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"
13 #include "journal_types.h"
14 #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 #define BTREE_ITER_SLOTS (1 << 0)
198 #define 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 #define BTREE_ITER_INTENT (1 << 2)
205 * Causes the btree iterator code to prefetch additional btree nodes from disk:
207 #define 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 #define BTREE_ITER_IS_EXTENTS (1 << 4)
213 #define BTREE_ITER_NOT_EXTENTS (1 << 5)
214 #define BTREE_ITER_CACHED (1 << 6)
215 #define BTREE_ITER_WITH_KEY_CACHE (1 << 7)
216 #define BTREE_ITER_WITH_UPDATES (1 << 8)
217 #define BTREE_ITER_WITH_JOURNAL (1 << 9)
218 #define __BTREE_ITER_ALL_SNAPSHOTS (1 << 10)
219 #define BTREE_ITER_ALL_SNAPSHOTS (1 << 11)
220 #define BTREE_ITER_FILTER_SNAPSHOTS (1 << 12)
221 #define BTREE_ITER_NOPRESERVE (1 << 13)
222 #define BTREE_ITER_CACHED_NOFILL (1 << 14)
223 #define BTREE_ITER_KEY_CACHE_FILL (1 << 15)
225 enum btree_path_uptodate {
226 BTREE_ITER_UPTODATE = 0,
227 BTREE_ITER_NEED_RELOCK = 1,
228 BTREE_ITER_NEED_TRAVERSE = 2,
231 #if defined(CONFIG_BCACHEFS_LOCK_TIME_STATS) || defined(CONFIG_BCACHEFS_DEBUG)
232 #define TRACK_PATH_ALLOCATED
241 /* btree_iter_copy starts here: */
244 enum btree_id btree_id:5;
247 enum btree_path_uptodate uptodate:2;
249 * When true, failing to relock this path will cause the transaction to
252 bool should_be_locked:1;
257 struct btree_path_level {
259 struct btree_node_iter iter;
261 #ifdef CONFIG_BCACHEFS_LOCK_TIME_STATS
264 } l[BTREE_MAX_DEPTH];
265 #ifdef TRACK_PATH_ALLOCATED
266 unsigned long ip_allocated;
270 static inline struct btree_path_level *path_l(struct btree_path *path)
272 return path->l + path->level;
275 static inline unsigned long btree_path_ip_allocated(struct btree_path *path)
277 #ifdef TRACK_PATH_ALLOCATED
278 return path->ip_allocated;
285 * @pos - iterator's current position
286 * @level - current btree depth
287 * @locks_want - btree level below which we start taking intent locks
288 * @nodes_locked - bitmask indicating which nodes in @nodes are locked
289 * @nodes_intent_locked - bitmask indicating which locks are intent locks
292 struct btree_trans *trans;
293 struct btree_path *path;
294 struct btree_path *update_path;
295 struct btree_path *key_cache_path;
297 enum btree_id btree_id:8;
298 unsigned min_depth:3;
301 /* btree_iter_copy starts here: */
304 /* When we're filtering by snapshot, the snapshot ID we're looking for: */
309 * Current unpacked key - so that bch2_btree_iter_next()/
310 * bch2_btree_iter_next_slot() can correctly advance pos.
314 /* BTREE_ITER_WITH_JOURNAL: */
316 struct bpos journal_pos;
317 #ifdef TRACK_PATH_ALLOCATED
318 unsigned long ip_allocated;
322 struct btree_key_cache_freelist {
323 struct bkey_cached *objs[16];
327 struct btree_key_cache {
329 struct rhashtable table;
330 bool table_init_done;
331 struct list_head freed_pcpu;
332 struct list_head freed_nonpcpu;
333 struct shrinker shrink;
334 unsigned shrink_iter;
335 struct btree_key_cache_freelist __percpu *pcpu_freed;
337 atomic_long_t nr_freed;
338 atomic_long_t nr_keys;
339 atomic_long_t nr_dirty;
342 struct bkey_cached_key {
345 } __packed __aligned(4);
347 #define BKEY_CACHED_ACCESSED 0
348 #define BKEY_CACHED_DIRTY 1
351 struct btree_bkey_cached_common c;
356 u32 btree_trans_barrier_seq;
357 struct bkey_cached_key key;
359 struct rhash_head hash;
360 struct list_head list;
362 struct journal_preres res;
363 struct journal_entry_pin journal;
369 static inline struct bpos btree_node_pos(struct btree_bkey_cached_common *b)
372 ? container_of(b, struct btree, c)->key.k.p
373 : container_of(b, struct bkey_cached, c)->key.pos;
376 struct btree_insert_entry {
379 enum btree_id btree_id:8;
382 bool insert_trigger_run:1;
383 bool overwrite_trigger_run:1;
384 bool key_cache_already_flushed:1;
386 * @old_k may be a key from the journal; @old_btree_u64s always refers
387 * to the size of the key being overwritten in the btree:
391 struct btree_path *path;
392 /* key being overwritten: */
394 const struct bch_val *old_v;
395 unsigned long ip_allocated;
398 #ifndef CONFIG_LOCKDEP
399 #define BTREE_ITER_MAX 64
401 #define BTREE_ITER_MAX 32
404 struct btree_trans_commit_hook;
405 typedef int (btree_trans_commit_hook_fn)(struct btree_trans *, struct btree_trans_commit_hook *);
407 struct btree_trans_commit_hook {
408 btree_trans_commit_hook_fn *fn;
409 struct btree_trans_commit_hook *next;
412 #define BTREE_TRANS_MEM_MAX (1U << 16)
414 #define BTREE_TRANS_MAX_LOCK_HOLD_TIME_NS 10000
420 struct list_head list;
423 u8 lock_may_not_fail;
425 struct btree_bkey_cached_common *locking;
426 struct six_lock_waiter locking_wait;
436 bool in_traverse_all:1;
438 bool memory_allocation_failure:1;
439 bool journal_transaction_names:1;
440 bool journal_replay_not_finished:1;
441 bool is_initial_gc:1;
442 bool notrace_relock_fail:1;
443 enum bch_errcode restarted:16;
445 unsigned long last_begin_ip;
446 unsigned long last_restarted_ip;
447 unsigned long srcu_lock_time;
450 * For when bch2_trans_update notices we'll be splitting a compressed
453 unsigned extra_journal_res;
454 unsigned nr_max_paths;
463 u8 sorted[BTREE_ITER_MAX + 8];
464 struct btree_path *paths;
465 struct btree_insert_entry *updates;
466 struct btree_write_buffered_key *wb_updates;
469 struct btree_trans_commit_hook *hooks;
470 darray_u64 extra_journal_entries;
471 struct journal_entry_pin *journal_pin;
473 struct journal_res journal_res;
474 struct journal_preres journal_preres;
476 struct disk_reservation *disk_res;
477 unsigned journal_u64s;
478 unsigned journal_preres_u64s;
479 struct replicas_delta_list *fs_usage_deltas;
482 #define BCH_BTREE_WRITE_TYPES() \
484 x(init_next_bset, 1) \
485 x(cache_reclaim, 2) \
486 x(journal_reclaim, 3) \
489 enum btree_write_type {
490 #define x(t, n) BTREE_WRITE_##t,
491 BCH_BTREE_WRITE_TYPES()
496 #define BTREE_WRITE_TYPE_MASK (roundup_pow_of_two(BTREE_WRITE_TYPE_NR) - 1)
497 #define BTREE_WRITE_TYPE_BITS ilog2(roundup_pow_of_two(BTREE_WRITE_TYPE_NR))
499 #define BTREE_FLAGS() \
505 x(will_make_reachable) \
510 x(write_in_flight_inner) \
518 /* First bits for btree node write type */
519 BTREE_NODE_FLAGS_START = BTREE_WRITE_TYPE_BITS - 1,
520 #define x(flag) BTREE_NODE_##flag,
526 static inline bool btree_node_ ## flag(struct btree *b) \
527 { return test_bit(BTREE_NODE_ ## flag, &b->flags); } \
529 static inline void set_btree_node_ ## flag(struct btree *b) \
530 { set_bit(BTREE_NODE_ ## flag, &b->flags); } \
532 static inline void clear_btree_node_ ## flag(struct btree *b) \
533 { clear_bit(BTREE_NODE_ ## flag, &b->flags); }
538 static inline struct btree_write *btree_current_write(struct btree *b)
540 return b->writes + btree_node_write_idx(b);
543 static inline struct btree_write *btree_prev_write(struct btree *b)
545 return b->writes + (btree_node_write_idx(b) ^ 1);
548 static inline struct bset_tree *bset_tree_last(struct btree *b)
551 return b->set + b->nsets - 1;
555 __btree_node_offset_to_ptr(const struct btree *b, u16 offset)
557 return (void *) ((u64 *) b->data + 1 + offset);
561 __btree_node_ptr_to_offset(const struct btree *b, const void *p)
563 u16 ret = (u64 *) p - 1 - (u64 *) b->data;
565 EBUG_ON(__btree_node_offset_to_ptr(b, ret) != p);
569 static inline struct bset *bset(const struct btree *b,
570 const struct bset_tree *t)
572 return __btree_node_offset_to_ptr(b, t->data_offset);
575 static inline void set_btree_bset_end(struct btree *b, struct bset_tree *t)
578 __btree_node_ptr_to_offset(b, vstruct_last(bset(b, t)));
581 static inline void set_btree_bset(struct btree *b, struct bset_tree *t,
582 const struct bset *i)
584 t->data_offset = __btree_node_ptr_to_offset(b, i);
585 set_btree_bset_end(b, t);
588 static inline struct bset *btree_bset_first(struct btree *b)
590 return bset(b, b->set);
593 static inline struct bset *btree_bset_last(struct btree *b)
595 return bset(b, bset_tree_last(b));
599 __btree_node_key_to_offset(const struct btree *b, const struct bkey_packed *k)
601 return __btree_node_ptr_to_offset(b, k);
604 static inline struct bkey_packed *
605 __btree_node_offset_to_key(const struct btree *b, u16 k)
607 return __btree_node_offset_to_ptr(b, k);
610 static inline unsigned btree_bkey_first_offset(const struct bset_tree *t)
612 return t->data_offset + offsetof(struct bset, _data) / sizeof(u64);
615 #define btree_bkey_first(_b, _t) \
617 EBUG_ON(bset(_b, _t)->start != \
618 __btree_node_offset_to_key(_b, btree_bkey_first_offset(_t)));\
620 bset(_b, _t)->start; \
623 #define btree_bkey_last(_b, _t) \
625 EBUG_ON(__btree_node_offset_to_key(_b, (_t)->end_offset) != \
626 vstruct_last(bset(_b, _t))); \
628 __btree_node_offset_to_key(_b, (_t)->end_offset); \
631 static inline unsigned bset_u64s(struct bset_tree *t)
633 return t->end_offset - t->data_offset -
634 sizeof(struct bset) / sizeof(u64);
637 static inline unsigned bset_dead_u64s(struct btree *b, struct bset_tree *t)
639 return bset_u64s(t) - b->nr.bset_u64s[t - b->set];
642 static inline unsigned bset_byte_offset(struct btree *b, void *i)
644 return i - (void *) b->data;
647 enum btree_node_type {
648 #define x(kwd, val) BKEY_TYPE_##kwd = val,
654 /* Type of a key in btree @id at level @level: */
655 static inline enum btree_node_type __btree_node_type(unsigned level, enum btree_id id)
657 return level ? BKEY_TYPE_btree : (enum btree_node_type) id;
660 /* Type of keys @b contains: */
661 static inline enum btree_node_type btree_node_type(struct btree *b)
663 return __btree_node_type(b->c.level, b->c.btree_id);
666 #define BTREE_NODE_TYPE_HAS_TRANS_TRIGGERS \
667 ((1U << BKEY_TYPE_extents)| \
668 (1U << BKEY_TYPE_alloc)| \
669 (1U << BKEY_TYPE_inodes)| \
670 (1U << BKEY_TYPE_stripes)| \
671 (1U << BKEY_TYPE_reflink)| \
672 (1U << BKEY_TYPE_btree))
674 #define BTREE_NODE_TYPE_HAS_MEM_TRIGGERS \
675 ((1U << BKEY_TYPE_alloc)| \
676 (1U << BKEY_TYPE_inodes)| \
677 (1U << BKEY_TYPE_stripes)| \
678 (1U << BKEY_TYPE_snapshots))
680 #define BTREE_NODE_TYPE_HAS_TRIGGERS \
681 (BTREE_NODE_TYPE_HAS_TRANS_TRIGGERS| \
682 BTREE_NODE_TYPE_HAS_MEM_TRIGGERS)
684 #define BTREE_ID_IS_EXTENTS \
685 ((1U << BTREE_ID_extents)| \
686 (1U << BTREE_ID_reflink)| \
687 (1U << BTREE_ID_freespace))
689 static inline bool btree_node_type_is_extents(enum btree_node_type type)
691 return (1U << type) & BTREE_ID_IS_EXTENTS;
694 #define BTREE_ID_HAS_SNAPSHOTS \
695 ((1U << BTREE_ID_extents)| \
696 (1U << BTREE_ID_inodes)| \
697 (1U << BTREE_ID_dirents)| \
698 (1U << BTREE_ID_xattrs))
700 #define BTREE_ID_HAS_PTRS \
701 ((1U << BTREE_ID_extents)| \
702 (1U << BTREE_ID_reflink))
704 static inline bool btree_type_has_snapshots(enum btree_id id)
706 return (1 << id) & BTREE_ID_HAS_SNAPSHOTS;
709 static inline bool btree_type_has_ptrs(enum btree_id id)
711 return (1 << id) & BTREE_ID_HAS_PTRS;
714 static inline bool btree_node_type_needs_gc(enum btree_node_type type)
716 return BTREE_NODE_TYPE_HAS_TRIGGERS & (1U << type);
722 /* On disk root - see async splits: */
723 __BKEY_PADDED(key, BKEY_BTREE_PTR_VAL_U64s_MAX);
729 enum btree_gc_coalesce_fail_reason {
730 BTREE_GC_COALESCE_FAIL_RESERVE_GET,
731 BTREE_GC_COALESCE_FAIL_KEYLIST_REALLOC,
732 BTREE_GC_COALESCE_FAIL_FORMAT_FITS,
735 enum btree_node_sibling {
740 #endif /* _BCACHEFS_BTREE_TYPES_H */