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 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;
176 struct shrinker shrink;
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
184 struct task_struct *alloc_lock;
185 struct closure_waitlist alloc_wait;
188 struct btree_node_iter {
189 struct btree_node_iter_set {
195 * Iterate over all possible positions, synthesizing deleted keys for holes:
197 static const __maybe_unused u16 BTREE_ITER_SLOTS = 1 << 0;
198 static const __maybe_unused u16 BTREE_ITER_ALL_LEVELS = 1 << 1;
200 * Indicates that intent locks should be taken on leaf nodes, because we expect
201 * to be doing updates:
203 static const __maybe_unused u16 BTREE_ITER_INTENT = 1 << 2;
205 * Causes the btree iterator code to prefetch additional btree nodes from disk:
207 static const __maybe_unused u16 BTREE_ITER_PREFETCH = 1 << 3;
209 * Used in bch2_btree_iter_traverse(), to indicate whether we're searching for
210 * @pos or the first key strictly greater than @pos
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
226 enum btree_path_uptodate {
227 BTREE_ITER_UPTODATE = 0,
228 BTREE_ITER_NEED_RELOCK = 1,
229 BTREE_ITER_NEED_TRAVERSE = 2,
232 #if defined(CONFIG_BCACHEFS_LOCK_TIME_STATS) || defined(CONFIG_BCACHEFS_DEBUG)
233 #define TRACK_PATH_ALLOCATED
244 /* btree_iter_copy starts here: */
247 enum btree_id btree_id:5;
250 enum btree_path_uptodate uptodate:2;
252 * When true, failing to relock this path will cause the transaction to
255 bool should_be_locked:1;
260 struct btree_path_level {
262 struct btree_node_iter iter;
264 #ifdef CONFIG_BCACHEFS_LOCK_TIME_STATS
267 } l[BTREE_MAX_DEPTH];
268 #ifdef TRACK_PATH_ALLOCATED
269 unsigned long ip_allocated;
273 static inline struct btree_path_level *path_l(struct btree_path *path)
275 return path->l + path->level;
278 static inline unsigned long btree_path_ip_allocated(struct btree_path *path)
280 #ifdef TRACK_PATH_ALLOCATED
281 return path->ip_allocated;
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
295 struct btree_trans *trans;
296 struct btree_path *path;
297 struct btree_path *update_path;
298 struct btree_path *key_cache_path;
300 enum btree_id btree_id:8;
301 unsigned min_depth:3;
304 /* btree_iter_copy starts here: */
307 /* When we're filtering by snapshot, the snapshot ID we're looking for: */
312 * Current unpacked key - so that bch2_btree_iter_next()/
313 * bch2_btree_iter_next_slot() can correctly advance pos.
317 /* BTREE_ITER_WITH_JOURNAL: */
319 struct bpos journal_pos;
320 #ifdef TRACK_PATH_ALLOCATED
321 unsigned long ip_allocated;
325 #define BKEY_CACHED_ACCESSED 0
326 #define BKEY_CACHED_DIRTY 1
329 struct btree_bkey_cached_common c;
334 u32 btree_trans_barrier_seq;
335 struct bkey_cached_key key;
337 struct rhash_head hash;
338 struct list_head list;
340 struct journal_entry_pin journal;
346 static inline struct bpos btree_node_pos(struct btree_bkey_cached_common *b)
349 ? container_of(b, struct btree, c)->key.k.p
350 : container_of(b, struct bkey_cached, c)->key.pos;
353 struct btree_insert_entry {
356 enum btree_id btree_id:8;
359 bool insert_trigger_run:1;
360 bool overwrite_trigger_run:1;
361 bool key_cache_already_flushed:1;
363 * @old_k may be a key from the journal; @old_btree_u64s always refers
364 * to the size of the key being overwritten in the btree:
368 struct btree_path *path;
369 /* key being overwritten: */
371 const struct bch_val *old_v;
372 unsigned long ip_allocated;
375 #define BTREE_ITER_MAX 64
377 struct btree_trans_commit_hook;
378 typedef int (btree_trans_commit_hook_fn)(struct btree_trans *, struct btree_trans_commit_hook *);
380 struct btree_trans_commit_hook {
381 btree_trans_commit_hook_fn *fn;
382 struct btree_trans_commit_hook *next;
385 #define BTREE_TRANS_MEM_MAX (1U << 16)
387 #define BTREE_TRANS_MAX_LOCK_HOLD_TIME_NS 10000
393 struct list_head list;
396 u8 lock_may_not_fail;
398 struct btree_bkey_cached_common *locking;
399 struct six_lock_waiter locking_wait;
410 bool in_traverse_all:1;
412 bool memory_allocation_failure:1;
413 bool journal_transaction_names:1;
414 bool journal_replay_not_finished:1;
415 bool is_initial_gc:1;
416 bool notrace_relock_fail:1;
418 enum bch_errcode restarted:16;
420 unsigned long last_begin_ip;
421 unsigned long last_restarted_ip;
422 unsigned long srcu_lock_time;
425 * For when bch2_trans_update notices we'll be splitting a compressed
428 unsigned extra_journal_res;
429 unsigned nr_max_paths;
438 u8 sorted[BTREE_ITER_MAX + 8];
439 struct btree_path paths[BTREE_ITER_MAX];
440 struct btree_insert_entry updates[BTREE_ITER_MAX];
441 struct btree_write_buffered_key *wb_updates;
444 struct btree_trans_commit_hook *hooks;
445 darray_u64 extra_journal_entries;
446 struct journal_entry_pin *journal_pin;
448 struct journal_res journal_res;
450 struct disk_reservation *disk_res;
451 unsigned journal_u64s;
452 struct replicas_delta_list *fs_usage_deltas;
455 #define BCH_BTREE_WRITE_TYPES() \
457 x(init_next_bset, 1) \
458 x(cache_reclaim, 2) \
459 x(journal_reclaim, 3) \
462 enum btree_write_type {
463 #define x(t, n) BTREE_WRITE_##t,
464 BCH_BTREE_WRITE_TYPES()
469 #define BTREE_WRITE_TYPE_MASK (roundup_pow_of_two(BTREE_WRITE_TYPE_NR) - 1)
470 #define BTREE_WRITE_TYPE_BITS ilog2(roundup_pow_of_two(BTREE_WRITE_TYPE_NR))
472 #define BTREE_FLAGS() \
478 x(will_make_reachable) \
483 x(write_in_flight_inner) \
491 /* First bits for btree node write type */
492 BTREE_NODE_FLAGS_START = BTREE_WRITE_TYPE_BITS - 1,
493 #define x(flag) BTREE_NODE_##flag,
499 static inline bool btree_node_ ## flag(struct btree *b) \
500 { return test_bit(BTREE_NODE_ ## flag, &b->flags); } \
502 static inline void set_btree_node_ ## flag(struct btree *b) \
503 { set_bit(BTREE_NODE_ ## flag, &b->flags); } \
505 static inline void clear_btree_node_ ## flag(struct btree *b) \
506 { clear_bit(BTREE_NODE_ ## flag, &b->flags); }
511 static inline struct btree_write *btree_current_write(struct btree *b)
513 return b->writes + btree_node_write_idx(b);
516 static inline struct btree_write *btree_prev_write(struct btree *b)
518 return b->writes + (btree_node_write_idx(b) ^ 1);
521 static inline struct bset_tree *bset_tree_last(struct btree *b)
524 return b->set + b->nsets - 1;
528 __btree_node_offset_to_ptr(const struct btree *b, u16 offset)
530 return (void *) ((u64 *) b->data + 1 + offset);
534 __btree_node_ptr_to_offset(const struct btree *b, const void *p)
536 u16 ret = (u64 *) p - 1 - (u64 *) b->data;
538 EBUG_ON(__btree_node_offset_to_ptr(b, ret) != p);
542 static inline struct bset *bset(const struct btree *b,
543 const struct bset_tree *t)
545 return __btree_node_offset_to_ptr(b, t->data_offset);
548 static inline void set_btree_bset_end(struct btree *b, struct bset_tree *t)
551 __btree_node_ptr_to_offset(b, vstruct_last(bset(b, t)));
554 static inline void set_btree_bset(struct btree *b, struct bset_tree *t,
555 const struct bset *i)
557 t->data_offset = __btree_node_ptr_to_offset(b, i);
558 set_btree_bset_end(b, t);
561 static inline struct bset *btree_bset_first(struct btree *b)
563 return bset(b, b->set);
566 static inline struct bset *btree_bset_last(struct btree *b)
568 return bset(b, bset_tree_last(b));
572 __btree_node_key_to_offset(const struct btree *b, const struct bkey_packed *k)
574 return __btree_node_ptr_to_offset(b, k);
577 static inline struct bkey_packed *
578 __btree_node_offset_to_key(const struct btree *b, u16 k)
580 return __btree_node_offset_to_ptr(b, k);
583 static inline unsigned btree_bkey_first_offset(const struct bset_tree *t)
585 return t->data_offset + offsetof(struct bset, _data) / sizeof(u64);
588 #define btree_bkey_first(_b, _t) \
590 EBUG_ON(bset(_b, _t)->start != \
591 __btree_node_offset_to_key(_b, btree_bkey_first_offset(_t)));\
593 bset(_b, _t)->start; \
596 #define btree_bkey_last(_b, _t) \
598 EBUG_ON(__btree_node_offset_to_key(_b, (_t)->end_offset) != \
599 vstruct_last(bset(_b, _t))); \
601 __btree_node_offset_to_key(_b, (_t)->end_offset); \
604 static inline unsigned bset_u64s(struct bset_tree *t)
606 return t->end_offset - t->data_offset -
607 sizeof(struct bset) / sizeof(u64);
610 static inline unsigned bset_dead_u64s(struct btree *b, struct bset_tree *t)
612 return bset_u64s(t) - b->nr.bset_u64s[t - b->set];
615 static inline unsigned bset_byte_offset(struct btree *b, void *i)
617 return i - (void *) b->data;
620 enum btree_node_type {
622 #define x(kwd, val, ...) BKEY_TYPE_##kwd = val + 1,
628 /* Type of a key in btree @id at level @level: */
629 static inline enum btree_node_type __btree_node_type(unsigned level, enum btree_id id)
631 return level ? BKEY_TYPE_btree : (unsigned) id + 1;
634 /* Type of keys @b contains: */
635 static inline enum btree_node_type btree_node_type(struct btree *b)
637 return __btree_node_type(b->c.level, b->c.btree_id);
640 const char *bch2_btree_node_type_str(enum btree_node_type);
642 #define BTREE_NODE_TYPE_HAS_TRANS_TRIGGERS \
643 (BIT_ULL(BKEY_TYPE_extents)| \
644 BIT_ULL(BKEY_TYPE_alloc)| \
645 BIT_ULL(BKEY_TYPE_inodes)| \
646 BIT_ULL(BKEY_TYPE_stripes)| \
647 BIT_ULL(BKEY_TYPE_reflink)| \
648 BIT_ULL(BKEY_TYPE_btree))
650 #define BTREE_NODE_TYPE_HAS_MEM_TRIGGERS \
651 (BIT_ULL(BKEY_TYPE_alloc)| \
652 BIT_ULL(BKEY_TYPE_inodes)| \
653 BIT_ULL(BKEY_TYPE_stripes)| \
654 BIT_ULL(BKEY_TYPE_snapshots))
656 #define BTREE_NODE_TYPE_HAS_TRIGGERS \
657 (BTREE_NODE_TYPE_HAS_TRANS_TRIGGERS| \
658 BTREE_NODE_TYPE_HAS_MEM_TRIGGERS)
660 static inline bool btree_node_type_needs_gc(enum btree_node_type type)
662 return BTREE_NODE_TYPE_HAS_TRIGGERS & BIT_ULL(type);
665 static inline bool btree_node_type_is_extents(enum btree_node_type type)
667 const unsigned mask = 0
668 #define x(name, nr, flags, ...) |((!!((flags) & BTREE_ID_EXTENTS)) << (nr + 1))
673 return (1U << type) & mask;
676 static inline bool btree_id_is_extents(enum btree_id btree)
678 return btree_node_type_is_extents(__btree_node_type(0, btree));
681 static inline bool btree_type_has_snapshots(enum btree_id id)
683 const unsigned mask = 0
684 #define x(name, nr, flags, ...) |((!!((flags) & BTREE_ID_SNAPSHOTS)) << nr)
689 return (1U << id) & mask;
692 static inline bool btree_type_has_snapshot_field(enum btree_id id)
694 const unsigned mask = 0
695 #define x(name, nr, flags, ...) |((!!((flags) & (BTREE_ID_SNAPSHOT_FIELD|BTREE_ID_SNAPSHOTS))) << nr)
700 return (1U << id) & mask;
703 static inline bool btree_type_has_ptrs(enum btree_id id)
705 const unsigned mask = 0
706 #define x(name, nr, flags, ...) |((!!((flags) & BTREE_ID_DATA)) << nr)
711 return (1U << id) & mask;
717 /* On disk root - see async splits: */
718 __BKEY_PADDED(key, BKEY_BTREE_PTR_VAL_U64s_MAX);
724 enum btree_gc_coalesce_fail_reason {
725 BTREE_GC_COALESCE_FAIL_RESERVE_GET,
726 BTREE_GC_COALESCE_FAIL_KEYLIST_REALLOC,
727 BTREE_GC_COALESCE_FAIL_FORMAT_FITS,
730 enum btree_node_sibling {
735 #endif /* _BCACHEFS_BTREE_TYPES_H */