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 static inline unsigned btree_ptr_sectors_written(struct bkey_i *k)
  36 {
  37         return k->k.type == KEY_TYPE_btree_ptr_v2
  38                 ? le16_to_cpu(bkey_i_to_btree_ptr_v2(k)->v.sectors_written)
  39                 : 0;
  40 }
  41
  42 struct btree_read_bio {
  43         struct bch_fs           *c;
  44         struct btree            *b;
  45         struct btree_node_read_all *ra;
  46         u64                     start_time;
  47         unsigned                have_ioref:1;
  48         unsigned                idx:7;
  49         struct extent_ptr_decoded       pick;
  50         struct work_struct      work;
  51         struct bio              bio;
  52 };
  53
  54 struct btree_write_bio {
  55         struct work_struct      work;
  56         __BKEY_PADDED(key, BKEY_BTREE_PTR_VAL_U64s_MAX);
  57         void                    *data;
  58         unsigned                data_bytes;
  59         unsigned                sector_offset;
  60         struct bch_write_bio    wbio;
  61 };
  62
  63 void bch2_btree_node_io_unlock(struct btree *);
  64 void bch2_btree_node_io_lock(struct btree *);
  65 void __bch2_btree_node_wait_on_read(struct btree *);
  66 void __bch2_btree_node_wait_on_write(struct btree *);
  67 void bch2_btree_node_wait_on_read(struct btree *);
  68 void bch2_btree_node_wait_on_write(struct btree *);
  69
  70 static inline bool btree_node_may_write(struct btree *b)
  71 {
  72         return list_empty_careful(&b->write_blocked) &&
  73                 (!b->written || !b->will_make_reachable);
  74 }
  75
  76 enum compact_mode {
  77         COMPACT_LAZY,
  78         COMPACT_ALL,
  79 };
  80
  81 bool bch2_compact_whiteouts(struct bch_fs *, struct btree *,
  82                             enum compact_mode);
  83
  84 static inline bool should_compact_bset_lazy(struct btree *b,
  85                                             struct bset_tree *t)
  86 {
  87         unsigned total_u64s = bset_u64s(t);
  88         unsigned dead_u64s = bset_dead_u64s(b, t);
  89
  90         return dead_u64s > 64 && dead_u64s * 3 > total_u64s;
  91 }
  92
  93 static inline bool bch2_maybe_compact_whiteouts(struct bch_fs *c, struct btree *b)
  94 {
  95         struct bset_tree *t;
  96
  97         for_each_bset(b, t)
  98                 if (should_compact_bset_lazy(b, t))
  99                         return bch2_compact_whiteouts(c, b, COMPACT_LAZY);
 100
 101         return false;
 102 }
 103
 104 static inline struct nonce btree_nonce(struct bset *i, unsigned offset)
 105 {
 106         return (struct nonce) {{
 107                 [0] = cpu_to_le32(offset),
 108                 [1] = ((__le32 *) &i->seq)[0],
 109                 [2] = ((__le32 *) &i->seq)[1],
 110                 [3] = ((__le32 *) &i->journal_seq)[0]^BCH_NONCE_BTREE,
 111         }};
 112 }
 113
 114 static inline void bset_encrypt(struct bch_fs *c, struct bset *i, unsigned offset)
 115 {
 116         struct nonce nonce = btree_nonce(i, offset);
 117
 118         if (!offset) {
 119                 struct btree_node *bn = container_of(i, struct btree_node, keys);
 120                 unsigned bytes = (void *) &bn->keys - (void *) &bn->flags;
 121
 122                 bch2_encrypt(c, BSET_CSUM_TYPE(i), nonce, &bn->flags,
 123                              bytes);
 124
 125                 nonce = nonce_add(nonce, round_up(bytes, CHACHA_BLOCK_SIZE));
 126         }
 127
 128         bch2_encrypt(c, BSET_CSUM_TYPE(i), nonce, i->_data,
 129                      vstruct_end(i) - (void *) i->_data);
 130 }
 131
 132 void bch2_btree_sort_into(struct bch_fs *, struct btree *, struct btree *);
 133
 134 void bch2_btree_node_drop_keys_outside_node(struct btree *);
 135
 136 void bch2_btree_build_aux_trees(struct btree *);
 137 void bch2_btree_init_next(struct btree_trans *, struct btree_iter *,
 138                           struct btree *);
 139
 140 int bch2_btree_node_read_done(struct bch_fs *, struct bch_dev *,
 141                               struct btree *, bool);
 142 void bch2_btree_node_read(struct bch_fs *, struct btree *, bool);
 143 int bch2_btree_root_read(struct bch_fs *, enum btree_id,
 144                          const struct bkey_i *, unsigned);
 145
 146 void bch2_btree_complete_write(struct bch_fs *, struct btree *,
 147                               struct btree_write *);
 148
 149 void __bch2_btree_node_write(struct bch_fs *, struct btree *, bool);
 150 bool bch2_btree_post_write_cleanup(struct bch_fs *, struct btree *);
 151
 152 void bch2_btree_node_write(struct bch_fs *, struct btree *,
 153                           enum six_lock_type);
 154
 155 static inline void btree_node_write_if_need(struct bch_fs *c, struct btree *b,
 156                                             enum six_lock_type lock_held)
 157 {
 158         if (b->written &&
 159             btree_node_need_write(b) &&
 160             btree_node_may_write(b) &&
 161             !btree_node_write_in_flight(b))
 162                 bch2_btree_node_write(c, b, lock_held);
 163 }
 164
 165 #define bch2_btree_node_write_cond(_c, _b, cond)                        \
 166 do {                                                                    \
 167         unsigned long old, new, v = READ_ONCE((_b)->flags);             \
 168                                                                         \
 169         do {                                                            \
 170                 old = new = v;                                          \
 171                                                                         \
 172                 if (!(old & (1 << BTREE_NODE_dirty)) || !(cond))        \
 173                         break;                                          \
 174                                                                         \
 175                 new |= (1 << BTREE_NODE_need_write);                    \
 176         } while ((v = cmpxchg(&(_b)->flags, old, new)) != old);         \
 177                                                                         \
 178         btree_node_write_if_need(_c, _b, SIX_LOCK_read);                \
 179 } while (0)
 180
 181 void bch2_btree_flush_all_reads(struct bch_fs *);
 182 void bch2_btree_flush_all_writes(struct bch_fs *);
 183 void bch2_dirty_btree_nodes_to_text(struct printbuf *, struct bch_fs *);
 184
 185 static inline void compat_bformat(unsigned level, enum btree_id btree_id,
 186                                   unsigned version, unsigned big_endian,
 187                                   int write, struct bkey_format *f)
 188 {
 189         if (version < bcachefs_metadata_version_inode_btree_change &&
 190             btree_id == BTREE_ID_inodes) {
 191                 swap(f->bits_per_field[BKEY_FIELD_INODE],
 192                      f->bits_per_field[BKEY_FIELD_OFFSET]);
 193                 swap(f->field_offset[BKEY_FIELD_INODE],
 194                      f->field_offset[BKEY_FIELD_OFFSET]);
 195         }
 196
 197         if (version < bcachefs_metadata_version_snapshot &&
 198             (level || btree_type_has_snapshots(btree_id))) {
 199                 u64 max_packed =
 200                         ~(~0ULL << f->bits_per_field[BKEY_FIELD_SNAPSHOT]);
 201
 202                 f->field_offset[BKEY_FIELD_SNAPSHOT] = write
 203                         ? 0
 204                         : U32_MAX - max_packed;
 205         }
 206 }
 207
 208 static inline void compat_bpos(unsigned level, enum btree_id btree_id,
 209                                unsigned version, unsigned big_endian,
 210                                int write, struct bpos *p)
 211 {
 212         if (big_endian != CPU_BIG_ENDIAN)
 213                 bch2_bpos_swab(p);
 214
 215         if (version < bcachefs_metadata_version_inode_btree_change &&
 216             btree_id == BTREE_ID_inodes)
 217                 swap(p->inode, p->offset);
 218 }
 219
 220 static inline void compat_btree_node(unsigned level, enum btree_id btree_id,
 221                                      unsigned version, unsigned big_endian,
 222                                      int write,
 223                                      struct btree_node *bn)
 224 {
 225         if (version < bcachefs_metadata_version_inode_btree_change &&
 226             btree_node_type_is_extents(btree_id) &&
 227             bpos_cmp(bn->min_key, POS_MIN) &&
 228             write)
 229                 bn->min_key = bpos_nosnap_predecessor(bn->min_key);
 230
 231         if (version < bcachefs_metadata_version_snapshot &&
 232             write)
 233                 bn->max_key.snapshot = 0;
 234
 235         compat_bpos(level, btree_id, version, big_endian, write, &bn->min_key);
 236         compat_bpos(level, btree_id, version, big_endian, write, &bn->max_key);
 237
 238         if (version < bcachefs_metadata_version_snapshot &&
 239             !write)
 240                 bn->max_key.snapshot = U32_MAX;
 241
 242         if (version < bcachefs_metadata_version_inode_btree_change &&
 243             btree_node_type_is_extents(btree_id) &&
 244             bpos_cmp(bn->min_key, POS_MIN) &&
 245             !write)
 246                 bn->min_key = bpos_nosnap_successor(bn->min_key);
 247 }
 248
 249 #endif /* _BCACHEFS_BTREE_IO_H */