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 *gc_bucket(struct bch_dev *ca, size_t b) 84 { 85 return genradix_ptr(&ca->buckets_gc, b); 86 } 87 88 static inline struct bucket_gens *bucket_gens(struct bch_dev *ca) 89 { 90 return rcu_dereference_check(ca->bucket_gens, 91 !ca->fs || 92 percpu_rwsem_is_held(&ca->fs->mark_lock) || 93 lockdep_is_held(&ca->fs->state_lock) || 94 lockdep_is_held(&ca->bucket_lock)); 95 } 96 97 static inline u8 *bucket_gen(struct bch_dev *ca, size_t b) 98 { 99 struct bucket_gens *gens = bucket_gens(ca); 100 101 if (b - gens->first_bucket >= gens->nbuckets_minus_first) 102 return NULL; 103 return gens->b + b; 104 } 105 106 static inline u8 bucket_gen_get(struct bch_dev *ca, size_t b) 107 { 108 rcu_read_lock(); 109 u8 gen = *bucket_gen(ca, b); 110 rcu_read_unlock(); 111 return gen; 112 } 113 114 static inline size_t PTR_BUCKET_NR(const struct bch_dev *ca, 115 const struct bch_extent_ptr *ptr) 116 { 117 return sector_to_bucket(ca, ptr->offset); 118 } 119 120 static inline struct bpos PTR_BUCKET_POS(const struct bch_dev *ca, 121 const struct bch_extent_ptr *ptr) 122 { 123 return POS(ptr->dev, PTR_BUCKET_NR(ca, ptr)); 124 } 125 126 static inline struct bpos PTR_BUCKET_POS_OFFSET(const struct bch_dev *ca, 127 const struct bch_extent_ptr *ptr, 128 u32 *bucket_offset) 129 { 130 return POS(ptr->dev, sector_to_bucket_and_offset(ca, ptr->offset, bucket_offset)); 131 } 132 133 static inline struct bucket *PTR_GC_BUCKET(struct bch_dev *ca, 134 const struct bch_extent_ptr *ptr) 135 { 136 return gc_bucket(ca, PTR_BUCKET_NR(ca, ptr)); 137 } 138 139 static inline enum bch_data_type ptr_data_type(const struct bkey *k, 140 const struct bch_extent_ptr *ptr) 141 { 142 if (bkey_is_btree_ptr(k)) 143 return BCH_DATA_btree; 144 145 return ptr->cached ? BCH_DATA_cached : BCH_DATA_user; 146 } 147 148 static inline s64 ptr_disk_sectors(s64 sectors, struct extent_ptr_decoded p) 149 { 150 EBUG_ON(sectors < 0); 151 152 return crc_is_compressed(p.crc) 153 ? DIV_ROUND_UP_ULL(sectors * p.crc.compressed_size, 154 p.crc.uncompressed_size) 155 : sectors; 156 } 157 158 static inline int gen_cmp(u8 a, u8 b) 159 { 160 return (s8) (a - b); 161 } 162 163 static inline int gen_after(u8 a, u8 b) 164 { 165 int r = gen_cmp(a, b); 166 167 return r > 0 ? r : 0; 168 } 169 170 static inline int dev_ptr_stale_rcu(struct bch_dev *ca, const struct bch_extent_ptr *ptr) 171 { 172 u8 *gen = bucket_gen(ca, PTR_BUCKET_NR(ca, ptr)); 173 if (!gen) 174 return -1; 175 return gen_after(*gen, ptr->gen); 176 } 177 178 /** 179 * dev_ptr_stale() - check if a pointer points into a bucket that has been 180 * invalidated. 181 */ 182 static inline int dev_ptr_stale(struct bch_dev *ca, const struct bch_extent_ptr *ptr) 183 { 184 rcu_read_lock(); 185 int ret = dev_ptr_stale_rcu(ca, ptr); 186 rcu_read_unlock(); 187 188 return ret; 189 } 190 191 /* Device usage: */ 192 193 void bch2_dev_usage_read_fast(struct bch_dev *, struct bch_dev_usage *); 194 static inline struct bch_dev_usage bch2_dev_usage_read(struct bch_dev *ca) 195 { 196 struct bch_dev_usage ret; 197 198 bch2_dev_usage_read_fast(ca, &ret); 199 return ret; 200 } 201 202 void bch2_dev_usage_to_text(struct printbuf *, struct bch_dev *, struct bch_dev_usage *); 203 204 static inline u64 bch2_dev_buckets_reserved(struct bch_dev *ca, enum bch_watermark watermark) 205 { 206 s64 reserved = 0; 207 208 switch (watermark) { 209 case BCH_WATERMARK_NR: 210 BUG(); 211 case BCH_WATERMARK_stripe: 212 reserved += ca->mi.nbuckets >> 6; 213 fallthrough; 214 case BCH_WATERMARK_normal: 215 reserved += ca->mi.nbuckets >> 6; 216 fallthrough; 217 case BCH_WATERMARK_copygc: 218 reserved += ca->nr_btree_reserve; 219 fallthrough; 220 case BCH_WATERMARK_btree: 221 reserved += ca->nr_btree_reserve; 222 fallthrough; 223 case BCH_WATERMARK_btree_copygc: 224 case BCH_WATERMARK_reclaim: 225 case BCH_WATERMARK_interior_updates: 226 break; 227 } 228 229 return reserved; 230 } 231 232 static inline u64 dev_buckets_free(struct bch_dev *ca, 233 struct bch_dev_usage usage, 234 enum bch_watermark watermark) 235 { 236 return max_t(s64, 0, 237 usage.d[BCH_DATA_free].buckets - 238 ca->nr_open_buckets - 239 bch2_dev_buckets_reserved(ca, watermark)); 240 } 241 242 static inline u64 __dev_buckets_available(struct bch_dev *ca, 243 struct bch_dev_usage usage, 244 enum bch_watermark watermark) 245 { 246 return max_t(s64, 0, 247 usage.d[BCH_DATA_free].buckets 248 + usage.d[BCH_DATA_cached].buckets 249 + usage.d[BCH_DATA_need_gc_gens].buckets 250 + usage.d[BCH_DATA_need_discard].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 enum bch_watermark watermark) 257 { 258 return __dev_buckets_available(ca, bch2_dev_usage_read(ca), watermark); 259 } 260 261 /* Filesystem usage: */ 262 263 static inline unsigned dev_usage_u64s(void) 264 { 265 return sizeof(struct bch_dev_usage) / sizeof(u64); 266 } 267 268 struct bch_fs_usage_short 269 bch2_fs_usage_read_short(struct bch_fs *); 270 271 int bch2_bucket_ref_update(struct btree_trans *, struct bch_dev *, 272 struct bkey_s_c, const struct bch_extent_ptr *, 273 s64, enum bch_data_type, u8, u8, u32 *); 274 275 int bch2_check_fix_ptrs(struct btree_trans *, 276 enum btree_id, unsigned, struct bkey_s_c, 277 enum btree_iter_update_trigger_flags); 278 279 int bch2_trigger_extent(struct btree_trans *, enum btree_id, unsigned, 280 struct bkey_s_c, struct bkey_s, 281 enum btree_iter_update_trigger_flags); 282 int bch2_trigger_reservation(struct btree_trans *, enum btree_id, unsigned, 283 struct bkey_s_c, struct bkey_s, 284 enum btree_iter_update_trigger_flags); 285 286 #define trigger_run_overwrite_then_insert(_fn, _trans, _btree_id, _level, _old, _new, _flags)\ 287 ({ \ 288 int ret = 0; \ 289 \ 290 if (_old.k->type) \ 291 ret = _fn(_trans, _btree_id, _level, _old, _flags & ~BTREE_TRIGGER_insert); \ 292 if (!ret && _new.k->type) \ 293 ret = _fn(_trans, _btree_id, _level, _new.s_c, _flags & ~BTREE_TRIGGER_overwrite);\ 294 ret; \ 295 }) 296 297 void bch2_trans_account_disk_usage_change(struct btree_trans *); 298 299 int bch2_trans_mark_metadata_bucket(struct btree_trans *, struct bch_dev *, u64, 300 enum bch_data_type, unsigned, 301 enum btree_iter_update_trigger_flags); 302 int bch2_trans_mark_dev_sb(struct bch_fs *, struct bch_dev *, 303 enum btree_iter_update_trigger_flags); 304 int bch2_trans_mark_dev_sbs_flags(struct bch_fs *, 305 enum btree_iter_update_trigger_flags); 306 int bch2_trans_mark_dev_sbs(struct bch_fs *); 307 308 static inline bool is_superblock_bucket(struct bch_dev *ca, u64 b) 309 { 310 struct bch_sb_layout *layout = &ca->disk_sb.sb->layout; 311 u64 b_offset = bucket_to_sector(ca, b); 312 u64 b_end = bucket_to_sector(ca, b + 1); 313 unsigned i; 314 315 if (!b) 316 return true; 317 318 for (i = 0; i < layout->nr_superblocks; i++) { 319 u64 offset = le64_to_cpu(layout->sb_offset[i]); 320 u64 end = offset + (1 << layout->sb_max_size_bits); 321 322 if (!(offset >= b_end || end <= b_offset)) 323 return true; 324 } 325 326 return false; 327 } 328 329 static inline const char *bch2_data_type_str(enum bch_data_type type) 330 { 331 return type < BCH_DATA_NR 332 ? __bch2_data_types[type] 333 : "(invalid data type)"; 334 } 335 336 /* disk reservations: */ 337 338 static inline void bch2_disk_reservation_put(struct bch_fs *c, 339 struct disk_reservation *res) 340 { 341 if (res->sectors) { 342 this_cpu_sub(*c->online_reserved, res->sectors); 343 res->sectors = 0; 344 } 345 } 346 347 #define BCH_DISK_RESERVATION_NOFAIL (1 << 0) 348 349 int __bch2_disk_reservation_add(struct bch_fs *, 350 struct disk_reservation *, 351 u64, int); 352 353 static inline int bch2_disk_reservation_add(struct bch_fs *c, struct disk_reservation *res, 354 u64 sectors, int flags) 355 { 356 #ifdef __KERNEL__ 357 u64 old, new; 358 359 old = this_cpu_read(c->pcpu->sectors_available); 360 do { 361 if (sectors > old) 362 return __bch2_disk_reservation_add(c, res, sectors, flags); 363 364 new = old - sectors; 365 } while (!this_cpu_try_cmpxchg(c->pcpu->sectors_available, &old, new)); 366 367 this_cpu_add(*c->online_reserved, sectors); 368 res->sectors += sectors; 369 return 0; 370 #else 371 return __bch2_disk_reservation_add(c, res, sectors, flags); 372 #endif 373 } 374 375 static inline struct disk_reservation 376 bch2_disk_reservation_init(struct bch_fs *c, unsigned nr_replicas) 377 { 378 return (struct disk_reservation) { 379 .sectors = 0, 380 #if 0 381 /* not used yet: */ 382 .gen = c->capacity_gen, 383 #endif 384 .nr_replicas = nr_replicas, 385 }; 386 } 387 388 static inline int bch2_disk_reservation_get(struct bch_fs *c, 389 struct disk_reservation *res, 390 u64 sectors, unsigned nr_replicas, 391 int flags) 392 { 393 *res = bch2_disk_reservation_init(c, nr_replicas); 394 395 return bch2_disk_reservation_add(c, res, sectors * nr_replicas, flags); 396 } 397 398 #define RESERVE_FACTOR 6 399 400 static inline u64 avail_factor(u64 r) 401 { 402 return div_u64(r << RESERVE_FACTOR, (1 << RESERVE_FACTOR) + 1); 403 } 404 405 void bch2_buckets_nouse_free(struct bch_fs *); 406 int bch2_buckets_nouse_alloc(struct bch_fs *); 407 408 int bch2_dev_buckets_resize(struct bch_fs *, struct bch_dev *, u64); 409 void bch2_dev_buckets_free(struct bch_dev *); 410 int bch2_dev_buckets_alloc(struct bch_fs *, struct bch_dev *); 411 412 #endif /* _BUCKETS_H */ 413