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 }; 162 163 BTREE_FLAG(io_error); 164 BTREE_FLAG(dirty); 165 BTREE_FLAG(write_idx); 166 167 static inline struct btree_write *btree_current_write(struct btree *b) 168 { 169 return b->writes + btree_node_write_idx(b); 170 } 171 172 static inline struct btree_write *btree_prev_write(struct btree *b) 173 { 174 return b->writes + (btree_node_write_idx(b) ^ 1); 175 } 176 177 static inline struct bset *btree_bset_first(struct btree *b) 178 { 179 return b->keys.set->data; 180 } 181 182 static inline struct bset *btree_bset_last(struct btree *b) 183 { 184 return bset_tree_last(&b->keys)->data; 185 } 186 187 static inline unsigned int bset_block_offset(struct btree *b, struct bset *i) 188 { 189 return bset_sector_offset(&b->keys, i) >> b->c->block_bits; 190 } 191 192 static inline void set_gc_sectors(struct cache_set *c) 193 { 194 atomic_set(&c->sectors_to_gc, c->sb.bucket_size * c->nbuckets / 16); 195 } 196 197 void bkey_put(struct cache_set *c, struct bkey *k); 198 199 /* 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) 206 207 /* Recursing down the btree */ 208 209 struct btree_op { 210 /* for waiting on btree reserve in btree_split() */ 211 wait_queue_entry_t wait; 212 213 /* Btree level at which we start taking write locks */ 214 short lock; 215 216 unsigned int insert_collision:1; 217 }; 218 219 static inline void bch_btree_op_init(struct btree_op *op, int write_lock_level) 220 { 221 memset(op, 0, sizeof(struct btree_op)); 222 init_wait(&op->wait); 223 op->lock = write_lock_level; 224 } 225 226 static 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 } 233 234 static inline void rw_unlock(bool w, struct btree *b) 235 { 236 if (w) 237 b->seq++; 238 (w ? up_write : up_read)(&b->lock); 239 } 240 241 void bch_btree_node_read_done(struct btree *b); 242 void __bch_btree_node_write(struct btree *b, struct closure *parent); 243 void bch_btree_node_write(struct btree *b, struct closure *parent); 244 245 void bch_btree_set_root(struct btree *b); 246 struct btree *__bch_btree_node_alloc(struct cache_set *c, struct btree_op *op, 247 int level, bool wait, 248 struct btree *parent); 249 struct 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); 252 253 int bch_btree_insert_check_key(struct btree *b, struct btree_op *op, 254 struct bkey *check_key); 255 int bch_btree_insert(struct cache_set *c, struct keylist *keys, 256 atomic_t *journal_ref, struct bkey *replace_key); 257 258 int bch_gc_thread_start(struct cache_set *c); 259 void bch_initial_gc_finish(struct cache_set *c); 260 void bch_moving_gc(struct cache_set *c); 261 int bch_btree_check(struct cache_set *c); 262 void bch_initial_mark_key(struct cache_set *c, int level, struct bkey *k); 263 264 static inline void wake_up_gc(struct cache_set *c) 265 { 266 wake_up(&c->gc_wait); 267 } 268 269 static 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 } 286 287 #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 1 294 295 typedef int (btree_map_nodes_fn)(struct btree_op *b_op, struct btree *b); 296 int __bch_btree_map_nodes(struct btree_op *op, struct cache_set *c, 297 struct bkey *from, btree_map_nodes_fn *fn, int flags); 298 299 static 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 } 304 305 static 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 } 312 313 typedef int (btree_map_keys_fn)(struct btree_op *op, struct btree *b, 314 struct bkey *k); 315 int bch_btree_map_keys(struct btree_op *op, struct cache_set *c, 316 struct bkey *from, btree_map_keys_fn *fn, int flags); 317 318 typedef bool (keybuf_pred_fn)(struct keybuf *buf, struct bkey *k); 319 320 void bch_keybuf_init(struct keybuf *buf); 321 void bch_refill_keybuf(struct cache_set *c, struct keybuf *buf, 322 struct bkey *end, keybuf_pred_fn *pred); 323 bool bch_keybuf_check_overlapping(struct keybuf *buf, struct bkey *start, 324 struct bkey *end); 325 void bch_keybuf_del(struct keybuf *buf, struct keybuf_key *w); 326 struct keybuf_key *bch_keybuf_next(struct keybuf *buf); 327 struct keybuf_key *bch_keybuf_next_rescan(struct cache_set *c, 328 struct keybuf *buf, 329 struct bkey *end, 330 keybuf_pred_fn *pred); 331 void bch_update_bucket_in_use(struct cache_set *c, struct gc_stat *stats); 332 #endif 333