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->gc_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 BUG_ON(!bucket_valid(ca, b)); 97 return buckets->b + b; 98 } 99 100 static inline struct bucket_gens *bucket_gens(struct bch_dev *ca) 101 { 102 return rcu_dereference_check(ca->bucket_gens, 103 !ca->fs || 104 percpu_rwsem_is_held(&ca->fs->mark_lock) || 105 lockdep_is_held(&ca->fs->gc_lock) || 106 lockdep_is_held(&ca->bucket_lock)); 107 } 108 109 static inline u8 *bucket_gen(struct bch_dev *ca, size_t b) 110 { 111 struct bucket_gens *gens = bucket_gens(ca); 112 113 BUG_ON(!bucket_valid(ca, b)); 114 return gens->b + b; 115 } 116 117 static inline size_t PTR_BUCKET_NR(const struct bch_dev *ca, 118 const struct bch_extent_ptr *ptr) 119 { 120 return sector_to_bucket(ca, ptr->offset); 121 } 122 123 static inline struct bpos PTR_BUCKET_POS(const struct bch_dev *ca, 124 const struct bch_extent_ptr *ptr) 125 { 126 return POS(ptr->dev, PTR_BUCKET_NR(ca, ptr)); 127 } 128 129 static inline struct bpos PTR_BUCKET_POS_OFFSET(const struct bch_dev *ca, 130 const struct bch_extent_ptr *ptr, 131 u32 *bucket_offset) 132 { 133 return POS(ptr->dev, sector_to_bucket_and_offset(ca, ptr->offset, bucket_offset)); 134 } 135 136 static inline struct bucket *PTR_GC_BUCKET(struct bch_dev *ca, 137 const struct bch_extent_ptr *ptr) 138 { 139 return gc_bucket(ca, PTR_BUCKET_NR(ca, ptr)); 140 } 141 142 static inline enum bch_data_type ptr_data_type(const struct bkey *k, 143 const struct bch_extent_ptr *ptr) 144 { 145 if (bkey_is_btree_ptr(k)) 146 return BCH_DATA_btree; 147 148 return ptr->cached ? BCH_DATA_cached : BCH_DATA_user; 149 } 150 151 static inline s64 ptr_disk_sectors(s64 sectors, struct extent_ptr_decoded p) 152 { 153 EBUG_ON(sectors < 0); 154 155 return crc_is_compressed(p.crc) 156 ? DIV_ROUND_UP_ULL(sectors * p.crc.compressed_size, 157 p.crc.uncompressed_size) 158 : sectors; 159 } 160 161 static inline int gen_cmp(u8 a, u8 b) 162 { 163 return (s8) (a - b); 164 } 165 166 static inline int gen_after(u8 a, u8 b) 167 { 168 int r = gen_cmp(a, b); 169 170 return r > 0 ? r : 0; 171 } 172 173 static inline u8 dev_ptr_stale_rcu(struct bch_dev *ca, const struct bch_extent_ptr *ptr) 174 { 175 return gen_after(*bucket_gen(ca, PTR_BUCKET_NR(ca, ptr)), 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 u8 dev_ptr_stale(struct bch_dev *ca, const struct bch_extent_ptr *ptr) 183 { 184 rcu_read_lock(); 185 u8 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_init(struct bch_dev *); 203 void bch2_dev_usage_to_text(struct printbuf *, struct bch_dev_usage *); 204 205 static inline u64 bch2_dev_buckets_reserved(struct bch_dev *ca, enum bch_watermark watermark) 206 { 207 s64 reserved = 0; 208 209 switch (watermark) { 210 case BCH_WATERMARK_NR: 211 BUG(); 212 case BCH_WATERMARK_stripe: 213 reserved += ca->mi.nbuckets >> 6; 214 fallthrough; 215 case BCH_WATERMARK_normal: 216 reserved += ca->mi.nbuckets >> 6; 217 fallthrough; 218 case BCH_WATERMARK_copygc: 219 reserved += ca->nr_btree_reserve; 220 fallthrough; 221 case BCH_WATERMARK_btree: 222 reserved += ca->nr_btree_reserve; 223 fallthrough; 224 case BCH_WATERMARK_btree_copygc: 225 case BCH_WATERMARK_reclaim: 226 case BCH_WATERMARK_interior_updates: 227 break; 228 } 229 230 return reserved; 231 } 232 233 static inline u64 dev_buckets_free(struct bch_dev *ca, 234 struct bch_dev_usage usage, 235 enum bch_watermark watermark) 236 { 237 return max_t(s64, 0, 238 usage.d[BCH_DATA_free].buckets - 239 ca->nr_open_buckets - 240 bch2_dev_buckets_reserved(ca, watermark)); 241 } 242 243 static inline u64 __dev_buckets_available(struct bch_dev *ca, 244 struct bch_dev_usage usage, 245 enum bch_watermark watermark) 246 { 247 return max_t(s64, 0, 248 usage.d[BCH_DATA_free].buckets 249 + usage.d[BCH_DATA_cached].buckets 250 + usage.d[BCH_DATA_need_gc_gens].buckets 251 + usage.d[BCH_DATA_need_discard].buckets 252 - ca->nr_open_buckets 253 - bch2_dev_buckets_reserved(ca, watermark)); 254 } 255 256 static inline u64 dev_buckets_available(struct bch_dev *ca, 257 enum bch_watermark watermark) 258 { 259 return __dev_buckets_available(ca, bch2_dev_usage_read(ca), watermark); 260 } 261 262 /* Filesystem usage: */ 263 264 static inline unsigned __fs_usage_u64s(unsigned nr_replicas) 265 { 266 return sizeof(struct bch_fs_usage) / sizeof(u64) + nr_replicas; 267 } 268 269 static inline unsigned fs_usage_u64s(struct bch_fs *c) 270 { 271 return __fs_usage_u64s(READ_ONCE(c->replicas.nr)); 272 } 273 274 static inline unsigned __fs_usage_online_u64s(unsigned nr_replicas) 275 { 276 return sizeof(struct bch_fs_usage_online) / sizeof(u64) + nr_replicas; 277 } 278 279 static inline unsigned fs_usage_online_u64s(struct bch_fs *c) 280 { 281 return __fs_usage_online_u64s(READ_ONCE(c->replicas.nr)); 282 } 283 284 static inline unsigned dev_usage_u64s(void) 285 { 286 return sizeof(struct bch_dev_usage) / sizeof(u64); 287 } 288 289 u64 bch2_fs_usage_read_one(struct bch_fs *, u64 *); 290 291 struct bch_fs_usage_online *bch2_fs_usage_read(struct bch_fs *); 292 293 void bch2_fs_usage_acc_to_base(struct bch_fs *, unsigned); 294 295 void bch2_fs_usage_to_text(struct printbuf *, 296 struct bch_fs *, struct bch_fs_usage_online *); 297 298 u64 bch2_fs_sectors_used(struct bch_fs *, struct bch_fs_usage_online *); 299 300 struct bch_fs_usage_short 301 bch2_fs_usage_read_short(struct bch_fs *); 302 303 void bch2_dev_usage_update(struct bch_fs *, struct bch_dev *, 304 const struct bch_alloc_v4 *, 305 const struct bch_alloc_v4 *, u64, bool); 306 307 /* key/bucket marking: */ 308 309 static inline struct bch_fs_usage *fs_usage_ptr(struct bch_fs *c, 310 unsigned journal_seq, 311 bool gc) 312 { 313 percpu_rwsem_assert_held(&c->mark_lock); 314 BUG_ON(!gc && !journal_seq); 315 316 return this_cpu_ptr(gc 317 ? c->usage_gc 318 : c->usage[journal_seq & JOURNAL_BUF_MASK]); 319 } 320 321 int bch2_update_replicas(struct bch_fs *, struct bkey_s_c, 322 struct bch_replicas_entry_v1 *, s64, 323 unsigned, bool); 324 int bch2_update_replicas_list(struct btree_trans *, 325 struct bch_replicas_entry_v1 *, s64); 326 int bch2_update_cached_sectors_list(struct btree_trans *, unsigned, s64); 327 int bch2_replicas_deltas_realloc(struct btree_trans *, unsigned); 328 329 void bch2_fs_usage_initialize(struct bch_fs *); 330 331 int bch2_bucket_ref_update(struct btree_trans *, struct bch_dev *, 332 struct bkey_s_c, const struct bch_extent_ptr *, 333 s64, enum bch_data_type, u8, u8, u32 *); 334 335 int bch2_check_fix_ptrs(struct btree_trans *, 336 enum btree_id, unsigned, struct bkey_s_c, 337 enum btree_iter_update_trigger_flags); 338 339 int bch2_trigger_extent(struct btree_trans *, enum btree_id, unsigned, 340 struct bkey_s_c, struct bkey_s, 341 enum btree_iter_update_trigger_flags); 342 int bch2_trigger_reservation(struct btree_trans *, enum btree_id, unsigned, 343 struct bkey_s_c, struct bkey_s, 344 enum btree_iter_update_trigger_flags); 345 346 #define trigger_run_overwrite_then_insert(_fn, _trans, _btree_id, _level, _old, _new, _flags)\ 347 ({ \ 348 int ret = 0; \ 349 \ 350 if (_old.k->type) \ 351 ret = _fn(_trans, _btree_id, _level, _old, _flags & ~BTREE_TRIGGER_insert); \ 352 if (!ret && _new.k->type) \ 353 ret = _fn(_trans, _btree_id, _level, _new.s_c, _flags & ~BTREE_TRIGGER_overwrite);\ 354 ret; \ 355 }) 356 357 void bch2_trans_account_disk_usage_change(struct btree_trans *); 358 359 void bch2_trans_fs_usage_revert(struct btree_trans *, struct replicas_delta_list *); 360 int bch2_trans_fs_usage_apply(struct btree_trans *, struct replicas_delta_list *); 361 362 int bch2_trans_mark_metadata_bucket(struct btree_trans *, struct bch_dev *, u64, 363 enum bch_data_type, unsigned, 364 enum btree_iter_update_trigger_flags); 365 int bch2_trans_mark_dev_sb(struct bch_fs *, struct bch_dev *, 366 enum btree_iter_update_trigger_flags); 367 int bch2_trans_mark_dev_sbs_flags(struct bch_fs *, 368 enum btree_iter_update_trigger_flags); 369 int bch2_trans_mark_dev_sbs(struct bch_fs *); 370 371 static inline bool is_superblock_bucket(struct bch_dev *ca, u64 b) 372 { 373 struct bch_sb_layout *layout = &ca->disk_sb.sb->layout; 374 u64 b_offset = bucket_to_sector(ca, b); 375 u64 b_end = bucket_to_sector(ca, b + 1); 376 unsigned i; 377 378 if (!b) 379 return true; 380 381 for (i = 0; i < layout->nr_superblocks; i++) { 382 u64 offset = le64_to_cpu(layout->sb_offset[i]); 383 u64 end = offset + (1 << layout->sb_max_size_bits); 384 385 if (!(offset >= b_end || end <= b_offset)) 386 return true; 387 } 388 389 return false; 390 } 391 392 static inline const char *bch2_data_type_str(enum bch_data_type type) 393 { 394 return type < BCH_DATA_NR 395 ? __bch2_data_types[type] 396 : "(invalid data type)"; 397 } 398 399 /* disk reservations: */ 400 401 static inline void bch2_disk_reservation_put(struct bch_fs *c, 402 struct disk_reservation *res) 403 { 404 if (res->sectors) { 405 this_cpu_sub(*c->online_reserved, res->sectors); 406 res->sectors = 0; 407 } 408 } 409 410 #define BCH_DISK_RESERVATION_NOFAIL (1 << 0) 411 412 int __bch2_disk_reservation_add(struct bch_fs *, 413 struct disk_reservation *, 414 u64, int); 415 416 static inline int bch2_disk_reservation_add(struct bch_fs *c, struct disk_reservation *res, 417 u64 sectors, int flags) 418 { 419 #ifdef __KERNEL__ 420 u64 old, new; 421 422 do { 423 old = this_cpu_read(c->pcpu->sectors_available); 424 if (sectors > old) 425 return __bch2_disk_reservation_add(c, res, sectors, flags); 426 427 new = old - sectors; 428 } while (this_cpu_cmpxchg(c->pcpu->sectors_available, old, new) != old); 429 430 this_cpu_add(*c->online_reserved, sectors); 431 res->sectors += sectors; 432 return 0; 433 #else 434 return __bch2_disk_reservation_add(c, res, sectors, flags); 435 #endif 436 } 437 438 static inline struct disk_reservation 439 bch2_disk_reservation_init(struct bch_fs *c, unsigned nr_replicas) 440 { 441 return (struct disk_reservation) { 442 .sectors = 0, 443 #if 0 444 /* not used yet: */ 445 .gen = c->capacity_gen, 446 #endif 447 .nr_replicas = nr_replicas, 448 }; 449 } 450 451 static inline int bch2_disk_reservation_get(struct bch_fs *c, 452 struct disk_reservation *res, 453 u64 sectors, unsigned nr_replicas, 454 int flags) 455 { 456 *res = bch2_disk_reservation_init(c, nr_replicas); 457 458 return bch2_disk_reservation_add(c, res, sectors * nr_replicas, flags); 459 } 460 461 #define RESERVE_FACTOR 6 462 463 static inline u64 avail_factor(u64 r) 464 { 465 return div_u64(r << RESERVE_FACTOR, (1 << RESERVE_FACTOR) + 1); 466 } 467 468 void bch2_buckets_nouse_free(struct bch_fs *); 469 int bch2_buckets_nouse_alloc(struct bch_fs *); 470 471 int bch2_dev_buckets_resize(struct bch_fs *, struct bch_dev *, u64); 472 void bch2_dev_buckets_free(struct bch_dev *); 473 int bch2_dev_buckets_alloc(struct bch_fs *, struct bch_dev *); 474 475 #endif /* _BUCKETS_H */ 476