1 #ifndef _BCACHEFS_BTREE_TYPES_H
2 #define _BCACHEFS_BTREE_TYPES_H
4 #include <linux/list.h>
5 #include <linux/rhashtable.h>
7 #include "bkey_methods.h"
8 #include "journal_types.h"
16 struct btree_nr_keys {
19 * Amount of live metadata (i.e. size of node after a compaction) in
23 u16 bset_u64s[MAX_BSETS];
32 * We construct a binary tree in an array as if the array
33 * started at 1, so that things line up on the same cachelines
34 * better: see comments in bset.c at cacheline_to_bkey() for
38 /* size of the binary tree and prev array */
41 /* function of size - precalculated for to_inorder() */
52 struct journal_entry_pin journal;
53 struct closure_waitlist wait;
57 struct open_buckets ob;
62 /* Hottest entries first */
63 struct rhash_head hash;
65 /* Key/pointer for this btree node */
66 __BKEY_PADDED(key, BKEY_BTREE_PTR_VAL_U64s_MAX);
77 struct bkey_format format;
79 struct btree_node *data;
83 * Sets of sorted keys - the real btree node - plus a binary search tree
85 * set[0] is special; set[0]->tree, set[0]->prev and set[0]->data point
86 * to the memory we have allocated for this btree node. Additionally,
87 * set[0]->data points to the entire btree node as it exists on disk.
89 struct bset_tree set[MAX_BSETS];
91 struct btree_nr_keys nr;
94 u16 uncompacted_whiteout_u64s;
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;
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 */
160 struct shrinker shrink;
163 * If we need to allocate memory for a new btree node and that
164 * allocation fails, we can cannibalize another node in the btree cache
165 * to satisfy the allocation - lock to guarantee only one thread does
168 struct task_struct *alloc_lock;
169 struct closure_waitlist alloc_wait;
172 struct btree_node_iter {
173 struct btree_node_iter_set {
178 enum btree_iter_type {
184 #define BTREE_ITER_TYPE ((1 << 2) - 1)
186 #define BTREE_ITER_INTENT (1 << 2)
187 #define BTREE_ITER_PREFETCH (1 << 3)
189 * Used in bch2_btree_iter_traverse(), to indicate whether we're searching for
190 * @pos or the first key strictly greater than @pos
192 #define BTREE_ITER_IS_EXTENTS (1 << 4)
193 #define BTREE_ITER_ERROR (1 << 5)
195 enum btree_iter_uptodate {
196 BTREE_ITER_UPTODATE = 0,
197 BTREE_ITER_NEED_PEEK = 1,
198 BTREE_ITER_NEED_RELOCK = 2,
199 BTREE_ITER_NEED_TRAVERSE = 3,
203 * @pos - iterator's current position
204 * @level - current btree depth
205 * @locks_want - btree level below which we start taking intent locks
206 * @nodes_locked - bitmask indicating which nodes in @nodes are locked
207 * @nodes_intent_locked - bitmask indicating which locks are intent locks
214 enum btree_iter_uptodate uptodate:4;
215 enum btree_id btree_id:4;
219 nodes_intent_locked:4;
221 struct btree_iter_level {
223 struct btree_node_iter iter;
225 } l[BTREE_MAX_DEPTH];
228 * Current unpacked key - so that bch2_btree_iter_next()/
229 * bch2_btree_iter_next_slot() can correctly advance pos.
234 * Circular linked list of linked iterators: linked iterators share
235 * locks (e.g. two linked iterators may have the same node intent
236 * locked, or read and write locked, at the same time), and insertions
237 * through one iterator won't invalidate the other linked iterators.
240 /* Must come last: */
241 struct btree_iter *next;
244 #define BTREE_ITER_MAX 8
246 struct btree_insert_entry {
247 struct btree_iter *iter;
264 struct btree_iter *iters;
265 u64 iter_ids[BTREE_ITER_MAX];
267 struct btree_insert_entry updates[BTREE_ITER_MAX];
269 struct btree_iter iters_onstack[2];
272 #define BTREE_FLAG(flag) \
273 static inline bool btree_node_ ## flag(struct btree *b) \
274 { return test_bit(BTREE_NODE_ ## flag, &b->flags); } \
276 static inline void set_btree_node_ ## flag(struct btree *b) \
277 { set_bit(BTREE_NODE_ ## flag, &b->flags); } \
279 static inline void clear_btree_node_ ## flag(struct btree *b) \
280 { clear_bit(BTREE_NODE_ ## flag, &b->flags); }
283 BTREE_NODE_read_in_flight,
284 BTREE_NODE_read_error,
286 BTREE_NODE_need_write,
288 BTREE_NODE_write_idx,
290 BTREE_NODE_write_in_flight,
291 BTREE_NODE_just_written,
296 BTREE_FLAG(read_in_flight);
297 BTREE_FLAG(read_error);
299 BTREE_FLAG(need_write);
301 BTREE_FLAG(write_idx);
302 BTREE_FLAG(accessed);
303 BTREE_FLAG(write_in_flight);
304 BTREE_FLAG(just_written);
308 static inline struct btree_write *btree_current_write(struct btree *b)
310 return b->writes + btree_node_write_idx(b);
313 static inline struct btree_write *btree_prev_write(struct btree *b)
315 return b->writes + (btree_node_write_idx(b) ^ 1);
318 static inline struct bset_tree *bset_tree_last(struct btree *b)
321 return b->set + b->nsets - 1;
325 __btree_node_offset_to_ptr(const struct btree *b, u16 offset)
327 return (void *) ((u64 *) b->data + 1 + offset);
331 __btree_node_ptr_to_offset(const struct btree *b, const void *p)
333 u16 ret = (u64 *) p - 1 - (u64 *) b->data;
335 EBUG_ON(__btree_node_offset_to_ptr(b, ret) != p);
339 static inline struct bset *bset(const struct btree *b,
340 const struct bset_tree *t)
342 return __btree_node_offset_to_ptr(b, t->data_offset);
345 static inline void set_btree_bset_end(struct btree *b, struct bset_tree *t)
348 __btree_node_ptr_to_offset(b, vstruct_last(bset(b, t)));
351 static inline void set_btree_bset(struct btree *b, struct bset_tree *t,
352 const struct bset *i)
354 t->data_offset = __btree_node_ptr_to_offset(b, i);
355 set_btree_bset_end(b, t);
358 static inline struct bset *btree_bset_first(struct btree *b)
360 return bset(b, b->set);
363 static inline struct bset *btree_bset_last(struct btree *b)
365 return bset(b, bset_tree_last(b));
369 __btree_node_key_to_offset(const struct btree *b, const struct bkey_packed *k)
371 return __btree_node_ptr_to_offset(b, k);
374 static inline struct bkey_packed *
375 __btree_node_offset_to_key(const struct btree *b, u16 k)
377 return __btree_node_offset_to_ptr(b, k);
380 static inline unsigned btree_bkey_first_offset(const struct bset_tree *t)
382 return t->data_offset + offsetof(struct bset, _data) / sizeof(u64);
385 #define btree_bkey_first(_b, _t) \
387 EBUG_ON(bset(_b, _t)->start != \
388 __btree_node_offset_to_key(_b, btree_bkey_first_offset(_t)));\
390 bset(_b, _t)->start; \
393 #define btree_bkey_last(_b, _t) \
395 EBUG_ON(__btree_node_offset_to_key(_b, (_t)->end_offset) != \
396 vstruct_last(bset(_b, _t))); \
398 __btree_node_offset_to_key(_b, (_t)->end_offset); \
401 static inline unsigned bset_byte_offset(struct btree *b, void *i)
403 return i - (void *) b->data;
406 /* Type of keys @b contains: */
407 static inline enum bkey_type btree_node_type(struct btree *b)
409 return b->level ? BKEY_TYPE_BTREE : b->btree_id;
412 static inline const struct bkey_ops *btree_node_ops(struct btree *b)
414 return &bch2_bkey_ops[btree_node_type(b)];
417 static inline bool btree_node_is_extents(struct btree *b)
419 return btree_node_type(b) == BKEY_TYPE_EXTENTS;
425 struct btree_update *as;
427 /* On disk root - see async splits: */
428 __BKEY_PADDED(key, BKEY_BTREE_PTR_VAL_U64s_MAX);
434 * Optional hook that will be called just prior to a btree node update, when
435 * we're holding the write lock and we know what key is about to be overwritten:
438 enum btree_insert_ret {
440 /* extent spanned multiple leaf nodes: have to traverse to next node: */
441 BTREE_INSERT_NEED_TRAVERSE,
442 /* write lock held for too long */
443 /* leaf node needs to be split */
444 BTREE_INSERT_BTREE_NODE_FULL,
446 BTREE_INSERT_NEED_GC_LOCK,
449 enum btree_gc_coalesce_fail_reason {
450 BTREE_GC_COALESCE_FAIL_RESERVE_GET,
451 BTREE_GC_COALESCE_FAIL_KEYLIST_REALLOC,
452 BTREE_GC_COALESCE_FAIL_FORMAT_FITS,
455 enum btree_node_sibling {
460 typedef struct btree_nr_keys (*sort_fix_overlapping_fn)(struct bset *,
462 struct btree_node_iter *);
464 #endif /* _BCACHEFS_BTREE_TYPES_H */