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 void bch2_btree_node_io_unlock(struct btree *);
  56 void bch2_btree_node_io_lock(struct btree *);
  57 void __bch2_btree_node_wait_on_read(struct btree *);
  58 void __bch2_btree_node_wait_on_write(struct btree *);
  59 void bch2_btree_node_wait_on_read(struct btree *);
  60 void bch2_btree_node_wait_on_write(struct btree *);
  61
  62 static inline bool btree_node_may_write(struct btree *b)
  63 {
  64         return list_empty_careful(&b->write_blocked) &&
  65                 (!b->written || !b->will_make_reachable);
  66 }
  67
  68 enum compact_mode {
  69         COMPACT_LAZY,
  70         COMPACT_ALL,
  71 };
  72
  73 bool bch2_compact_whiteouts(struct bch_fs *, struct btree *,
  74                             enum compact_mode);
  75
  76 static inline bool should_compact_bset_lazy(struct btree *b,
  77                                             struct bset_tree *t)
  78 {
  79         unsigned total_u64s = bset_u64s(t);
  80         unsigned dead_u64s = bset_dead_u64s(b, t);
  81
  82         return dead_u64s > 64 && dead_u64s * 3 > total_u64s;
  83 }
  84
  85 static inline bool bch2_maybe_compact_whiteouts(struct bch_fs *c, struct btree *b)
  86 {
  87         struct bset_tree *t;
  88
  89         for_each_bset(b, t)
  90                 if (should_compact_bset_lazy(b, t))
  91                         return bch2_compact_whiteouts(c, b, COMPACT_LAZY);
  92
  93         return false;
  94 }
  95
  96 static inline struct nonce btree_nonce(struct bset *i, unsigned offset)
  97 {
  98         return (struct nonce) {{
  99                 [0] = cpu_to_le32(offset),
 100                 [1] = ((__le32 *) &i->seq)[0],
 101                 [2] = ((__le32 *) &i->seq)[1],
 102                 [3] = ((__le32 *) &i->journal_seq)[0]^BCH_NONCE_BTREE,
 103         }};
 104 }
 105
 106 static inline void bset_encrypt(struct bch_fs *c, struct bset *i, unsigned offset)
 107 {
 108         struct nonce nonce = btree_nonce(i, offset);
 109
 110         if (!offset) {
 111                 struct btree_node *bn = container_of(i, struct btree_node, keys);
 112                 unsigned bytes = (void *) &bn->keys - (void *) &bn->flags;
 113
 114                 bch2_encrypt(c, BSET_CSUM_TYPE(i), nonce, &bn->flags,
 115                              bytes);
 116
 117                 nonce = nonce_add(nonce, round_up(bytes, CHACHA_BLOCK_SIZE));
 118         }
 119
 120         bch2_encrypt(c, BSET_CSUM_TYPE(i), nonce, i->_data,
 121                      vstruct_end(i) - (void *) i->_data);
 122 }
 123
 124 void bch2_btree_sort_into(struct bch_fs *, struct btree *, struct btree *);
 125
 126 void bch2_btree_node_drop_keys_outside_node(struct btree *);
 127
 128 void bch2_btree_build_aux_trees(struct btree *);
 129 void bch2_btree_init_next(struct btree_trans *, struct btree_iter *,
 130                           struct btree *);
 131
 132 int bch2_btree_node_read_done(struct bch_fs *, struct bch_dev *,
 133                               struct btree *, bool);
 134 void bch2_btree_node_read(struct bch_fs *, struct btree *, bool);
 135 int bch2_btree_root_read(struct bch_fs *, enum btree_id,
 136                          const struct bkey_i *, unsigned);
 137
 138 void bch2_btree_complete_write(struct bch_fs *, struct btree *,
 139                               struct btree_write *);
 140 void bch2_btree_write_error_work(struct work_struct *);
 141
 142 void __bch2_btree_node_write(struct bch_fs *, struct btree *, bool);
 143 bool bch2_btree_post_write_cleanup(struct bch_fs *, struct btree *);
 144
 145 void bch2_btree_node_write(struct bch_fs *, struct btree *,
 146                           enum six_lock_type);
 147
 148 static inline void btree_node_write_if_need(struct bch_fs *c, struct btree *b,
 149                                             enum six_lock_type lock_held)
 150 {
 151         if (b->written &&
 152             btree_node_need_write(b) &&
 153             btree_node_may_write(b) &&
 154             !btree_node_write_in_flight(b))
 155                 bch2_btree_node_write(c, b, lock_held);
 156 }
 157
 158 #define bch2_btree_node_write_cond(_c, _b, cond)                        \
 159 do {                                                                    \
 160         unsigned long old, new, v = READ_ONCE((_b)->flags);             \
 161                                                                         \
 162         do {                                                            \
 163                 old = new = v;                                          \
 164                                                                         \
 165                 if (!(old & (1 << BTREE_NODE_dirty)) || !(cond))        \
 166                         break;                                          \
 167                                                                         \
 168                 new |= (1 << BTREE_NODE_need_write);                    \
 169         } while ((v = cmpxchg(&(_b)->flags, old, new)) != old);         \
 170                                                                         \
 171         btree_node_write_if_need(_c, _b, SIX_LOCK_read);                \
 172 } while (0)
 173
 174 void bch2_btree_flush_all_reads(struct bch_fs *);
 175 void bch2_btree_flush_all_writes(struct bch_fs *);
 176 void bch2_dirty_btree_nodes_to_text(struct printbuf *, struct bch_fs *);
 177
 178 static inline void compat_bformat(unsigned level, enum btree_id btree_id,
 179                                   unsigned version, unsigned big_endian,
 180                                   int write, struct bkey_format *f)
 181 {
 182         if (version < bcachefs_metadata_version_inode_btree_change &&
 183             btree_id == BTREE_ID_inodes) {
 184                 swap(f->bits_per_field[BKEY_FIELD_INODE],
 185                      f->bits_per_field[BKEY_FIELD_OFFSET]);
 186                 swap(f->field_offset[BKEY_FIELD_INODE],
 187                      f->field_offset[BKEY_FIELD_OFFSET]);
 188         }
 189
 190         if (version < bcachefs_metadata_version_snapshot &&
 191             (level || btree_type_has_snapshots(btree_id))) {
 192                 u64 max_packed =
 193                         ~(~0ULL << f->bits_per_field[BKEY_FIELD_SNAPSHOT]);
 194
 195                 f->field_offset[BKEY_FIELD_SNAPSHOT] = write
 196                         ? 0
 197                         : U32_MAX - max_packed;
 198         }
 199 }
 200
 201 static inline void compat_bpos(unsigned level, enum btree_id btree_id,
 202                                unsigned version, unsigned big_endian,
 203                                int write, struct bpos *p)
 204 {
 205         if (big_endian != CPU_BIG_ENDIAN)
 206                 bch2_bpos_swab(p);
 207
 208         if (version < bcachefs_metadata_version_inode_btree_change &&
 209             btree_id == BTREE_ID_inodes)
 210                 swap(p->inode, p->offset);
 211 }
 212
 213 static inline void compat_btree_node(unsigned level, enum btree_id btree_id,
 214                                      unsigned version, unsigned big_endian,
 215                                      int write,
 216                                      struct btree_node *bn)
 217 {
 218         if (version < bcachefs_metadata_version_inode_btree_change &&
 219             btree_node_type_is_extents(btree_id) &&
 220             bpos_cmp(bn->min_key, POS_MIN) &&
 221             write)
 222                 bn->min_key = bpos_nosnap_predecessor(bn->min_key);
 223
 224         if (version < bcachefs_metadata_version_snapshot &&
 225             write)
 226                 bn->max_key.snapshot = 0;
 227
 228         compat_bpos(level, btree_id, version, big_endian, write, &bn->min_key);
 229         compat_bpos(level, btree_id, version, big_endian, write, &bn->max_key);
 230
 231         if (version < bcachefs_metadata_version_snapshot &&
 232             !write)
 233                 bn->max_key.snapshot = U32_MAX;
 234
 235         if (version < bcachefs_metadata_version_inode_btree_change &&
 236             btree_node_type_is_extents(btree_id) &&
 237             bpos_cmp(bn->min_key, POS_MIN) &&
 238             !write)
 239                 bn->min_key = bpos_nosnap_successor(bn->min_key);
 240 }
 241
 242 #endif /* _BCACHEFS_BTREE_IO_H */