4 #include <linux/slab.h>
9 * A bkey contains a key, a size field, a variable number of pointers, and some
10 * ancillary flag bits.
12 * We use two different functions for validating bkeys, bch_ptr_invalid and
15 * bch_ptr_invalid() primarily filters out keys and pointers that would be
16 * invalid due to some sort of bug, whereas bch_ptr_bad() filters out keys and
17 * pointer that occur in normal practice but don't point to real data.
19 * The one exception to the rule that ptr_invalid() filters out invalid keys is
20 * that it also filters out keys of size 0 - these are keys that have been
21 * completely overwritten. It'd be safe to delete these in memory while leaving
22 * them on disk, just unnecessary work - so we filter them out when resorting
25 * We can't filter out stale keys when we're resorting, because garbage
26 * collection needs to find them to ensure bucket gens don't wrap around -
27 * unless we're rewriting the btree node those stale keys still exist on disk.
29 * We also implement functions here for removing some number of sectors from the
30 * front or the back of a bkey - this is mainly used for fixing overlapping
31 * extents, by removing the overlapping sectors from the older key.
35 * A bset is an array of bkeys laid out contiguously in memory in sorted order,
36 * along with a header. A btree node is made up of a number of these, written at
39 * There could be many of them on disk, but we never allow there to be more than
40 * 4 in memory - we lazily resort as needed.
42 * We implement code here for creating and maintaining auxiliary search trees
43 * (described below) for searching an individial bset, and on top of that we
44 * implement a btree iterator.
48 * Most of the code in bcache doesn't care about an individual bset - it needs
49 * to search entire btree nodes and iterate over them in sorted order.
51 * The btree iterator code serves both functions; it iterates through the keys
52 * in a btree node in sorted order, starting from either keys after a specific
53 * point (if you pass it a search key) or the start of the btree node.
55 * AUXILIARY SEARCH TREES:
57 * Since keys are variable length, we can't use a binary search on a bset - we
58 * wouldn't be able to find the start of the next key. But binary searches are
59 * slow anyways, due to terrible cache behaviour; bcache originally used binary
60 * searches and that code topped out at under 50k lookups/second.
62 * So we need to construct some sort of lookup table. Since we only insert keys
63 * into the last (unwritten) set, most of the keys within a given btree node are
64 * usually in sets that are mostly constant. We use two different types of
65 * lookup tables to take advantage of this.
67 * Both lookup tables share in common that they don't index every key in the
68 * set; they index one key every BSET_CACHELINE bytes, and then a linear search
69 * is used for the rest.
71 * For sets that have been written to disk and are no longer being inserted
72 * into, we construct a binary search tree in an array - traversing a binary
73 * search tree in an array gives excellent locality of reference and is very
74 * fast, since both children of any node are adjacent to each other in memory
75 * (and their grandchildren, and great grandchildren...) - this means
76 * prefetching can be used to great effect.
78 * It's quite useful performance wise to keep these nodes small - not just
79 * because they're more likely to be in L2, but also because we can prefetch
80 * more nodes on a single cacheline and thus prefetch more iterations in advance
81 * when traversing this tree.
83 * Nodes in the auxiliary search tree must contain both a key to compare against
84 * (we don't want to fetch the key from the set, that would defeat the purpose),
85 * and a pointer to the key. We use a few tricks to compress both of these.
87 * To compress the pointer, we take advantage of the fact that one node in the
88 * search tree corresponds to precisely BSET_CACHELINE bytes in the set. We have
89 * a function (to_inorder()) that takes the index of a node in a binary tree and
90 * returns what its index would be in an inorder traversal, so we only have to
91 * store the low bits of the offset.
93 * The key is 84 bits (KEY_DEV + key->key, the offset on the device). To
94 * compress that, we take advantage of the fact that when we're traversing the
95 * search tree at every iteration we know that both our search key and the key
96 * we're looking for lie within some range - bounded by our previous
97 * comparisons. (We special case the start of a search so that this is true even
98 * at the root of the tree).
100 * So we know the key we're looking for is between a and b, and a and b don't
101 * differ higher than bit 50, we don't need to check anything higher than bit
104 * We don't usually need the rest of the bits, either; we only need enough bits
105 * to partition the key range we're currently checking. Consider key n - the
106 * key our auxiliary search tree node corresponds to, and key p, the key
107 * immediately preceding n. The lowest bit we need to store in the auxiliary
108 * search tree is the highest bit that differs between n and p.
110 * Note that this could be bit 0 - we might sometimes need all 80 bits to do the
111 * comparison. But we'd really like our nodes in the auxiliary search tree to be
114 * The solution is to make them fixed size, and when we're constructing a node
115 * check if p and n differed in the bits we needed them to. If they don't we
116 * flag that node, and when doing lookups we fallback to comparing against the
117 * real key. As long as this doesn't happen to often (and it seems to reliably
118 * happen a bit less than 1% of the time), we win - even on failures, that key
119 * is then more likely to be in cache than if we were doing binary searches all
120 * the way, since we're touching so much less memory.
122 * The keys in the auxiliary search tree are stored in (software) floating
123 * point, with an exponent and a mantissa. The exponent needs to be big enough
124 * to address all the bits in the original key, but the number of bits in the
125 * mantissa is somewhat arbitrary; more bits just gets us fewer failures.
127 * We need 7 bits for the exponent and 3 bits for the key's offset (since keys
128 * are 8 byte aligned); using 22 bits for the mantissa means a node is 4 bytes.
129 * We need one node per 128 bytes in the btree node, which means the auxiliary
130 * search trees take up 3% as much memory as the btree itself.
132 * Constructing these auxiliary search trees is moderately expensive, and we
133 * don't want to be constantly rebuilding the search tree for the last set
134 * whenever we insert another key into it. For the unwritten set, we use a much
135 * simpler lookup table - it's just a flat array, so index i in the lookup table
136 * corresponds to the i range of BSET_CACHELINE bytes in the set. Indexing
137 * within each byte range works the same as with the auxiliary search trees.
139 * These are much easier to keep up to date when we insert a key - we do it
140 * somewhat lazily; when we shift a key up we usually just increment the pointer
141 * to it, only when it would overflow do we go to the trouble of finding the
142 * first key in that range of bytes again.
147 /* Btree key comparison/iteration */
153 #ifdef CONFIG_BCACHE_DEBUG
156 struct btree_iter_set {
157 struct bkey *k, *end;
163 * We construct a binary tree in an array as if the array
164 * started at 1, so that things line up on the same cachelines
165 * better: see comments in bset.c at cacheline_to_bkey() for
169 /* size of the binary tree and prev array */
172 /* function of size - precalculated for to_inorder() */
175 /* copy of the last key in the set */
177 struct bkey_float *tree;
180 * The nodes in the bset tree point to specific keys - this
181 * array holds the sizes of the previous key.
183 * Conceptually it's a member of struct bkey_float, but we want
184 * to keep bkey_float to 4 bytes and prev isn't used in the fast
189 /* The actual btree node, with pointers to each sorted set */
193 static __always_inline int64_t bkey_cmp(const struct bkey *l,
194 const struct bkey *r)
196 return unlikely(KEY_INODE(l) != KEY_INODE(r))
197 ? (int64_t) KEY_INODE(l) - (int64_t) KEY_INODE(r)
198 : (int64_t) KEY_OFFSET(l) - (int64_t) KEY_OFFSET(r);
213 /* Enough room for btree_split's keys without realloc */
214 #define KEYLIST_INLINE 16
215 uint64_t inline_keys[KEYLIST_INLINE];
218 static inline void bch_keylist_init(struct keylist *l)
220 l->top_p = l->keys_p = l->inline_keys;
223 static inline void bch_keylist_push(struct keylist *l)
225 l->top = bkey_next(l->top);
228 static inline void bch_keylist_add(struct keylist *l, struct bkey *k)
230 bkey_copy(l->top, k);
234 static inline bool bch_keylist_empty(struct keylist *l)
236 return l->top == l->keys;
239 static inline void bch_keylist_reset(struct keylist *l)
244 static inline void bch_keylist_free(struct keylist *l)
246 if (l->keys_p != l->inline_keys)
250 static inline size_t bch_keylist_nkeys(struct keylist *l)
252 return l->top_p - l->keys_p;
255 static inline size_t bch_keylist_bytes(struct keylist *l)
257 return bch_keylist_nkeys(l) * sizeof(uint64_t);
260 struct bkey *bch_keylist_pop(struct keylist *);
261 void bch_keylist_pop_front(struct keylist *);
262 int __bch_keylist_realloc(struct keylist *, unsigned);
264 void bch_bkey_copy_single_ptr(struct bkey *, const struct bkey *,
266 bool __bch_cut_front(const struct bkey *, struct bkey *);
267 bool __bch_cut_back(const struct bkey *, struct bkey *);
269 static inline bool bch_cut_front(const struct bkey *where, struct bkey *k)
271 BUG_ON(bkey_cmp(where, k) > 0);
272 return __bch_cut_front(where, k);
275 static inline bool bch_cut_back(const struct bkey *where, struct bkey *k)
277 BUG_ON(bkey_cmp(where, &START_KEY(k)) < 0);
278 return __bch_cut_back(where, k);
281 const char *bch_ptr_status(struct cache_set *, const struct bkey *);
282 bool bch_btree_ptr_invalid(struct cache_set *, const struct bkey *);
283 bool bch_extent_ptr_invalid(struct cache_set *, const struct bkey *);
285 bool bch_ptr_bad(struct btree *, const struct bkey *);
287 typedef bool (*ptr_filter_fn)(struct btree *, const struct bkey *);
289 struct bkey *bch_btree_iter_next(struct btree_iter *);
290 struct bkey *bch_btree_iter_next_filter(struct btree_iter *,
291 struct btree *, ptr_filter_fn);
293 void bch_btree_iter_push(struct btree_iter *, struct bkey *, struct bkey *);
294 struct bkey *__bch_btree_iter_init(struct btree *, struct btree_iter *,
295 struct bkey *, struct bset_tree *);
298 #define BKEY_MID_BITS 3
299 #define BKEY_EXPONENT_BITS 7
300 #define BKEY_MANTISSA_BITS 22
301 #define BKEY_MANTISSA_MASK ((1 << BKEY_MANTISSA_BITS) - 1)
304 unsigned exponent:BKEY_EXPONENT_BITS;
305 unsigned m:BKEY_MID_BITS;
306 unsigned mantissa:BKEY_MANTISSA_BITS;
310 * BSET_CACHELINE was originally intended to match the hardware cacheline size -
311 * it used to be 64, but I realized the lookup code would touch slightly less
312 * memory if it was 128.
314 * It definites the number of bytes (in struct bset) per struct bkey_float in
315 * the auxiliar search tree - when we're done searching the bset_float tree we
316 * have this many bytes left that we do a linear search over.
318 * Since (after level 5) every level of the bset_tree is on a new cacheline,
319 * we're touching one fewer cacheline in the bset tree in exchange for one more
320 * cacheline in the linear search - but the linear search might stop before it
321 * gets to the second cacheline.
324 #define BSET_CACHELINE 128
325 #define bset_tree_space(b) (btree_data_space(b) / BSET_CACHELINE)
327 #define bset_tree_bytes(b) (bset_tree_space(b) * sizeof(struct bkey_float))
328 #define bset_prev_bytes(b) (bset_tree_space(b) * sizeof(uint8_t))
330 void bch_bset_init_next(struct btree *);
332 void bch_bset_fix_invalidated_key(struct btree *, struct bkey *);
333 void bch_bset_fix_lookup_table(struct btree *, struct bkey *);
335 struct bkey *__bch_bset_search(struct btree *, struct bset_tree *,
336 const struct bkey *);
339 * Returns the first key that is strictly greater than search
341 static inline struct bkey *bch_bset_search(struct btree *b, struct bset_tree *t,
342 const struct bkey *search)
344 return search ? __bch_bset_search(b, t, search) : t->data->start;
347 #define PRECEDING_KEY(_k) \
349 struct bkey *_ret = NULL; \
351 if (KEY_INODE(_k) || KEY_OFFSET(_k)) { \
352 _ret = &KEY(KEY_INODE(_k), KEY_OFFSET(_k), 0); \
362 bool bch_bkey_try_merge(struct btree *, struct bkey *, struct bkey *);
363 void bch_btree_sort_lazy(struct btree *);
364 void bch_btree_sort_into(struct btree *, struct btree *);
365 void bch_btree_sort_and_fix_extents(struct btree *, struct btree_iter *);
366 void bch_btree_sort_partial(struct btree *, unsigned);
368 static inline void bch_btree_sort(struct btree *b)
370 bch_btree_sort_partial(b, 0);
373 int bch_bset_print_stats(struct cache_set *, char *);