1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _BCACHE_BTREE_H
3 #define _BCACHE_BTREE_H
8 * At a high level, bcache's btree is relatively standard b+ tree. All keys and
9 * pointers are in the leaves; interior nodes only have pointers to the child
12 * In the interior nodes, a struct bkey always points to a child btree node, and
13 * the key is the highest key in the child node - except that the highest key in
14 * an interior node is always MAX_KEY. The size field refers to the size on disk
15 * of the child node - this would allow us to have variable sized btree nodes
16 * (handy for keeping the depth of the btree 1 by expanding just the root).
18 * Btree nodes are themselves log structured, but this is hidden fairly
19 * thoroughly. Btree nodes on disk will in practice have extents that overlap
20 * (because they were written at different times), but in memory we never have
21 * overlapping extents - when we read in a btree node from disk, the first thing
22 * we do is resort all the sets of keys with a mergesort, and in the same pass
23 * we check for overlapping extents and adjust them appropriately.
25 * struct btree_op is a central interface to the btree code. It's used for
26 * specifying read vs. write locking, and the embedded closure is used for
27 * waiting on IO or reserve memory.
31 * Btree nodes are cached in memory; traversing the btree might require reading
32 * in btree nodes which is handled mostly transparently.
34 * bch_btree_node_get() looks up a btree node in the cache and reads it in from
35 * disk if necessary. This function is almost never called directly though - the
36 * btree() macro is used to get a btree node, call some function on it, and
37 * unlock the node after the function returns.
39 * The root is special cased - it's taken out of the cache's lru (thus pinning
40 * it in memory), so we can find the root of the btree by just dereferencing a
41 * pointer instead of looking it up in the cache. This makes locking a bit
42 * tricky, since the root pointer is protected by the lock in the btree node it
43 * points to - the btree_root() macro handles this.
45 * In various places we must be able to allocate memory for multiple btree nodes
46 * in order to make forward progress. To do this we use the btree cache itself
47 * as a reserve; if __get_free_pages() fails, we'll find a node in the btree
48 * cache we can reuse. We can't allow more than one thread to be doing this at a
49 * time, so there's a lock, implemented by a pointer to the btree_op closure -
50 * this allows the btree_root() macro to implicitly release this lock.
54 * Btree nodes never have to be explicitly read in; bch_btree_node_get() handles
57 * For writing, we have two btree_write structs embeddded in struct btree - one
58 * write in flight, and one being set up, and we toggle between them.
60 * Writing is done with a single function - bch_btree_write() really serves two
61 * different purposes and should be broken up into two different functions. When
62 * passing now = false, it merely indicates that the node is now dirty - calling
63 * it ensures that the dirty keys will be written at some point in the future.
65 * When passing now = true, bch_btree_write() causes a write to happen
66 * "immediately" (if there was already a write in flight, it'll cause the write
67 * to happen as soon as the previous write completes). It returns immediately
68 * though - but it takes a refcount on the closure in struct btree_op you passed
69 * to it, so a closure_sync() later can be used to wait for the write to
72 * This is handy because btree_split() and garbage collection can issue writes
73 * in parallel, reducing the amount of time they have to hold write locks.
77 * When traversing the btree, we may need write locks starting at some level -
78 * inserting a key into the btree will typically only require a write lock on
81 * This is specified with the lock field in struct btree_op; lock = 0 means we
82 * take write locks at level <= 0, i.e. only leaf nodes. bch_btree_node_get()
83 * checks this field and returns the node with the appropriate lock held.
85 * If, after traversing the btree, the insertion code discovers it has to split
86 * then it must restart from the root and take new locks - to do this it changes
87 * the lock field and returns -EINTR, which causes the btree_root() macro to
90 * Handling cache misses require a different mechanism for upgrading to a write
91 * lock. We do cache lookups with only a read lock held, but if we get a cache
92 * miss and we wish to insert this data into the cache, we have to insert a
93 * placeholder key to detect races - otherwise, we could race with a write and
94 * overwrite the data that was just written to the cache with stale data from
97 * For this we use a sequence number that write locks and unlocks increment - to
98 * insert the check key it unlocks the btree node and then takes a write lock,
99 * and fails if the sequence number doesn't match.
108 /* If btree_split() frees a btree node, it writes a new pointer to that
109 * btree node indicating it was freed; it takes a refcount on
110 * c->prio_blocked because we can't write the gens until the new
111 * pointer is on disk. This allows btree_write_endio() to release the
112 * refcount that btree_split() took.
118 /* Hottest entries first */
119 struct hlist_node hash;
121 /* Key/pointer for this btree node */
124 /* Single bit - set when accessed, cleared by shrinker */
125 unsigned long accessed;
127 struct rw_semaphore lock;
129 struct btree *parent;
131 struct mutex write_lock;
134 uint16_t written; /* would be nice to kill */
137 struct btree_keys keys;
139 /* For outstanding btree writes, used as a lock - protects write_idx */
141 struct semaphore io_mutex;
143 struct list_head list;
144 struct delayed_work work;
146 struct btree_write writes[2];
150 #define BTREE_FLAG(flag) \
151 static inline bool btree_node_ ## flag(struct btree *b) \
152 { return test_bit(BTREE_NODE_ ## flag, &b->flags); } \
154 static inline void set_btree_node_ ## flag(struct btree *b) \
155 { set_bit(BTREE_NODE_ ## flag, &b->flags); }
160 BTREE_NODE_write_idx,
161 BTREE_NODE_journal_flush,
164 BTREE_FLAG(io_error);
166 BTREE_FLAG(write_idx);
167 BTREE_FLAG(journal_flush);
169 static inline struct btree_write *btree_current_write(struct btree *b)
171 return b->writes + btree_node_write_idx(b);
174 static inline struct btree_write *btree_prev_write(struct btree *b)
176 return b->writes + (btree_node_write_idx(b) ^ 1);
179 static inline struct bset *btree_bset_first(struct btree *b)
181 return b->keys.set->data;
184 static inline struct bset *btree_bset_last(struct btree *b)
186 return bset_tree_last(&b->keys)->data;
189 static inline unsigned int bset_block_offset(struct btree *b, struct bset *i)
191 return bset_sector_offset(&b->keys, i) >> b->c->block_bits;
194 static inline void set_gc_sectors(struct cache_set *c)
196 atomic_set(&c->sectors_to_gc, c->sb.bucket_size * c->nbuckets / 16);
199 void bkey_put(struct cache_set *c, struct bkey *k);
203 #define for_each_cached_btree(b, c, iter) \
205 iter < ARRAY_SIZE((c)->bucket_hash); \
207 hlist_for_each_entry_rcu((b), (c)->bucket_hash + iter, hash)
209 /* Recursing down the btree */
212 /* for waiting on btree reserve in btree_split() */
213 wait_queue_entry_t wait;
215 /* Btree level at which we start taking write locks */
218 unsigned int insert_collision:1;
221 static inline void bch_btree_op_init(struct btree_op *op, int write_lock_level)
223 memset(op, 0, sizeof(struct btree_op));
224 init_wait(&op->wait);
225 op->lock = write_lock_level;
228 static inline void rw_lock(bool w, struct btree *b, int level)
230 w ? down_write_nested(&b->lock, level + 1)
231 : down_read_nested(&b->lock, level + 1);
236 static inline void rw_unlock(bool w, struct btree *b)
240 (w ? up_write : up_read)(&b->lock);
243 void bch_btree_node_read_done(struct btree *b);
244 void __bch_btree_node_write(struct btree *b, struct closure *parent);
245 void bch_btree_node_write(struct btree *b, struct closure *parent);
247 void bch_btree_set_root(struct btree *b);
248 struct btree *__bch_btree_node_alloc(struct cache_set *c, struct btree_op *op,
249 int level, bool wait,
250 struct btree *parent);
251 struct btree *bch_btree_node_get(struct cache_set *c, struct btree_op *op,
252 struct bkey *k, int level, bool write,
253 struct btree *parent);
255 int bch_btree_insert_check_key(struct btree *b, struct btree_op *op,
256 struct bkey *check_key);
257 int bch_btree_insert(struct cache_set *c, struct keylist *keys,
258 atomic_t *journal_ref, struct bkey *replace_key);
260 int bch_gc_thread_start(struct cache_set *c);
261 void bch_initial_gc_finish(struct cache_set *c);
262 void bch_moving_gc(struct cache_set *c);
263 int bch_btree_check(struct cache_set *c);
264 void bch_initial_mark_key(struct cache_set *c, int level, struct bkey *k);
266 static inline void wake_up_gc(struct cache_set *c)
268 wake_up(&c->gc_wait);
272 #define MAP_CONTINUE 1
274 #define MAP_ALL_NODES 0
275 #define MAP_LEAF_NODES 1
277 #define MAP_END_KEY 1
279 typedef int (btree_map_nodes_fn)(struct btree_op *b_op, struct btree *b);
280 int __bch_btree_map_nodes(struct btree_op *op, struct cache_set *c,
281 struct bkey *from, btree_map_nodes_fn *fn, int flags);
283 static inline int bch_btree_map_nodes(struct btree_op *op, struct cache_set *c,
284 struct bkey *from, btree_map_nodes_fn *fn)
286 return __bch_btree_map_nodes(op, c, from, fn, MAP_ALL_NODES);
289 static inline int bch_btree_map_leaf_nodes(struct btree_op *op,
292 btree_map_nodes_fn *fn)
294 return __bch_btree_map_nodes(op, c, from, fn, MAP_LEAF_NODES);
297 typedef int (btree_map_keys_fn)(struct btree_op *op, struct btree *b,
299 int bch_btree_map_keys(struct btree_op *op, struct cache_set *c,
300 struct bkey *from, btree_map_keys_fn *fn, int flags);
302 typedef bool (keybuf_pred_fn)(struct keybuf *buf, struct bkey *k);
304 void bch_keybuf_init(struct keybuf *buf);
305 void bch_refill_keybuf(struct cache_set *c, struct keybuf *buf,
306 struct bkey *end, keybuf_pred_fn *pred);
307 bool bch_keybuf_check_overlapping(struct keybuf *buf, struct bkey *start,
309 void bch_keybuf_del(struct keybuf *buf, struct keybuf_key *w);
310 struct keybuf_key *bch_keybuf_next(struct keybuf *buf);
311 struct keybuf_key *bch_keybuf_next_rescan(struct cache_set *c,
314 keybuf_pred_fn *pred);
315 void bch_update_bucket_in_use(struct cache_set *c, struct gc_stat *stats);