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"
11 #include "journal_types.h"
19 struct btree_nr_keys {
22 * Amount of live metadata (i.e. size of node after a compaction) in
26 u16 bset_u64s[MAX_BSETS];
35 * We construct a binary tree in an array as if the array
36 * started at 1, so that things line up on the same cachelines
37 * better: see comments in bset.c at cacheline_to_bkey() for
41 /* size of the binary tree and prev array */
44 /* function of size - precalculated for to_inorder() */
53 struct journal_entry_pin journal;
57 struct open_buckets ob;
58 __BKEY_PADDED(k, BKEY_BTREE_PTR_VAL_U64s_MAX);
61 struct btree_bkey_cached_common {
68 struct btree_bkey_cached_common c;
70 struct rhash_head hash;
79 struct bkey_format format;
81 struct btree_node *data;
85 * Sets of sorted keys - the real btree node - plus a binary search tree
87 * set[0] is special; set[0]->tree, set[0]->prev and set[0]->data point
88 * to the memory we have allocated for this btree node. Additionally,
89 * set[0]->data points to the entire btree node as it exists on disk.
91 struct bset_tree set[MAX_BSETS];
93 struct btree_nr_keys nr;
99 struct btree_write writes[2];
101 /* Key/pointer for this btree node */
102 __BKEY_PADDED(key, BKEY_BTREE_PTR_VAL_U64s_MAX);
105 * XXX: add a delete sequence number, so when bch2_btree_node_relock()
106 * fails because the lock sequence number has changed - i.e. the
107 * contents were modified - we can still relock the node if it's still
108 * the one we want, without redoing the traversal
112 * For asynchronous splits/interior node updates:
113 * When we do a split, we allocate new child nodes and update the parent
114 * node to point to them: we update the parent in memory immediately,
115 * but then we must wait until the children have been written out before
116 * the update to the parent can be written - this is a list of the
117 * btree_updates that are blocking this node from being
120 struct list_head write_blocked;
123 * Also for asynchronous splits/interior node updates:
124 * If a btree node isn't reachable yet, we don't want to kick off
125 * another write - because that write also won't yet be reachable and
126 * marking it as completed before it's reachable would be incorrect:
128 unsigned long will_make_reachable;
130 struct open_buckets ob;
133 struct list_head list;
137 struct rhashtable table;
138 bool table_init_done;
140 * We never free a struct btree, except on shutdown - we just put it on
141 * the btree_cache_freed list and reuse it later. This simplifies the
142 * code, and it doesn't cost us much memory as the memory usage is
143 * dominated by buffers that hold the actual btree node data and those
144 * can be freed - and the number of struct btrees allocated is
145 * effectively bounded.
147 * btree_cache_freeable effectively is a small cache - we use it because
148 * high order page allocations can be rather expensive, and it's quite
149 * common to delete and allocate btree nodes in quick succession. It
150 * should never grow past ~2-3 nodes in practice.
153 struct list_head live;
154 struct list_head freeable;
155 struct list_head freed;
157 /* Number of elements in live + freeable lists */
161 struct shrinker shrink;
164 * If we need to allocate memory for a new btree node and that
165 * allocation fails, we can cannibalize another node in the btree cache
166 * to satisfy the allocation - lock to guarantee only one thread does
169 struct task_struct *alloc_lock;
170 struct closure_waitlist alloc_wait;
173 struct btree_node_iter {
174 struct btree_node_iter_set {
180 * Iterate over all possible positions, synthesizing deleted keys for holes:
182 #define BTREE_ITER_SLOTS (1 << 0)
184 * Indicates that intent locks should be taken on leaf nodes, because we expect
185 * to be doing updates:
187 #define BTREE_ITER_INTENT (1 << 1)
189 * Causes the btree iterator code to prefetch additional btree nodes from disk:
191 #define BTREE_ITER_PREFETCH (1 << 2)
193 * Indicates that this iterator should not be reused until transaction commit,
194 * either because a pending update references it or because the update depends
195 * on that particular key being locked (e.g. by the str_hash code, for hash
198 #define BTREE_ITER_KEEP_UNTIL_COMMIT (1 << 3)
200 * Used in bch2_btree_iter_traverse(), to indicate whether we're searching for
201 * @pos or the first key strictly greater than @pos
203 #define BTREE_ITER_IS_EXTENTS (1 << 4)
204 #define BTREE_ITER_NOT_EXTENTS (1 << 5)
205 #define BTREE_ITER_CACHED (1 << 6)
206 #define BTREE_ITER_CACHED_NOFILL (1 << 7)
207 #define BTREE_ITER_CACHED_NOCREATE (1 << 8)
208 #define BTREE_ITER_WITH_KEY_CACHE (1 << 9)
209 #define BTREE_ITER_WITH_UPDATES (1 << 10)
210 #define BTREE_ITER_WITH_JOURNAL (1 << 11)
211 #define __BTREE_ITER_ALL_SNAPSHOTS (1 << 12)
212 #define BTREE_ITER_ALL_SNAPSHOTS (1 << 13)
213 #define BTREE_ITER_FILTER_SNAPSHOTS (1 << 14)
214 #define BTREE_ITER_NOPRESERVE (1 << 15)
216 enum btree_path_uptodate {
217 BTREE_ITER_UPTODATE = 0,
218 BTREE_ITER_NEED_RELOCK = 1,
219 BTREE_ITER_NEED_TRAVERSE = 2,
222 #define BTREE_ITER_NO_NODE_GET_LOCKS ((struct btree *) 1)
223 #define BTREE_ITER_NO_NODE_DROP ((struct btree *) 2)
224 #define BTREE_ITER_NO_NODE_LOCK_ROOT ((struct btree *) 3)
225 #define BTREE_ITER_NO_NODE_UP ((struct btree *) 4)
226 #define BTREE_ITER_NO_NODE_DOWN ((struct btree *) 5)
227 #define BTREE_ITER_NO_NODE_INIT ((struct btree *) 6)
228 #define BTREE_ITER_NO_NODE_ERROR ((struct btree *) 7)
229 #define BTREE_ITER_NO_NODE_CACHED ((struct btree *) 8)
237 /* btree_iter_copy starts here: */
240 enum btree_id btree_id:4;
243 enum btree_path_uptodate uptodate:2;
245 * When true, failing to relock this path will cause the transaction to
248 bool should_be_locked:1;
252 nodes_intent_locked:4;
254 struct btree_path_level {
256 struct btree_node_iter iter;
258 } l[BTREE_MAX_DEPTH];
259 #ifdef CONFIG_BCACHEFS_DEBUG
260 unsigned long ip_allocated;
264 static inline struct btree_path_level *path_l(struct btree_path *path)
266 return path->l + path->level;
270 * @pos - iterator's current position
271 * @level - current btree depth
272 * @locks_want - btree level below which we start taking intent locks
273 * @nodes_locked - bitmask indicating which nodes in @nodes are locked
274 * @nodes_intent_locked - bitmask indicating which locks are intent locks
277 struct btree_trans *trans;
278 struct btree_path *path;
279 struct btree_path *update_path;
280 struct btree_path *key_cache_path;
282 enum btree_id btree_id:4;
283 unsigned min_depth:4;
285 /* btree_iter_copy starts here: */
288 /* When we're filtering by snapshot, the snapshot ID we're looking for: */
292 struct bpos pos_after_commit;
294 * Current unpacked key - so that bch2_btree_iter_next()/
295 * bch2_btree_iter_next_slot() can correctly advance pos.
298 #ifdef CONFIG_BCACHEFS_DEBUG
299 unsigned long ip_allocated;
303 struct btree_key_cache {
305 struct rhashtable table;
306 bool table_init_done;
307 struct list_head freed;
308 struct shrinker shrink;
309 unsigned shrink_iter;
312 atomic_long_t nr_keys;
313 atomic_long_t nr_dirty;
316 struct bkey_cached_key {
319 } __attribute__((packed, aligned(4)));
321 #define BKEY_CACHED_ACCESSED 0
322 #define BKEY_CACHED_DIRTY 1
325 struct btree_bkey_cached_common c;
330 u32 btree_trans_barrier_seq;
331 struct bkey_cached_key key;
333 struct rhash_head hash;
334 struct list_head list;
336 struct journal_preres res;
337 struct journal_entry_pin journal;
342 struct btree_insert_entry {
345 enum btree_id btree_id:8;
348 bool insert_trigger_run:1;
349 bool overwrite_trigger_run:1;
351 * @old_k may be a key from the journal; @old_btree_u64s always refers
352 * to the size of the key being overwritten in the btree:
356 struct btree_path *path;
357 /* key being overwritten: */
359 const struct bch_val *old_v;
360 unsigned long ip_allocated;
363 #define BTREE_ITER_MAX 64
365 struct btree_trans_commit_hook;
366 typedef int (btree_trans_commit_hook_fn)(struct btree_trans *, struct btree_trans_commit_hook *);
368 struct btree_trans_commit_hook {
369 btree_trans_commit_hook_fn *fn;
370 struct btree_trans_commit_hook *next;
373 #define BTREE_TRANS_MEM_MAX (1U << 14)
378 struct list_head list;
379 struct btree *locking;
380 unsigned locking_path_idx;
381 struct bpos locking_pos;
384 u8 locking_lock_type;
391 bool in_traverse_all:1;
393 bool memory_allocation_failure:1;
394 bool journal_transaction_names:1;
396 * For when bch2_trans_update notices we'll be splitting a compressed
399 unsigned extra_journal_res;
407 u8 sorted[BTREE_ITER_MAX];
408 struct btree_path *paths;
409 struct btree_insert_entry *updates;
412 struct btree_trans_commit_hook *hooks;
413 struct jset_entry *extra_journal_entries;
414 unsigned extra_journal_entry_u64s;
415 struct journal_entry_pin *journal_pin;
417 struct journal_res journal_res;
418 struct journal_preres journal_preres;
420 struct disk_reservation *disk_res;
422 unsigned journal_u64s;
423 unsigned journal_preres_u64s;
424 struct replicas_delta_list *fs_usage_deltas;
427 #define BTREE_FLAG(flag) \
428 static inline bool btree_node_ ## flag(struct btree *b) \
429 { return test_bit(BTREE_NODE_ ## flag, &b->flags); } \
431 static inline void set_btree_node_ ## flag(struct btree *b) \
432 { set_bit(BTREE_NODE_ ## flag, &b->flags); } \
434 static inline void clear_btree_node_ ## flag(struct btree *b) \
435 { clear_bit(BTREE_NODE_ ## flag, &b->flags); }
438 BTREE_NODE_read_in_flight,
439 BTREE_NODE_read_error,
441 BTREE_NODE_need_write,
443 BTREE_NODE_write_idx,
445 BTREE_NODE_write_in_flight,
446 BTREE_NODE_write_in_flight_inner,
447 BTREE_NODE_just_written,
450 BTREE_NODE_need_rewrite,
451 BTREE_NODE_never_write,
454 BTREE_FLAG(read_in_flight);
455 BTREE_FLAG(read_error);
456 BTREE_FLAG(need_write);
458 BTREE_FLAG(write_idx);
459 BTREE_FLAG(accessed);
460 BTREE_FLAG(write_in_flight);
461 BTREE_FLAG(write_in_flight_inner);
462 BTREE_FLAG(just_written);
465 BTREE_FLAG(need_rewrite);
466 BTREE_FLAG(never_write);
468 static inline struct btree_write *btree_current_write(struct btree *b)
470 return b->writes + btree_node_write_idx(b);
473 static inline struct btree_write *btree_prev_write(struct btree *b)
475 return b->writes + (btree_node_write_idx(b) ^ 1);
478 static inline struct bset_tree *bset_tree_last(struct btree *b)
481 return b->set + b->nsets - 1;
485 __btree_node_offset_to_ptr(const struct btree *b, u16 offset)
487 return (void *) ((u64 *) b->data + 1 + offset);
491 __btree_node_ptr_to_offset(const struct btree *b, const void *p)
493 u16 ret = (u64 *) p - 1 - (u64 *) b->data;
495 EBUG_ON(__btree_node_offset_to_ptr(b, ret) != p);
499 static inline struct bset *bset(const struct btree *b,
500 const struct bset_tree *t)
502 return __btree_node_offset_to_ptr(b, t->data_offset);
505 static inline void set_btree_bset_end(struct btree *b, struct bset_tree *t)
508 __btree_node_ptr_to_offset(b, vstruct_last(bset(b, t)));
511 static inline void set_btree_bset(struct btree *b, struct bset_tree *t,
512 const struct bset *i)
514 t->data_offset = __btree_node_ptr_to_offset(b, i);
515 set_btree_bset_end(b, t);
518 static inline struct bset *btree_bset_first(struct btree *b)
520 return bset(b, b->set);
523 static inline struct bset *btree_bset_last(struct btree *b)
525 return bset(b, bset_tree_last(b));
529 __btree_node_key_to_offset(const struct btree *b, const struct bkey_packed *k)
531 return __btree_node_ptr_to_offset(b, k);
534 static inline struct bkey_packed *
535 __btree_node_offset_to_key(const struct btree *b, u16 k)
537 return __btree_node_offset_to_ptr(b, k);
540 static inline unsigned btree_bkey_first_offset(const struct bset_tree *t)
542 return t->data_offset + offsetof(struct bset, _data) / sizeof(u64);
545 #define btree_bkey_first(_b, _t) \
547 EBUG_ON(bset(_b, _t)->start != \
548 __btree_node_offset_to_key(_b, btree_bkey_first_offset(_t)));\
550 bset(_b, _t)->start; \
553 #define btree_bkey_last(_b, _t) \
555 EBUG_ON(__btree_node_offset_to_key(_b, (_t)->end_offset) != \
556 vstruct_last(bset(_b, _t))); \
558 __btree_node_offset_to_key(_b, (_t)->end_offset); \
561 static inline unsigned bset_u64s(struct bset_tree *t)
563 return t->end_offset - t->data_offset -
564 sizeof(struct bset) / sizeof(u64);
567 static inline unsigned bset_dead_u64s(struct btree *b, struct bset_tree *t)
569 return bset_u64s(t) - b->nr.bset_u64s[t - b->set];
572 static inline unsigned bset_byte_offset(struct btree *b, void *i)
574 return i - (void *) b->data;
577 enum btree_node_type {
578 #define x(kwd, val) BKEY_TYPE_##kwd = val,
584 /* Type of a key in btree @id at level @level: */
585 static inline enum btree_node_type __btree_node_type(unsigned level, enum btree_id id)
587 return level ? BKEY_TYPE_btree : (enum btree_node_type) id;
590 /* Type of keys @b contains: */
591 static inline enum btree_node_type btree_node_type(struct btree *b)
593 return __btree_node_type(b->c.level, b->c.btree_id);
596 static inline bool btree_node_type_is_extents(enum btree_node_type type)
599 case BKEY_TYPE_extents:
600 case BKEY_TYPE_reflink:
607 static inline bool btree_node_is_extents(struct btree *b)
609 return btree_node_type_is_extents(btree_node_type(b));
612 #define BTREE_NODE_TYPE_HAS_TRANS_TRIGGERS \
613 ((1U << BKEY_TYPE_extents)| \
614 (1U << BKEY_TYPE_inodes)| \
615 (1U << BKEY_TYPE_stripes)| \
616 (1U << BKEY_TYPE_reflink)| \
617 (1U << BKEY_TYPE_btree))
619 #define BTREE_NODE_TYPE_HAS_MEM_TRIGGERS \
620 ((1U << BKEY_TYPE_alloc)| \
621 (1U << BKEY_TYPE_inodes)| \
622 (1U << BKEY_TYPE_stripes)| \
623 (1U << BKEY_TYPE_snapshots))
625 #define BTREE_NODE_TYPE_HAS_TRIGGERS \
626 (BTREE_NODE_TYPE_HAS_TRANS_TRIGGERS| \
627 BTREE_NODE_TYPE_HAS_MEM_TRIGGERS)
629 #define BTREE_ID_HAS_SNAPSHOTS \
630 ((1U << BTREE_ID_extents)| \
631 (1U << BTREE_ID_inodes)| \
632 (1U << BTREE_ID_dirents)| \
633 (1U << BTREE_ID_xattrs))
635 #define BTREE_ID_HAS_PTRS \
636 ((1U << BTREE_ID_extents)| \
637 (1U << BTREE_ID_reflink))
639 static inline bool btree_type_has_snapshots(enum btree_id id)
641 return (1 << id) & BTREE_ID_HAS_SNAPSHOTS;
644 enum btree_update_flags {
645 __BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE,
646 __BTREE_UPDATE_KEY_CACHE_RECLAIM,
648 __BTREE_TRIGGER_NORUN, /* Don't run triggers at all */
650 __BTREE_TRIGGER_INSERT,
651 __BTREE_TRIGGER_OVERWRITE,
654 __BTREE_TRIGGER_BUCKET_INVALIDATE,
655 __BTREE_TRIGGER_NOATOMIC,
658 #define BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE (1U << __BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE)
659 #define BTREE_UPDATE_KEY_CACHE_RECLAIM (1U << __BTREE_UPDATE_KEY_CACHE_RECLAIM)
661 #define BTREE_TRIGGER_NORUN (1U << __BTREE_TRIGGER_NORUN)
663 #define BTREE_TRIGGER_INSERT (1U << __BTREE_TRIGGER_INSERT)
664 #define BTREE_TRIGGER_OVERWRITE (1U << __BTREE_TRIGGER_OVERWRITE)
666 #define BTREE_TRIGGER_GC (1U << __BTREE_TRIGGER_GC)
667 #define BTREE_TRIGGER_BUCKET_INVALIDATE (1U << __BTREE_TRIGGER_BUCKET_INVALIDATE)
668 #define BTREE_TRIGGER_NOATOMIC (1U << __BTREE_TRIGGER_NOATOMIC)
670 #define BTREE_TRIGGER_WANTS_OLD_AND_NEW \
671 ((1U << KEY_TYPE_alloc)| \
672 (1U << KEY_TYPE_alloc_v2)| \
673 (1U << KEY_TYPE_alloc_v3)| \
674 (1U << KEY_TYPE_stripe)| \
675 (1U << KEY_TYPE_inode)| \
676 (1U << KEY_TYPE_inode_v2)| \
677 (1U << KEY_TYPE_snapshot))
679 static inline bool btree_node_type_needs_gc(enum btree_node_type type)
681 return BTREE_NODE_TYPE_HAS_TRIGGERS & (1U << type);
687 /* On disk root - see async splits: */
688 __BKEY_PADDED(key, BKEY_BTREE_PTR_VAL_U64s_MAX);
694 enum btree_insert_ret {
696 /* leaf node needs to be split */
697 BTREE_INSERT_BTREE_NODE_FULL,
698 BTREE_INSERT_NEED_MARK_REPLICAS,
699 BTREE_INSERT_NEED_JOURNAL_RES,
700 BTREE_INSERT_NEED_JOURNAL_RECLAIM,
703 enum btree_gc_coalesce_fail_reason {
704 BTREE_GC_COALESCE_FAIL_RESERVE_GET,
705 BTREE_GC_COALESCE_FAIL_KEYLIST_REALLOC,
706 BTREE_GC_COALESCE_FAIL_FORMAT_FITS,
709 enum btree_node_sibling {
714 #endif /* _BCACHEFS_BTREE_TYPES_H */