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"
20 struct btree_nr_keys {
23 * Amount of live metadata (i.e. size of node after a compaction) in
27 u16 bset_u64s[MAX_BSETS];
36 * We construct a binary tree in an array as if the array
37 * started at 1, so that things line up on the same cachelines
38 * better: see comments in bset.c at cacheline_to_bkey() for
42 /* size of the binary tree and prev array */
45 /* function of size - precalculated for to_inorder() */
54 struct journal_entry_pin journal;
58 struct open_buckets ob;
59 __BKEY_PADDED(k, BKEY_BTREE_PTR_VAL_U64s_MAX);
62 struct btree_bkey_cached_common {
70 struct btree_bkey_cached_common c;
72 struct rhash_head hash;
81 struct bkey_format format;
83 struct btree_node *data;
87 * Sets of sorted keys - the real btree node - plus a binary search tree
89 * set[0] is special; set[0]->tree, set[0]->prev and set[0]->data point
90 * to the memory we have allocated for this btree node. Additionally,
91 * set[0]->data points to the entire btree node as it exists on disk.
93 struct bset_tree set[MAX_BSETS];
95 struct btree_nr_keys nr;
101 struct btree_write writes[2];
103 /* Key/pointer for this btree node */
104 __BKEY_PADDED(key, BKEY_BTREE_PTR_VAL_U64s_MAX);
107 * XXX: add a delete sequence number, so when bch2_btree_node_relock()
108 * fails because the lock sequence number has changed - i.e. the
109 * contents were modified - we can still relock the node if it's still
110 * the one we want, without redoing the traversal
114 * For asynchronous splits/interior node updates:
115 * When we do a split, we allocate new child nodes and update the parent
116 * node to point to them: we update the parent in memory immediately,
117 * but then we must wait until the children have been written out before
118 * the update to the parent can be written - this is a list of the
119 * btree_updates that are blocking this node from being
122 struct list_head write_blocked;
125 * Also for asynchronous splits/interior node updates:
126 * If a btree node isn't reachable yet, we don't want to kick off
127 * another write - because that write also won't yet be reachable and
128 * marking it as completed before it's reachable would be incorrect:
130 unsigned long will_make_reachable;
132 struct open_buckets ob;
135 struct list_head list;
139 struct rhashtable table;
140 bool table_init_done;
142 * We never free a struct btree, except on shutdown - we just put it on
143 * the btree_cache_freed list and reuse it later. This simplifies the
144 * code, and it doesn't cost us much memory as the memory usage is
145 * dominated by buffers that hold the actual btree node data and those
146 * can be freed - and the number of struct btrees allocated is
147 * effectively bounded.
149 * btree_cache_freeable effectively is a small cache - we use it because
150 * high order page allocations can be rather expensive, and it's quite
151 * common to delete and allocate btree nodes in quick succession. It
152 * should never grow past ~2-3 nodes in practice.
155 struct list_head live;
156 struct list_head freeable;
157 struct list_head freed_pcpu;
158 struct list_head freed_nonpcpu;
160 /* Number of elements in live + freeable lists */
164 struct shrinker shrink;
167 * If we need to allocate memory for a new btree node and that
168 * allocation fails, we can cannibalize another node in the btree cache
169 * to satisfy the allocation - lock to guarantee only one thread does
172 struct task_struct *alloc_lock;
173 struct closure_waitlist alloc_wait;
176 struct btree_node_iter {
177 struct btree_node_iter_set {
183 * Iterate over all possible positions, synthesizing deleted keys for holes:
185 #define BTREE_ITER_SLOTS (1 << 0)
186 #define BTREE_ITER_ALL_LEVELS (1 << 1)
188 * Indicates that intent locks should be taken on leaf nodes, because we expect
189 * to be doing updates:
191 #define BTREE_ITER_INTENT (1 << 2)
193 * Causes the btree iterator code to prefetch additional btree nodes from disk:
195 #define BTREE_ITER_PREFETCH (1 << 3)
197 * Used in bch2_btree_iter_traverse(), to indicate whether we're searching for
198 * @pos or the first key strictly greater than @pos
200 #define BTREE_ITER_IS_EXTENTS (1 << 4)
201 #define BTREE_ITER_NOT_EXTENTS (1 << 5)
202 #define BTREE_ITER_CACHED (1 << 6)
203 #define BTREE_ITER_WITH_KEY_CACHE (1 << 7)
204 #define BTREE_ITER_WITH_UPDATES (1 << 8)
205 #define BTREE_ITER_WITH_JOURNAL (1 << 9)
206 #define __BTREE_ITER_ALL_SNAPSHOTS (1 << 10)
207 #define BTREE_ITER_ALL_SNAPSHOTS (1 << 11)
208 #define BTREE_ITER_FILTER_SNAPSHOTS (1 << 12)
209 #define BTREE_ITER_NOPRESERVE (1 << 13)
211 enum btree_path_uptodate {
212 BTREE_ITER_UPTODATE = 0,
213 BTREE_ITER_NEED_RELOCK = 1,
214 BTREE_ITER_NEED_TRAVERSE = 2,
223 /* btree_iter_copy starts here: */
226 enum btree_id btree_id:4;
229 enum btree_path_uptodate uptodate:2;
231 * When true, failing to relock this path will cause the transaction to
234 bool should_be_locked:1;
239 struct btree_path_level {
241 struct btree_node_iter iter;
243 #ifdef CONFIG_BCACHEFS_LOCK_TIME_STATS
246 } l[BTREE_MAX_DEPTH];
247 #ifdef CONFIG_BCACHEFS_DEBUG
248 unsigned long ip_allocated;
252 static inline struct btree_path_level *path_l(struct btree_path *path)
254 return path->l + path->level;
258 * @pos - iterator's current position
259 * @level - current btree depth
260 * @locks_want - btree level below which we start taking intent locks
261 * @nodes_locked - bitmask indicating which nodes in @nodes are locked
262 * @nodes_intent_locked - bitmask indicating which locks are intent locks
265 struct btree_trans *trans;
266 struct btree_path *path;
267 struct btree_path *update_path;
268 struct btree_path *key_cache_path;
270 enum btree_id btree_id:4;
271 unsigned min_depth:3;
274 /* btree_iter_copy starts here: */
277 /* When we're filtering by snapshot, the snapshot ID we're looking for: */
281 struct bpos pos_after_commit;
283 * Current unpacked key - so that bch2_btree_iter_next()/
284 * bch2_btree_iter_next_slot() can correctly advance pos.
288 /* BTREE_ITER_WITH_JOURNAL: */
290 struct bpos journal_pos;
291 #ifdef CONFIG_BCACHEFS_DEBUG
292 unsigned long ip_allocated;
296 struct btree_key_cache_freelist {
297 struct bkey_cached *objs[16];
301 struct btree_key_cache {
303 struct rhashtable table;
304 bool table_init_done;
305 struct list_head freed_pcpu;
306 struct list_head freed_nonpcpu;
307 struct shrinker shrink;
308 unsigned shrink_iter;
309 struct btree_key_cache_freelist __percpu *pcpu_freed;
311 atomic_long_t nr_freed;
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 static inline struct bpos btree_node_pos(struct btree_bkey_cached_common *b)
345 ? container_of(b, struct btree, c)->key.k.p
346 : container_of(b, struct bkey_cached, c)->key.pos;
349 struct btree_insert_entry {
352 enum btree_id btree_id:8;
355 bool insert_trigger_run:1;
356 bool overwrite_trigger_run:1;
357 bool key_cache_already_flushed:1;
359 * @old_k may be a key from the journal; @old_btree_u64s always refers
360 * to the size of the key being overwritten in the btree:
364 struct btree_path *path;
365 /* key being overwritten: */
367 const struct bch_val *old_v;
368 unsigned long ip_allocated;
371 #ifndef CONFIG_LOCKDEP
372 #define BTREE_ITER_MAX 64
374 #define BTREE_ITER_MAX 32
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;
408 bool in_traverse_all:1;
409 bool memory_allocation_failure:1;
410 bool is_initial_gc:1;
411 enum bch_errcode restarted:16;
413 unsigned long last_restarted_ip;
416 * For when bch2_trans_update notices we'll be splitting a compressed
419 unsigned extra_journal_res;
420 unsigned nr_max_paths;
429 u8 sorted[BTREE_ITER_MAX];
430 struct btree_path *paths;
431 struct btree_insert_entry *updates;
434 struct btree_trans_commit_hook *hooks;
435 DARRAY(u64) extra_journal_entries;
436 struct journal_entry_pin *journal_pin;
438 struct journal_res journal_res;
439 struct journal_preres journal_preres;
441 struct disk_reservation *disk_res;
443 unsigned journal_u64s;
444 unsigned journal_preres_u64s;
445 struct replicas_delta_list *fs_usage_deltas;
448 #define BTREE_FLAGS() \
454 x(will_make_reachable) \
459 x(write_in_flight_inner) \
467 #define x(flag) BTREE_NODE_##flag,
473 static inline bool btree_node_ ## flag(struct btree *b) \
474 { return test_bit(BTREE_NODE_ ## flag, &b->flags); } \
476 static inline void set_btree_node_ ## flag(struct btree *b) \
477 { set_bit(BTREE_NODE_ ## flag, &b->flags); } \
479 static inline void clear_btree_node_ ## flag(struct btree *b) \
480 { clear_bit(BTREE_NODE_ ## flag, &b->flags); }
485 static inline struct btree_write *btree_current_write(struct btree *b)
487 return b->writes + btree_node_write_idx(b);
490 static inline struct btree_write *btree_prev_write(struct btree *b)
492 return b->writes + (btree_node_write_idx(b) ^ 1);
495 static inline struct bset_tree *bset_tree_last(struct btree *b)
498 return b->set + b->nsets - 1;
502 __btree_node_offset_to_ptr(const struct btree *b, u16 offset)
504 return (void *) ((u64 *) b->data + 1 + offset);
508 __btree_node_ptr_to_offset(const struct btree *b, const void *p)
510 u16 ret = (u64 *) p - 1 - (u64 *) b->data;
512 EBUG_ON(__btree_node_offset_to_ptr(b, ret) != p);
516 static inline struct bset *bset(const struct btree *b,
517 const struct bset_tree *t)
519 return __btree_node_offset_to_ptr(b, t->data_offset);
522 static inline void set_btree_bset_end(struct btree *b, struct bset_tree *t)
525 __btree_node_ptr_to_offset(b, vstruct_last(bset(b, t)));
528 static inline void set_btree_bset(struct btree *b, struct bset_tree *t,
529 const struct bset *i)
531 t->data_offset = __btree_node_ptr_to_offset(b, i);
532 set_btree_bset_end(b, t);
535 static inline struct bset *btree_bset_first(struct btree *b)
537 return bset(b, b->set);
540 static inline struct bset *btree_bset_last(struct btree *b)
542 return bset(b, bset_tree_last(b));
546 __btree_node_key_to_offset(const struct btree *b, const struct bkey_packed *k)
548 return __btree_node_ptr_to_offset(b, k);
551 static inline struct bkey_packed *
552 __btree_node_offset_to_key(const struct btree *b, u16 k)
554 return __btree_node_offset_to_ptr(b, k);
557 static inline unsigned btree_bkey_first_offset(const struct bset_tree *t)
559 return t->data_offset + offsetof(struct bset, _data) / sizeof(u64);
562 #define btree_bkey_first(_b, _t) \
564 EBUG_ON(bset(_b, _t)->start != \
565 __btree_node_offset_to_key(_b, btree_bkey_first_offset(_t)));\
567 bset(_b, _t)->start; \
570 #define btree_bkey_last(_b, _t) \
572 EBUG_ON(__btree_node_offset_to_key(_b, (_t)->end_offset) != \
573 vstruct_last(bset(_b, _t))); \
575 __btree_node_offset_to_key(_b, (_t)->end_offset); \
578 static inline unsigned bset_u64s(struct bset_tree *t)
580 return t->end_offset - t->data_offset -
581 sizeof(struct bset) / sizeof(u64);
584 static inline unsigned bset_dead_u64s(struct btree *b, struct bset_tree *t)
586 return bset_u64s(t) - b->nr.bset_u64s[t - b->set];
589 static inline unsigned bset_byte_offset(struct btree *b, void *i)
591 return i - (void *) b->data;
594 enum btree_node_type {
595 #define x(kwd, val) BKEY_TYPE_##kwd = val,
601 /* Type of a key in btree @id at level @level: */
602 static inline enum btree_node_type __btree_node_type(unsigned level, enum btree_id id)
604 return level ? BKEY_TYPE_btree : (enum btree_node_type) id;
607 /* Type of keys @b contains: */
608 static inline enum btree_node_type btree_node_type(struct btree *b)
610 return __btree_node_type(b->c.level, b->c.btree_id);
613 #define BTREE_NODE_TYPE_HAS_TRANS_TRIGGERS \
614 ((1U << BKEY_TYPE_extents)| \
615 (1U << BKEY_TYPE_alloc)| \
616 (1U << BKEY_TYPE_inodes)| \
617 (1U << BKEY_TYPE_stripes)| \
618 (1U << BKEY_TYPE_reflink)| \
619 (1U << BKEY_TYPE_btree))
621 #define BTREE_NODE_TYPE_HAS_MEM_TRIGGERS \
622 ((1U << BKEY_TYPE_alloc)| \
623 (1U << BKEY_TYPE_inodes)| \
624 (1U << BKEY_TYPE_stripes)| \
625 (1U << BKEY_TYPE_snapshots))
627 #define BTREE_NODE_TYPE_HAS_TRIGGERS \
628 (BTREE_NODE_TYPE_HAS_TRANS_TRIGGERS| \
629 BTREE_NODE_TYPE_HAS_MEM_TRIGGERS)
631 #define BTREE_ID_IS_EXTENTS \
632 ((1U << BTREE_ID_extents)| \
633 (1U << BTREE_ID_reflink)| \
634 (1U << BTREE_ID_freespace))
636 static inline bool btree_node_type_is_extents(enum btree_node_type type)
638 return (1U << type) & BTREE_ID_IS_EXTENTS;
641 #define BTREE_ID_HAS_SNAPSHOTS \
642 ((1U << BTREE_ID_extents)| \
643 (1U << BTREE_ID_inodes)| \
644 (1U << BTREE_ID_dirents)| \
645 (1U << BTREE_ID_xattrs))
647 #define BTREE_ID_HAS_PTRS \
648 ((1U << BTREE_ID_extents)| \
649 (1U << BTREE_ID_reflink))
651 static inline bool btree_type_has_snapshots(enum btree_id id)
653 return (1 << id) & BTREE_ID_HAS_SNAPSHOTS;
656 static inline bool btree_type_has_ptrs(enum btree_id id)
658 return (1 << id) & BTREE_ID_HAS_PTRS;
661 static inline bool btree_node_type_needs_gc(enum btree_node_type type)
663 return BTREE_NODE_TYPE_HAS_TRIGGERS & (1U << type);
669 /* On disk root - see async splits: */
670 __BKEY_PADDED(key, BKEY_BTREE_PTR_VAL_U64s_MAX);
676 enum btree_insert_ret {
678 /* leaf node needs to be split */
679 BTREE_INSERT_BTREE_NODE_FULL,
680 BTREE_INSERT_NEED_MARK_REPLICAS,
681 BTREE_INSERT_NEED_JOURNAL_RES,
682 BTREE_INSERT_NEED_JOURNAL_RECLAIM,
685 enum btree_gc_coalesce_fail_reason {
686 BTREE_GC_COALESCE_FAIL_RESERVE_GET,
687 BTREE_GC_COALESCE_FAIL_KEYLIST_REALLOC,
688 BTREE_GC_COALESCE_FAIL_FORMAT_FITS,
691 enum btree_node_sibling {
696 #endif /* _BCACHEFS_BTREE_TYPES_H */