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>
9 //#include "bkey_methods.h"
10 #include "buckets_types.h"
12 #include "journal_types.h"
13 #include "replicas_types.h"
21 struct btree_nr_keys {
24 * Amount of live metadata (i.e. size of node after a compaction) in
28 u16 bset_u64s[MAX_BSETS];
37 * We construct a binary tree in an array as if the array
38 * started at 1, so that things line up on the same cachelines
39 * better: see comments in bset.c at cacheline_to_bkey() for
43 /* size of the binary tree and prev array */
46 /* function of size - precalculated for to_inorder() */
55 struct journal_entry_pin journal;
59 struct open_buckets ob;
60 __BKEY_PADDED(k, BKEY_BTREE_PTR_VAL_U64s_MAX);
63 struct btree_bkey_cached_common {
71 struct btree_bkey_cached_common c;
73 struct rhash_head hash;
82 struct bkey_format format;
84 struct btree_node *data;
88 * Sets of sorted keys - the real btree node - plus a binary search tree
90 * set[0] is special; set[0]->tree, set[0]->prev and set[0]->data point
91 * to the memory we have allocated for this btree node. Additionally,
92 * set[0]->data points to the entire btree node as it exists on disk.
94 struct bset_tree set[MAX_BSETS];
96 struct btree_nr_keys nr;
102 struct btree_write writes[2];
104 /* Key/pointer for this btree node */
105 __BKEY_PADDED(key, BKEY_BTREE_PTR_VAL_U64s_MAX);
108 * XXX: add a delete sequence number, so when bch2_btree_node_relock()
109 * fails because the lock sequence number has changed - i.e. the
110 * contents were modified - we can still relock the node if it's still
111 * the one we want, without redoing the traversal
115 * For asynchronous splits/interior node updates:
116 * When we do a split, we allocate new child nodes and update the parent
117 * node to point to them: we update the parent in memory immediately,
118 * but then we must wait until the children have been written out before
119 * the update to the parent can be written - this is a list of the
120 * btree_updates that are blocking this node from being
123 struct list_head write_blocked;
126 * Also for asynchronous splits/interior node updates:
127 * If a btree node isn't reachable yet, we don't want to kick off
128 * another write - because that write also won't yet be reachable and
129 * marking it as completed before it's reachable would be incorrect:
131 unsigned long will_make_reachable;
133 struct open_buckets ob;
136 struct list_head list;
140 struct rhashtable table;
141 bool table_init_done;
143 * We never free a struct btree, except on shutdown - we just put it on
144 * the btree_cache_freed list and reuse it later. This simplifies the
145 * code, and it doesn't cost us much memory as the memory usage is
146 * dominated by buffers that hold the actual btree node data and those
147 * can be freed - and the number of struct btrees allocated is
148 * effectively bounded.
150 * btree_cache_freeable effectively is a small cache - we use it because
151 * high order page allocations can be rather expensive, and it's quite
152 * common to delete and allocate btree nodes in quick succession. It
153 * should never grow past ~2-3 nodes in practice.
156 struct list_head live;
157 struct list_head freeable;
158 struct list_head freed_pcpu;
159 struct list_head freed_nonpcpu;
161 /* Number of elements in live + freeable lists */
165 unsigned not_freed_lock_intent;
166 unsigned not_freed_lock_write;
167 unsigned not_freed_dirty;
168 unsigned not_freed_read_in_flight;
169 unsigned not_freed_write_in_flight;
170 unsigned not_freed_noevict;
171 unsigned not_freed_write_blocked;
172 unsigned not_freed_will_make_reachable;
173 unsigned not_freed_access_bit;
175 struct shrinker shrink;
178 * If we need to allocate memory for a new btree node and that
179 * allocation fails, we can cannibalize another node in the btree cache
180 * to satisfy the allocation - lock to guarantee only one thread does
183 struct task_struct *alloc_lock;
184 struct closure_waitlist alloc_wait;
187 struct btree_node_iter {
188 struct btree_node_iter_set {
194 * Iterate over all possible positions, synthesizing deleted keys for holes:
196 #define BTREE_ITER_SLOTS (1 << 0)
197 #define BTREE_ITER_ALL_LEVELS (1 << 1)
199 * Indicates that intent locks should be taken on leaf nodes, because we expect
200 * to be doing updates:
202 #define BTREE_ITER_INTENT (1 << 2)
204 * Causes the btree iterator code to prefetch additional btree nodes from disk:
206 #define BTREE_ITER_PREFETCH (1 << 3)
208 * Used in bch2_btree_iter_traverse(), to indicate whether we're searching for
209 * @pos or the first key strictly greater than @pos
211 #define BTREE_ITER_IS_EXTENTS (1 << 4)
212 #define BTREE_ITER_NOT_EXTENTS (1 << 5)
213 #define BTREE_ITER_CACHED (1 << 6)
214 #define BTREE_ITER_WITH_KEY_CACHE (1 << 7)
215 #define BTREE_ITER_WITH_UPDATES (1 << 8)
216 #define BTREE_ITER_WITH_JOURNAL (1 << 9)
217 #define __BTREE_ITER_ALL_SNAPSHOTS (1 << 10)
218 #define BTREE_ITER_ALL_SNAPSHOTS (1 << 11)
219 #define BTREE_ITER_FILTER_SNAPSHOTS (1 << 12)
220 #define BTREE_ITER_NOPRESERVE (1 << 13)
221 #define BTREE_ITER_CACHED_NOFILL (1 << 14)
222 #define BTREE_ITER_KEY_CACHE_FILL (1 << 15)
224 enum btree_path_uptodate {
225 BTREE_ITER_UPTODATE = 0,
226 BTREE_ITER_NEED_RELOCK = 1,
227 BTREE_ITER_NEED_TRAVERSE = 2,
230 #if defined(CONFIG_BCACHEFS_LOCK_TIME_STATS) || defined(CONFIG_BCACHEFS_DEBUG)
231 #define TRACK_PATH_ALLOCATED
240 /* btree_iter_copy starts here: */
243 enum btree_id btree_id:5;
246 enum btree_path_uptodate uptodate:2;
248 * When true, failing to relock this path will cause the transaction to
251 bool should_be_locked:1;
256 struct btree_path_level {
258 struct btree_node_iter iter;
260 #ifdef CONFIG_BCACHEFS_LOCK_TIME_STATS
263 } l[BTREE_MAX_DEPTH];
264 #ifdef TRACK_PATH_ALLOCATED
265 unsigned long ip_allocated;
269 static inline struct btree_path_level *path_l(struct btree_path *path)
271 return path->l + path->level;
274 static inline unsigned long btree_path_ip_allocated(struct btree_path *path)
276 #ifdef TRACK_PATH_ALLOCATED
277 return path->ip_allocated;
284 * @pos - iterator's current position
285 * @level - current btree depth
286 * @locks_want - btree level below which we start taking intent locks
287 * @nodes_locked - bitmask indicating which nodes in @nodes are locked
288 * @nodes_intent_locked - bitmask indicating which locks are intent locks
291 struct btree_trans *trans;
292 struct btree_path *path;
293 struct btree_path *update_path;
294 struct btree_path *key_cache_path;
296 enum btree_id btree_id:8;
297 unsigned min_depth:3;
300 /* btree_iter_copy starts here: */
303 /* When we're filtering by snapshot, the snapshot ID we're looking for: */
308 * Current unpacked key - so that bch2_btree_iter_next()/
309 * bch2_btree_iter_next_slot() can correctly advance pos.
313 /* BTREE_ITER_WITH_JOURNAL: */
315 struct bpos journal_pos;
316 #ifdef TRACK_PATH_ALLOCATED
317 unsigned long ip_allocated;
321 struct btree_key_cache_freelist {
322 struct bkey_cached *objs[16];
326 struct btree_key_cache {
328 struct rhashtable table;
329 bool table_init_done;
330 struct list_head freed_pcpu;
331 struct list_head freed_nonpcpu;
332 struct shrinker shrink;
333 unsigned shrink_iter;
334 struct btree_key_cache_freelist __percpu *pcpu_freed;
336 atomic_long_t nr_freed;
337 atomic_long_t nr_keys;
338 atomic_long_t nr_dirty;
341 struct bkey_cached_key {
344 } __packed __aligned(4);
346 #define BKEY_CACHED_ACCESSED 0
347 #define BKEY_CACHED_DIRTY 1
350 struct btree_bkey_cached_common c;
355 u32 btree_trans_barrier_seq;
356 struct bkey_cached_key key;
358 struct rhash_head hash;
359 struct list_head list;
361 struct journal_preres res;
362 struct journal_entry_pin journal;
368 static inline struct bpos btree_node_pos(struct btree_bkey_cached_common *b)
371 ? container_of(b, struct btree, c)->key.k.p
372 : container_of(b, struct bkey_cached, c)->key.pos;
375 struct btree_insert_entry {
378 enum btree_id btree_id:8;
381 bool insert_trigger_run:1;
382 bool overwrite_trigger_run:1;
383 bool key_cache_already_flushed:1;
385 * @old_k may be a key from the journal; @old_btree_u64s always refers
386 * to the size of the key being overwritten in the btree:
390 struct btree_path *path;
391 /* key being overwritten: */
393 const struct bch_val *old_v;
394 unsigned long ip_allocated;
397 #ifndef CONFIG_LOCKDEP
398 #define BTREE_ITER_MAX 64
400 #define BTREE_ITER_MAX 32
403 struct btree_trans_commit_hook;
404 typedef int (btree_trans_commit_hook_fn)(struct btree_trans *, struct btree_trans_commit_hook *);
406 struct btree_trans_commit_hook {
407 btree_trans_commit_hook_fn *fn;
408 struct btree_trans_commit_hook *next;
411 #define BTREE_TRANS_MEM_MAX (1U << 16)
413 #define BTREE_TRANS_MAX_LOCK_HOLD_TIME_NS 10000
419 struct list_head list;
422 u8 lock_may_not_fail;
424 struct btree_bkey_cached_common *locking;
425 struct six_lock_waiter locking_wait;
433 bool in_traverse_all:1;
435 bool memory_allocation_failure:1;
436 bool journal_transaction_names:1;
437 bool journal_replay_not_finished:1;
438 bool is_initial_gc:1;
439 bool notrace_relock_fail:1;
440 enum bch_errcode restarted:16;
442 unsigned long last_restarted_ip;
443 unsigned long srcu_lock_time;
446 * For when bch2_trans_update notices we'll be splitting a compressed
449 unsigned extra_journal_res;
450 unsigned nr_max_paths;
459 u8 sorted[BTREE_ITER_MAX + 8];
460 struct btree_path *paths;
461 struct btree_insert_entry *updates;
464 struct btree_trans_commit_hook *hooks;
465 darray_u64 extra_journal_entries;
466 struct journal_entry_pin *journal_pin;
468 struct journal_res journal_res;
469 struct journal_preres journal_preres;
471 struct disk_reservation *disk_res;
472 unsigned journal_u64s;
473 unsigned journal_preres_u64s;
474 struct replicas_delta_list *fs_usage_deltas;
477 #define BCH_BTREE_WRITE_TYPES() \
479 x(init_next_bset, 1) \
480 x(cache_reclaim, 2) \
481 x(journal_reclaim, 3) \
484 enum btree_write_type {
485 #define x(t, n) BTREE_WRITE_##t,
486 BCH_BTREE_WRITE_TYPES()
491 #define BTREE_WRITE_TYPE_MASK (roundup_pow_of_two(BTREE_WRITE_TYPE_NR) - 1)
492 #define BTREE_WRITE_TYPE_BITS ilog2(roundup_pow_of_two(BTREE_WRITE_TYPE_NR))
494 #define BTREE_FLAGS() \
500 x(will_make_reachable) \
505 x(write_in_flight_inner) \
513 /* First bits for btree node write type */
514 BTREE_NODE_FLAGS_START = BTREE_WRITE_TYPE_BITS - 1,
515 #define x(flag) BTREE_NODE_##flag,
521 static inline bool btree_node_ ## flag(struct btree *b) \
522 { return test_bit(BTREE_NODE_ ## flag, &b->flags); } \
524 static inline void set_btree_node_ ## flag(struct btree *b) \
525 { set_bit(BTREE_NODE_ ## flag, &b->flags); } \
527 static inline void clear_btree_node_ ## flag(struct btree *b) \
528 { clear_bit(BTREE_NODE_ ## flag, &b->flags); }
533 static inline struct btree_write *btree_current_write(struct btree *b)
535 return b->writes + btree_node_write_idx(b);
538 static inline struct btree_write *btree_prev_write(struct btree *b)
540 return b->writes + (btree_node_write_idx(b) ^ 1);
543 static inline struct bset_tree *bset_tree_last(struct btree *b)
546 return b->set + b->nsets - 1;
550 __btree_node_offset_to_ptr(const struct btree *b, u16 offset)
552 return (void *) ((u64 *) b->data + 1 + offset);
556 __btree_node_ptr_to_offset(const struct btree *b, const void *p)
558 u16 ret = (u64 *) p - 1 - (u64 *) b->data;
560 EBUG_ON(__btree_node_offset_to_ptr(b, ret) != p);
564 static inline struct bset *bset(const struct btree *b,
565 const struct bset_tree *t)
567 return __btree_node_offset_to_ptr(b, t->data_offset);
570 static inline void set_btree_bset_end(struct btree *b, struct bset_tree *t)
573 __btree_node_ptr_to_offset(b, vstruct_last(bset(b, t)));
576 static inline void set_btree_bset(struct btree *b, struct bset_tree *t,
577 const struct bset *i)
579 t->data_offset = __btree_node_ptr_to_offset(b, i);
580 set_btree_bset_end(b, t);
583 static inline struct bset *btree_bset_first(struct btree *b)
585 return bset(b, b->set);
588 static inline struct bset *btree_bset_last(struct btree *b)
590 return bset(b, bset_tree_last(b));
594 __btree_node_key_to_offset(const struct btree *b, const struct bkey_packed *k)
596 return __btree_node_ptr_to_offset(b, k);
599 static inline struct bkey_packed *
600 __btree_node_offset_to_key(const struct btree *b, u16 k)
602 return __btree_node_offset_to_ptr(b, k);
605 static inline unsigned btree_bkey_first_offset(const struct bset_tree *t)
607 return t->data_offset + offsetof(struct bset, _data) / sizeof(u64);
610 #define btree_bkey_first(_b, _t) \
612 EBUG_ON(bset(_b, _t)->start != \
613 __btree_node_offset_to_key(_b, btree_bkey_first_offset(_t)));\
615 bset(_b, _t)->start; \
618 #define btree_bkey_last(_b, _t) \
620 EBUG_ON(__btree_node_offset_to_key(_b, (_t)->end_offset) != \
621 vstruct_last(bset(_b, _t))); \
623 __btree_node_offset_to_key(_b, (_t)->end_offset); \
626 static inline unsigned bset_u64s(struct bset_tree *t)
628 return t->end_offset - t->data_offset -
629 sizeof(struct bset) / sizeof(u64);
632 static inline unsigned bset_dead_u64s(struct btree *b, struct bset_tree *t)
634 return bset_u64s(t) - b->nr.bset_u64s[t - b->set];
637 static inline unsigned bset_byte_offset(struct btree *b, void *i)
639 return i - (void *) b->data;
642 enum btree_node_type {
643 #define x(kwd, val) BKEY_TYPE_##kwd = val,
649 /* Type of a key in btree @id at level @level: */
650 static inline enum btree_node_type __btree_node_type(unsigned level, enum btree_id id)
652 return level ? BKEY_TYPE_btree : (enum btree_node_type) id;
655 /* Type of keys @b contains: */
656 static inline enum btree_node_type btree_node_type(struct btree *b)
658 return __btree_node_type(b->c.level, b->c.btree_id);
661 #define BTREE_NODE_TYPE_HAS_TRANS_TRIGGERS \
662 ((1U << BKEY_TYPE_extents)| \
663 (1U << BKEY_TYPE_alloc)| \
664 (1U << BKEY_TYPE_inodes)| \
665 (1U << BKEY_TYPE_stripes)| \
666 (1U << BKEY_TYPE_reflink)| \
667 (1U << BKEY_TYPE_btree))
669 #define BTREE_NODE_TYPE_HAS_MEM_TRIGGERS \
670 ((1U << BKEY_TYPE_alloc)| \
671 (1U << BKEY_TYPE_inodes)| \
672 (1U << BKEY_TYPE_stripes)| \
673 (1U << BKEY_TYPE_snapshots))
675 #define BTREE_NODE_TYPE_HAS_TRIGGERS \
676 (BTREE_NODE_TYPE_HAS_TRANS_TRIGGERS| \
677 BTREE_NODE_TYPE_HAS_MEM_TRIGGERS)
679 #define BTREE_ID_IS_EXTENTS \
680 ((1U << BTREE_ID_extents)| \
681 (1U << BTREE_ID_reflink)| \
682 (1U << BTREE_ID_freespace))
684 static inline bool btree_node_type_is_extents(enum btree_node_type type)
686 return (1U << type) & BTREE_ID_IS_EXTENTS;
689 #define BTREE_ID_HAS_SNAPSHOTS \
690 ((1U << BTREE_ID_extents)| \
691 (1U << BTREE_ID_inodes)| \
692 (1U << BTREE_ID_dirents)| \
693 (1U << BTREE_ID_xattrs))
695 #define BTREE_ID_HAS_PTRS \
696 ((1U << BTREE_ID_extents)| \
697 (1U << BTREE_ID_reflink))
699 static inline bool btree_type_has_snapshots(enum btree_id id)
701 return (1 << id) & BTREE_ID_HAS_SNAPSHOTS;
704 static inline bool btree_type_has_ptrs(enum btree_id id)
706 return (1 << id) & BTREE_ID_HAS_PTRS;
709 static inline bool btree_node_type_needs_gc(enum btree_node_type type)
711 return BTREE_NODE_TYPE_HAS_TRIGGERS & (1U << type);
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 */