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;
56 struct closure_waitlist wait;
60 struct open_buckets ob;
65 /* Hottest entries first */
66 struct rhash_head hash;
78 struct bkey_format format;
80 struct btree_node *data;
84 * Sets of sorted keys - the real btree node - plus a binary search tree
86 * set[0] is special; set[0]->tree, set[0]->prev and set[0]->data point
87 * to the memory we have allocated for this btree node. Additionally,
88 * set[0]->data points to the entire btree node as it exists on disk.
90 struct bset_tree set[MAX_BSETS];
92 struct btree_nr_keys nr;
99 * XXX: add a delete sequence number, so when bch2_btree_node_relock()
100 * fails because the lock sequence number has changed - i.e. the
101 * contents were modified - we can still relock the node if it's still
102 * the one we want, without redoing the traversal
106 * For asynchronous splits/interior node updates:
107 * When we do a split, we allocate new child nodes and update the parent
108 * node to point to them: we update the parent in memory immediately,
109 * but then we must wait until the children have been written out before
110 * the update to the parent can be written - this is a list of the
111 * btree_updates that are blocking this node from being
114 struct list_head write_blocked;
117 * Also for asynchronous splits/interior node updates:
118 * If a btree node isn't reachable yet, we don't want to kick off
119 * another write - because that write also won't yet be reachable and
120 * marking it as completed before it's reachable would be incorrect:
122 unsigned long will_make_reachable;
124 struct open_buckets ob;
127 struct list_head list;
129 struct btree_write writes[2];
131 #ifdef CONFIG_BCACHEFS_DEBUG
132 bool *expensive_debug_checks;
135 /* Key/pointer for this btree node */
136 __BKEY_PADDED(key, BKEY_BTREE_PTR_VAL_U64s_MAX);
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;
160 /* Number of elements in live + freeable lists */
163 struct shrinker shrink;
166 * If we need to allocate memory for a new btree node and that
167 * allocation fails, we can cannibalize another node in the btree cache
168 * to satisfy the allocation - lock to guarantee only one thread does
171 struct task_struct *alloc_lock;
172 struct closure_waitlist alloc_wait;
175 struct btree_node_iter {
176 struct btree_node_iter_set {
181 enum btree_iter_type {
186 #define BTREE_ITER_TYPE ((1 << 2) - 1)
189 * Iterate over all possible positions, synthesizing deleted keys for holes:
191 #define BTREE_ITER_SLOTS (1 << 2)
193 * Indicates that intent locks should be taken on leaf nodes, because we expect
194 * to be doing updates:
196 #define BTREE_ITER_INTENT (1 << 3)
198 * Causes the btree iterator code to prefetch additional btree nodes from disk:
200 #define BTREE_ITER_PREFETCH (1 << 4)
202 * Indicates that this iterator should not be reused until transaction commit,
203 * either because a pending update references it or because the update depends
204 * on that particular key being locked (e.g. by the str_hash code, for hash
207 #define BTREE_ITER_KEEP_UNTIL_COMMIT (1 << 5)
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 << 6)
213 #define BTREE_ITER_ERROR (1 << 7)
214 #define BTREE_ITER_SET_POS_AFTER_COMMIT (1 << 8)
216 enum btree_iter_uptodate {
217 BTREE_ITER_UPTODATE = 0,
218 BTREE_ITER_NEED_PEEK = 1,
219 BTREE_ITER_NEED_RELOCK = 2,
220 BTREE_ITER_NEED_TRAVERSE = 3,
224 * @pos - iterator's current position
225 * @level - current btree depth
226 * @locks_want - btree level below which we start taking intent locks
227 * @nodes_locked - bitmask indicating which nodes in @nodes are locked
228 * @nodes_intent_locked - bitmask indicating which locks are intent locks
231 struct btree_trans *trans;
233 struct bpos pos_after_commit;
238 enum btree_id btree_id:4;
239 enum btree_iter_uptodate uptodate:4;
244 nodes_intent_locked:4;
246 struct btree_iter_level {
248 struct btree_node_iter iter;
250 } l[BTREE_MAX_DEPTH];
253 * Current unpacked key - so that bch2_btree_iter_next()/
254 * bch2_btree_iter_next_slot() can correctly advance pos.
259 static inline enum btree_iter_type btree_iter_type(struct btree_iter *iter)
261 return iter->flags & BTREE_ITER_TYPE;
264 struct btree_insert_entry {
265 unsigned trigger_flags;
266 unsigned trans_triggers_run:1;
268 struct btree_iter *iter;
271 #define BTREE_ITER_MAX 64
285 unsigned used_mempool:1;
288 unsigned need_reset:1;
294 struct btree_iter *iters;
295 struct btree_insert_entry *updates;
296 struct btree_insert_entry *updates2;
299 struct journal_res journal_res;
300 struct journal_preres journal_preres;
302 struct disk_reservation *disk_res;
304 unsigned journal_u64s;
305 unsigned journal_preres_u64s;
306 struct replicas_delta_list *fs_usage_deltas;
308 struct btree_iter iters_onstack[2];
309 struct btree_insert_entry updates_onstack[2];
310 struct btree_insert_entry updates2_onstack[2];
313 #define BTREE_FLAG(flag) \
314 static inline bool btree_node_ ## flag(struct btree *b) \
315 { return test_bit(BTREE_NODE_ ## flag, &b->flags); } \
317 static inline void set_btree_node_ ## flag(struct btree *b) \
318 { set_bit(BTREE_NODE_ ## flag, &b->flags); } \
320 static inline void clear_btree_node_ ## flag(struct btree *b) \
321 { clear_bit(BTREE_NODE_ ## flag, &b->flags); }
324 BTREE_NODE_read_in_flight,
325 BTREE_NODE_read_error,
327 BTREE_NODE_need_write,
329 BTREE_NODE_write_idx,
331 BTREE_NODE_write_in_flight,
332 BTREE_NODE_just_written,
335 BTREE_NODE_old_extent_overwrite,
338 BTREE_FLAG(read_in_flight);
339 BTREE_FLAG(read_error);
341 BTREE_FLAG(need_write);
343 BTREE_FLAG(write_idx);
344 BTREE_FLAG(accessed);
345 BTREE_FLAG(write_in_flight);
346 BTREE_FLAG(just_written);
349 BTREE_FLAG(old_extent_overwrite);
351 static inline struct btree_write *btree_current_write(struct btree *b)
353 return b->writes + btree_node_write_idx(b);
356 static inline struct btree_write *btree_prev_write(struct btree *b)
358 return b->writes + (btree_node_write_idx(b) ^ 1);
361 static inline struct bset_tree *bset_tree_last(struct btree *b)
364 return b->set + b->nsets - 1;
368 __btree_node_offset_to_ptr(const struct btree *b, u16 offset)
370 return (void *) ((u64 *) b->data + 1 + offset);
374 __btree_node_ptr_to_offset(const struct btree *b, const void *p)
376 u16 ret = (u64 *) p - 1 - (u64 *) b->data;
378 EBUG_ON(__btree_node_offset_to_ptr(b, ret) != p);
382 static inline struct bset *bset(const struct btree *b,
383 const struct bset_tree *t)
385 return __btree_node_offset_to_ptr(b, t->data_offset);
388 static inline void set_btree_bset_end(struct btree *b, struct bset_tree *t)
391 __btree_node_ptr_to_offset(b, vstruct_last(bset(b, t)));
394 static inline void set_btree_bset(struct btree *b, struct bset_tree *t,
395 const struct bset *i)
397 t->data_offset = __btree_node_ptr_to_offset(b, i);
398 set_btree_bset_end(b, t);
401 static inline struct bset *btree_bset_first(struct btree *b)
403 return bset(b, b->set);
406 static inline struct bset *btree_bset_last(struct btree *b)
408 return bset(b, bset_tree_last(b));
412 __btree_node_key_to_offset(const struct btree *b, const struct bkey_packed *k)
414 return __btree_node_ptr_to_offset(b, k);
417 static inline struct bkey_packed *
418 __btree_node_offset_to_key(const struct btree *b, u16 k)
420 return __btree_node_offset_to_ptr(b, k);
423 static inline unsigned btree_bkey_first_offset(const struct bset_tree *t)
425 return t->data_offset + offsetof(struct bset, _data) / sizeof(u64);
428 #define btree_bkey_first(_b, _t) \
430 EBUG_ON(bset(_b, _t)->start != \
431 __btree_node_offset_to_key(_b, btree_bkey_first_offset(_t)));\
433 bset(_b, _t)->start; \
436 #define btree_bkey_last(_b, _t) \
438 EBUG_ON(__btree_node_offset_to_key(_b, (_t)->end_offset) != \
439 vstruct_last(bset(_b, _t))); \
441 __btree_node_offset_to_key(_b, (_t)->end_offset); \
444 static inline unsigned bset_u64s(struct bset_tree *t)
446 return t->end_offset - t->data_offset -
447 sizeof(struct bset) / sizeof(u64);
450 static inline unsigned bset_dead_u64s(struct btree *b, struct bset_tree *t)
452 return bset_u64s(t) - b->nr.bset_u64s[t - b->set];
455 static inline unsigned bset_byte_offset(struct btree *b, void *i)
457 return i - (void *) b->data;
460 enum btree_node_type {
461 #define x(kwd, val, name) BKEY_TYPE_##kwd = val,
467 /* Type of a key in btree @id at level @level: */
468 static inline enum btree_node_type __btree_node_type(unsigned level, enum btree_id id)
470 return level ? BKEY_TYPE_BTREE : (enum btree_node_type) id;
473 /* Type of keys @b contains: */
474 static inline enum btree_node_type btree_node_type(struct btree *b)
476 return __btree_node_type(b->level, b->btree_id);
479 static inline bool btree_node_type_is_extents(enum btree_node_type type)
482 case BKEY_TYPE_EXTENTS:
483 case BKEY_TYPE_REFLINK:
490 static inline bool btree_node_is_extents(struct btree *b)
492 return btree_node_type_is_extents(btree_node_type(b));
495 #define BTREE_NODE_TYPE_HAS_TRIGGERS \
496 ((1U << BKEY_TYPE_EXTENTS)| \
497 (1U << BKEY_TYPE_ALLOC)| \
498 (1U << BKEY_TYPE_INODES)| \
499 (1U << BKEY_TYPE_REFLINK)| \
500 (1U << BKEY_TYPE_EC)| \
501 (1U << BKEY_TYPE_BTREE))
503 #define BTREE_NODE_TYPE_HAS_TRANS_TRIGGERS \
504 ((1U << BKEY_TYPE_EXTENTS)| \
505 (1U << BKEY_TYPE_INODES)| \
506 (1U << BKEY_TYPE_REFLINK))
508 enum btree_trigger_flags {
509 __BTREE_TRIGGER_NORUN, /* Don't run triggers at all */
510 __BTREE_TRIGGER_NOOVERWRITES, /* Don't run triggers on overwrites */
512 __BTREE_TRIGGER_INSERT,
513 __BTREE_TRIGGER_OVERWRITE,
514 __BTREE_TRIGGER_OVERWRITE_SPLIT,
517 __BTREE_TRIGGER_BUCKET_INVALIDATE,
518 __BTREE_TRIGGER_ALLOC_READ,
519 __BTREE_TRIGGER_NOATOMIC,
522 #define BTREE_TRIGGER_NORUN (1U << __BTREE_TRIGGER_NORUN)
523 #define BTREE_TRIGGER_NOOVERWRITES (1U << __BTREE_TRIGGER_NOOVERWRITES)
525 #define BTREE_TRIGGER_INSERT (1U << __BTREE_TRIGGER_INSERT)
526 #define BTREE_TRIGGER_OVERWRITE (1U << __BTREE_TRIGGER_OVERWRITE)
527 #define BTREE_TRIGGER_OVERWRITE_SPLIT (1U << __BTREE_TRIGGER_OVERWRITE_SPLIT)
529 #define BTREE_TRIGGER_GC (1U << __BTREE_TRIGGER_GC)
530 #define BTREE_TRIGGER_BUCKET_INVALIDATE (1U << __BTREE_TRIGGER_BUCKET_INVALIDATE)
531 #define BTREE_TRIGGER_ALLOC_READ (1U << __BTREE_TRIGGER_ALLOC_READ)
532 #define BTREE_TRIGGER_NOATOMIC (1U << __BTREE_TRIGGER_NOATOMIC)
534 static inline bool btree_node_type_needs_gc(enum btree_node_type type)
536 return BTREE_NODE_TYPE_HAS_TRIGGERS & (1U << type);
542 struct btree_update *as;
544 /* On disk root - see async splits: */
545 __BKEY_PADDED(key, BKEY_BTREE_PTR_VAL_U64s_MAX);
552 * Optional hook that will be called just prior to a btree node update, when
553 * we're holding the write lock and we know what key is about to be overwritten:
556 enum btree_insert_ret {
558 /* leaf node needs to be split */
559 BTREE_INSERT_BTREE_NODE_FULL,
561 BTREE_INSERT_NEED_MARK_REPLICAS,
562 BTREE_INSERT_NEED_JOURNAL_RES,
565 enum btree_gc_coalesce_fail_reason {
566 BTREE_GC_COALESCE_FAIL_RESERVE_GET,
567 BTREE_GC_COALESCE_FAIL_KEYLIST_REALLOC,
568 BTREE_GC_COALESCE_FAIL_FORMAT_FITS,
571 enum btree_node_sibling {
576 typedef struct btree_nr_keys (*sort_fix_overlapping_fn)(struct bset *,
578 struct btree_node_iter *);
580 #endif /* _BCACHEFS_BTREE_TYPES_H */