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 "bkey_methods.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 u16 BTREE_ITER_SLOTS = 1 << 0;
198 static const 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 u16 BTREE_ITER_INTENT = 1 << 2;
205 * Causes the btree iterator code to prefetch additional btree nodes from disk:
207 static const 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 u16 BTREE_ITER_IS_EXTENTS = 1 << 4;
213 static const u16 BTREE_ITER_NOT_EXTENTS = 1 << 5;
214 static const u16 BTREE_ITER_CACHED = 1 << 6;
215 static const u16 BTREE_ITER_WITH_KEY_CACHE = 1 << 7;
216 static const u16 BTREE_ITER_WITH_UPDATES = 1 << 8;
217 static const u16 BTREE_ITER_WITH_JOURNAL = 1 << 9;
218 static const u16 __BTREE_ITER_ALL_SNAPSHOTS = 1 << 10;
219 static const u16 BTREE_ITER_ALL_SNAPSHOTS = 1 << 11;
220 static const u16 BTREE_ITER_FILTER_SNAPSHOTS = 1 << 12;
221 static const u16 BTREE_ITER_NOPRESERVE = 1 << 13;
222 static const u16 BTREE_ITER_CACHED_NOFILL = 1 << 14;
223 static const 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
242 /* btree_iter_copy starts here: */
245 enum btree_id btree_id:5;
248 enum btree_path_uptodate uptodate:2;
250 * When true, failing to relock this path will cause the transaction to
253 bool should_be_locked:1;
258 struct btree_path_level {
260 struct btree_node_iter iter;
262 #ifdef CONFIG_BCACHEFS_LOCK_TIME_STATS
265 } l[BTREE_MAX_DEPTH];
266 #ifdef TRACK_PATH_ALLOCATED
267 unsigned long ip_allocated;
271 static inline struct btree_path_level *path_l(struct btree_path *path)
273 return path->l + path->level;
276 static inline unsigned long btree_path_ip_allocated(struct btree_path *path)
278 #ifdef TRACK_PATH_ALLOCATED
279 return path->ip_allocated;
286 * @pos - iterator's current position
287 * @level - current btree depth
288 * @locks_want - btree level below which we start taking intent locks
289 * @nodes_locked - bitmask indicating which nodes in @nodes are locked
290 * @nodes_intent_locked - bitmask indicating which locks are intent locks
293 struct btree_trans *trans;
294 struct btree_path *path;
295 struct btree_path *update_path;
296 struct btree_path *key_cache_path;
298 enum btree_id btree_id:8;
299 unsigned min_depth:3;
302 /* btree_iter_copy starts here: */
305 /* When we're filtering by snapshot, the snapshot ID we're looking for: */
310 * Current unpacked key - so that bch2_btree_iter_next()/
311 * bch2_btree_iter_next_slot() can correctly advance pos.
315 /* BTREE_ITER_WITH_JOURNAL: */
317 struct bpos journal_pos;
318 #ifdef TRACK_PATH_ALLOCATED
319 unsigned long ip_allocated;
323 struct btree_key_cache_freelist {
324 struct bkey_cached *objs[16];
328 struct btree_key_cache {
330 struct rhashtable table;
331 bool table_init_done;
332 struct list_head freed_pcpu;
333 struct list_head freed_nonpcpu;
334 struct shrinker shrink;
335 unsigned shrink_iter;
336 struct btree_key_cache_freelist __percpu *pcpu_freed;
338 atomic_long_t nr_freed;
339 atomic_long_t nr_keys;
340 atomic_long_t nr_dirty;
343 struct bkey_cached_key {
346 } __packed __aligned(4);
348 #define BKEY_CACHED_ACCESSED 0
349 #define BKEY_CACHED_DIRTY 1
352 struct btree_bkey_cached_common c;
357 u32 btree_trans_barrier_seq;
358 struct bkey_cached_key key;
360 struct rhash_head hash;
361 struct list_head list;
363 struct journal_preres res;
364 struct journal_entry_pin journal;
370 static inline struct bpos btree_node_pos(struct btree_bkey_cached_common *b)
373 ? container_of(b, struct btree, c)->key.k.p
374 : container_of(b, struct bkey_cached, c)->key.pos;
377 struct btree_insert_entry {
380 enum btree_id btree_id:8;
383 bool insert_trigger_run:1;
384 bool overwrite_trigger_run:1;
385 bool key_cache_already_flushed:1;
387 * @old_k may be a key from the journal; @old_btree_u64s always refers
388 * to the size of the key being overwritten in the btree:
392 struct btree_path *path;
394 /* key being overwritten: */
396 const struct bch_val *old_v;
397 unsigned long ip_allocated;
400 #define BTREE_ITER_MAX 64
402 struct btree_trans_commit_hook;
403 typedef int (btree_trans_commit_hook_fn)(struct btree_trans *, struct btree_trans_commit_hook *);
405 struct btree_trans_commit_hook {
406 btree_trans_commit_hook_fn *fn;
407 struct btree_trans_commit_hook *next;
410 #define BTREE_TRANS_MEM_MAX (1U << 16)
412 #define BTREE_TRANS_MAX_LOCK_HOLD_TIME_NS 10000
418 struct list_head list;
421 u8 lock_may_not_fail;
423 struct btree_bkey_cached_common *locking;
424 struct six_lock_waiter locking_wait;
434 bool in_traverse_all:1;
436 bool memory_allocation_failure:1;
437 bool journal_transaction_names:1;
438 bool journal_replay_not_finished:1;
439 bool is_initial_gc:1;
440 bool notrace_relock_fail:1;
441 enum bch_errcode restarted:16;
443 unsigned long last_begin_ip;
444 unsigned long last_restarted_ip;
445 unsigned long srcu_lock_time;
448 * For when bch2_trans_update notices we'll be splitting a compressed
451 unsigned extra_journal_res;
452 unsigned nr_max_paths;
461 u8 sorted[BTREE_ITER_MAX + 8];
462 struct btree_path *paths;
463 struct btree_insert_entry *updates;
464 struct btree_write_buffered_key *wb_updates;
467 struct btree_trans_commit_hook *hooks;
468 darray_u64 extra_journal_entries;
469 struct journal_entry_pin *journal_pin;
471 struct journal_res journal_res;
472 struct journal_preres journal_preres;
474 struct disk_reservation *disk_res;
475 unsigned journal_u64s;
476 unsigned journal_preres_u64s;
477 struct replicas_delta_list *fs_usage_deltas;
480 #define BCH_BTREE_WRITE_TYPES() \
482 x(init_next_bset, 1) \
483 x(cache_reclaim, 2) \
484 x(journal_reclaim, 3) \
487 enum btree_write_type {
488 #define x(t, n) BTREE_WRITE_##t,
489 BCH_BTREE_WRITE_TYPES()
494 #define BTREE_WRITE_TYPE_MASK (roundup_pow_of_two(BTREE_WRITE_TYPE_NR) - 1)
495 #define BTREE_WRITE_TYPE_BITS ilog2(roundup_pow_of_two(BTREE_WRITE_TYPE_NR))
497 #define BTREE_FLAGS() \
503 x(will_make_reachable) \
508 x(write_in_flight_inner) \
516 /* First bits for btree node write type */
517 BTREE_NODE_FLAGS_START = BTREE_WRITE_TYPE_BITS - 1,
518 #define x(flag) BTREE_NODE_##flag,
524 static inline bool btree_node_ ## flag(struct btree *b) \
525 { return test_bit(BTREE_NODE_ ## flag, &b->flags); } \
527 static inline void set_btree_node_ ## flag(struct btree *b) \
528 { set_bit(BTREE_NODE_ ## flag, &b->flags); } \
530 static inline void clear_btree_node_ ## flag(struct btree *b) \
531 { clear_bit(BTREE_NODE_ ## flag, &b->flags); }
536 static inline struct btree_write *btree_current_write(struct btree *b)
538 return b->writes + btree_node_write_idx(b);
541 static inline struct btree_write *btree_prev_write(struct btree *b)
543 return b->writes + (btree_node_write_idx(b) ^ 1);
546 static inline struct bset_tree *bset_tree_last(struct btree *b)
549 return b->set + b->nsets - 1;
553 __btree_node_offset_to_ptr(const struct btree *b, u16 offset)
555 return (void *) ((u64 *) b->data + 1 + offset);
559 __btree_node_ptr_to_offset(const struct btree *b, const void *p)
561 u16 ret = (u64 *) p - 1 - (u64 *) b->data;
563 EBUG_ON(__btree_node_offset_to_ptr(b, ret) != p);
567 static inline struct bset *bset(const struct btree *b,
568 const struct bset_tree *t)
570 return __btree_node_offset_to_ptr(b, t->data_offset);
573 static inline void set_btree_bset_end(struct btree *b, struct bset_tree *t)
576 __btree_node_ptr_to_offset(b, vstruct_last(bset(b, t)));
579 static inline void set_btree_bset(struct btree *b, struct bset_tree *t,
580 const struct bset *i)
582 t->data_offset = __btree_node_ptr_to_offset(b, i);
583 set_btree_bset_end(b, t);
586 static inline struct bset *btree_bset_first(struct btree *b)
588 return bset(b, b->set);
591 static inline struct bset *btree_bset_last(struct btree *b)
593 return bset(b, bset_tree_last(b));
597 __btree_node_key_to_offset(const struct btree *b, const struct bkey_packed *k)
599 return __btree_node_ptr_to_offset(b, k);
602 static inline struct bkey_packed *
603 __btree_node_offset_to_key(const struct btree *b, u16 k)
605 return __btree_node_offset_to_ptr(b, k);
608 static inline unsigned btree_bkey_first_offset(const struct bset_tree *t)
610 return t->data_offset + offsetof(struct bset, _data) / sizeof(u64);
613 #define btree_bkey_first(_b, _t) \
615 EBUG_ON(bset(_b, _t)->start != \
616 __btree_node_offset_to_key(_b, btree_bkey_first_offset(_t)));\
618 bset(_b, _t)->start; \
621 #define btree_bkey_last(_b, _t) \
623 EBUG_ON(__btree_node_offset_to_key(_b, (_t)->end_offset) != \
624 vstruct_last(bset(_b, _t))); \
626 __btree_node_offset_to_key(_b, (_t)->end_offset); \
629 static inline unsigned bset_u64s(struct bset_tree *t)
631 return t->end_offset - t->data_offset -
632 sizeof(struct bset) / sizeof(u64);
635 static inline unsigned bset_dead_u64s(struct btree *b, struct bset_tree *t)
637 return bset_u64s(t) - b->nr.bset_u64s[t - b->set];
640 static inline unsigned bset_byte_offset(struct btree *b, void *i)
642 return i - (void *) b->data;
645 enum btree_node_type {
646 #define x(kwd, val, ...) BKEY_TYPE_##kwd = val,
652 /* Type of a key in btree @id at level @level: */
653 static inline enum btree_node_type __btree_node_type(unsigned level, enum btree_id id)
655 return level ? BKEY_TYPE_btree : (enum btree_node_type) id;
658 /* Type of keys @b contains: */
659 static inline enum btree_node_type btree_node_type(struct btree *b)
661 return __btree_node_type(b->c.level, b->c.btree_id);
664 #define BTREE_NODE_TYPE_HAS_TRANS_TRIGGERS \
665 (BIT(BKEY_TYPE_extents)| \
666 BIT(BKEY_TYPE_alloc)| \
667 BIT(BKEY_TYPE_inodes)| \
668 BIT(BKEY_TYPE_stripes)| \
669 BIT(BKEY_TYPE_reflink)| \
670 BIT(BKEY_TYPE_btree))
672 #define BTREE_NODE_TYPE_HAS_MEM_TRIGGERS \
673 (BIT(BKEY_TYPE_alloc)| \
674 BIT(BKEY_TYPE_inodes)| \
675 BIT(BKEY_TYPE_stripes)| \
676 BIT(BKEY_TYPE_snapshots))
678 #define BTREE_NODE_TYPE_HAS_TRIGGERS \
679 (BTREE_NODE_TYPE_HAS_TRANS_TRIGGERS| \
680 BTREE_NODE_TYPE_HAS_MEM_TRIGGERS)
682 static inline bool btree_node_type_needs_gc(enum btree_node_type type)
684 return BTREE_NODE_TYPE_HAS_TRIGGERS & (1U << type);
687 static inline bool btree_node_type_is_extents(enum btree_node_type type)
689 const unsigned mask = 0
690 #define x(name, nr, flags, ...) |((!!((flags) & BTREE_ID_EXTENTS)) << nr)
695 return (1U << type) & mask;
698 static inline bool btree_id_is_extents(enum btree_id btree)
700 return btree_node_type_is_extents((enum btree_node_type) btree);
703 static inline bool btree_type_has_snapshots(enum btree_id id)
705 const unsigned mask = 0
706 #define x(name, nr, flags, ...) |((!!((flags) & BTREE_ID_SNAPSHOTS)) << nr)
711 return (1U << id) & mask;
714 static inline bool btree_type_has_ptrs(enum btree_id id)
716 const unsigned mask = 0
717 #define x(name, nr, flags, ...) |((!!((flags) & BTREE_ID_DATA)) << nr)
722 return (1U << id) & mask;
728 /* On disk root - see async splits: */
729 __BKEY_PADDED(key, BKEY_BTREE_PTR_VAL_U64s_MAX);
735 enum btree_gc_coalesce_fail_reason {
736 BTREE_GC_COALESCE_FAIL_RESERVE_GET,
737 BTREE_GC_COALESCE_FAIL_KEYLIST_REALLOC,
738 BTREE_GC_COALESCE_FAIL_FORMAT_FITS,
741 enum btree_node_sibling {
746 #endif /* _BCACHEFS_BTREE_TYPES_H */