git.sesse.net Git - bcachefs-tools-debian/blob - libbcachefs/btree_io.h

   1 /* SPDX-License-Identifier: GPL-2.0 */
   2 #ifndef _BCACHEFS_BTREE_IO_H
   3 #define _BCACHEFS_BTREE_IO_H
   4
   5 #include "bkey_methods.h"
   6 #include "bset.h"
   7 #include "btree_locking.h"
   8 #include "checksum.h"
   9 #include "extents.h"
  10 #include "io_types.h"
  11
  12 struct bch_fs;
  13 struct btree_write;
  14 struct btree;
  15 struct btree_iter;
  16 struct btree_node_read_all;
  17
  18 static inline bool btree_node_dirty(struct btree *b)
  19 {
  20         return test_bit(BTREE_NODE_dirty, &b->flags);
  21 }
  22
  23 static inline void set_btree_node_dirty(struct bch_fs *c, struct btree *b)
  24 {
  25         if (!test_and_set_bit(BTREE_NODE_dirty, &b->flags))
  26                 atomic_inc(&c->btree_cache.dirty);
  27 }
  28
  29 static inline void clear_btree_node_dirty(struct bch_fs *c, struct btree *b)
  30 {
  31         if (test_and_clear_bit(BTREE_NODE_dirty, &b->flags))
  32                 atomic_dec(&c->btree_cache.dirty);
  33 }
  34
  35 struct btree_read_bio {
  36         struct bch_fs           *c;
  37         struct btree            *b;
  38         struct btree_node_read_all *ra;
  39         u64                     start_time;
  40         unsigned                have_ioref:1;
  41         unsigned                idx:7;
  42         struct extent_ptr_decoded       pick;
  43         struct work_struct      work;
  44         struct bio              bio;
  45 };
  46
  47 struct btree_write_bio {
  48         struct work_struct      work;
  49         __BKEY_PADDED(key, BKEY_BTREE_PTR_VAL_U64s_MAX);
  50         void                    *data;
  51         unsigned                bytes;
  52         struct bch_write_bio    wbio;
  53 };
  54
  55 static inline void btree_node_io_unlock(struct btree *b)
  56 {
  57         EBUG_ON(!btree_node_write_in_flight(b));
  58         clear_btree_node_write_in_flight(b);
  59         wake_up_bit(&b->flags, BTREE_NODE_write_in_flight);
  60 }
  61
  62 static inline void btree_node_io_lock(struct btree *b)
  63 {
  64         wait_on_bit_lock_io(&b->flags, BTREE_NODE_write_in_flight,
  65                             TASK_UNINTERRUPTIBLE);
  66 }
  67
  68 static inline void btree_node_wait_on_io(struct btree *b)
  69 {
  70         wait_on_bit_io(&b->flags, BTREE_NODE_write_in_flight,
  71                        TASK_UNINTERRUPTIBLE);
  72 }
  73
  74 static inline bool btree_node_may_write(struct btree *b)
  75 {
  76         return list_empty_careful(&b->write_blocked) &&
  77                 (!b->written || !b->will_make_reachable);
  78 }
  79
  80 enum compact_mode {
  81         COMPACT_LAZY,
  82         COMPACT_ALL,
  83 };
  84
  85 bool bch2_compact_whiteouts(struct bch_fs *, struct btree *,
  86                             enum compact_mode);
  87
  88 static inline bool should_compact_bset_lazy(struct btree *b,
  89                                             struct bset_tree *t)
  90 {
  91         unsigned total_u64s = bset_u64s(t);
  92         unsigned dead_u64s = bset_dead_u64s(b, t);
  93
  94         return dead_u64s > 64 && dead_u64s * 3 > total_u64s;
  95 }
  96
  97 static inline bool bch2_maybe_compact_whiteouts(struct bch_fs *c, struct btree *b)
  98 {
  99         struct bset_tree *t;
 100
 101         for_each_bset(b, t)
 102                 if (should_compact_bset_lazy(b, t))
 103                         return bch2_compact_whiteouts(c, b, COMPACT_LAZY);
 104
 105         return false;
 106 }
 107
 108 static inline struct nonce btree_nonce(struct bset *i, unsigned offset)
 109 {
 110         return (struct nonce) {{
 111                 [0] = cpu_to_le32(offset),
 112                 [1] = ((__le32 *) &i->seq)[0],
 113                 [2] = ((__le32 *) &i->seq)[1],
 114                 [3] = ((__le32 *) &i->journal_seq)[0]^BCH_NONCE_BTREE,
 115         }};
 116 }
 117
 118 static inline void bset_encrypt(struct bch_fs *c, struct bset *i, unsigned offset)
 119 {
 120         struct nonce nonce = btree_nonce(i, offset);
 121
 122         if (!offset) {
 123                 struct btree_node *bn = container_of(i, struct btree_node, keys);
 124                 unsigned bytes = (void *) &bn->keys - (void *) &bn->flags;
 125
 126                 bch2_encrypt(c, BSET_CSUM_TYPE(i), nonce, &bn->flags,
 127                              bytes);
 128
 129                 nonce = nonce_add(nonce, round_up(bytes, CHACHA_BLOCK_SIZE));
 130         }
 131
 132         bch2_encrypt(c, BSET_CSUM_TYPE(i), nonce, i->_data,
 133                      vstruct_end(i) - (void *) i->_data);
 134 }
 135
 136 void bch2_btree_sort_into(struct bch_fs *, struct btree *, struct btree *);
 137
 138 void bch2_btree_node_drop_keys_outside_node(struct btree *);
 139
 140 void bch2_btree_build_aux_trees(struct btree *);
 141 void bch2_btree_init_next(struct bch_fs *, struct btree *,
 142                          struct btree_iter *);
 143
 144 int bch2_btree_node_read_done(struct bch_fs *, struct bch_dev *,
 145                               struct btree *, bool);
 146 void bch2_btree_node_read(struct bch_fs *, struct btree *, bool);
 147 int bch2_btree_root_read(struct bch_fs *, enum btree_id,
 148                          const struct bkey_i *, unsigned);
 149
 150 void bch2_btree_complete_write(struct bch_fs *, struct btree *,
 151                               struct btree_write *);
 152 void bch2_btree_write_error_work(struct work_struct *);
 153
 154 void __bch2_btree_node_write(struct bch_fs *, struct btree *);
 155 bool bch2_btree_post_write_cleanup(struct bch_fs *, struct btree *);
 156
 157 void bch2_btree_node_write(struct bch_fs *, struct btree *,
 158                           enum six_lock_type);
 159
 160 static inline void btree_node_write_if_need(struct bch_fs *c, struct btree *b,
 161                                             enum six_lock_type lock_held)
 162 {
 163         while (b->written &&
 164                btree_node_need_write(b) &&
 165                btree_node_may_write(b)) {
 166                 if (!btree_node_write_in_flight(b)) {
 167                         bch2_btree_node_write(c, b, lock_held);
 168                         break;
 169                 }
 170
 171                 six_unlock_type(&b->c.lock, lock_held);
 172                 btree_node_wait_on_io(b);
 173                 btree_node_lock_type(c, b, lock_held);
 174         }
 175 }
 176
 177 #define bch2_btree_node_write_cond(_c, _b, cond)                        \
 178 do {                                                                    \
 179         unsigned long old, new, v = READ_ONCE((_b)->flags);             \
 180                                                                         \
 181         do {                                                            \
 182                 old = new = v;                                          \
 183                                                                         \
 184                 if (!(old & (1 << BTREE_NODE_dirty)) || !(cond))        \
 185                         break;                                          \
 186                                                                         \
 187                 new |= (1 << BTREE_NODE_need_write);                    \
 188         } while ((v = cmpxchg(&(_b)->flags, old, new)) != old);         \
 189                                                                         \
 190         btree_node_write_if_need(_c, _b, SIX_LOCK_read);                \
 191 } while (0)
 192
 193 void bch2_btree_flush_all_reads(struct bch_fs *);
 194 void bch2_btree_flush_all_writes(struct bch_fs *);
 195 void bch2_dirty_btree_nodes_to_text(struct printbuf *, struct bch_fs *);
 196
 197 static inline void compat_bformat(unsigned level, enum btree_id btree_id,
 198                                   unsigned version, unsigned big_endian,
 199                                   int write, struct bkey_format *f)
 200 {
 201         if (version < bcachefs_metadata_version_inode_btree_change &&
 202             btree_id == BTREE_ID_inodes) {
 203                 swap(f->bits_per_field[BKEY_FIELD_INODE],
 204                      f->bits_per_field[BKEY_FIELD_OFFSET]);
 205                 swap(f->field_offset[BKEY_FIELD_INODE],
 206                      f->field_offset[BKEY_FIELD_OFFSET]);
 207         }
 208
 209         if (version < bcachefs_metadata_version_snapshot &&
 210             (level || btree_type_has_snapshots(btree_id))) {
 211                 u64 max_packed =
 212                         ~(~0ULL << f->bits_per_field[BKEY_FIELD_SNAPSHOT]);
 213
 214                 f->field_offset[BKEY_FIELD_SNAPSHOT] = write
 215                         ? 0
 216                         : U32_MAX - max_packed;
 217         }
 218 }
 219
 220 static inline void compat_bpos(unsigned level, enum btree_id btree_id,
 221                                unsigned version, unsigned big_endian,
 222                                int write, struct bpos *p)
 223 {
 224         if (big_endian != CPU_BIG_ENDIAN)
 225                 bch2_bpos_swab(p);
 226
 227         if (version < bcachefs_metadata_version_inode_btree_change &&
 228             btree_id == BTREE_ID_inodes)
 229                 swap(p->inode, p->offset);
 230 }
 231
 232 static inline void compat_btree_node(unsigned level, enum btree_id btree_id,
 233                                      unsigned version, unsigned big_endian,
 234                                      int write,
 235                                      struct btree_node *bn)
 236 {
 237         if (version < bcachefs_metadata_version_inode_btree_change &&
 238             btree_node_type_is_extents(btree_id) &&
 239             bpos_cmp(bn->min_key, POS_MIN) &&
 240             write)
 241                 bn->min_key = bpos_nosnap_predecessor(bn->min_key);
 242
 243         if (version < bcachefs_metadata_version_snapshot &&
 244             write)
 245                 bn->max_key.snapshot = 0;
 246
 247         compat_bpos(level, btree_id, version, big_endian, write, &bn->min_key);
 248         compat_bpos(level, btree_id, version, big_endian, write, &bn->max_key);
 249
 250         if (version < bcachefs_metadata_version_snapshot &&
 251             !write)
 252                 bn->max_key.snapshot = U32_MAX;
 253
 254         if (version < bcachefs_metadata_version_inode_btree_change &&
 255             btree_node_type_is_extents(btree_id) &&
 256             bpos_cmp(bn->min_key, POS_MIN) &&
 257             !write)
 258                 bn->min_key = bpos_nosnap_successor(bn->min_key);
 259 }
 260
 261 #endif /* _BCACHEFS_BTREE_IO_H */