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 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->gc_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_init(struct bch_dev *); 216 void bch2_dev_usage_to_text(struct printbuf *, struct bch_dev_usage *); 217 218 static inline u64 bch2_dev_buckets_reserved(struct bch_dev *ca, enum bch_watermark watermark) 219 { 220 s64 reserved = 0; 221 222 switch (watermark) { 223 case BCH_WATERMARK_NR: 224 BUG(); 225 case BCH_WATERMARK_stripe: 226 reserved += ca->mi.nbuckets >> 6; 227 fallthrough; 228 case BCH_WATERMARK_normal: 229 reserved += ca->mi.nbuckets >> 6; 230 fallthrough; 231 case BCH_WATERMARK_copygc: 232 reserved += ca->nr_btree_reserve; 233 fallthrough; 234 case BCH_WATERMARK_btree: 235 reserved += ca->nr_btree_reserve; 236 fallthrough; 237 case BCH_WATERMARK_btree_copygc: 238 case BCH_WATERMARK_reclaim: 239 case BCH_WATERMARK_interior_updates: 240 break; 241 } 242 243 return reserved; 244 } 245 246 static inline u64 dev_buckets_free(struct bch_dev *ca, 247 struct bch_dev_usage usage, 248 enum bch_watermark watermark) 249 { 250 return max_t(s64, 0, 251 usage.d[BCH_DATA_free].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 struct bch_dev_usage usage, 258 enum bch_watermark watermark) 259 { 260 return max_t(s64, 0, 261 usage.d[BCH_DATA_free].buckets 262 + usage.d[BCH_DATA_cached].buckets 263 + usage.d[BCH_DATA_need_gc_gens].buckets 264 + usage.d[BCH_DATA_need_discard].buckets 265 - ca->nr_open_buckets 266 - bch2_dev_buckets_reserved(ca, watermark)); 267 } 268 269 static inline u64 dev_buckets_available(struct bch_dev *ca, 270 enum bch_watermark watermark) 271 { 272 return __dev_buckets_available(ca, bch2_dev_usage_read(ca), watermark); 273 } 274 275 /* Filesystem usage: */ 276 277 static inline unsigned __fs_usage_u64s(unsigned nr_replicas) 278 { 279 return sizeof(struct bch_fs_usage) / sizeof(u64) + nr_replicas; 280 } 281 282 static inline unsigned fs_usage_u64s(struct bch_fs *c) 283 { 284 return __fs_usage_u64s(READ_ONCE(c->replicas.nr)); 285 } 286 287 static inline unsigned __fs_usage_online_u64s(unsigned nr_replicas) 288 { 289 return sizeof(struct bch_fs_usage_online) / sizeof(u64) + nr_replicas; 290 } 291 292 static inline unsigned fs_usage_online_u64s(struct bch_fs *c) 293 { 294 return __fs_usage_online_u64s(READ_ONCE(c->replicas.nr)); 295 } 296 297 static inline unsigned dev_usage_u64s(void) 298 { 299 return sizeof(struct bch_dev_usage) / sizeof(u64); 300 } 301 302 u64 bch2_fs_usage_read_one(struct bch_fs *, u64 *); 303 304 struct bch_fs_usage_online *bch2_fs_usage_read(struct bch_fs *); 305 306 void bch2_fs_usage_acc_to_base(struct bch_fs *, unsigned); 307 308 void bch2_fs_usage_to_text(struct printbuf *, 309 struct bch_fs *, struct bch_fs_usage_online *); 310 311 u64 bch2_fs_sectors_used(struct bch_fs *, struct bch_fs_usage_online *); 312 313 struct bch_fs_usage_short 314 bch2_fs_usage_read_short(struct bch_fs *); 315 316 void bch2_dev_usage_update(struct bch_fs *, struct bch_dev *, 317 const struct bch_alloc_v4 *, 318 const struct bch_alloc_v4 *, u64, bool); 319 320 /* key/bucket marking: */ 321 322 static inline struct bch_fs_usage *fs_usage_ptr(struct bch_fs *c, 323 unsigned journal_seq, 324 bool gc) 325 { 326 percpu_rwsem_assert_held(&c->mark_lock); 327 BUG_ON(!gc && !journal_seq); 328 329 return this_cpu_ptr(gc 330 ? c->usage_gc 331 : c->usage[journal_seq & JOURNAL_BUF_MASK]); 332 } 333 334 int bch2_update_replicas(struct bch_fs *, struct bkey_s_c, 335 struct bch_replicas_entry_v1 *, s64, 336 unsigned, bool); 337 int bch2_update_replicas_list(struct btree_trans *, 338 struct bch_replicas_entry_v1 *, s64); 339 int bch2_update_cached_sectors_list(struct btree_trans *, unsigned, s64); 340 int bch2_replicas_deltas_realloc(struct btree_trans *, unsigned); 341 342 void bch2_fs_usage_initialize(struct bch_fs *); 343 344 int bch2_bucket_ref_update(struct btree_trans *, struct bch_dev *, 345 struct bkey_s_c, const struct bch_extent_ptr *, 346 s64, enum bch_data_type, u8, u8, u32 *); 347 348 int bch2_check_fix_ptrs(struct btree_trans *, 349 enum btree_id, unsigned, struct bkey_s_c, 350 enum btree_iter_update_trigger_flags); 351 352 int bch2_trigger_extent(struct btree_trans *, enum btree_id, unsigned, 353 struct bkey_s_c, struct bkey_s, 354 enum btree_iter_update_trigger_flags); 355 int bch2_trigger_reservation(struct btree_trans *, enum btree_id, unsigned, 356 struct bkey_s_c, struct bkey_s, 357 enum btree_iter_update_trigger_flags); 358 359 #define trigger_run_overwrite_then_insert(_fn, _trans, _btree_id, _level, _old, _new, _flags)\ 360 ({ \ 361 int ret = 0; \ 362 \ 363 if (_old.k->type) \ 364 ret = _fn(_trans, _btree_id, _level, _old, _flags & ~BTREE_TRIGGER_insert); \ 365 if (!ret && _new.k->type) \ 366 ret = _fn(_trans, _btree_id, _level, _new.s_c, _flags & ~BTREE_TRIGGER_overwrite);\ 367 ret; \ 368 }) 369 370 void bch2_trans_account_disk_usage_change(struct btree_trans *); 371 372 void bch2_trans_fs_usage_revert(struct btree_trans *, struct replicas_delta_list *); 373 int bch2_trans_fs_usage_apply(struct btree_trans *, struct replicas_delta_list *); 374 375 int bch2_trans_mark_metadata_bucket(struct btree_trans *, struct bch_dev *, u64, 376 enum bch_data_type, unsigned, 377 enum btree_iter_update_trigger_flags); 378 int bch2_trans_mark_dev_sb(struct bch_fs *, struct bch_dev *, 379 enum btree_iter_update_trigger_flags); 380 int bch2_trans_mark_dev_sbs_flags(struct bch_fs *, 381 enum btree_iter_update_trigger_flags); 382 int bch2_trans_mark_dev_sbs(struct bch_fs *); 383 384 static inline bool is_superblock_bucket(struct bch_dev *ca, u64 b) 385 { 386 struct bch_sb_layout *layout = &ca->disk_sb.sb->layout; 387 u64 b_offset = bucket_to_sector(ca, b); 388 u64 b_end = bucket_to_sector(ca, b + 1); 389 unsigned i; 390 391 if (!b) 392 return true; 393 394 for (i = 0; i < layout->nr_superblocks; i++) { 395 u64 offset = le64_to_cpu(layout->sb_offset[i]); 396 u64 end = offset + (1 << layout->sb_max_size_bits); 397 398 if (!(offset >= b_end || end <= b_offset)) 399 return true; 400 } 401 402 return false; 403 } 404 405 static inline const char *bch2_data_type_str(enum bch_data_type type) 406 { 407 return type < BCH_DATA_NR 408 ? __bch2_data_types[type] 409 : "(invalid data type)"; 410 } 411 412 /* disk reservations: */ 413 414 static inline void bch2_disk_reservation_put(struct bch_fs *c, 415 struct disk_reservation *res) 416 { 417 if (res->sectors) { 418 this_cpu_sub(*c->online_reserved, res->sectors); 419 res->sectors = 0; 420 } 421 } 422 423 #define BCH_DISK_RESERVATION_NOFAIL (1 << 0) 424 425 int __bch2_disk_reservation_add(struct bch_fs *, 426 struct disk_reservation *, 427 u64, int); 428 429 static inline int bch2_disk_reservation_add(struct bch_fs *c, struct disk_reservation *res, 430 u64 sectors, int flags) 431 { 432 #ifdef __KERNEL__ 433 u64 old, new; 434 435 do { 436 old = this_cpu_read(c->pcpu->sectors_available); 437 if (sectors > old) 438 return __bch2_disk_reservation_add(c, res, sectors, flags); 439 440 new = old - sectors; 441 } while (this_cpu_cmpxchg(c->pcpu->sectors_available, old, new) != old); 442 443 this_cpu_add(*c->online_reserved, sectors); 444 res->sectors += sectors; 445 return 0; 446 #else 447 return __bch2_disk_reservation_add(c, res, sectors, flags); 448 #endif 449 } 450 451 static inline struct disk_reservation 452 bch2_disk_reservation_init(struct bch_fs *c, unsigned nr_replicas) 453 { 454 return (struct disk_reservation) { 455 .sectors = 0, 456 #if 0 457 /* not used yet: */ 458 .gen = c->capacity_gen, 459 #endif 460 .nr_replicas = nr_replicas, 461 }; 462 } 463 464 static inline int bch2_disk_reservation_get(struct bch_fs *c, 465 struct disk_reservation *res, 466 u64 sectors, unsigned nr_replicas, 467 int flags) 468 { 469 *res = bch2_disk_reservation_init(c, nr_replicas); 470 471 return bch2_disk_reservation_add(c, res, sectors * nr_replicas, flags); 472 } 473 474 #define RESERVE_FACTOR 6 475 476 static inline u64 avail_factor(u64 r) 477 { 478 return div_u64(r << RESERVE_FACTOR, (1 << RESERVE_FACTOR) + 1); 479 } 480 481 void bch2_buckets_nouse_free(struct bch_fs *); 482 int bch2_buckets_nouse_alloc(struct bch_fs *); 483 484 int bch2_dev_buckets_resize(struct bch_fs *, struct bch_dev *, u64); 485 void bch2_dev_buckets_free(struct bch_dev *); 486 int bch2_dev_buckets_alloc(struct bch_fs *, struct bch_dev *); 487 488 #endif /* _BUCKETS_H */ 489