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;
68 /* Key/pointer for this btree node */
69 __BKEY_PADDED(key, BKEY_BTREE_PTR_VAL_U64s_MAX);
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 #ifdef CONFIG_BCACHEFS_DEBUG
134 bool *expensive_debug_checks;
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;
159 /* Number of elements in live + freeable lists */
162 struct shrinker shrink;
165 * If we need to allocate memory for a new btree node and that
166 * allocation fails, we can cannibalize another node in the btree cache
167 * to satisfy the allocation - lock to guarantee only one thread does
170 struct task_struct *alloc_lock;
171 struct closure_waitlist alloc_wait;
174 struct btree_node_iter {
175 struct btree_node_iter_set {
180 enum btree_iter_type {
185 #define BTREE_ITER_TYPE ((1 << 2) - 1)
187 #define BTREE_ITER_SLOTS (1 << 2)
188 #define BTREE_ITER_INTENT (1 << 3)
189 #define BTREE_ITER_PREFETCH (1 << 4)
190 #define BTREE_ITER_KEEP_UNTIL_COMMIT (1 << 5)
192 * Used in bch2_btree_iter_traverse(), to indicate whether we're searching for
193 * @pos or the first key strictly greater than @pos
195 #define BTREE_ITER_IS_EXTENTS (1 << 6)
196 #define BTREE_ITER_ERROR (1 << 7)
198 enum btree_iter_uptodate {
199 BTREE_ITER_UPTODATE = 0,
200 BTREE_ITER_NEED_PEEK = 1,
201 BTREE_ITER_NEED_RELOCK = 2,
202 BTREE_ITER_NEED_TRAVERSE = 3,
206 * @pos - iterator's current position
207 * @level - current btree depth
208 * @locks_want - btree level below which we start taking intent locks
209 * @nodes_locked - bitmask indicating which nodes in @nodes are locked
210 * @nodes_intent_locked - bitmask indicating which locks are intent locks
215 struct btree_trans *trans;
219 enum btree_iter_uptodate uptodate:4;
220 enum btree_id btree_id:4;
224 nodes_intent_locked:4;
226 struct btree_iter_level {
228 struct btree_node_iter iter;
230 } l[BTREE_MAX_DEPTH];
233 * Current unpacked key - so that bch2_btree_iter_next()/
234 * bch2_btree_iter_next_slot() can correctly advance pos.
239 static inline enum btree_iter_type btree_iter_type(struct btree_iter *iter)
241 return iter->flags & BTREE_ITER_TYPE;
244 struct btree_insert_entry {
246 struct btree_iter *iter;
249 #define BTREE_ITER_MAX 64
262 unsigned used_mempool:1;
270 struct btree_iter *iters;
271 struct btree_insert_entry *updates;
275 struct journal_res journal_res;
276 struct journal_preres journal_preres;
278 struct disk_reservation *disk_res;
280 unsigned journal_u64s;
281 unsigned journal_preres_u64s;
282 struct replicas_delta_list *fs_usage_deltas;
284 struct btree_iter iters_onstack[2];
285 struct btree_insert_entry updates_onstack[6];
286 u8 updates_sorted_onstack[6];
289 #define BTREE_FLAG(flag) \
290 static inline bool btree_node_ ## flag(struct btree *b) \
291 { return test_bit(BTREE_NODE_ ## flag, &b->flags); } \
293 static inline void set_btree_node_ ## flag(struct btree *b) \
294 { set_bit(BTREE_NODE_ ## flag, &b->flags); } \
296 static inline void clear_btree_node_ ## flag(struct btree *b) \
297 { clear_bit(BTREE_NODE_ ## flag, &b->flags); }
300 BTREE_NODE_read_in_flight,
301 BTREE_NODE_read_error,
303 BTREE_NODE_need_write,
305 BTREE_NODE_write_idx,
307 BTREE_NODE_write_in_flight,
308 BTREE_NODE_just_written,
313 BTREE_FLAG(read_in_flight);
314 BTREE_FLAG(read_error);
316 BTREE_FLAG(need_write);
318 BTREE_FLAG(write_idx);
319 BTREE_FLAG(accessed);
320 BTREE_FLAG(write_in_flight);
321 BTREE_FLAG(just_written);
325 static inline struct btree_write *btree_current_write(struct btree *b)
327 return b->writes + btree_node_write_idx(b);
330 static inline struct btree_write *btree_prev_write(struct btree *b)
332 return b->writes + (btree_node_write_idx(b) ^ 1);
335 static inline struct bset_tree *bset_tree_last(struct btree *b)
338 return b->set + b->nsets - 1;
342 __btree_node_offset_to_ptr(const struct btree *b, u16 offset)
344 return (void *) ((u64 *) b->data + 1 + offset);
348 __btree_node_ptr_to_offset(const struct btree *b, const void *p)
350 u16 ret = (u64 *) p - 1 - (u64 *) b->data;
352 EBUG_ON(__btree_node_offset_to_ptr(b, ret) != p);
356 static inline struct bset *bset(const struct btree *b,
357 const struct bset_tree *t)
359 return __btree_node_offset_to_ptr(b, t->data_offset);
362 static inline void set_btree_bset_end(struct btree *b, struct bset_tree *t)
365 __btree_node_ptr_to_offset(b, vstruct_last(bset(b, t)));
368 static inline void set_btree_bset(struct btree *b, struct bset_tree *t,
369 const struct bset *i)
371 t->data_offset = __btree_node_ptr_to_offset(b, i);
372 set_btree_bset_end(b, t);
375 static inline struct bset *btree_bset_first(struct btree *b)
377 return bset(b, b->set);
380 static inline struct bset *btree_bset_last(struct btree *b)
382 return bset(b, bset_tree_last(b));
386 __btree_node_key_to_offset(const struct btree *b, const struct bkey_packed *k)
388 return __btree_node_ptr_to_offset(b, k);
391 static inline struct bkey_packed *
392 __btree_node_offset_to_key(const struct btree *b, u16 k)
394 return __btree_node_offset_to_ptr(b, k);
397 static inline unsigned btree_bkey_first_offset(const struct bset_tree *t)
399 return t->data_offset + offsetof(struct bset, _data) / sizeof(u64);
402 #define btree_bkey_first(_b, _t) \
404 EBUG_ON(bset(_b, _t)->start != \
405 __btree_node_offset_to_key(_b, btree_bkey_first_offset(_t)));\
407 bset(_b, _t)->start; \
410 #define btree_bkey_last(_b, _t) \
412 EBUG_ON(__btree_node_offset_to_key(_b, (_t)->end_offset) != \
413 vstruct_last(bset(_b, _t))); \
415 __btree_node_offset_to_key(_b, (_t)->end_offset); \
418 static inline unsigned bset_u64s(struct bset_tree *t)
420 return t->end_offset - t->data_offset -
421 sizeof(struct bset) / sizeof(u64);
424 static inline unsigned bset_dead_u64s(struct btree *b, struct bset_tree *t)
426 return bset_u64s(t) - b->nr.bset_u64s[t - b->set];
429 static inline unsigned bset_byte_offset(struct btree *b, void *i)
431 return i - (void *) b->data;
434 enum btree_node_type {
435 #define x(kwd, val, name) BKEY_TYPE_##kwd = val,
441 /* Type of a key in btree @id at level @level: */
442 static inline enum btree_node_type __btree_node_type(unsigned level, enum btree_id id)
444 return level ? BKEY_TYPE_BTREE : (enum btree_node_type) id;
447 /* Type of keys @b contains: */
448 static inline enum btree_node_type btree_node_type(struct btree *b)
450 return __btree_node_type(b->level, b->btree_id);
453 static inline bool btree_node_type_is_extents(enum btree_node_type type)
456 case BKEY_TYPE_EXTENTS:
457 case BKEY_TYPE_REFLINK:
464 static inline bool btree_node_is_extents(struct btree *b)
466 return btree_node_type_is_extents(btree_node_type(b));
469 #define BTREE_NODE_TYPE_HAS_TRIGGERS \
470 ((1U << BKEY_TYPE_EXTENTS)| \
471 (1U << BKEY_TYPE_ALLOC)| \
472 (1U << BKEY_TYPE_INODES)| \
473 (1U << BKEY_TYPE_REFLINK)| \
474 (1U << BKEY_TYPE_EC)| \
475 (1U << BKEY_TYPE_BTREE))
477 #define BTREE_NODE_TYPE_HAS_TRANS_TRIGGERS \
478 ((1U << BKEY_TYPE_EXTENTS)| \
479 (1U << BKEY_TYPE_INODES)| \
480 (1U << BKEY_TYPE_REFLINK))
482 static inline bool btree_node_type_needs_gc(enum btree_node_type type)
484 return BTREE_NODE_TYPE_HAS_TRIGGERS & (1U << type);
490 struct btree_update *as;
492 /* On disk root - see async splits: */
493 __BKEY_PADDED(key, BKEY_BTREE_PTR_VAL_U64s_MAX);
500 * Optional hook that will be called just prior to a btree node update, when
501 * we're holding the write lock and we know what key is about to be overwritten:
504 enum btree_insert_ret {
506 /* leaf node needs to be split */
507 BTREE_INSERT_BTREE_NODE_FULL,
509 BTREE_INSERT_NEED_MARK_REPLICAS,
510 BTREE_INSERT_NEED_JOURNAL_RES,
513 enum btree_gc_coalesce_fail_reason {
514 BTREE_GC_COALESCE_FAIL_RESERVE_GET,
515 BTREE_GC_COALESCE_FAIL_KEYLIST_REALLOC,
516 BTREE_GC_COALESCE_FAIL_FORMAT_FITS,
519 enum btree_node_sibling {
524 typedef struct btree_nr_keys (*sort_fix_overlapping_fn)(struct bset *,
526 struct btree_node_iter *);
528 #endif /* _BCACHEFS_BTREE_TYPES_H */