blob: d1c72ef64edf5a1611fe309cd64b9524b2b0b9d9 [file] [log] [blame]
Greg Kroah-Hartmanb2441312017-11-01 15:07:57 +01001/* SPDX-License-Identifier: GPL-2.0 */
Kent Overstreetcafe5632013-03-23 16:11:31 -07002#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
105struct btree_write {
Kent Overstreetcafe5632013-03-23 16:11:31 -0700106 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
117struct 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;
Kent Overstreetd6fd3b12013-07-24 17:20:19 -0700129 struct btree *parent;
Kent Overstreetcafe5632013-03-23 16:11:31 -0700130
Kent Overstreet2a285682014-03-04 16:42:42 -0800131 struct mutex write_lock;
132
Kent Overstreetcafe5632013-03-23 16:11:31 -0700133 unsigned long flags;
134 uint16_t written; /* would be nice to kill */
135 uint8_t level;
Kent Overstreetcafe5632013-03-23 16:11:31 -0700136
Kent Overstreeta85e9682013-12-20 17:28:16 -0800137 struct btree_keys keys;
Kent Overstreetcafe5632013-03-23 16:11:31 -0700138
Kent Overstreet57943512013-04-25 13:58:35 -0700139 /* For outstanding btree writes, used as a lock - protects write_idx */
Kent Overstreetcb7a5832013-12-16 15:27:25 -0800140 struct closure io;
141 struct semaphore io_mutex;
Kent Overstreetcafe5632013-03-23 16:11:31 -0700142
Kent Overstreetcafe5632013-03-23 16:11:31 -0700143 struct list_head list;
144 struct delayed_work work;
145
Kent Overstreetcafe5632013-03-23 16:11:31 -0700146 struct btree_write writes[2];
147 struct bio *bio;
148};
149
150#define BTREE_FLAG(flag) \
151static inline bool btree_node_ ## flag(struct btree *b) \
152{ return test_bit(BTREE_NODE_ ## flag, &b->flags); } \
153 \
154static inline void set_btree_node_ ## flag(struct btree *b) \
Shenghui Wangcbb751c2018-08-09 15:48:51 +0800155{ set_bit(BTREE_NODE_ ## flag, &b->flags); }
Kent Overstreetcafe5632013-03-23 16:11:31 -0700156
157enum btree_flags {
Kent Overstreetcafe5632013-03-23 16:11:31 -0700158 BTREE_NODE_io_error,
159 BTREE_NODE_dirty,
160 BTREE_NODE_write_idx,
161};
162
Kent Overstreetcafe5632013-03-23 16:11:31 -0700163BTREE_FLAG(io_error);
164BTREE_FLAG(dirty);
165BTREE_FLAG(write_idx);
166
167static inline struct btree_write *btree_current_write(struct btree *b)
168{
169 return b->writes + btree_node_write_idx(b);
170}
171
172static inline struct btree_write *btree_prev_write(struct btree *b)
173{
174 return b->writes + (btree_node_write_idx(b) ^ 1);
175}
176
Kent Overstreet88b9f8c2013-12-17 21:46:35 -0800177static inline struct bset *btree_bset_first(struct btree *b)
178{
Kent Overstreeta85e9682013-12-20 17:28:16 -0800179 return b->keys.set->data;
Kent Overstreet88b9f8c2013-12-17 21:46:35 -0800180}
181
Kent Overstreetee811282013-12-17 23:49:49 -0800182static inline struct bset *btree_bset_last(struct btree *b)
183{
Kent Overstreeta85e9682013-12-20 17:28:16 -0800184 return bset_tree_last(&b->keys)->data;
Kent Overstreet88b9f8c2013-12-17 21:46:35 -0800185}
186
Coly Li6f10f7d2018-08-11 13:19:44 +0800187static inline unsigned int bset_block_offset(struct btree *b, struct bset *i)
Kent Overstreet88b9f8c2013-12-17 21:46:35 -0800188{
Kent Overstreeta85e9682013-12-20 17:28:16 -0800189 return bset_sector_offset(&b->keys, i) >> b->c->block_bits;
Kent Overstreetcafe5632013-03-23 16:11:31 -0700190}
191
192static inline void set_gc_sectors(struct cache_set *c)
193{
Kent Overstreeta1f03582013-09-10 19:07:00 -0700194 atomic_set(&c->sectors_to_gc, c->sb.bucket_size * c->nbuckets / 16);
Kent Overstreetcafe5632013-03-23 16:11:31 -0700195}
196
Kent Overstreet3a3b6a42013-07-24 16:46:42 -0700197void bkey_put(struct cache_set *c, struct bkey *k);
Kent Overstreete7c590e2013-09-10 18:39:16 -0700198
Kent Overstreetcafe5632013-03-23 16:11:31 -0700199/* Looping macros */
200
201#define for_each_cached_btree(b, c, iter) \
202 for (iter = 0; \
203 iter < ARRAY_SIZE((c)->bucket_hash); \
204 iter++) \
205 hlist_for_each_entry_rcu((b), (c)->bucket_hash + iter, hash)
206
Kent Overstreetcafe5632013-03-23 16:11:31 -0700207/* Recursing down the btree */
208
209struct btree_op {
Kent Overstreet78365412013-12-17 01:29:34 -0800210 /* for waiting on btree reserve in btree_split() */
Ingo Molnarac6424b2017-06-20 12:06:13 +0200211 wait_queue_entry_t wait;
Kent Overstreet78365412013-12-17 01:29:34 -0800212
Kent Overstreetcafe5632013-03-23 16:11:31 -0700213 /* Btree level at which we start taking write locks */
214 short lock;
215
Coly Li6f10f7d2018-08-11 13:19:44 +0800216 unsigned int insert_collision:1;
Kent Overstreetcafe5632013-03-23 16:11:31 -0700217};
218
Kent Overstreetb54d6932013-07-24 18:04:18 -0700219static inline void bch_btree_op_init(struct btree_op *op, int write_lock_level)
220{
221 memset(op, 0, sizeof(struct btree_op));
Kent Overstreet78365412013-12-17 01:29:34 -0800222 init_wait(&op->wait);
Kent Overstreetb54d6932013-07-24 18:04:18 -0700223 op->lock = write_lock_level;
224}
Kent Overstreetcafe5632013-03-23 16:11:31 -0700225
226static inline void rw_lock(bool w, struct btree *b, int level)
227{
228 w ? down_write_nested(&b->lock, level + 1)
229 : down_read_nested(&b->lock, level + 1);
230 if (w)
231 b->seq++;
232}
233
234static inline void rw_unlock(bool w, struct btree *b)
235{
Kent Overstreetcafe5632013-03-23 16:11:31 -0700236 if (w)
237 b->seq++;
238 (w ? up_write : up_read)(&b->lock);
239}
240
Coly Lifc2d5982018-08-11 13:19:46 +0800241void bch_btree_node_read_done(struct btree *b);
242void __bch_btree_node_write(struct btree *b, struct closure *parent);
243void bch_btree_node_write(struct btree *b, struct closure *parent);
Kent Overstreetcafe5632013-03-23 16:11:31 -0700244
Coly Lifc2d5982018-08-11 13:19:46 +0800245void bch_btree_set_root(struct btree *b);
246struct btree *__bch_btree_node_alloc(struct cache_set *c, struct btree_op *op,
247 int level, bool wait,
248 struct btree *parent);
249struct btree *bch_btree_node_get(struct cache_set *c, struct btree_op *op,
250 struct bkey *k, int level, bool write,
251 struct btree *parent);
Kent Overstreetcafe5632013-03-23 16:11:31 -0700252
Coly Lifc2d5982018-08-11 13:19:46 +0800253int bch_btree_insert_check_key(struct btree *b, struct btree_op *op,
254 struct bkey *check_key);
255int bch_btree_insert(struct cache_set *c, struct keylist *keys,
256 atomic_t *journal_ref, struct bkey *replace_key);
Kent Overstreetcafe5632013-03-23 16:11:31 -0700257
Coly Lifc2d5982018-08-11 13:19:46 +0800258int bch_gc_thread_start(struct cache_set *c);
259void bch_initial_gc_finish(struct cache_set *c);
260void bch_moving_gc(struct cache_set *c);
261int bch_btree_check(struct cache_set *c);
262void bch_initial_mark_key(struct cache_set *c, int level, struct bkey *k);
Kent Overstreetcafe5632013-03-23 16:11:31 -0700263
Kent Overstreet72a44512013-10-24 17:19:26 -0700264static inline void wake_up_gc(struct cache_set *c)
265{
Kent Overstreetbe628be2016-10-26 20:31:17 -0700266 wake_up(&c->gc_wait);
Kent Overstreet72a44512013-10-24 17:19:26 -0700267}
268
Coly Licb07ad62018-12-13 22:53:52 +0800269static inline void force_wake_up_gc(struct cache_set *c)
270{
271 /*
272 * Garbage collection thread only works when sectors_to_gc < 0,
273 * calling wake_up_gc() won't start gc thread if sectors_to_gc is
274 * not a nagetive value.
275 * Therefore sectors_to_gc is set to -1 here, before waking up
276 * gc thread by calling wake_up_gc(). Then gc_should_run() will
277 * give a chance to permit gc thread to run. "Give a chance" means
278 * before going into gc_should_run(), there is still possibility
279 * that c->sectors_to_gc being set to other positive value. So
280 * this routine won't 100% make sure gc thread will be woken up
281 * to run.
282 */
283 atomic_set(&c->sectors_to_gc, -1);
284 wake_up_gc(c);
285}
286
Kent Overstreet48dad8b2013-09-10 18:48:51 -0700287#define MAP_DONE 0
288#define MAP_CONTINUE 1
289
290#define MAP_ALL_NODES 0
291#define MAP_LEAF_NODES 1
292
293#define MAP_END_KEY 1
294
Coly Lifc2d5982018-08-11 13:19:46 +0800295typedef int (btree_map_nodes_fn)(struct btree_op *b_op, struct btree *b);
296int __bch_btree_map_nodes(struct btree_op *op, struct cache_set *c,
297 struct bkey *from, btree_map_nodes_fn *fn, int flags);
Kent Overstreet48dad8b2013-09-10 18:48:51 -0700298
299static inline int bch_btree_map_nodes(struct btree_op *op, struct cache_set *c,
300 struct bkey *from, btree_map_nodes_fn *fn)
301{
302 return __bch_btree_map_nodes(op, c, from, fn, MAP_ALL_NODES);
303}
304
305static inline int bch_btree_map_leaf_nodes(struct btree_op *op,
306 struct cache_set *c,
307 struct bkey *from,
308 btree_map_nodes_fn *fn)
309{
310 return __bch_btree_map_nodes(op, c, from, fn, MAP_LEAF_NODES);
311}
312
Coly Lifc2d5982018-08-11 13:19:46 +0800313typedef int (btree_map_keys_fn)(struct btree_op *op, struct btree *b,
314 struct bkey *k);
315int bch_btree_map_keys(struct btree_op *op, struct cache_set *c,
316 struct bkey *from, btree_map_keys_fn *fn, int flags);
Kent Overstreet48dad8b2013-09-10 18:48:51 -0700317
Coly Lifc2d5982018-08-11 13:19:46 +0800318typedef bool (keybuf_pred_fn)(struct keybuf *buf, struct bkey *k);
Kent Overstreet48dad8b2013-09-10 18:48:51 -0700319
Coly Lifc2d5982018-08-11 13:19:46 +0800320void bch_keybuf_init(struct keybuf *buf);
321void bch_refill_keybuf(struct cache_set *c, struct keybuf *buf,
322 struct bkey *end, keybuf_pred_fn *pred);
323bool bch_keybuf_check_overlapping(struct keybuf *buf, struct bkey *start,
324 struct bkey *end);
325void bch_keybuf_del(struct keybuf *buf, struct keybuf_key *w);
326struct keybuf_key *bch_keybuf_next(struct keybuf *buf);
Coly Lib0d30982018-08-11 13:19:47 +0800327struct keybuf_key *bch_keybuf_next_rescan(struct cache_set *c,
328 struct keybuf *buf,
329 struct bkey *end,
330 keybuf_pred_fn *pred);
Tang Junhuid44c2f92017-10-30 14:46:33 -0700331void bch_update_bucket_in_use(struct cache_set *c, struct gc_stat *stats);
Kent Overstreetcafe5632013-03-23 16:11:31 -0700332#endif