1 #ifndef _BCACHE_BTREE_H
2 #define _BCACHE_BTREE_H
7 * At a high level, bcache's btree is relatively standard b+ tree. All keys and
8 * pointers are in the leaves; interior nodes only have pointers to the child
11 * In the interior nodes, a struct bkey always points to a child btree node, and
12 * the key is the highest key in the child node - except that the highest key in
13 * an interior node is always MAX_KEY. The size field refers to the size on disk
14 * of the child node - this would allow us to have variable sized btree nodes
15 * (handy for keeping the depth of the btree 1 by expanding just the root).
17 * Btree nodes are themselves log structured, but this is hidden fairly
18 * thoroughly. Btree nodes on disk will in practice have extents that overlap
19 * (because they were written at different times), but in memory we never have
20 * overlapping extents - when we read in a btree node from disk, the first thing
21 * we do is resort all the sets of keys with a mergesort, and in the same pass
22 * we check for overlapping extents and adjust them appropriately.
24 * struct btree_op is a central interface to the btree code. It's used for
25 * specifying read vs. write locking, and the embedded closure is used for
26 * waiting on IO or reserve memory.
30 * Btree nodes are cached in memory; traversing the btree might require reading
31 * in btree nodes which is handled mostly transparently.
33 * bch_btree_node_get() looks up a btree node in the cache and reads it in from
34 * disk if necessary. This function is almost never called directly though - the
35 * btree() macro is used to get a btree node, call some function on it, and
36 * unlock the node after the function returns.
38 * The root is special cased - it's taken out of the cache's lru (thus pinning
39 * it in memory), so we can find the root of the btree by just dereferencing a
40 * pointer instead of looking it up in the cache. This makes locking a bit
41 * tricky, since the root pointer is protected by the lock in the btree node it
42 * points to - the btree_root() macro handles this.
44 * In various places we must be able to allocate memory for multiple btree nodes
45 * in order to make forward progress. To do this we use the btree cache itself
46 * as a reserve; if __get_free_pages() fails, we'll find a node in the btree
47 * cache we can reuse. We can't allow more than one thread to be doing this at a
48 * time, so there's a lock, implemented by a pointer to the btree_op closure -
49 * this allows the btree_root() macro to implicitly release this lock.
53 * Btree nodes never have to be explicitly read in; bch_btree_node_get() handles
56 * For writing, we have two btree_write structs embeddded in struct btree - one
57 * write in flight, and one being set up, and we toggle between them.
59 * Writing is done with a single function - bch_btree_write() really serves two
60 * different purposes and should be broken up into two different functions. When
61 * passing now = false, it merely indicates that the node is now dirty - calling
62 * it ensures that the dirty keys will be written at some point in the future.
64 * When passing now = true, bch_btree_write() causes a write to happen
65 * "immediately" (if there was already a write in flight, it'll cause the write
66 * to happen as soon as the previous write completes). It returns immediately
67 * though - but it takes a refcount on the closure in struct btree_op you passed
68 * to it, so a closure_sync() later can be used to wait for the write to
71 * This is handy because btree_split() and garbage collection can issue writes
72 * in parallel, reducing the amount of time they have to hold write locks.
76 * When traversing the btree, we may need write locks starting at some level -
77 * inserting a key into the btree will typically only require a write lock on
80 * This is specified with the lock field in struct btree_op; lock = 0 means we
81 * take write locks at level <= 0, i.e. only leaf nodes. bch_btree_node_get()
82 * checks this field and returns the node with the appropriate lock held.
84 * If, after traversing the btree, the insertion code discovers it has to split
85 * then it must restart from the root and take new locks - to do this it changes
86 * the lock field and returns -EINTR, which causes the btree_root() macro to
89 * Handling cache misses require a different mechanism for upgrading to a write
90 * lock. We do cache lookups with only a read lock held, but if we get a cache
91 * miss and we wish to insert this data into the cache, we have to insert a
92 * placeholder key to detect races - otherwise, we could race with a write and
93 * overwrite the data that was just written to the cache with stale data from
96 * For this we use a sequence number that write locks and unlocks increment - to
97 * insert the check key it unlocks the btree node and then takes a write lock,
98 * and fails if the sequence number doesn't match.
107 /* If btree_split() frees a btree node, it writes a new pointer to that
108 * btree node indicating it was freed; it takes a refcount on
109 * c->prio_blocked because we can't write the gens until the new
110 * pointer is on disk. This allows btree_write_endio() to release the
111 * refcount that btree_split() took.
117 /* Hottest entries first */
118 struct hlist_node hash;
120 /* Key/pointer for this btree node */
123 /* Single bit - set when accessed, cleared by shrinker */
124 unsigned long accessed;
126 struct rw_semaphore lock;
128 struct btree *parent;
131 uint16_t written; /* would be nice to kill */
137 * Set of sorted keys - the real btree node - plus a binary search tree
139 * sets[0] is special; set[0]->tree, set[0]->prev and set[0]->data point
140 * to the memory we have allocated for this btree node. Additionally,
141 * set[0]->data points to the entire btree node as it exists on disk.
143 struct bset_tree sets[MAX_BSETS];
145 /* For outstanding btree writes, used as a lock - protects write_idx */
146 struct closure_with_waitlist io;
148 struct list_head list;
149 struct delayed_work work;
151 struct btree_write writes[2];
155 #define BTREE_FLAG(flag) \
156 static inline bool btree_node_ ## flag(struct btree *b) \
157 { return test_bit(BTREE_NODE_ ## flag, &b->flags); } \
159 static inline void set_btree_node_ ## flag(struct btree *b) \
160 { set_bit(BTREE_NODE_ ## flag, &b->flags); } \
165 BTREE_NODE_write_idx,
168 BTREE_FLAG(io_error);
170 BTREE_FLAG(write_idx);
172 static inline struct btree_write *btree_current_write(struct btree *b)
174 return b->writes + btree_node_write_idx(b);
177 static inline struct btree_write *btree_prev_write(struct btree *b)
179 return b->writes + (btree_node_write_idx(b) ^ 1);
182 static inline unsigned bset_offset(struct btree *b, struct bset *i)
184 return (((size_t) i) - ((size_t) b->sets->data)) >> 9;
187 static inline struct bset *write_block(struct btree *b)
189 return ((void *) b->sets[0].data) + b->written * block_bytes(b->c);
192 static inline bool bset_written(struct btree *b, struct bset_tree *t)
194 return t->data < write_block(b);
197 static inline bool bkey_written(struct btree *b, struct bkey *k)
199 return k < write_block(b)->start;
202 static inline void set_gc_sectors(struct cache_set *c)
204 atomic_set(&c->sectors_to_gc, c->sb.bucket_size * c->nbuckets / 8);
207 static inline bool bch_ptr_invalid(struct btree *b, const struct bkey *k)
209 return __bch_ptr_invalid(b->c, b->level, k);
212 static inline struct bkey *bch_btree_iter_init(struct btree *b,
213 struct btree_iter *iter,
216 return __bch_btree_iter_init(b, iter, search, b->sets);
219 void __bkey_put(struct cache_set *c, struct bkey *k);
223 #define for_each_cached_btree(b, c, iter) \
225 iter < ARRAY_SIZE((c)->bucket_hash); \
227 hlist_for_each_entry_rcu((b), (c)->bucket_hash + iter, hash)
229 #define for_each_key_filter(b, k, iter, filter) \
230 for (bch_btree_iter_init((b), (iter), NULL); \
231 ((k) = bch_btree_iter_next_filter((iter), b, filter));)
233 #define for_each_key(b, k, iter) \
234 for (bch_btree_iter_init((b), (iter), NULL); \
235 ((k) = bch_btree_iter_next(iter));)
237 /* Recursing down the btree */
243 /* Journal entry we have a refcount on */
246 /* Bio to be inserted into the cache */
247 struct bio *cache_bio;
253 /* Btree level at which we start taking write locks */
256 /* Btree insertion type */
264 unsigned flush_journal:1;
266 unsigned insert_data_done:1;
267 unsigned lookup_done:1;
268 unsigned insert_collision:1;
270 BKEY_PADDED(replace);
274 BTREE_INSERT_STATUS_INSERT,
275 BTREE_INSERT_STATUS_BACK_MERGE,
276 BTREE_INSERT_STATUS_OVERWROTE,
277 BTREE_INSERT_STATUS_FRONT_MERGE,
280 void bch_btree_op_init_stack(struct btree_op *);
282 static inline void rw_lock(bool w, struct btree *b, int level)
284 w ? down_write_nested(&b->lock, level + 1)
285 : down_read_nested(&b->lock, level + 1);
290 static inline void rw_unlock(bool w, struct btree *b)
292 #ifdef CONFIG_BCACHE_EDEBUG
295 if (w && b->key.ptr[0])
296 for (i = 0; i <= b->nsets; i++)
297 bch_check_key_order(b, b->sets[i].data);
302 (w ? up_write : up_read)(&b->lock);
305 #define insert_lock(s, b) ((b)->level <= (s)->lock)
308 * These macros are for recursing down the btree - they handle the details of
309 * locking and looking up nodes in the cache for you. They're best treated as
310 * mere syntax when reading code that uses them.
312 * op->lock determines whether we take a read or a write lock at a given depth.
313 * If you've got a read lock and find that you need a write lock (i.e. you're
314 * going to have to split), set op->lock and return -EINTR; btree_root() will
315 * call you again and you'll have the correct lock.
319 * btree - recurse down the btree on a specified key
320 * @fn: function to call, which will be passed the child node
321 * @key: key to recurse on
322 * @b: parent btree node
323 * @op: pointer to struct btree_op
325 #define btree(fn, key, b, op, ...) \
327 int _r, l = (b)->level - 1; \
328 bool _w = l <= (op)->lock; \
329 struct btree *_child = bch_btree_node_get((b)->c, key, l, _w); \
330 if (!IS_ERR(_child)) { \
331 _child->parent = (b); \
332 _r = bch_btree_ ## fn(_child, op, ##__VA_ARGS__); \
333 rw_unlock(_w, _child); \
335 _r = PTR_ERR(_child); \
340 * btree_root - call a function on the root of the btree
341 * @fn: function to call, which will be passed the child node
343 * @op: pointer to struct btree_op
345 #define btree_root(fn, c, op, ...) \
349 struct btree *_b = (c)->root; \
350 bool _w = insert_lock(op, _b); \
351 rw_lock(_w, _b, _b->level); \
352 if (_b == (c)->root && \
353 _w == insert_lock(op, _b)) { \
355 _r = bch_btree_ ## fn(_b, op, ##__VA_ARGS__); \
358 bch_cannibalize_unlock(c); \
359 if (_r == -ENOSPC) { \
360 wait_event((c)->try_wait, \
364 } while (_r == -EINTR); \
369 static inline bool should_split(struct btree *b)
371 struct bset *i = write_block(b);
372 return b->written >= btree_blocks(b) ||
373 (b->written + __set_blocks(i, i->keys + 15, b->c)
377 void bch_btree_node_read(struct btree *);
378 void bch_btree_node_write(struct btree *, struct closure *);
380 void bch_cannibalize_unlock(struct cache_set *);
381 void bch_btree_set_root(struct btree *);
382 struct btree *bch_btree_node_alloc(struct cache_set *, int);
383 struct btree *bch_btree_node_get(struct cache_set *, struct bkey *, int, bool);
385 int bch_btree_insert_check_key(struct btree *, struct btree_op *,
387 int bch_btree_insert(struct btree_op *, struct cache_set *, struct keylist *);
389 int bch_btree_search_recurse(struct btree *, struct btree_op *);
391 int bch_gc_thread_start(struct cache_set *);
392 size_t bch_btree_gc_finish(struct cache_set *);
393 void bch_moving_gc(struct cache_set *);
394 int bch_btree_check(struct cache_set *, struct btree_op *);
395 uint8_t __bch_btree_mark_key(struct cache_set *, int, struct bkey *);
397 static inline void wake_up_gc(struct cache_set *c)
400 wake_up_process(c->gc_thread);
403 void bch_keybuf_init(struct keybuf *);
404 void bch_refill_keybuf(struct cache_set *, struct keybuf *, struct bkey *,
406 bool bch_keybuf_check_overlapping(struct keybuf *, struct bkey *,
408 void bch_keybuf_del(struct keybuf *, struct keybuf_key *);
409 struct keybuf_key *bch_keybuf_next(struct keybuf *);
410 struct keybuf_key *bch_keybuf_next_rescan(struct cache_set *, struct keybuf *,
411 struct bkey *, keybuf_pred_fn *);