xref: /linux/fs/bcachefs/alloc_background.h (revision c532de5a67a70f8533d495f8f2aaa9a0491c3ad0)
1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _BCACHEFS_ALLOC_BACKGROUND_H
3 #define _BCACHEFS_ALLOC_BACKGROUND_H
4 
5 #include "bcachefs.h"
6 #include "alloc_types.h"
7 #include "buckets.h"
8 #include "debug.h"
9 #include "super.h"
10 
11 enum bch_validate_flags;
12 
13 /* How out of date a pointer gen is allowed to be: */
14 #define BUCKET_GC_GEN_MAX	96U
15 
16 static inline bool bch2_dev_bucket_exists(struct bch_fs *c, struct bpos pos)
17 {
18 	rcu_read_lock();
19 	struct bch_dev *ca = bch2_dev_rcu_noerror(c, pos.inode);
20 	bool ret = ca && bucket_valid(ca, pos.offset);
21 	rcu_read_unlock();
22 	return ret;
23 }
24 
25 static inline u64 bucket_to_u64(struct bpos bucket)
26 {
27 	return (bucket.inode << 48) | bucket.offset;
28 }
29 
30 static inline struct bpos u64_to_bucket(u64 bucket)
31 {
32 	return POS(bucket >> 48, bucket & ~(~0ULL << 48));
33 }
34 
35 static inline u8 alloc_gc_gen(struct bch_alloc_v4 a)
36 {
37 	return a.gen - a.oldest_gen;
38 }
39 
40 static inline void alloc_to_bucket(struct bucket *dst, struct bch_alloc_v4 src)
41 {
42 	dst->gen		= src.gen;
43 	dst->data_type		= src.data_type;
44 	dst->stripe_sectors	= src.stripe_sectors;
45 	dst->dirty_sectors	= src.dirty_sectors;
46 	dst->cached_sectors	= src.cached_sectors;
47 	dst->stripe		= src.stripe;
48 }
49 
50 static inline void __bucket_m_to_alloc(struct bch_alloc_v4 *dst, struct bucket src)
51 {
52 	dst->gen		= src.gen;
53 	dst->data_type		= src.data_type;
54 	dst->stripe_sectors	= src.stripe_sectors;
55 	dst->dirty_sectors	= src.dirty_sectors;
56 	dst->cached_sectors	= src.cached_sectors;
57 	dst->stripe		= src.stripe;
58 }
59 
60 static inline struct bch_alloc_v4 bucket_m_to_alloc(struct bucket b)
61 {
62 	struct bch_alloc_v4 ret = {};
63 	__bucket_m_to_alloc(&ret, b);
64 	return ret;
65 }
66 
67 static inline enum bch_data_type bucket_data_type(enum bch_data_type data_type)
68 {
69 	switch (data_type) {
70 	case BCH_DATA_cached:
71 	case BCH_DATA_stripe:
72 		return BCH_DATA_user;
73 	default:
74 		return data_type;
75 	}
76 }
77 
78 static inline bool bucket_data_type_mismatch(enum bch_data_type bucket,
79 					     enum bch_data_type ptr)
80 {
81 	return !data_type_is_empty(bucket) &&
82 		bucket_data_type(bucket) != bucket_data_type(ptr);
83 }
84 
85 /*
86  * It is my general preference to use unsigned types for unsigned quantities -
87  * however, these helpers are used in disk accounting calculations run by
88  * triggers where the output will be negated and added to an s64. unsigned is
89  * right out even though all these quantities will fit in 32 bits, since it
90  * won't be sign extended correctly; u64 will negate "correctly", but s64 is the
91  * simpler option here.
92  */
93 static inline s64 bch2_bucket_sectors_total(struct bch_alloc_v4 a)
94 {
95 	return a.stripe_sectors + a.dirty_sectors + a.cached_sectors;
96 }
97 
98 static inline s64 bch2_bucket_sectors_dirty(struct bch_alloc_v4 a)
99 {
100 	return a.stripe_sectors + a.dirty_sectors;
101 }
102 
103 static inline s64 bch2_bucket_sectors(struct bch_alloc_v4 a)
104 {
105 	return a.data_type == BCH_DATA_cached
106 		? a.cached_sectors
107 		: bch2_bucket_sectors_dirty(a);
108 }
109 
110 static inline s64 bch2_bucket_sectors_fragmented(struct bch_dev *ca,
111 						 struct bch_alloc_v4 a)
112 {
113 	int d = bch2_bucket_sectors(a);
114 
115 	return d ? max(0, ca->mi.bucket_size - d) : 0;
116 }
117 
118 static inline s64 bch2_gc_bucket_sectors_fragmented(struct bch_dev *ca, struct bucket a)
119 {
120 	int d = a.stripe_sectors + a.dirty_sectors;
121 
122 	return d ? max(0, ca->mi.bucket_size - d) : 0;
123 }
124 
125 static inline s64 bch2_bucket_sectors_unstriped(struct bch_alloc_v4 a)
126 {
127 	return a.data_type == BCH_DATA_stripe ? a.dirty_sectors : 0;
128 }
129 
130 static inline enum bch_data_type alloc_data_type(struct bch_alloc_v4 a,
131 						 enum bch_data_type data_type)
132 {
133 	if (a.stripe)
134 		return data_type == BCH_DATA_parity ? data_type : BCH_DATA_stripe;
135 	if (bch2_bucket_sectors_dirty(a))
136 		return data_type;
137 	if (a.cached_sectors)
138 		return BCH_DATA_cached;
139 	if (BCH_ALLOC_V4_NEED_DISCARD(&a))
140 		return BCH_DATA_need_discard;
141 	if (alloc_gc_gen(a) >= BUCKET_GC_GEN_MAX)
142 		return BCH_DATA_need_gc_gens;
143 	return BCH_DATA_free;
144 }
145 
146 static inline void alloc_data_type_set(struct bch_alloc_v4 *a, enum bch_data_type data_type)
147 {
148 	a->data_type = alloc_data_type(*a, data_type);
149 }
150 
151 static inline u64 alloc_lru_idx_read(struct bch_alloc_v4 a)
152 {
153 	return a.data_type == BCH_DATA_cached
154 		? a.io_time[READ] & LRU_TIME_MAX
155 		: 0;
156 }
157 
158 #define DATA_TYPES_MOVABLE		\
159 	((1U << BCH_DATA_btree)|	\
160 	 (1U << BCH_DATA_user)|		\
161 	 (1U << BCH_DATA_stripe))
162 
163 static inline bool data_type_movable(enum bch_data_type type)
164 {
165 	return (1U << type) & DATA_TYPES_MOVABLE;
166 }
167 
168 static inline u64 alloc_lru_idx_fragmentation(struct bch_alloc_v4 a,
169 					      struct bch_dev *ca)
170 {
171 	if (a.data_type >= BCH_DATA_NR)
172 		return 0;
173 
174 	if (!data_type_movable(a.data_type) ||
175 	    !bch2_bucket_sectors_fragmented(ca, a))
176 		return 0;
177 
178 	/*
179 	 * avoid overflowing LRU_TIME_BITS on a corrupted fs, when
180 	 * bucket_sectors_dirty is (much) bigger than bucket_size
181 	 */
182 	u64 d = min_t(s64, bch2_bucket_sectors_dirty(a),
183 		      ca->mi.bucket_size);
184 
185 	return div_u64(d * (1ULL << 31), ca->mi.bucket_size);
186 }
187 
188 static inline u64 alloc_freespace_genbits(struct bch_alloc_v4 a)
189 {
190 	return ((u64) alloc_gc_gen(a) >> 4) << 56;
191 }
192 
193 static inline struct bpos alloc_freespace_pos(struct bpos pos, struct bch_alloc_v4 a)
194 {
195 	pos.offset |= alloc_freespace_genbits(a);
196 	return pos;
197 }
198 
199 static inline unsigned alloc_v4_u64s_noerror(const struct bch_alloc_v4 *a)
200 {
201 	return (BCH_ALLOC_V4_BACKPOINTERS_START(a) ?:
202 			BCH_ALLOC_V4_U64s_V0) +
203 		BCH_ALLOC_V4_NR_BACKPOINTERS(a) *
204 		(sizeof(struct bch_backpointer) / sizeof(u64));
205 }
206 
207 static inline unsigned alloc_v4_u64s(const struct bch_alloc_v4 *a)
208 {
209 	unsigned ret = alloc_v4_u64s_noerror(a);
210 	BUG_ON(ret > U8_MAX - BKEY_U64s);
211 	return ret;
212 }
213 
214 static inline void set_alloc_v4_u64s(struct bkey_i_alloc_v4 *a)
215 {
216 	set_bkey_val_u64s(&a->k, alloc_v4_u64s(&a->v));
217 }
218 
219 struct bkey_i_alloc_v4 *
220 bch2_trans_start_alloc_update_noupdate(struct btree_trans *, struct btree_iter *, struct bpos);
221 struct bkey_i_alloc_v4 *
222 bch2_trans_start_alloc_update(struct btree_trans *, struct bpos,
223 			      enum btree_iter_update_trigger_flags);
224 
225 void __bch2_alloc_to_v4(struct bkey_s_c, struct bch_alloc_v4 *);
226 
227 static inline const struct bch_alloc_v4 *bch2_alloc_to_v4(struct bkey_s_c k, struct bch_alloc_v4 *convert)
228 {
229 	const struct bch_alloc_v4 *ret;
230 
231 	if (unlikely(k.k->type != KEY_TYPE_alloc_v4))
232 		goto slowpath;
233 
234 	ret = bkey_s_c_to_alloc_v4(k).v;
235 	if (BCH_ALLOC_V4_BACKPOINTERS_START(ret) != BCH_ALLOC_V4_U64s)
236 		goto slowpath;
237 
238 	return ret;
239 slowpath:
240 	__bch2_alloc_to_v4(k, convert);
241 	return convert;
242 }
243 
244 struct bkey_i_alloc_v4 *bch2_alloc_to_v4_mut(struct btree_trans *, struct bkey_s_c);
245 
246 int bch2_bucket_io_time_reset(struct btree_trans *, unsigned, size_t, int);
247 
248 int bch2_alloc_v1_validate(struct bch_fs *, struct bkey_s_c, enum bch_validate_flags);
249 int bch2_alloc_v2_validate(struct bch_fs *, struct bkey_s_c, enum bch_validate_flags);
250 int bch2_alloc_v3_validate(struct bch_fs *, struct bkey_s_c, enum bch_validate_flags);
251 int bch2_alloc_v4_validate(struct bch_fs *, struct bkey_s_c, enum bch_validate_flags);
252 void bch2_alloc_v4_swab(struct bkey_s);
253 void bch2_alloc_to_text(struct printbuf *, struct bch_fs *, struct bkey_s_c);
254 
255 #define bch2_bkey_ops_alloc ((struct bkey_ops) {	\
256 	.key_validate	= bch2_alloc_v1_validate,	\
257 	.val_to_text	= bch2_alloc_to_text,		\
258 	.trigger	= bch2_trigger_alloc,		\
259 	.min_val_size	= 8,				\
260 })
261 
262 #define bch2_bkey_ops_alloc_v2 ((struct bkey_ops) {	\
263 	.key_validate	= bch2_alloc_v2_validate,	\
264 	.val_to_text	= bch2_alloc_to_text,		\
265 	.trigger	= bch2_trigger_alloc,		\
266 	.min_val_size	= 8,				\
267 })
268 
269 #define bch2_bkey_ops_alloc_v3 ((struct bkey_ops) {	\
270 	.key_validate	= bch2_alloc_v3_validate,	\
271 	.val_to_text	= bch2_alloc_to_text,		\
272 	.trigger	= bch2_trigger_alloc,		\
273 	.min_val_size	= 16,				\
274 })
275 
276 #define bch2_bkey_ops_alloc_v4 ((struct bkey_ops) {	\
277 	.key_validate	= bch2_alloc_v4_validate,	\
278 	.val_to_text	= bch2_alloc_to_text,		\
279 	.swab		= bch2_alloc_v4_swab,		\
280 	.trigger	= bch2_trigger_alloc,		\
281 	.min_val_size	= 48,				\
282 })
283 
284 int bch2_bucket_gens_validate(struct bch_fs *, struct bkey_s_c,
285 			     enum bch_validate_flags);
286 void bch2_bucket_gens_to_text(struct printbuf *, struct bch_fs *, struct bkey_s_c);
287 
288 #define bch2_bkey_ops_bucket_gens ((struct bkey_ops) {	\
289 	.key_validate	= bch2_bucket_gens_validate,	\
290 	.val_to_text	= bch2_bucket_gens_to_text,	\
291 })
292 
293 int bch2_bucket_gens_init(struct bch_fs *);
294 
295 static inline bool bkey_is_alloc(const struct bkey *k)
296 {
297 	return  k->type == KEY_TYPE_alloc ||
298 		k->type == KEY_TYPE_alloc_v2 ||
299 		k->type == KEY_TYPE_alloc_v3;
300 }
301 
302 int bch2_alloc_read(struct bch_fs *);
303 
304 int bch2_alloc_key_to_dev_counters(struct btree_trans *, struct bch_dev *,
305 				   const struct bch_alloc_v4 *,
306 				   const struct bch_alloc_v4 *, unsigned);
307 int bch2_trigger_alloc(struct btree_trans *, enum btree_id, unsigned,
308 		       struct bkey_s_c, struct bkey_s,
309 		       enum btree_iter_update_trigger_flags);
310 int bch2_check_alloc_info(struct bch_fs *);
311 int bch2_check_alloc_to_lru_refs(struct bch_fs *);
312 void bch2_dev_do_discards(struct bch_dev *);
313 void bch2_do_discards(struct bch_fs *);
314 
315 static inline u64 should_invalidate_buckets(struct bch_dev *ca,
316 					    struct bch_dev_usage u)
317 {
318 	u64 want_free = ca->mi.nbuckets >> 7;
319 	u64 free = max_t(s64, 0,
320 			   u.d[BCH_DATA_free].buckets
321 			 + u.d[BCH_DATA_need_discard].buckets
322 			 - bch2_dev_buckets_reserved(ca, BCH_WATERMARK_stripe));
323 
324 	return clamp_t(s64, want_free - free, 0, u.d[BCH_DATA_cached].buckets);
325 }
326 
327 void bch2_dev_do_invalidates(struct bch_dev *);
328 void bch2_do_invalidates(struct bch_fs *);
329 
330 static inline struct bch_backpointer *alloc_v4_backpointers(struct bch_alloc_v4 *a)
331 {
332 	return (void *) ((u64 *) &a->v +
333 			 (BCH_ALLOC_V4_BACKPOINTERS_START(a) ?:
334 			  BCH_ALLOC_V4_U64s_V0));
335 }
336 
337 static inline const struct bch_backpointer *alloc_v4_backpointers_c(const struct bch_alloc_v4 *a)
338 {
339 	return (void *) ((u64 *) &a->v + BCH_ALLOC_V4_BACKPOINTERS_START(a));
340 }
341 
342 int bch2_dev_freespace_init(struct bch_fs *, struct bch_dev *, u64, u64);
343 int bch2_fs_freespace_init(struct bch_fs *);
344 int bch2_dev_remove_alloc(struct bch_fs *, struct bch_dev *);
345 
346 void bch2_recalc_capacity(struct bch_fs *);
347 u64 bch2_min_rw_member_capacity(struct bch_fs *);
348 
349 void bch2_dev_allocator_remove(struct bch_fs *, struct bch_dev *);
350 void bch2_dev_allocator_add(struct bch_fs *, struct bch_dev *);
351 
352 void bch2_dev_allocator_background_exit(struct bch_dev *);
353 void bch2_dev_allocator_background_init(struct bch_dev *);
354 
355 void bch2_fs_allocator_background_init(struct bch_fs *);
356 
357 #endif /* _BCACHEFS_ALLOC_BACKGROUND_H */
358