fs/bcachefs/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_write_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 void set_btree_node_dirty_acct(struct bch_fs *c, struct btree *b)
  19 {
  20         if (!test_and_set_bit(BTREE_NODE_dirty, &b->flags))
  21                 atomic_long_inc(&c->btree_cache.nr_dirty);
  22 }
  23
  24 static inline void clear_btree_node_dirty_acct(struct bch_fs *c, struct btree *b)
  25 {
  26         if (test_and_clear_bit(BTREE_NODE_dirty, &b->flags))
  27                 atomic_long_dec(&c->btree_cache.nr_dirty);
  28 }
  29
  30 static inline unsigned btree_ptr_sectors_written(struct bkey_s_c k)
  31 {
  32         return k.k->type == KEY_TYPE_btree_ptr_v2
  33                 ? le16_to_cpu(bkey_s_c_to_btree_ptr_v2(k).v->sectors_written)
  34                 : 0;
  35 }
  36
  37 struct btree_read_bio {
  38         struct bch_fs           *c;
  39         struct btree            *b;
  40         struct btree_node_read_all *ra;
  41         u64                     start_time;
  42         unsigned                have_ioref:1;
  43         unsigned                idx:7;
  44         struct extent_ptr_decoded       pick;
  45         struct work_struct      work;
  46         struct bio              bio;
  47 };
  48
  49 struct btree_write_bio {
  50         struct work_struct      work;
  51         __BKEY_PADDED(key, BKEY_BTREE_PTR_VAL_U64s_MAX);
  52         void                    *data;
  53         unsigned                data_bytes;
  54         unsigned                sector_offset;
  55         struct bch_write_bio    wbio;
  56 };
  57
  58 void bch2_btree_node_io_unlock(struct btree *);
  59 void bch2_btree_node_io_lock(struct btree *);
  60 void __bch2_btree_node_wait_on_read(struct btree *);
  61 void __bch2_btree_node_wait_on_write(struct btree *);
  62 void bch2_btree_node_wait_on_read(struct btree *);
  63 void bch2_btree_node_wait_on_write(struct btree *);
  64
  65 enum compact_mode {
  66         COMPACT_LAZY,
  67         COMPACT_ALL,
  68 };
  69
  70 bool bch2_compact_whiteouts(struct bch_fs *, struct btree *,
  71                             enum compact_mode);
  72
  73 static inline bool should_compact_bset_lazy(struct btree *b,
  74                                             struct bset_tree *t)
  75 {
  76         unsigned total_u64s = bset_u64s(t);
  77         unsigned dead_u64s = bset_dead_u64s(b, t);
  78
  79         return dead_u64s > 64 && dead_u64s * 3 > total_u64s;
  80 }
  81
  82 static inline bool bch2_maybe_compact_whiteouts(struct bch_fs *c, struct btree *b)
  83 {
  84         for_each_bset(b, t)
  85                 if (should_compact_bset_lazy(b, t))
  86                         return bch2_compact_whiteouts(c, b, COMPACT_LAZY);
  87
  88         return false;
  89 }
  90
  91 static inline struct nonce btree_nonce(struct bset *i, unsigned offset)
  92 {
  93         return (struct nonce) {{
  94                 [0] = cpu_to_le32(offset),
  95                 [1] = ((__le32 *) &i->seq)[0],
  96                 [2] = ((__le32 *) &i->seq)[1],
  97                 [3] = ((__le32 *) &i->journal_seq)[0]^BCH_NONCE_BTREE,
  98         }};
  99 }
 100
 101 static inline int bset_encrypt(struct bch_fs *c, struct bset *i, unsigned offset)
 102 {
 103         struct nonce nonce = btree_nonce(i, offset);
 104         int ret;
 105
 106         if (!offset) {
 107                 struct btree_node *bn = container_of(i, struct btree_node, keys);
 108                 unsigned bytes = (void *) &bn->keys - (void *) &bn->flags;
 109
 110                 ret = bch2_encrypt(c, BSET_CSUM_TYPE(i), nonce,
 111                                    &bn->flags, bytes);
 112                 if (ret)
 113                         return ret;
 114
 115                 nonce = nonce_add(nonce, round_up(bytes, CHACHA_BLOCK_SIZE));
 116         }
 117
 118         return bch2_encrypt(c, BSET_CSUM_TYPE(i), nonce, i->_data,
 119                             vstruct_end(i) - (void *) i->_data);
 120 }
 121
 122 void bch2_btree_sort_into(struct bch_fs *, struct btree *, struct btree *);
 123
 124 void bch2_btree_node_drop_keys_outside_node(struct btree *);
 125
 126 void bch2_btree_build_aux_trees(struct btree *);
 127 void bch2_btree_init_next(struct btree_trans *, struct btree *);
 128
 129 int bch2_btree_node_read_done(struct bch_fs *, struct bch_dev *,
 130                               struct btree *, bool, bool *);
 131 void bch2_btree_node_read(struct btree_trans *, struct btree *, bool);
 132 int bch2_btree_root_read(struct bch_fs *, enum btree_id,
 133                          const struct bkey_i *, unsigned);
 134
 135 bool bch2_btree_post_write_cleanup(struct bch_fs *, struct btree *);
 136
 137 enum btree_write_flags {
 138         __BTREE_WRITE_ONLY_IF_NEED = BTREE_WRITE_TYPE_BITS,
 139         __BTREE_WRITE_ALREADY_STARTED,
 140 };
 141 #define BTREE_WRITE_ONLY_IF_NEED        BIT(__BTREE_WRITE_ONLY_IF_NEED)
 142 #define BTREE_WRITE_ALREADY_STARTED     BIT(__BTREE_WRITE_ALREADY_STARTED)
 143
 144 void __bch2_btree_node_write(struct bch_fs *, struct btree *, unsigned);
 145 void bch2_btree_node_write(struct bch_fs *, struct btree *,
 146                            enum six_lock_type, unsigned);
 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         bch2_btree_node_write(c, b, lock_held, BTREE_WRITE_ONLY_IF_NEED);
 152 }
 153
 154 bool bch2_btree_flush_all_reads(struct bch_fs *);
 155 bool bch2_btree_flush_all_writes(struct bch_fs *);
 156
 157 static inline void compat_bformat(unsigned level, enum btree_id btree_id,
 158                                   unsigned version, unsigned big_endian,
 159                                   int write, struct bkey_format *f)
 160 {
 161         if (version < bcachefs_metadata_version_inode_btree_change &&
 162             btree_id == BTREE_ID_inodes) {
 163                 swap(f->bits_per_field[BKEY_FIELD_INODE],
 164                      f->bits_per_field[BKEY_FIELD_OFFSET]);
 165                 swap(f->field_offset[BKEY_FIELD_INODE],
 166                      f->field_offset[BKEY_FIELD_OFFSET]);
 167         }
 168
 169         if (version < bcachefs_metadata_version_snapshot &&
 170             (level || btree_type_has_snapshots(btree_id))) {
 171                 u64 max_packed =
 172                         ~(~0ULL << f->bits_per_field[BKEY_FIELD_SNAPSHOT]);
 173
 174                 f->field_offset[BKEY_FIELD_SNAPSHOT] = write
 175                         ? 0
 176                         : cpu_to_le64(U32_MAX - max_packed);
 177         }
 178 }
 179
 180 static inline void compat_bpos(unsigned level, enum btree_id btree_id,
 181                                unsigned version, unsigned big_endian,
 182                                int write, struct bpos *p)
 183 {
 184         if (big_endian != CPU_BIG_ENDIAN)
 185                 bch2_bpos_swab(p);
 186
 187         if (version < bcachefs_metadata_version_inode_btree_change &&
 188             btree_id == BTREE_ID_inodes)
 189                 swap(p->inode, p->offset);
 190 }
 191
 192 static inline void compat_btree_node(unsigned level, enum btree_id btree_id,
 193                                      unsigned version, unsigned big_endian,
 194                                      int write,
 195                                      struct btree_node *bn)
 196 {
 197         if (version < bcachefs_metadata_version_inode_btree_change &&
 198             btree_id_is_extents(btree_id) &&
 199             !bpos_eq(bn->min_key, POS_MIN) &&
 200             write)
 201                 bn->min_key = bpos_nosnap_predecessor(bn->min_key);
 202
 203         if (version < bcachefs_metadata_version_snapshot &&
 204             write)
 205                 bn->max_key.snapshot = 0;
 206
 207         compat_bpos(level, btree_id, version, big_endian, write, &bn->min_key);
 208         compat_bpos(level, btree_id, version, big_endian, write, &bn->max_key);
 209
 210         if (version < bcachefs_metadata_version_snapshot &&
 211             !write)
 212                 bn->max_key.snapshot = U32_MAX;
 213
 214         if (version < bcachefs_metadata_version_inode_btree_change &&
 215             btree_id_is_extents(btree_id) &&
 216             !bpos_eq(bn->min_key, POS_MIN) &&
 217             !write)
 218                 bn->min_key = bpos_nosnap_successor(bn->min_key);
 219 }
 220
 221 void bch2_btree_write_stats_to_text(struct printbuf *, struct bch_fs *);
 222
 223 #endif /* _BCACHEFS_BTREE_IO_H */