1 /* SPDX-License-Identifier: GPL-2.0 */
2 /*
3 * Code for manipulating bucket marks for garbage collection.
4 *
5 * Copyright 2014 Datera, Inc.
6 */
7
8 #ifndef _BUCKETS_H
9 #define _BUCKETS_H
10
11 #include "buckets_types.h"
12 #include "extents.h"
13 #include "sb-members.h"
14
15 static inline u64 sector_to_bucket(const struct bch_dev *ca, sector_t s)
16 {
17 return div_u64(s, ca->mi.bucket_size);
18 }
19
20 static inline sector_t bucket_to_sector(const struct bch_dev *ca, size_t b)
21 {
22 return ((sector_t) b) * ca->mi.bucket_size;
23 }
24
25 static inline sector_t bucket_remainder(const struct bch_dev *ca, sector_t s)
26 {
27 u32 remainder;
28
29 div_u64_rem(s, ca->mi.bucket_size, &remainder);
30 return remainder;
31 }
32
33 static inline u64 sector_to_bucket_and_offset(const struct bch_dev *ca, sector_t s, u32 *offset)
34 {
35 return div_u64_rem(s, ca->mi.bucket_size, offset);
36 }
37
38 #define for_each_bucket(_b, _buckets) \
39 for (_b = (_buckets)->b + (_buckets)->first_bucket; \
40 _b < (_buckets)->b + (_buckets)->nbuckets; _b++)
41
42 /*
43 * Ugly hack alert:
44 *
45 * We need to cram a spinlock in a single byte, because that's what we have left
46 * in struct bucket, and we care about the size of these - during fsck, we need
47 * in memory state for every single bucket on every device.
48 *
49 * We used to do
50 * while (xchg(&b->lock, 1) cpu_relax();
51 * but, it turns out not all architectures support xchg on a single byte.
52 *
53 * So now we use bit_spin_lock(), with fun games since we can't burn a whole
54 * ulong for this - we just need to make sure the lock bit always ends up in the
55 * first byte.
56 */
57
58 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
59 #define BUCKET_LOCK_BITNR 0
60 #else
61 #define BUCKET_LOCK_BITNR (BITS_PER_LONG - 1)
62 #endif
63
64 union ulong_byte_assert {
65 ulong ulong;
66 u8 byte;
67 };
68
69 static inline void bucket_unlock(struct bucket *b)
70 {
71 BUILD_BUG_ON(!((union ulong_byte_assert) { .ulong = 1UL << BUCKET_LOCK_BITNR }).byte);
72
73 clear_bit_unlock(BUCKET_LOCK_BITNR, (void *) &b->lock);
74 wake_up_bit((void *) &b->lock, BUCKET_LOCK_BITNR);
75 }
76
77 static inline void bucket_lock(struct bucket *b)
78 {
79 wait_on_bit_lock((void *) &b->lock, BUCKET_LOCK_BITNR,
80 TASK_UNINTERRUPTIBLE);
81 }
82
83 static inline struct bucket_array *gc_bucket_array(struct bch_dev *ca)
84 {
85 return rcu_dereference_check(ca->buckets_gc,
86 !ca->fs ||
87 percpu_rwsem_is_held(&ca->fs->mark_lock) ||
88 lockdep_is_held(&ca->fs->state_lock) ||
89 lockdep_is_held(&ca->bucket_lock));
90 }
91
92 static inline struct bucket *gc_bucket(struct bch_dev *ca, size_t b)
93 {
94 struct bucket_array *buckets = gc_bucket_array(ca);
95
96 if (b - buckets->first_bucket >= buckets->nbuckets_minus_first)
97 return NULL;
98 return buckets->b + b;
99 }
100
101 static inline struct bucket_gens *bucket_gens(struct bch_dev *ca)
102 {
103 return rcu_dereference_check(ca->bucket_gens,
104 !ca->fs ||
105 percpu_rwsem_is_held(&ca->fs->mark_lock) ||
106 lockdep_is_held(&ca->fs->state_lock) ||
107 lockdep_is_held(&ca->bucket_lock));
108 }
109
110 static inline u8 *bucket_gen(struct bch_dev *ca, size_t b)
111 {
112 struct bucket_gens *gens = bucket_gens(ca);
113
114 if (b - gens->first_bucket >= gens->nbuckets_minus_first)
115 return NULL;
116 return gens->b + b;
117 }
118
119 static inline u8 bucket_gen_get(struct bch_dev *ca, size_t b)
120 {
121 rcu_read_lock();
122 u8 gen = *bucket_gen(ca, b);
123 rcu_read_unlock();
124 return gen;
125 }
126
127 static inline size_t PTR_BUCKET_NR(const struct bch_dev *ca,
128 const struct bch_extent_ptr *ptr)
129 {
130 return sector_to_bucket(ca, ptr->offset);
131 }
132
133 static inline struct bpos PTR_BUCKET_POS(const struct bch_dev *ca,
134 const struct bch_extent_ptr *ptr)
135 {
136 return POS(ptr->dev, PTR_BUCKET_NR(ca, ptr));
137 }
138
139 static inline struct bpos PTR_BUCKET_POS_OFFSET(const struct bch_dev *ca,
140 const struct bch_extent_ptr *ptr,
141 u32 *bucket_offset)
142 {
143 return POS(ptr->dev, sector_to_bucket_and_offset(ca, ptr->offset, bucket_offset));
144 }
145
146 static inline struct bucket *PTR_GC_BUCKET(struct bch_dev *ca,
147 const struct bch_extent_ptr *ptr)
148 {
149 return gc_bucket(ca, PTR_BUCKET_NR(ca, ptr));
150 }
151
152 static inline enum bch_data_type ptr_data_type(const struct bkey *k,
153 const struct bch_extent_ptr *ptr)
154 {
155 if (bkey_is_btree_ptr(k))
156 return BCH_DATA_btree;
157
158 return ptr->cached ? BCH_DATA_cached : BCH_DATA_user;
159 }
160
161 static inline s64 ptr_disk_sectors(s64 sectors, struct extent_ptr_decoded p)
162 {
163 EBUG_ON(sectors < 0);
164
165 return crc_is_compressed(p.crc)
166 ? DIV_ROUND_UP_ULL(sectors * p.crc.compressed_size,
167 p.crc.uncompressed_size)
168 : sectors;
169 }
170
171 static inline int gen_cmp(u8 a, u8 b)
172 {
173 return (s8) (a - b);
174 }
175
176 static inline int gen_after(u8 a, u8 b)
177 {
178 int r = gen_cmp(a, b);
179
180 return r > 0 ? r : 0;
181 }
182
183 static inline int dev_ptr_stale_rcu(struct bch_dev *ca, const struct bch_extent_ptr *ptr)
184 {
185 u8 *gen = bucket_gen(ca, PTR_BUCKET_NR(ca, ptr));
186 if (!gen)
187 return -1;
188 return gen_after(*gen, ptr->gen);
189 }
190
191 /**
192 * dev_ptr_stale() - check if a pointer points into a bucket that has been
193 * invalidated.
194 */
195 static inline int dev_ptr_stale(struct bch_dev *ca, const struct bch_extent_ptr *ptr)
196 {
197 rcu_read_lock();
198 int ret = dev_ptr_stale_rcu(ca, ptr);
199 rcu_read_unlock();
200
201 return ret;
202 }
203
204 /* Device usage: */
205
206 void bch2_dev_usage_read_fast(struct bch_dev *, struct bch_dev_usage *);
207 static inline struct bch_dev_usage bch2_dev_usage_read(struct bch_dev *ca)
208 {
209 struct bch_dev_usage ret;
210
211 bch2_dev_usage_read_fast(ca, &ret);
212 return ret;
213 }
214
215 void bch2_dev_usage_to_text(struct printbuf *, struct bch_dev *, struct bch_dev_usage *);
216
217 static inline u64 bch2_dev_buckets_reserved(struct bch_dev *ca, enum bch_watermark watermark)
218 {
219 s64 reserved = 0;
220
221 switch (watermark) {
222 case BCH_WATERMARK_NR:
223 BUG();
224 case BCH_WATERMARK_stripe:
225 reserved += ca->mi.nbuckets >> 6;
226 fallthrough;
227 case BCH_WATERMARK_normal:
228 reserved += ca->mi.nbuckets >> 6;
229 fallthrough;
230 case BCH_WATERMARK_copygc:
231 reserved += ca->nr_btree_reserve;
232 fallthrough;
233 case BCH_WATERMARK_btree:
234 reserved += ca->nr_btree_reserve;
235 fallthrough;
236 case BCH_WATERMARK_btree_copygc:
237 case BCH_WATERMARK_reclaim:
238 case BCH_WATERMARK_interior_updates:
239 break;
240 }
241
242 return reserved;
243 }
244
245 static inline u64 dev_buckets_free(struct bch_dev *ca,
246 struct bch_dev_usage usage,
247 enum bch_watermark watermark)
248 {
249 return max_t(s64, 0,
250 usage.d[BCH_DATA_free].buckets -
251 ca->nr_open_buckets -
252 bch2_dev_buckets_reserved(ca, watermark));
253 }
254
255 static inline u64 __dev_buckets_available(struct bch_dev *ca,
256 struct bch_dev_usage usage,
257 enum bch_watermark watermark)
258 {
259 return max_t(s64, 0,
260 usage.d[BCH_DATA_free].buckets
261 + usage.d[BCH_DATA_cached].buckets
262 + usage.d[BCH_DATA_need_gc_gens].buckets
263 + usage.d[BCH_DATA_need_discard].buckets
264 - ca->nr_open_buckets
265 - bch2_dev_buckets_reserved(ca, watermark));
266 }
267
268 static inline u64 dev_buckets_available(struct bch_dev *ca,
269 enum bch_watermark watermark)
270 {
271 return __dev_buckets_available(ca, bch2_dev_usage_read(ca), watermark);
272 }
273
274 /* Filesystem usage: */
275
276 static inline unsigned dev_usage_u64s(void)
277 {
278 return sizeof(struct bch_dev_usage) / sizeof(u64);
279 }
280
281 struct bch_fs_usage_short
282 bch2_fs_usage_read_short(struct bch_fs *);
283
284 int bch2_bucket_ref_update(struct btree_trans *, struct bch_dev *,
285 struct bkey_s_c, const struct bch_extent_ptr *,
286 s64, enum bch_data_type, u8, u8, u32 *);
287
288 int bch2_check_fix_ptrs(struct btree_trans *,
289 enum btree_id, unsigned, struct bkey_s_c,
290 enum btree_iter_update_trigger_flags);
291
292 int bch2_trigger_extent(struct btree_trans *, enum btree_id, unsigned,
293 struct bkey_s_c, struct bkey_s,
294 enum btree_iter_update_trigger_flags);
295 int bch2_trigger_reservation(struct btree_trans *, enum btree_id, unsigned,
296 struct bkey_s_c, struct bkey_s,
297 enum btree_iter_update_trigger_flags);
298
299 #define trigger_run_overwrite_then_insert(_fn, _trans, _btree_id, _level, _old, _new, _flags)\
300 ({ \
301 int ret = 0; \
302 \
303 if (_old.k->type) \
304 ret = _fn(_trans, _btree_id, _level, _old, _flags & ~BTREE_TRIGGER_insert); \
305 if (!ret && _new.k->type) \
306 ret = _fn(_trans, _btree_id, _level, _new.s_c, _flags & ~BTREE_TRIGGER_overwrite);\
307 ret; \
308 })
309
310 void bch2_trans_account_disk_usage_change(struct btree_trans *);
311
312 int bch2_trans_mark_metadata_bucket(struct btree_trans *, struct bch_dev *, u64,
313 enum bch_data_type, unsigned,
314 enum btree_iter_update_trigger_flags);
315 int bch2_trans_mark_dev_sb(struct bch_fs *, struct bch_dev *,
316 enum btree_iter_update_trigger_flags);
317 int bch2_trans_mark_dev_sbs_flags(struct bch_fs *,
318 enum btree_iter_update_trigger_flags);
319 int bch2_trans_mark_dev_sbs(struct bch_fs *);
320
321 static inline bool is_superblock_bucket(struct bch_dev *ca, u64 b)
322 {
323 struct bch_sb_layout *layout = &ca->disk_sb.sb->layout;
324 u64 b_offset = bucket_to_sector(ca, b);
325 u64 b_end = bucket_to_sector(ca, b + 1);
326 unsigned i;
327
328 if (!b)
329 return true;
330
331 for (i = 0; i < layout->nr_superblocks; i++) {
332 u64 offset = le64_to_cpu(layout->sb_offset[i]);
333 u64 end = offset + (1 << layout->sb_max_size_bits);
334
335 if (!(offset >= b_end || end <= b_offset))
336 return true;
337 }
338
339 return false;
340 }
341
342 static inline const char *bch2_data_type_str(enum bch_data_type type)
343 {
344 return type < BCH_DATA_NR
345 ? __bch2_data_types[type]
346 : "(invalid data type)";
347 }
348
349 /* disk reservations: */
350
351 static inline void bch2_disk_reservation_put(struct bch_fs *c,
352 struct disk_reservation *res)
353 {
354 if (res->sectors) {
355 this_cpu_sub(*c->online_reserved, res->sectors);
356 res->sectors = 0;
357 }
358 }
359
360 #define BCH_DISK_RESERVATION_NOFAIL (1 << 0)
361
362 int __bch2_disk_reservation_add(struct bch_fs *,
363 struct disk_reservation *,
364 u64, int);
365
366 static inline int bch2_disk_reservation_add(struct bch_fs *c, struct disk_reservation *res,
367 u64 sectors, int flags)
368 {
369 #ifdef __KERNEL__
370 u64 old, new;
371
372 old = this_cpu_read(c->pcpu->sectors_available);
373 do {
374 if (sectors > old)
375 return __bch2_disk_reservation_add(c, res, sectors, flags);
376
377 new = old - sectors;
378 } while (!this_cpu_try_cmpxchg(c->pcpu->sectors_available, &old, new));
379
380 this_cpu_add(*c->online_reserved, sectors);
381 res->sectors += sectors;
382 return 0;
383 #else
384 return __bch2_disk_reservation_add(c, res, sectors, flags);
385 #endif
386 }
387
388 static inline struct disk_reservation
389 bch2_disk_reservation_init(struct bch_fs *c, unsigned nr_replicas)
390 {
391 return (struct disk_reservation) {
392 .sectors = 0,
393 #if 0
394 /* not used yet: */
395 .gen = c->capacity_gen,
396 #endif
397 .nr_replicas = nr_replicas,
398 };
399 }
400
401 static inline int bch2_disk_reservation_get(struct bch_fs *c,
402 struct disk_reservation *res,
403 u64 sectors, unsigned nr_replicas,
404 int flags)
405 {
406 *res = bch2_disk_reservation_init(c, nr_replicas);
407
408 return bch2_disk_reservation_add(c, res, sectors * nr_replicas, flags);
409 }
410
411 #define RESERVE_FACTOR 6
412
413 static inline u64 avail_factor(u64 r)
414 {
415 return div_u64(r << RESERVE_FACTOR, (1 << RESERVE_FACTOR) + 1);
416 }
417
418 void bch2_buckets_nouse_free(struct bch_fs *);
419 int bch2_buckets_nouse_alloc(struct bch_fs *);
420
421 int bch2_dev_buckets_resize(struct bch_fs *, struct bch_dev *, u64);
422 void bch2_dev_buckets_free(struct bch_dev *);
423 int bch2_dev_buckets_alloc(struct bch_fs *, struct bch_dev *);
424
425 #endif /* _BUCKETS_H */
426
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.