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