_bset_end(struct btree *b, struct bset_tree *t)
537 __btree_node_ptr_to_offset(b, vstruct_last(bset(b, t)));
540 static inline void set_btree_bset(struct btree *b, struct bset_tree *t,
541 const struct bset *i)
543 t->data_offset = __btree_node_ptr_to_offset(b, i);
544 set_btree_bset_end(b, t);
547 static inline struct bset *btree_bset_first(struct btree *b)
549 return bset(b, b->set);
552 static inline struct bset *btree_bset_last(struct btree *b)
554 return bset(b, bset_tree_last(b));
558 __btree_node_key_to_offset(const struct btree *b, const struct bkey_packed *k)
560 return __btree_node_ptr_to_offset(b, k);
563 static inline struct bkey_packed *
564 __btree_node_offset_to_key(const struct btree *b, u16 k)
566 return __btree_node_offset_to_ptr(b, k);
569 static inline unsigned btree_bkey_first_offset(const struct bset_tree *t)
571 return t->data_offset + offsetof(struct bset, _data) / sizeof(u64);
574 #define btree_bkey_first(_b, _t) \
576 EBUG_ON(bset(_b, _t)->start != \
577 __btree_node_offset_to_key(_b, btree_bkey_first_offset(_t)));\
579 bset(_b, _t)->start; \
582 #define btree_bkey_last(_b, _t) \
584 EBUG_ON(__btree_node_offset_to_key(_b, (_t)->end_offset) != \
585 vstruct_last(bset(_b, _t))); \
587 __btree_node_offset_to_key(_b, (_t)->end_offset); \
590 static inline unsigned bset_u64s(struct bset_tree *t)
592 return t->end_offset - t->data_offset -
593 sizeof(struct bset) / sizeof(u64);
596 static inline unsigned bset_dead_u64s(struct btree *b, struct bset_tree *t)
598 return bset_u64s(t) - b->nr.bset_u64s[t - b->set];
601 static inline unsigned bset_byte_offset(struct btree *b, void *i)
603 return i - (void *) b->data;
606 enum btree_node_type {
608 #define x(kwd, val, ...) BKEY_TYPE_##kwd = val + 1,
614 /* Type of a key in btree @id at level @level: */
615 static inline enum btree_node_type __btree_node_type(unsigned level, enum btree_id id)
617 return level ? BKEY_TYPE_btree : (unsigned) id + 1;
620 /* Type of keys @b contains: */
621 static inline enum btree_node_type btree_node_type(struct btree *b)
623 return __btree_node_type(b->c.level, b->c.btree_id);
626 const char *bch2_btree_node_type_str(enum btree_node_type);
628 #define BTREE_NODE_TYPE_HAS_TRANS_TRIGGERS \
629 (BIT_ULL(BKEY_TYPE_extents)| \
630 BIT_ULL(BKEY_TYPE_alloc)| \
631 BIT_ULL(BKEY_TYPE_inodes)| \
632 BIT_ULL(BKEY_TYPE_stripes)| \
633 BIT_ULL(BKEY_TYPE_reflink)| \
634 BIT_ULL(BKEY_TYPE_btree))
636 #define BTREE_NODE_TYPE_HAS_MEM_TRIGGERS \
637 (BIT_ULL(BKEY_TYPE_alloc)| \
638 BIT_ULL(BKEY_TYPE_inodes)| \
639 BIT_ULL(BKEY_TYPE_stripes)| \
640 BIT_ULL(BKEY_TYPE_snapshots))
642 #define BTREE_NODE_TYPE_HAS_TRIGGERS \
643 (BTREE_NODE_TYPE_HAS_TRANS_TRIGGERS| \
644 BTREE_NODE_TYPE_HAS_MEM_TRIGGERS)
646 static inline bool btree_node_type_needs_gc(enum btree_node_type type)
648 return BTREE_NODE_TYPE_HAS_TRIGGERS & BIT_ULL(type);
651 static inline bool btree_node_type_is_extents(enum btree_node_type type)
653 const unsigned mask = 0
654 #define x(name, nr, flags, ...) |((!!((flags) & BTREE_ID_EXTENTS)) << (nr + 1))
659 return (1U << type) & mask;
662 static inline bool btree_id_is_extents(enum btree_id btree)
664 return btree_node_type_is_extents(__btree_node_type(0, btree));
667 static inline bool btree_type_has_snapshots(enum btree_id id)
669 const unsigned mask = 0
670 #define x(name, nr, flags, ...) |((!!((flags) & BTREE_ID_SNAPSHOTS)) << nr)
675 return (1U << id) & mask;
678 static inline bool btree_type_has_snapshot_field(enum btree_id id)
680 const unsigned mask = 0
681 #define x(name, nr, flags, ...) |((!!((flags) & (BTREE_ID_SNAPSHOT_FIELD|BTREE_ID_SNAPSHOTS))) << nr)
686 return (1U << id) & mask;
689 static inline bool btree_type_has_ptrs(enum btree_id id)
691 const unsigned mask = 0
692 #define x(name, nr, flags, ...) |((!!((flags) & BTREE_ID_DATA)) << nr)
697 return (1U << id) & mask;
703 /* On disk root - see async splits: */
704 __BKEY_PADDED(key, BKEY_BTREE_PTR_VAL_U64s_MAX);
710 enum btree_gc_coalesce_fail_reason {
711 BTREE_GC_COALESCE_FAIL_RESERVE_GET,
712 BTREE_GC_COALESCE_FAIL_KEYLIST_REALLOC,
713 BTREE_GC_COALESCE_FAIL_FORMAT_FITS,
716 enum btree_node_sibling {
721 #endif /* _BCACHEFS_BTREE_TYPES_H */