git.sesse.net Git - bcachefs-tools-debian/blob - libbcachefs/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 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         bit_spin_unlock(BUCKET_LOCK_BITNR, (void *) &b->lock);
  74 }
  75
  76 static inline void bucket_lock(struct bucket *b)
  77 {
  78         bit_spin_lock(BUCKET_LOCK_BITNR, (void *) &b->lock);
  79 }
  80
  81 static inline struct bucket_array *gc_bucket_array(struct bch_dev *ca)
  82 {
  83         return rcu_dereference_check(ca->buckets_gc,
  84                                      !ca->fs ||
  85                                      percpu_rwsem_is_held(&ca->fs->mark_lock) ||
  86                                      lockdep_is_held(&ca->fs->gc_lock) ||
  87                                      lockdep_is_held(&ca->bucket_lock));
  88 }
  89
  90 static inline struct bucket *gc_bucket(struct bch_dev *ca, size_t b)
  91 {
  92         struct bucket_array *buckets = gc_bucket_array(ca);
  93
  94         BUG_ON(b < buckets->first_bucket || b >= buckets->nbuckets);
  95         return buckets->b + b;
  96 }
  97
  98 static inline struct bucket_gens *bucket_gens(struct bch_dev *ca)
  99 {
 100         return rcu_dereference_check(ca->bucket_gens,
 101                                      !ca->fs ||
 102                                      percpu_rwsem_is_held(&ca->fs->mark_lock) ||
 103                                      lockdep_is_held(&ca->fs->gc_lock) ||
 104                                      lockdep_is_held(&ca->bucket_lock));
 105 }
 106
 107 static inline u8 *bucket_gen(struct bch_dev *ca, size_t b)
 108 {
 109         struct bucket_gens *gens = bucket_gens(ca);
 110
 111         BUG_ON(b < gens->first_bucket || b >= gens->nbuckets);
 112         return gens->b + b;
 113 }
 114
 115 static inline size_t PTR_BUCKET_NR(const struct bch_dev *ca,
 116                                    const struct bch_extent_ptr *ptr)
 117 {
 118         return sector_to_bucket(ca, ptr->offset);
 119 }
 120
 121 static inline struct bpos PTR_BUCKET_POS(const struct bch_fs *c,
 122                                    const struct bch_extent_ptr *ptr)
 123 {
 124         struct bch_dev *ca = bch_dev_bkey_exists(c, ptr->dev);
 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_fs *c,
 130                                                 const struct bch_extent_ptr *ptr,
 131                                                 u32 *bucket_offset)
 132 {
 133         struct bch_dev *ca = bch_dev_bkey_exists(c, ptr->dev);
 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 /**
 176  * ptr_stale() - check if a pointer points into a bucket that has been
 177  * invalidated.
 178  */
 179 static inline u8 ptr_stale(struct bch_dev *ca,
 180                            const struct bch_extent_ptr *ptr)
 181 {
 182         u8 ret;
 183
 184         rcu_read_lock();
 185         ret = gen_after(*bucket_gen(ca, PTR_BUCKET_NR(ca, ptr)), ptr->gen);
 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
 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                 break;
 226         }
 227
 228         return reserved;
 229 }
 230
 231 static inline u64 dev_buckets_free(struct bch_dev *ca,
 232                                    struct bch_dev_usage usage,
 233                                    enum bch_watermark watermark)
 234 {
 235         return max_t(s64, 0,
 236                      usage.d[BCH_DATA_free].buckets -
 237                      ca->nr_open_buckets -
 238                      bch2_dev_buckets_reserved(ca, watermark));
 239 }
 240
 241 static inline u64 __dev_buckets_available(struct bch_dev *ca,
 242                                           struct bch_dev_usage usage,
 243                                           enum bch_watermark watermark)
 244 {
 245         return max_t(s64, 0,
 246                        usage.d[BCH_DATA_free].buckets
 247                      + usage.d[BCH_DATA_cached].buckets
 248                      + usage.d[BCH_DATA_need_gc_gens].buckets
 249                      + usage.d[BCH_DATA_need_discard].buckets
 250                      - ca->nr_open_buckets
 251                      - bch2_dev_buckets_reserved(ca, watermark));
 252 }
 253
 254 static inline u64 dev_buckets_available(struct bch_dev *ca,
 255                                         enum bch_watermark watermark)
 256 {
 257         return __dev_buckets_available(ca, bch2_dev_usage_read(ca), watermark);
 258 }
 259
 260 /* Filesystem usage: */
 261
 262 static inline unsigned __fs_usage_u64s(unsigned nr_replicas)
 263 {
 264         return sizeof(struct bch_fs_usage) / sizeof(u64) + nr_replicas;
 265 }
 266
 267 static inline unsigned fs_usage_u64s(struct bch_fs *c)
 268 {
 269         return __fs_usage_u64s(READ_ONCE(c->replicas.nr));
 270 }
 271
 272 static inline unsigned __fs_usage_online_u64s(unsigned nr_replicas)
 273 {
 274         return sizeof(struct bch_fs_usage_online) / sizeof(u64) + nr_replicas;
 275 }
 276
 277 static inline unsigned fs_usage_online_u64s(struct bch_fs *c)
 278 {
 279         return __fs_usage_online_u64s(READ_ONCE(c->replicas.nr));
 280 }
 281
 282 static inline unsigned dev_usage_u64s(void)
 283 {
 284         return sizeof(struct bch_dev_usage) / sizeof(u64);
 285 }
 286
 287 u64 bch2_fs_usage_read_one(struct bch_fs *, u64 *);
 288
 289 struct bch_fs_usage_online *bch2_fs_usage_read(struct bch_fs *);
 290
 291 void bch2_fs_usage_acc_to_base(struct bch_fs *, unsigned);
 292
 293 void bch2_fs_usage_to_text(struct printbuf *,
 294                            struct bch_fs *, struct bch_fs_usage_online *);
 295
 296 u64 bch2_fs_sectors_used(struct bch_fs *, struct bch_fs_usage_online *);
 297
 298 struct bch_fs_usage_short
 299 bch2_fs_usage_read_short(struct bch_fs *);
 300
 301 /* key/bucket marking: */
 302
 303 static inline struct bch_fs_usage *fs_usage_ptr(struct bch_fs *c,
 304                                                 unsigned journal_seq,
 305                                                 bool gc)
 306 {
 307         percpu_rwsem_assert_held(&c->mark_lock);
 308         BUG_ON(!gc && !journal_seq);
 309
 310         return this_cpu_ptr(gc
 311                             ? c->usage_gc
 312                             : c->usage[journal_seq & JOURNAL_BUF_MASK]);
 313 }
 314
 315 int bch2_replicas_deltas_realloc(struct btree_trans *, unsigned);
 316
 317 void bch2_fs_usage_initialize(struct bch_fs *);
 318
 319 int bch2_mark_metadata_bucket(struct bch_fs *, struct bch_dev *,
 320                               size_t, enum bch_data_type, unsigned,
 321                               struct gc_pos, unsigned);
 322
 323 int bch2_mark_alloc(struct btree_trans *, enum btree_id, unsigned,
 324                     struct bkey_s_c, struct bkey_s_c, unsigned);
 325 int bch2_mark_extent(struct btree_trans *, enum btree_id, unsigned,
 326                      struct bkey_s_c, struct bkey_s_c, unsigned);
 327 int bch2_mark_stripe(struct btree_trans *, enum btree_id, unsigned,
 328                      struct bkey_s_c, struct bkey_s_c, unsigned);
 329 int bch2_mark_reservation(struct btree_trans *, enum btree_id, unsigned,
 330                           struct bkey_s_c, struct bkey_s_c, unsigned);
 331 int bch2_mark_reflink_p(struct btree_trans *, enum btree_id, unsigned,
 332                         struct bkey_s_c, struct bkey_s_c, unsigned);
 333
 334 int bch2_trans_mark_extent(struct btree_trans *, enum btree_id, unsigned, struct bkey_s_c, struct bkey_i *, unsigned);
 335 int bch2_trans_mark_stripe(struct btree_trans *, enum btree_id, unsigned, struct bkey_s_c, struct bkey_i *, unsigned);
 336 int bch2_trans_mark_reservation(struct btree_trans *, enum btree_id, unsigned, struct bkey_s_c, struct bkey_i *, unsigned);
 337 int bch2_trans_mark_reflink_p(struct btree_trans *, enum btree_id, unsigned, struct bkey_s_c, struct bkey_i *, unsigned);
 338
 339 void bch2_trans_fs_usage_revert(struct btree_trans *, struct replicas_delta_list *);
 340 int bch2_trans_fs_usage_apply(struct btree_trans *, struct replicas_delta_list *);
 341
 342 int bch2_trans_mark_metadata_bucket(struct btree_trans *, struct bch_dev *,
 343                                     size_t, enum bch_data_type, unsigned);
 344 int bch2_trans_mark_dev_sb(struct bch_fs *, struct bch_dev *);
 345
 346 static inline bool is_superblock_bucket(struct bch_dev *ca, u64 b)
 347 {
 348         struct bch_sb_layout *layout = &ca->disk_sb.sb->layout;
 349         u64 b_offset    = bucket_to_sector(ca, b);
 350         u64 b_end       = bucket_to_sector(ca, b + 1);
 351         unsigned i;
 352
 353         if (!b)
 354                 return true;
 355
 356         for (i = 0; i < layout->nr_superblocks; i++) {
 357                 u64 offset = le64_to_cpu(layout->sb_offset[i]);
 358                 u64 end = offset + (1 << layout->sb_max_size_bits);
 359
 360                 if (!(offset >= b_end || end <= b_offset))
 361                         return true;
 362         }
 363
 364         return false;
 365 }
 366
 367 /* disk reservations: */
 368
 369 static inline void bch2_disk_reservation_put(struct bch_fs *c,
 370                                              struct disk_reservation *res)
 371 {
 372         if (res->sectors) {
 373                 this_cpu_sub(*c->online_reserved, res->sectors);
 374                 res->sectors = 0;
 375         }
 376 }
 377
 378 #define BCH_DISK_RESERVATION_NOFAIL             (1 << 0)
 379
 380 int __bch2_disk_reservation_add(struct bch_fs *,
 381                                 struct disk_reservation *,
 382                                 u64, int);
 383
 384 static inline int bch2_disk_reservation_add(struct bch_fs *c, struct disk_reservation *res,
 385                                             u64 sectors, int flags)
 386 {
 387 #ifdef __KERNEL__
 388         u64 old, new;
 389
 390         do {
 391                 old = this_cpu_read(c->pcpu->sectors_available);
 392                 if (sectors > old)
 393                         return __bch2_disk_reservation_add(c, res, sectors, flags);
 394
 395                 new = old - sectors;
 396         } while (this_cpu_cmpxchg(c->pcpu->sectors_available, old, new) != old);
 397
 398         this_cpu_add(*c->online_reserved, sectors);
 399         res->sectors                    += sectors;
 400         return 0;
 401 #else
 402         return __bch2_disk_reservation_add(c, res, sectors, flags);
 403 #endif
 404 }
 405
 406 static inline struct disk_reservation
 407 bch2_disk_reservation_init(struct bch_fs *c, unsigned nr_replicas)
 408 {
 409         return (struct disk_reservation) {
 410                 .sectors        = 0,
 411 #if 0
 412                 /* not used yet: */
 413                 .gen            = c->capacity_gen,
 414 #endif
 415                 .nr_replicas    = nr_replicas,
 416         };
 417 }
 418
 419 static inline int bch2_disk_reservation_get(struct bch_fs *c,
 420                                             struct disk_reservation *res,
 421                                             u64 sectors, unsigned nr_replicas,
 422                                             int flags)
 423 {
 424         *res = bch2_disk_reservation_init(c, nr_replicas);
 425
 426         return bch2_disk_reservation_add(c, res, sectors * nr_replicas, flags);
 427 }
 428
 429 #define RESERVE_FACTOR  6
 430
 431 static inline u64 avail_factor(u64 r)
 432 {
 433         return div_u64(r << RESERVE_FACTOR, (1 << RESERVE_FACTOR) + 1);
 434 }
 435
 436 int bch2_dev_buckets_resize(struct bch_fs *, struct bch_dev *, u64);
 437 void bch2_dev_buckets_free(struct bch_dev *);
 438 int bch2_dev_buckets_alloc(struct bch_fs *, struct bch_dev *);
 439
 440 #endif /* _BUCKETS_H */