xref: /linux/drivers/md/bcache/btree.h (revision 17cfcb68af3bc7d5e8ae08779b1853310a2949f3)
1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _BCACHE_BTREE_H
3 #define _BCACHE_BTREE_H
4 
5 /*
6  * THE BTREE:
7  *
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
10  * nodes.
11  *
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).
17  *
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.
24  *
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.
28  *
29  * BTREE CACHE:
30  *
31  * Btree nodes are cached in memory; traversing the btree might require reading
32  * in btree nodes which is handled mostly transparently.
33  *
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.
38  *
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.
44  *
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.
51  *
52  * BTREE IO:
53  *
54  * Btree nodes never have to be explicitly read in; bch_btree_node_get() handles
55  * this.
56  *
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.
59  *
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.
64  *
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
70  * complete.
71  *
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.
74  *
75  * LOCKING:
76  *
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
79  * the leaf node.
80  *
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.
84  *
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
88  * loop.
89  *
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
95  * the backing device.
96  *
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.
100  */
101 
102 #include "bset.h"
103 #include "debug.h"
104 
105 struct btree_write {
106 	atomic_t		*journal;
107 
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.
113 	 */
114 	int			prio_blocked;
115 };
116 
117 struct btree {
118 	/* Hottest entries first */
119 	struct hlist_node	hash;
120 
121 	/* Key/pointer for this btree node */
122 	BKEY_PADDED(key);
123 
124 	/* Single bit - set when accessed, cleared by shrinker */
125 	unsigned long		accessed;
126 	unsigned long		seq;
127 	struct rw_semaphore	lock;
128 	struct cache_set	*c;
129 	struct btree		*parent;
130 
131 	struct mutex		write_lock;
132 
133 	unsigned long		flags;
134 	uint16_t		written;	/* would be nice to kill */
135 	uint8_t			level;
136 
137 	struct btree_keys	keys;
138 
139 	/* For outstanding btree writes, used as a lock - protects write_idx */
140 	struct closure		io;
141 	struct semaphore	io_mutex;
142 
143 	struct list_head	list;
144 	struct delayed_work	work;
145 
146 	struct btree_write	writes[2];
147 	struct bio		*bio;
148 };
149 
150 #define BTREE_FLAG(flag)						\
151 static inline bool btree_node_ ## flag(struct btree *b)			\
152 {	return test_bit(BTREE_NODE_ ## flag, &b->flags); }		\
153 									\
154 static inline void set_btree_node_ ## flag(struct btree *b)		\
155 {	set_bit(BTREE_NODE_ ## flag, &b->flags); }
156 
157 enum btree_flags {
158 	BTREE_NODE_io_error,
159 	BTREE_NODE_dirty,
160 	BTREE_NODE_write_idx,
161 	BTREE_NODE_journal_flush,
162 };
163 
164 BTREE_FLAG(io_error);
165 BTREE_FLAG(dirty);
166 BTREE_FLAG(write_idx);
167 BTREE_FLAG(journal_flush);
168 
169 static inline struct btree_write *btree_current_write(struct btree *b)
170 {
171 	return b->writes + btree_node_write_idx(b);
172 }
173 
174 static inline struct btree_write *btree_prev_write(struct btree *b)
175 {
176 	return b->writes + (btree_node_write_idx(b) ^ 1);
177 }
178 
179 static inline struct bset *btree_bset_first(struct btree *b)
180 {
181 	return b->keys.set->data;
182 }
183 
184 static inline struct bset *btree_bset_last(struct btree *b)
185 {
186 	return bset_tree_last(&b->keys)->data;
187 }
188 
189 static inline unsigned int bset_block_offset(struct btree *b, struct bset *i)
190 {
191 	return bset_sector_offset(&b->keys, i) >> b->c->block_bits;
192 }
193 
194 static inline void set_gc_sectors(struct cache_set *c)
195 {
196 	atomic_set(&c->sectors_to_gc, c->sb.bucket_size * c->nbuckets / 16);
197 }
198 
199 void bkey_put(struct cache_set *c, struct bkey *k);
200 
201 /* Looping macros */
202 
203 #define for_each_cached_btree(b, c, iter)				\
204 	for (iter = 0;							\
205 	     iter < ARRAY_SIZE((c)->bucket_hash);			\
206 	     iter++)							\
207 		hlist_for_each_entry_rcu((b), (c)->bucket_hash + iter, hash)
208 
209 /* Recursing down the btree */
210 
211 struct btree_op {
212 	/* for waiting on btree reserve in btree_split() */
213 	wait_queue_entry_t		wait;
214 
215 	/* Btree level at which we start taking write locks */
216 	short			lock;
217 
218 	unsigned int		insert_collision:1;
219 };
220 
221 static inline void bch_btree_op_init(struct btree_op *op, int write_lock_level)
222 {
223 	memset(op, 0, sizeof(struct btree_op));
224 	init_wait(&op->wait);
225 	op->lock = write_lock_level;
226 }
227 
228 static inline void rw_lock(bool w, struct btree *b, int level)
229 {
230 	w ? down_write_nested(&b->lock, level + 1)
231 	  : down_read_nested(&b->lock, level + 1);
232 	if (w)
233 		b->seq++;
234 }
235 
236 static inline void rw_unlock(bool w, struct btree *b)
237 {
238 	if (w)
239 		b->seq++;
240 	(w ? up_write : up_read)(&b->lock);
241 }
242 
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);
246 
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);
254 
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);
259 
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);
265 
266 static inline void wake_up_gc(struct cache_set *c)
267 {
268 	wake_up(&c->gc_wait);
269 }
270 
271 static inline void force_wake_up_gc(struct cache_set *c)
272 {
273 	/*
274 	 * Garbage collection thread only works when sectors_to_gc < 0,
275 	 * calling wake_up_gc() won't start gc thread if sectors_to_gc is
276 	 * not a nagetive value.
277 	 * Therefore sectors_to_gc is set to -1 here, before waking up
278 	 * gc thread by calling wake_up_gc(). Then gc_should_run() will
279 	 * give a chance to permit gc thread to run. "Give a chance" means
280 	 * before going into gc_should_run(), there is still possibility
281 	 * that c->sectors_to_gc being set to other positive value. So
282 	 * this routine won't 100% make sure gc thread will be woken up
283 	 * to run.
284 	 */
285 	atomic_set(&c->sectors_to_gc, -1);
286 	wake_up_gc(c);
287 }
288 
289 #define MAP_DONE	0
290 #define MAP_CONTINUE	1
291 
292 #define MAP_ALL_NODES	0
293 #define MAP_LEAF_NODES	1
294 
295 #define MAP_END_KEY	1
296 
297 typedef int (btree_map_nodes_fn)(struct btree_op *b_op, struct btree *b);
298 int __bch_btree_map_nodes(struct btree_op *op, struct cache_set *c,
299 			  struct bkey *from, btree_map_nodes_fn *fn, int flags);
300 
301 static inline int bch_btree_map_nodes(struct btree_op *op, struct cache_set *c,
302 				      struct bkey *from, btree_map_nodes_fn *fn)
303 {
304 	return __bch_btree_map_nodes(op, c, from, fn, MAP_ALL_NODES);
305 }
306 
307 static inline int bch_btree_map_leaf_nodes(struct btree_op *op,
308 					   struct cache_set *c,
309 					   struct bkey *from,
310 					   btree_map_nodes_fn *fn)
311 {
312 	return __bch_btree_map_nodes(op, c, from, fn, MAP_LEAF_NODES);
313 }
314 
315 typedef int (btree_map_keys_fn)(struct btree_op *op, struct btree *b,
316 				struct bkey *k);
317 int bch_btree_map_keys(struct btree_op *op, struct cache_set *c,
318 		       struct bkey *from, btree_map_keys_fn *fn, int flags);
319 
320 typedef bool (keybuf_pred_fn)(struct keybuf *buf, struct bkey *k);
321 
322 void bch_keybuf_init(struct keybuf *buf);
323 void bch_refill_keybuf(struct cache_set *c, struct keybuf *buf,
324 		       struct bkey *end, keybuf_pred_fn *pred);
325 bool bch_keybuf_check_overlapping(struct keybuf *buf, struct bkey *start,
326 				  struct bkey *end);
327 void bch_keybuf_del(struct keybuf *buf, struct keybuf_key *w);
328 struct keybuf_key *bch_keybuf_next(struct keybuf *buf);
329 struct keybuf_key *bch_keybuf_next_rescan(struct cache_set *c,
330 					  struct keybuf *buf,
331 					  struct bkey *end,
332 					  keybuf_pred_fn *pred);
333 void bch_update_bucket_in_use(struct cache_set *c, struct gc_stat *stats);
334 #endif
335