fs/bcachefs/buckets.h

   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 bucket_valid(ca, b)
  86                 ? genradix_ptr(&ca->buckets_gc, b)
  87                 : NULL;
  88 }
  89
  90 static inline struct bucket_gens *bucket_gens(struct bch_dev *ca)
  91 {
  92         return rcu_dereference_check(ca->bucket_gens,
  93                                      lockdep_is_held(&ca->fs->state_lock));
  94 }
  95
  96 static inline u8 *bucket_gen(struct bch_dev *ca, size_t b)
  97 {
  98         struct bucket_gens *gens = bucket_gens(ca);
  99
 100         if (b - gens->first_bucket >= gens->nbuckets_minus_first)
 101                 return NULL;
 102         return gens->b + b;
 103 }
 104
 105 static inline int bucket_gen_get_rcu(struct bch_dev *ca, size_t b)
 106 {
 107         u8 *gen = bucket_gen(ca, b);
 108         return gen ? *gen : -1;
 109 }
 110
 111 static inline int bucket_gen_get(struct bch_dev *ca, size_t b)
 112 {
 113         rcu_read_lock();
 114         int ret = bucket_gen_get_rcu(ca, b);
 115         rcu_read_unlock();
 116         return ret;
 117 }
 118
 119 static inline size_t PTR_BUCKET_NR(const struct bch_dev *ca,
 120                                    const struct bch_extent_ptr *ptr)
 121 {
 122         return sector_to_bucket(ca, ptr->offset);
 123 }
 124
 125 static inline struct bpos PTR_BUCKET_POS(const struct bch_dev *ca,
 126                                          const struct bch_extent_ptr *ptr)
 127 {
 128         return POS(ptr->dev, PTR_BUCKET_NR(ca, ptr));
 129 }
 130
 131 static inline struct bpos PTR_BUCKET_POS_OFFSET(const struct bch_dev *ca,
 132                                                 const struct bch_extent_ptr *ptr,
 133                                                 u32 *bucket_offset)
 134 {
 135         return POS(ptr->dev, sector_to_bucket_and_offset(ca, ptr->offset, bucket_offset));
 136 }
 137
 138 static inline struct bucket *PTR_GC_BUCKET(struct bch_dev *ca,
 139                                            const struct bch_extent_ptr *ptr)
 140 {
 141         return gc_bucket(ca, PTR_BUCKET_NR(ca, ptr));
 142 }
 143
 144 static inline enum bch_data_type ptr_data_type(const struct bkey *k,
 145                                                const struct bch_extent_ptr *ptr)
 146 {
 147         if (bkey_is_btree_ptr(k))
 148                 return BCH_DATA_btree;
 149
 150         return ptr->cached ? BCH_DATA_cached : BCH_DATA_user;
 151 }
 152
 153 static inline s64 ptr_disk_sectors(s64 sectors, struct extent_ptr_decoded p)
 154 {
 155         EBUG_ON(sectors < 0);
 156
 157         return crc_is_compressed(p.crc)
 158                 ? DIV_ROUND_UP_ULL(sectors * p.crc.compressed_size,
 159                                    p.crc.uncompressed_size)
 160                 : sectors;
 161 }
 162
 163 static inline int gen_cmp(u8 a, u8 b)
 164 {
 165         return (s8) (a - b);
 166 }
 167
 168 static inline int gen_after(u8 a, u8 b)
 169 {
 170         int r = gen_cmp(a, b);
 171
 172         return r > 0 ? r : 0;
 173 }
 174
 175 static inline int dev_ptr_stale_rcu(struct bch_dev *ca, const struct bch_extent_ptr *ptr)
 176 {
 177         int gen = bucket_gen_get_rcu(ca, PTR_BUCKET_NR(ca, ptr));
 178         return gen < 0 ? gen : gen_after(gen, ptr->gen);
 179 }
 180
 181 /**
 182  * dev_ptr_stale() - check if a pointer points into a bucket that has been
 183  * invalidated.
 184  */
 185 static inline int dev_ptr_stale(struct bch_dev *ca, const struct bch_extent_ptr *ptr)
 186 {
 187         rcu_read_lock();
 188         int ret = dev_ptr_stale_rcu(ca, ptr);
 189         rcu_read_unlock();
 190         return ret;
 191 }
 192
 193 /* Device usage: */
 194
 195 void bch2_dev_usage_read_fast(struct bch_dev *, struct bch_dev_usage *);
 196 static inline struct bch_dev_usage bch2_dev_usage_read(struct bch_dev *ca)
 197 {
 198         struct bch_dev_usage ret;
 199
 200         bch2_dev_usage_read_fast(ca, &ret);
 201         return ret;
 202 }
 203
 204 void bch2_dev_usage_to_text(struct printbuf *, struct bch_dev *, struct bch_dev_usage *);
 205
 206 static inline u64 bch2_dev_buckets_reserved(struct bch_dev *ca, enum bch_watermark watermark)
 207 {
 208         s64 reserved = 0;
 209
 210         switch (watermark) {
 211         case BCH_WATERMARK_NR:
 212                 BUG();
 213         case BCH_WATERMARK_stripe:
 214                 reserved += ca->mi.nbuckets >> 6;
 215                 fallthrough;
 216         case BCH_WATERMARK_normal:
 217                 reserved += ca->mi.nbuckets >> 6;
 218                 fallthrough;
 219         case BCH_WATERMARK_copygc:
 220                 reserved += ca->nr_btree_reserve;
 221                 fallthrough;
 222         case BCH_WATERMARK_btree:
 223                 reserved += ca->nr_btree_reserve;
 224                 fallthrough;
 225         case BCH_WATERMARK_btree_copygc:
 226         case BCH_WATERMARK_reclaim:
 227         case BCH_WATERMARK_interior_updates:
 228                 break;
 229         }
 230
 231         return reserved;
 232 }
 233
 234 static inline u64 dev_buckets_free(struct bch_dev *ca,
 235                                    struct bch_dev_usage usage,
 236                                    enum bch_watermark watermark)
 237 {
 238         return max_t(s64, 0,
 239                      usage.d[BCH_DATA_free].buckets -
 240                      ca->nr_open_buckets -
 241                      bch2_dev_buckets_reserved(ca, watermark));
 242 }
 243
 244 static inline u64 __dev_buckets_available(struct bch_dev *ca,
 245                                           struct bch_dev_usage usage,
 246                                           enum bch_watermark watermark)
 247 {
 248         return max_t(s64, 0,
 249                        usage.d[BCH_DATA_free].buckets
 250                      + usage.d[BCH_DATA_cached].buckets
 251                      + usage.d[BCH_DATA_need_gc_gens].buckets
 252                      + usage.d[BCH_DATA_need_discard].buckets
 253                      - ca->nr_open_buckets
 254                      - bch2_dev_buckets_reserved(ca, watermark));
 255 }
 256
 257 static inline u64 dev_buckets_available(struct bch_dev *ca,
 258                                         enum bch_watermark watermark)
 259 {
 260         return __dev_buckets_available(ca, bch2_dev_usage_read(ca), watermark);
 261 }
 262
 263 /* Filesystem usage: */
 264
 265 static inline unsigned dev_usage_u64s(void)
 266 {
 267         return sizeof(struct bch_dev_usage) / sizeof(u64);
 268 }
 269
 270 struct bch_fs_usage_short
 271 bch2_fs_usage_read_short(struct bch_fs *);
 272
 273 int bch2_bucket_ref_update(struct btree_trans *, struct bch_dev *,
 274                            struct bkey_s_c, const struct bch_extent_ptr *,
 275                            s64, enum bch_data_type, u8, u8, u32 *);
 276
 277 int bch2_check_fix_ptrs(struct btree_trans *,
 278                         enum btree_id, unsigned, struct bkey_s_c,
 279                         enum btree_iter_update_trigger_flags);
 280
 281 int bch2_trigger_extent(struct btree_trans *, enum btree_id, unsigned,
 282                         struct bkey_s_c, struct bkey_s,
 283                         enum btree_iter_update_trigger_flags);
 284 int bch2_trigger_reservation(struct btree_trans *, enum btree_id, unsigned,
 285                           struct bkey_s_c, struct bkey_s,
 286                           enum btree_iter_update_trigger_flags);
 287
 288 #define trigger_run_overwrite_then_insert(_fn, _trans, _btree_id, _level, _old, _new, _flags)\
 289 ({                                                                                              \
 290         int ret = 0;                                                                            \
 291                                                                                                 \
 292         if (_old.k->type)                                                                       \
 293                 ret = _fn(_trans, _btree_id, _level, _old, _flags & ~BTREE_TRIGGER_insert);     \
 294         if (!ret && _new.k->type)                                                               \
 295                 ret = _fn(_trans, _btree_id, _level, _new.s_c, _flags & ~BTREE_TRIGGER_overwrite);\
 296         ret;                                                                                    \
 297 })
 298
 299 void bch2_trans_account_disk_usage_change(struct btree_trans *);
 300
 301 int bch2_trans_mark_metadata_bucket(struct btree_trans *, struct bch_dev *, u64,
 302                                     enum bch_data_type, unsigned,
 303                                     enum btree_iter_update_trigger_flags);
 304 int bch2_trans_mark_dev_sb(struct bch_fs *, struct bch_dev *,
 305                                     enum btree_iter_update_trigger_flags);
 306 int bch2_trans_mark_dev_sbs_flags(struct bch_fs *,
 307                                     enum btree_iter_update_trigger_flags);
 308 int bch2_trans_mark_dev_sbs(struct bch_fs *);
 309
 310 bool bch2_is_superblock_bucket(struct bch_dev *, u64);
 311
 312 static inline const char *bch2_data_type_str(enum bch_data_type type)
 313 {
 314         return type < BCH_DATA_NR
 315                 ? __bch2_data_types[type]
 316                 : "(invalid data type)";
 317 }
 318
 319 /* disk reservations: */
 320
 321 static inline void bch2_disk_reservation_put(struct bch_fs *c,
 322                                              struct disk_reservation *res)
 323 {
 324         if (res->sectors) {
 325                 this_cpu_sub(*c->online_reserved, res->sectors);
 326                 res->sectors = 0;
 327         }
 328 }
 329
 330 enum bch_reservation_flags {
 331         BCH_DISK_RESERVATION_NOFAIL     = 1 << 0,
 332         BCH_DISK_RESERVATION_PARTIAL    = 1 << 1,
 333 };
 334
 335 int __bch2_disk_reservation_add(struct bch_fs *, struct disk_reservation *,
 336                                 u64, enum bch_reservation_flags);
 337
 338 static inline int bch2_disk_reservation_add(struct bch_fs *c, struct disk_reservation *res,
 339                                             u64 sectors, enum bch_reservation_flags flags)
 340 {
 341 #ifdef __KERNEL__
 342         u64 old, new;
 343
 344         old = this_cpu_read(c->pcpu->sectors_available);
 345         do {
 346                 if (sectors > old)
 347                         return __bch2_disk_reservation_add(c, res, sectors, flags);
 348
 349                 new = old - sectors;
 350         } while (!this_cpu_try_cmpxchg(c->pcpu->sectors_available, &old, new));
 351
 352         this_cpu_add(*c->online_reserved, sectors);
 353         res->sectors                    += sectors;
 354         return 0;
 355 #else
 356         return __bch2_disk_reservation_add(c, res, sectors, flags);
 357 #endif
 358 }
 359
 360 static inline struct disk_reservation
 361 bch2_disk_reservation_init(struct bch_fs *c, unsigned nr_replicas)
 362 {
 363         return (struct disk_reservation) {
 364                 .sectors        = 0,
 365 #if 0
 366                 /* not used yet: */
 367                 .gen            = c->capacity_gen,
 368 #endif
 369                 .nr_replicas    = nr_replicas,
 370         };
 371 }
 372
 373 static inline int bch2_disk_reservation_get(struct bch_fs *c,
 374                                             struct disk_reservation *res,
 375                                             u64 sectors, unsigned nr_replicas,
 376                                             int flags)
 377 {
 378         *res = bch2_disk_reservation_init(c, nr_replicas);
 379
 380         return bch2_disk_reservation_add(c, res, sectors * nr_replicas, flags);
 381 }
 382
 383 #define RESERVE_FACTOR  6
 384
 385 static inline u64 avail_factor(u64 r)
 386 {
 387         return div_u64(r << RESERVE_FACTOR, (1 << RESERVE_FACTOR) + 1);
 388 }
 389
 390 void bch2_buckets_nouse_free(struct bch_fs *);
 391 int bch2_buckets_nouse_alloc(struct bch_fs *);
 392
 393 int bch2_dev_buckets_resize(struct bch_fs *, struct bch_dev *, u64);
 394 void bch2_dev_buckets_free(struct bch_dev *);
 395 int bch2_dev_buckets_alloc(struct bch_fs *, struct bch_dev *);
 396
 397 #endif /* _BUCKETS_H */