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 size_t 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 size_t sector_to_bucket_and_offset(const struct bch_dev *ca, sector_t s, 34 u32 *offset) 35 { 36 return div_u64_rem(s, ca->mi.bucket_size, offset); 37 } 38 39 #define for_each_bucket(_b, _buckets) \ 40 for (_b = (_buckets)->b + (_buckets)->first_bucket; \ 41 _b < (_buckets)->b + (_buckets)->nbuckets; _b++) 42 43 /* 44 * Ugly hack alert: 45 * 46 * We need to cram a spinlock in a single byte, because that's what we have left 47 * in struct bucket, and we care about the size of these - during fsck, we need 48 * in memory state for every single bucket on every device. 49 * 50 * We used to do 51 * while (xchg(&b->lock, 1) cpu_relax(); 52 * but, it turns out not all architectures support xchg on a single byte. 53 * 54 * So now we use bit_spin_lock(), with fun games since we can't burn a whole 55 * ulong for this - we just need to make sure the lock bit always ends up in the 56 * first byte. 57 */ 58 59 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ 60 #define BUCKET_LOCK_BITNR 0 61 #else 62 #define BUCKET_LOCK_BITNR (BITS_PER_LONG - 1) 63 #endif 64 65 union ulong_byte_assert { 66 ulong ulong; 67 u8 byte; 68 }; 69 70 static inline void bucket_unlock(struct bucket *b) 71 { 72 BUILD_BUG_ON(!((union ulong_byte_assert) { .ulong = 1UL << BUCKET_LOCK_BITNR }).byte); 73 74 clear_bit_unlock(BUCKET_LOCK_BITNR, (void *) &b->lock); 75 wake_up_bit((void *) &b->lock, BUCKET_LOCK_BITNR); 76 } 77 78 static inline void bucket_lock(struct bucket *b) 79 { 80 wait_on_bit_lock((void *) &b->lock, BUCKET_LOCK_BITNR, 81 TASK_UNINTERRUPTIBLE); 82 } 83 84 static inline struct bucket_array *gc_bucket_array(struct bch_dev *ca) 85 { 86 return rcu_dereference_check(ca->buckets_gc, 87 !ca->fs || 88 percpu_rwsem_is_held(&ca->fs->mark_lock) || 89 lockdep_is_held(&ca->fs->gc_lock) || 90 lockdep_is_held(&ca->bucket_lock)); 91 } 92 93 static inline struct bucket *gc_bucket(struct bch_dev *ca, size_t b) 94 { 95 struct bucket_array *buckets = gc_bucket_array(ca); 96 97 BUG_ON(b < buckets->first_bucket || b >= buckets->nbuckets); 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 BUG_ON(b < gens->first_bucket || b >= gens->nbuckets); 115 return gens->b + b; 116 } 117 118 static inline size_t PTR_BUCKET_NR(const struct bch_dev *ca, 119 const struct bch_extent_ptr *ptr) 120 { 121 return sector_to_bucket(ca, ptr->offset); 122 } 123 124 static inline struct bpos PTR_BUCKET_POS(const struct bch_fs *c, 125 const struct bch_extent_ptr *ptr) 126 { 127 struct bch_dev *ca = bch_dev_bkey_exists(c, ptr->dev); 128 129 return POS(ptr->dev, PTR_BUCKET_NR(ca, ptr)); 130 } 131 132 static inline struct bpos PTR_BUCKET_POS_OFFSET(const struct bch_fs *c, 133 const struct bch_extent_ptr *ptr, 134 u32 *bucket_offset) 135 { 136 struct bch_dev *ca = bch_dev_bkey_exists(c, ptr->dev); 137 138 return POS(ptr->dev, sector_to_bucket_and_offset(ca, ptr->offset, bucket_offset)); 139 } 140 141 static inline struct bucket *PTR_GC_BUCKET(struct bch_dev *ca, 142 const struct bch_extent_ptr *ptr) 143 { 144 return gc_bucket(ca, PTR_BUCKET_NR(ca, ptr)); 145 } 146 147 static inline enum bch_data_type ptr_data_type(const struct bkey *k, 148 const struct bch_extent_ptr *ptr) 149 { 150 if (bkey_is_btree_ptr(k)) 151 return BCH_DATA_btree; 152 153 return ptr->cached ? BCH_DATA_cached : BCH_DATA_user; 154 } 155 156 static inline s64 ptr_disk_sectors(s64 sectors, struct extent_ptr_decoded p) 157 { 158 EBUG_ON(sectors < 0); 159 160 return crc_is_compressed(p.crc) 161 ? DIV_ROUND_UP_ULL(sectors * p.crc.compressed_size, 162 p.crc.uncompressed_size) 163 : sectors; 164 } 165 166 static inline int gen_cmp(u8 a, u8 b) 167 { 168 return (s8) (a - b); 169 } 170 171 static inline int gen_after(u8 a, u8 b) 172 { 173 int r = gen_cmp(a, b); 174 175 return r > 0 ? r : 0; 176 } 177 178 /** 179 * ptr_stale() - check if a pointer points into a bucket that has been 180 * invalidated. 181 */ 182 static inline u8 ptr_stale(struct bch_dev *ca, 183 const struct bch_extent_ptr *ptr) 184 { 185 u8 ret; 186 187 rcu_read_lock(); 188 ret = gen_after(*bucket_gen(ca, PTR_BUCKET_NR(ca, ptr)), ptr->gen); 189 rcu_read_unlock(); 190 191 return ret; 192 } 193 194 /* Device usage: */ 195 196 void bch2_dev_usage_read_fast(struct bch_dev *, struct bch_dev_usage *); 197 static inline struct bch_dev_usage bch2_dev_usage_read(struct bch_dev *ca) 198 { 199 struct bch_dev_usage ret; 200 201 bch2_dev_usage_read_fast(ca, &ret); 202 return ret; 203 } 204 205 void bch2_dev_usage_init(struct bch_dev *); 206 void bch2_dev_usage_to_text(struct printbuf *, struct bch_dev_usage *); 207 208 static inline u64 bch2_dev_buckets_reserved(struct bch_dev *ca, enum bch_watermark watermark) 209 { 210 s64 reserved = 0; 211 212 switch (watermark) { 213 case BCH_WATERMARK_NR: 214 BUG(); 215 case BCH_WATERMARK_stripe: 216 reserved += ca->mi.nbuckets >> 6; 217 fallthrough; 218 case BCH_WATERMARK_normal: 219 reserved += ca->mi.nbuckets >> 6; 220 fallthrough; 221 case BCH_WATERMARK_copygc: 222 reserved += ca->nr_btree_reserve; 223 fallthrough; 224 case BCH_WATERMARK_btree: 225 reserved += ca->nr_btree_reserve; 226 fallthrough; 227 case BCH_WATERMARK_btree_copygc: 228 case BCH_WATERMARK_reclaim: 229 break; 230 } 231 232 return reserved; 233 } 234 235 static inline u64 dev_buckets_free(struct bch_dev *ca, 236 struct bch_dev_usage usage, 237 enum bch_watermark watermark) 238 { 239 return max_t(s64, 0, 240 usage.d[BCH_DATA_free].buckets - 241 ca->nr_open_buckets - 242 bch2_dev_buckets_reserved(ca, watermark)); 243 } 244 245 static inline u64 __dev_buckets_available(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 + usage.d[BCH_DATA_cached].buckets 252 + usage.d[BCH_DATA_need_gc_gens].buckets 253 + usage.d[BCH_DATA_need_discard].buckets 254 - ca->nr_open_buckets 255 - bch2_dev_buckets_reserved(ca, watermark)); 256 } 257 258 static inline u64 dev_buckets_available(struct bch_dev *ca, 259 enum bch_watermark watermark) 260 { 261 return __dev_buckets_available(ca, bch2_dev_usage_read(ca), watermark); 262 } 263 264 /* Filesystem usage: */ 265 266 static inline unsigned __fs_usage_u64s(unsigned nr_replicas) 267 { 268 return sizeof(struct bch_fs_usage) / sizeof(u64) + nr_replicas; 269 } 270 271 static inline unsigned fs_usage_u64s(struct bch_fs *c) 272 { 273 return __fs_usage_u64s(READ_ONCE(c->replicas.nr)); 274 } 275 276 static inline unsigned __fs_usage_online_u64s(unsigned nr_replicas) 277 { 278 return sizeof(struct bch_fs_usage_online) / sizeof(u64) + nr_replicas; 279 } 280 281 static inline unsigned fs_usage_online_u64s(struct bch_fs *c) 282 { 283 return __fs_usage_online_u64s(READ_ONCE(c->replicas.nr)); 284 } 285 286 static inline unsigned dev_usage_u64s(void) 287 { 288 return sizeof(struct bch_dev_usage) / sizeof(u64); 289 } 290 291 u64 bch2_fs_usage_read_one(struct bch_fs *, u64 *); 292 293 struct bch_fs_usage_online *bch2_fs_usage_read(struct bch_fs *); 294 295 void bch2_fs_usage_acc_to_base(struct bch_fs *, unsigned); 296 297 void bch2_fs_usage_to_text(struct printbuf *, 298 struct bch_fs *, struct bch_fs_usage_online *); 299 300 u64 bch2_fs_sectors_used(struct bch_fs *, struct bch_fs_usage_online *); 301 302 struct bch_fs_usage_short 303 bch2_fs_usage_read_short(struct bch_fs *); 304 305 void bch2_dev_usage_update(struct bch_fs *, struct bch_dev *, 306 const struct bch_alloc_v4 *, 307 const struct bch_alloc_v4 *, u64, bool); 308 void bch2_dev_usage_update_m(struct bch_fs *, struct bch_dev *, 309 struct bucket *, struct bucket *); 310 311 /* key/bucket marking: */ 312 313 static inline struct bch_fs_usage *fs_usage_ptr(struct bch_fs *c, 314 unsigned journal_seq, 315 bool gc) 316 { 317 percpu_rwsem_assert_held(&c->mark_lock); 318 BUG_ON(!gc && !journal_seq); 319 320 return this_cpu_ptr(gc 321 ? c->usage_gc 322 : c->usage[journal_seq & JOURNAL_BUF_MASK]); 323 } 324 325 int bch2_update_replicas(struct bch_fs *, struct bkey_s_c, 326 struct bch_replicas_entry_v1 *, s64, 327 unsigned, bool); 328 int bch2_update_replicas_list(struct btree_trans *, 329 struct bch_replicas_entry_v1 *, s64); 330 int bch2_update_cached_sectors_list(struct btree_trans *, unsigned, s64); 331 int bch2_replicas_deltas_realloc(struct btree_trans *, unsigned); 332 333 void bch2_fs_usage_initialize(struct bch_fs *); 334 335 int bch2_check_bucket_ref(struct btree_trans *, struct bkey_s_c, 336 const struct bch_extent_ptr *, 337 s64, enum bch_data_type, u8, u8, u32); 338 339 int bch2_mark_metadata_bucket(struct bch_fs *, struct bch_dev *, 340 size_t, enum bch_data_type, unsigned, 341 struct gc_pos, unsigned); 342 343 int bch2_trigger_extent(struct btree_trans *, enum btree_id, unsigned, 344 struct bkey_s_c, struct bkey_s, unsigned); 345 int bch2_trigger_reservation(struct btree_trans *, enum btree_id, unsigned, 346 struct bkey_s_c, struct bkey_s, unsigned); 347 348 #define trigger_run_overwrite_then_insert(_fn, _trans, _btree_id, _level, _old, _new, _flags)\ 349 ({ \ 350 int ret = 0; \ 351 \ 352 if (_old.k->type) \ 353 ret = _fn(_trans, _btree_id, _level, _old, _flags & ~BTREE_TRIGGER_INSERT); \ 354 if (!ret && _new.k->type) \ 355 ret = _fn(_trans, _btree_id, _level, _new.s_c, _flags & ~BTREE_TRIGGER_OVERWRITE);\ 356 ret; \ 357 }) 358 359 void bch2_trans_account_disk_usage_change(struct btree_trans *); 360 361 void bch2_trans_fs_usage_revert(struct btree_trans *, struct replicas_delta_list *); 362 int bch2_trans_fs_usage_apply(struct btree_trans *, struct replicas_delta_list *); 363 364 int bch2_trans_mark_metadata_bucket(struct btree_trans *, struct bch_dev *, 365 size_t, enum bch_data_type, unsigned); 366 int bch2_trans_mark_dev_sb(struct bch_fs *, struct bch_dev *); 367 int bch2_trans_mark_dev_sbs(struct bch_fs *); 368 369 static inline bool is_superblock_bucket(struct bch_dev *ca, u64 b) 370 { 371 struct bch_sb_layout *layout = &ca->disk_sb.sb->layout; 372 u64 b_offset = bucket_to_sector(ca, b); 373 u64 b_end = bucket_to_sector(ca, b + 1); 374 unsigned i; 375 376 if (!b) 377 return true; 378 379 for (i = 0; i < layout->nr_superblocks; i++) { 380 u64 offset = le64_to_cpu(layout->sb_offset[i]); 381 u64 end = offset + (1 << layout->sb_max_size_bits); 382 383 if (!(offset >= b_end || end <= b_offset)) 384 return true; 385 } 386 387 return false; 388 } 389 390 static inline const char *bch2_data_type_str(enum bch_data_type type) 391 { 392 return type < BCH_DATA_NR 393 ? __bch2_data_types[type] 394 : "(invalid data type)"; 395 } 396 397 static inline void bch2_prt_data_type(struct printbuf *out, enum bch_data_type type) 398 { 399 if (type < BCH_DATA_NR) 400 prt_str(out, __bch2_data_types[type]); 401 else 402 prt_printf(out, "(invalid data type %u)", type); 403 } 404 405 /* disk reservations: */ 406 407 static inline void bch2_disk_reservation_put(struct bch_fs *c, 408 struct disk_reservation *res) 409 { 410 if (res->sectors) { 411 this_cpu_sub(*c->online_reserved, res->sectors); 412 res->sectors = 0; 413 } 414 } 415 416 #define BCH_DISK_RESERVATION_NOFAIL (1 << 0) 417 418 int __bch2_disk_reservation_add(struct bch_fs *, 419 struct disk_reservation *, 420 u64, int); 421 422 static inline int bch2_disk_reservation_add(struct bch_fs *c, struct disk_reservation *res, 423 u64 sectors, int flags) 424 { 425 #ifdef __KERNEL__ 426 u64 old, new; 427 428 do { 429 old = this_cpu_read(c->pcpu->sectors_available); 430 if (sectors > old) 431 return __bch2_disk_reservation_add(c, res, sectors, flags); 432 433 new = old - sectors; 434 } while (this_cpu_cmpxchg(c->pcpu->sectors_available, old, new) != old); 435 436 this_cpu_add(*c->online_reserved, sectors); 437 res->sectors += sectors; 438 return 0; 439 #else 440 return __bch2_disk_reservation_add(c, res, sectors, flags); 441 #endif 442 } 443 444 static inline struct disk_reservation 445 bch2_disk_reservation_init(struct bch_fs *c, unsigned nr_replicas) 446 { 447 return (struct disk_reservation) { 448 .sectors = 0, 449 #if 0 450 /* not used yet: */ 451 .gen = c->capacity_gen, 452 #endif 453 .nr_replicas = nr_replicas, 454 }; 455 } 456 457 static inline int bch2_disk_reservation_get(struct bch_fs *c, 458 struct disk_reservation *res, 459 u64 sectors, unsigned nr_replicas, 460 int flags) 461 { 462 *res = bch2_disk_reservation_init(c, nr_replicas); 463 464 return bch2_disk_reservation_add(c, res, sectors * nr_replicas, flags); 465 } 466 467 #define RESERVE_FACTOR 6 468 469 static inline u64 avail_factor(u64 r) 470 { 471 return div_u64(r << RESERVE_FACTOR, (1 << RESERVE_FACTOR) + 1); 472 } 473 474 int bch2_dev_buckets_resize(struct bch_fs *, struct bch_dev *, u64); 475 void bch2_dev_buckets_free(struct bch_dev *); 476 int bch2_dev_buckets_alloc(struct bch_fs *, struct bch_dev *); 477 478 #endif /* _BUCKETS_H */ 479