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() */
55 struct journal_entry_pin journal;
59 struct open_buckets ob;
63 struct btree_bkey_cached_common {
70 struct btree_bkey_cached_common c;
72 struct rhash_head hash;
80 struct bkey_format format;
82 struct btree_node *data;
86 * Sets of sorted keys - the real btree node - plus a binary search tree
88 * set[0] is special; set[0]->tree, set[0]->prev and set[0]->data point
89 * to the memory we have allocated for this btree node. Additionally,
90 * set[0]->data points to the entire btree node as it exists on disk.
92 struct bset_tree set[MAX_BSETS];
94 struct btree_nr_keys nr;
101 * XXX: add a delete sequence number, so when bch2_btree_node_relock()
102 * fails because the lock sequence number has changed - i.e. the
103 * contents were modified - we can still relock the node if it's still
104 * the one we want, without redoing the traversal
108 * For asynchronous splits/interior node updates:
109 * When we do a split, we allocate new child nodes and update the parent
110 * node to point to them: we update the parent in memory immediately,
111 * but then we must wait until the children have been written out before
112 * the update to the parent can be written - this is a list of the
113 * btree_updates that are blocking this node from being
116 struct list_head write_blocked;
119 * Also for asynchronous splits/interior node updates:
120 * If a btree node isn't reachable yet, we don't want to kick off
121 * another write - because that write also won't yet be reachable and
122 * marking it as completed before it's reachable would be incorrect:
124 unsigned long will_make_reachable;
126 struct open_buckets ob;
129 struct list_head list;
131 struct btree_write writes[2];
133 /* Key/pointer for this btree node */
134 __BKEY_PADDED(key, BKEY_BTREE_PTR_VAL_U64s_MAX);
138 struct rhashtable table;
139 bool table_init_done;
141 * We never free a struct btree, except on shutdown - we just put it on
142 * the btree_cache_freed list and reuse it later. This simplifies the
143 * code, and it doesn't cost us much memory as the memory usage is
144 * dominated by buffers that hold the actual btree node data and those
145 * can be freed - and the number of struct btrees allocated is
146 * effectively bounded.
148 * btree_cache_freeable effectively is a small cache - we use it because
149 * high order page allocations can be rather expensive, and it's quite
150 * common to delete and allocate btree nodes in quick succession. It
151 * should never grow past ~2-3 nodes in practice.
154 struct list_head live;
155 struct list_head freeable;
156 struct list_head freed;
158 /* 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 {
179 enum btree_iter_type {
185 #define BTREE_ITER_TYPE ((1 << 2) - 1)
188 * Iterate over all possible positions, synthesizing deleted keys for holes:
190 #define BTREE_ITER_SLOTS (1 << 2)
192 * Indicates that intent locks should be taken on leaf nodes, because we expect
193 * to be doing updates:
195 #define BTREE_ITER_INTENT (1 << 3)
197 * Causes the btree iterator code to prefetch additional btree nodes from disk:
199 #define BTREE_ITER_PREFETCH (1 << 4)
201 * Indicates that this iterator should not be reused until transaction commit,
202 * either because a pending update references it or because the update depends
203 * on that particular key being locked (e.g. by the str_hash code, for hash
206 #define BTREE_ITER_KEEP_UNTIL_COMMIT (1 << 5)
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 << 6)
212 #define BTREE_ITER_ERROR (1 << 7)
213 #define BTREE_ITER_SET_POS_AFTER_COMMIT (1 << 8)
214 #define BTREE_ITER_CACHED_NOFILL (1 << 9)
215 #define BTREE_ITER_CACHED_NOCREATE (1 << 10)
217 #define BTREE_ITER_USER_FLAGS \
220 |BTREE_ITER_PREFETCH \
221 |BTREE_ITER_CACHED_NOFILL \
222 |BTREE_ITER_CACHED_NOCREATE)
224 enum btree_iter_uptodate {
225 BTREE_ITER_UPTODATE = 0,
226 BTREE_ITER_NEED_PEEK = 1,
227 BTREE_ITER_NEED_RELOCK = 2,
228 BTREE_ITER_NEED_TRAVERSE = 3,
231 #define BTREE_ITER_NO_NODE_GET_LOCKS ((struct btree *) 1)
232 #define BTREE_ITER_NO_NODE_DROP ((struct btree *) 2)
233 #define BTREE_ITER_NO_NODE_LOCK_ROOT ((struct btree *) 3)
234 #define BTREE_ITER_NO_NODE_UP ((struct btree *) 4)
235 #define BTREE_ITER_NO_NODE_DOWN ((struct btree *) 5)
236 #define BTREE_ITER_NO_NODE_INIT ((struct btree *) 6)
237 #define BTREE_ITER_NO_NODE_ERROR ((struct btree *) 7)
240 * @pos - iterator's current position
241 * @level - current btree depth
242 * @locks_want - btree level below which we start taking intent locks
243 * @nodes_locked - bitmask indicating which nodes in @nodes are locked
244 * @nodes_intent_locked - bitmask indicating which locks are intent locks
247 struct btree_trans *trans;
249 struct bpos pos_after_commit;
254 enum btree_id btree_id:4;
255 enum btree_iter_uptodate uptodate:4;
260 nodes_intent_locked:4;
262 struct btree_iter_level {
264 struct btree_node_iter iter;
266 } l[BTREE_MAX_DEPTH];
269 * Current unpacked key - so that bch2_btree_iter_next()/
270 * bch2_btree_iter_next_slot() can correctly advance pos.
273 unsigned long ip_allocated;
276 static inline enum btree_iter_type
277 btree_iter_type(const struct btree_iter *iter)
279 return iter->flags & BTREE_ITER_TYPE;
282 static inline bool btree_iter_is_cached(const struct btree_iter *iter)
284 return btree_iter_type(iter) == BTREE_ITER_CACHED;
287 static inline struct btree_iter_level *iter_l(struct btree_iter *iter)
289 return iter->l + iter->level;
292 struct btree_key_cache {
294 struct rhashtable table;
295 struct list_head freed;
296 struct list_head clean;
299 struct bkey_cached_key {
302 } __attribute__((packed, aligned(4)));
304 #define BKEY_CACHED_DIRTY 0
307 struct btree_bkey_cached_common c;
312 struct bkey_cached_key key;
314 struct rhash_head hash;
315 struct list_head list;
317 struct journal_preres res;
318 struct journal_entry_pin journal;
323 struct btree_insert_entry {
324 unsigned trigger_flags;
325 unsigned trans_triggers_run:1;
327 struct btree_iter *iter;
330 #ifndef CONFIG_LOCKDEP
331 #define BTREE_ITER_MAX 64
333 #define BTREE_ITER_MAX 32
338 #ifdef CONFIG_BCACHEFS_DEBUG
339 struct list_head list;
340 struct btree *locking;
341 unsigned locking_iter_idx;
342 struct bpos locking_pos;
357 unsigned used_mempool:1;
360 unsigned need_reset:1;
361 unsigned in_traverse_all:1;
367 struct btree_iter *iters;
368 struct btree_insert_entry *updates;
369 struct btree_insert_entry *updates2;
372 struct jset_entry *extra_journal_entries;
373 unsigned extra_journal_entry_u64s;
374 struct journal_entry_pin *journal_pin;
376 struct journal_res journal_res;
377 struct journal_preres journal_preres;
379 struct disk_reservation *disk_res;
381 unsigned journal_u64s;
382 unsigned journal_preres_u64s;
383 struct replicas_delta_list *fs_usage_deltas;
386 #define BTREE_FLAG(flag) \
387 static inline bool btree_node_ ## flag(struct btree *b) \
388 { return test_bit(BTREE_NODE_ ## flag, &b->flags); } \
390 static inline void set_btree_node_ ## flag(struct btree *b) \
391 { set_bit(BTREE_NODE_ ## flag, &b->flags); } \
393 static inline void clear_btree_node_ ## flag(struct btree *b) \
394 { clear_bit(BTREE_NODE_ ## flag, &b->flags); }
397 BTREE_NODE_read_in_flight,
398 BTREE_NODE_read_error,
400 BTREE_NODE_need_write,
402 BTREE_NODE_write_idx,
404 BTREE_NODE_write_in_flight,
405 BTREE_NODE_just_written,
408 BTREE_NODE_old_extent_overwrite,
409 BTREE_NODE_need_rewrite,
412 BTREE_FLAG(read_in_flight);
413 BTREE_FLAG(read_error);
415 BTREE_FLAG(need_write);
417 BTREE_FLAG(write_idx);
418 BTREE_FLAG(accessed);
419 BTREE_FLAG(write_in_flight);
420 BTREE_FLAG(just_written);
423 BTREE_FLAG(old_extent_overwrite);
424 BTREE_FLAG(need_rewrite);
426 static inline struct btree_write *btree_current_write(struct btree *b)
428 return b->writes + btree_node_write_idx(b);
431 static inline struct btree_write *btree_prev_write(struct btree *b)
433 return b->writes + (btree_node_write_idx(b) ^ 1);
436 static inline struct bset_tree *bset_tree_last(struct btree *b)
439 return b->set + b->nsets - 1;
443 __btree_node_offset_to_ptr(const struct btree *b, u16 offset)
445 return (void *) ((u64 *) b->data + 1 + offset);
449 __btree_node_ptr_to_offset(const struct btree *b, const void *p)
451 u16 ret = (u64 *) p - 1 - (u64 *) b->data;
453 EBUG_ON(__btree_node_offset_to_ptr(b, ret) != p);
457 static inline struct bset *bset(const struct btree *b,
458 const struct bset_tree *t)
460 return __btree_node_offset_to_ptr(b, t->data_offset);
463 static inline void set_btree_bset_end(struct btree *b, struct bset_tree *t)
466 __btree_node_ptr_to_offset(b, vstruct_last(bset(b, t)));
469 static inline void set_btree_bset(struct btree *b, struct bset_tree *t,
470 const struct bset *i)
472 t->data_offset = __btree_node_ptr_to_offset(b, i);
473 set_btree_bset_end(b, t);
476 static inline struct bset *btree_bset_first(struct btree *b)
478 return bset(b, b->set);
481 static inline struct bset *btree_bset_last(struct btree *b)
483 return bset(b, bset_tree_last(b));
487 __btree_node_key_to_offset(const struct btree *b, const struct bkey_packed *k)
489 return __btree_node_ptr_to_offset(b, k);
492 static inline struct bkey_packed *
493 __btree_node_offset_to_key(const struct btree *b, u16 k)
495 return __btree_node_offset_to_ptr(b, k);
498 static inline unsigned btree_bkey_first_offset(const struct bset_tree *t)
500 return t->data_offset + offsetof(struct bset, _data) / sizeof(u64);
503 #define btree_bkey_first(_b, _t) \
505 EBUG_ON(bset(_b, _t)->start != \
506 __btree_node_offset_to_key(_b, btree_bkey_first_offset(_t)));\
508 bset(_b, _t)->start; \
511 #define btree_bkey_last(_b, _t) \
513 EBUG_ON(__btree_node_offset_to_key(_b, (_t)->end_offset) != \
514 vstruct_last(bset(_b, _t))); \
516 __btree_node_offset_to_key(_b, (_t)->end_offset); \
519 static inline unsigned bset_u64s(struct bset_tree *t)
521 return t->end_offset - t->data_offset -
522 sizeof(struct bset) / sizeof(u64);
525 static inline unsigned bset_dead_u64s(struct btree *b, struct bset_tree *t)
527 return bset_u64s(t) - b->nr.bset_u64s[t - b->set];
530 static inline unsigned bset_byte_offset(struct btree *b, void *i)
532 return i - (void *) b->data;
535 enum btree_node_type {
536 #define x(kwd, val, name) BKEY_TYPE_##kwd = val,
542 /* Type of a key in btree @id at level @level: */
543 static inline enum btree_node_type __btree_node_type(unsigned level, enum btree_id id)
545 return level ? BKEY_TYPE_BTREE : (enum btree_node_type) id;
548 /* Type of keys @b contains: */
549 static inline enum btree_node_type btree_node_type(struct btree *b)
551 return __btree_node_type(b->c.level, b->c.btree_id);
554 static inline bool btree_node_type_is_extents(enum btree_node_type type)
557 case BKEY_TYPE_EXTENTS:
558 case BKEY_TYPE_REFLINK:
565 static inline bool btree_node_is_extents(struct btree *b)
567 return btree_node_type_is_extents(btree_node_type(b));
570 static inline enum btree_node_type btree_iter_key_type(struct btree_iter *iter)
572 return __btree_node_type(iter->level, iter->btree_id);
575 static inline bool btree_iter_is_extents(struct btree_iter *iter)
577 return btree_node_type_is_extents(btree_iter_key_type(iter));
580 #define BTREE_NODE_TYPE_HAS_TRIGGERS \
581 ((1U << BKEY_TYPE_EXTENTS)| \
582 (1U << BKEY_TYPE_ALLOC)| \
583 (1U << BKEY_TYPE_INODES)| \
584 (1U << BKEY_TYPE_REFLINK)| \
585 (1U << BKEY_TYPE_EC)| \
586 (1U << BKEY_TYPE_BTREE))
588 #define BTREE_NODE_TYPE_HAS_TRANS_TRIGGERS \
589 ((1U << BKEY_TYPE_EXTENTS)| \
590 (1U << BKEY_TYPE_INODES)| \
591 (1U << BKEY_TYPE_EC)| \
592 (1U << BKEY_TYPE_REFLINK))
594 enum btree_trigger_flags {
595 __BTREE_TRIGGER_NORUN, /* Don't run triggers at all */
597 __BTREE_TRIGGER_INSERT,
598 __BTREE_TRIGGER_OVERWRITE,
599 __BTREE_TRIGGER_OVERWRITE_SPLIT,
602 __BTREE_TRIGGER_BUCKET_INVALIDATE,
603 __BTREE_TRIGGER_NOATOMIC,
606 #define BTREE_TRIGGER_NORUN (1U << __BTREE_TRIGGER_NORUN)
608 #define BTREE_TRIGGER_INSERT (1U << __BTREE_TRIGGER_INSERT)
609 #define BTREE_TRIGGER_OVERWRITE (1U << __BTREE_TRIGGER_OVERWRITE)
610 #define BTREE_TRIGGER_OVERWRITE_SPLIT (1U << __BTREE_TRIGGER_OVERWRITE_SPLIT)
612 #define BTREE_TRIGGER_GC (1U << __BTREE_TRIGGER_GC)
613 #define BTREE_TRIGGER_BUCKET_INVALIDATE (1U << __BTREE_TRIGGER_BUCKET_INVALIDATE)
614 #define BTREE_TRIGGER_NOATOMIC (1U << __BTREE_TRIGGER_NOATOMIC)
616 static inline bool btree_node_type_needs_gc(enum btree_node_type type)
618 return BTREE_NODE_TYPE_HAS_TRIGGERS & (1U << type);
624 /* On disk root - see async splits: */
625 __BKEY_PADDED(key, BKEY_BTREE_PTR_VAL_U64s_MAX);
632 * Optional hook that will be called just prior to a btree node update, when
633 * we're holding the write lock and we know what key is about to be overwritten:
636 enum btree_insert_ret {
638 /* leaf node needs to be split */
639 BTREE_INSERT_BTREE_NODE_FULL,
641 BTREE_INSERT_NEED_MARK_REPLICAS,
642 BTREE_INSERT_NEED_JOURNAL_RES,
645 enum btree_gc_coalesce_fail_reason {
646 BTREE_GC_COALESCE_FAIL_RESERVE_GET,
647 BTREE_GC_COALESCE_FAIL_KEYLIST_REALLOC,
648 BTREE_GC_COALESCE_FAIL_FORMAT_FITS,
651 enum btree_node_sibling {
656 typedef struct btree_nr_keys (*sort_fix_overlapping_fn)(struct bset *,
658 struct btree_node_iter *);
660 #endif /* _BCACHEFS_BTREE_TYPES_H */