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 #ifdef CONFIG_BCACHEFS_ASYNC_OBJECT_LISTS 45 unsigned list_idx; 46 #endif 47 struct extent_ptr_decoded pick; 48 struct work_struct work; 49 struct bio bio; 50 }; 51 52 struct btree_write_bio { 53 struct work_struct work; 54 __BKEY_PADDED(key, BKEY_BTREE_PTR_VAL_U64s_MAX); 55 void *data; 56 unsigned data_bytes; 57 unsigned sector_offset; 58 u64 start_time; 59 #ifdef CONFIG_BCACHEFS_ASYNC_OBJECT_LISTS 60 unsigned list_idx; 61 #endif 62 struct bch_write_bio wbio; 63 }; 64 65 void bch2_btree_node_io_unlock(struct btree *); 66 void bch2_btree_node_io_lock(struct btree *); 67 void __bch2_btree_node_wait_on_read(struct btree *); 68 void __bch2_btree_node_wait_on_write(struct btree *); 69 void bch2_btree_node_wait_on_read(struct btree *); 70 void bch2_btree_node_wait_on_write(struct btree *); 71 72 enum compact_mode { 73 COMPACT_LAZY, 74 COMPACT_ALL, 75 }; 76 77 bool bch2_compact_whiteouts(struct bch_fs *, struct btree *, 78 enum compact_mode); 79 80 static inline bool should_compact_bset_lazy(struct btree *b, 81 struct bset_tree *t) 82 { 83 unsigned total_u64s = bset_u64s(t); 84 unsigned dead_u64s = bset_dead_u64s(b, t); 85 86 return dead_u64s > 64 && dead_u64s * 3 > total_u64s; 87 } 88 89 static inline bool bch2_maybe_compact_whiteouts(struct bch_fs *c, struct btree *b) 90 { 91 for_each_bset(b, t) 92 if (should_compact_bset_lazy(b, t)) 93 return bch2_compact_whiteouts(c, b, COMPACT_LAZY); 94 95 return false; 96 } 97 98 static inline struct nonce btree_nonce(struct bset *i, unsigned offset) 99 { 100 return (struct nonce) {{ 101 [0] = cpu_to_le32(offset), 102 [1] = ((__le32 *) &i->seq)[0], 103 [2] = ((__le32 *) &i->seq)[1], 104 [3] = ((__le32 *) &i->journal_seq)[0]^BCH_NONCE_BTREE, 105 }}; 106 } 107 108 static inline int bset_encrypt(struct bch_fs *c, struct bset *i, unsigned offset) 109 { 110 struct nonce nonce = btree_nonce(i, offset); 111 int ret; 112 113 if (!offset) { 114 struct btree_node *bn = container_of(i, struct btree_node, keys); 115 unsigned bytes = (void *) &bn->keys - (void *) &bn->flags; 116 117 ret = bch2_encrypt(c, BSET_CSUM_TYPE(i), nonce, 118 &bn->flags, bytes); 119 if (ret) 120 return ret; 121 122 nonce = nonce_add(nonce, round_up(bytes, CHACHA_BLOCK_SIZE)); 123 } 124 125 return bch2_encrypt(c, BSET_CSUM_TYPE(i), nonce, i->_data, 126 vstruct_end(i) - (void *) i->_data); 127 } 128 129 void bch2_btree_sort_into(struct bch_fs *, struct btree *, struct btree *); 130 131 void bch2_btree_node_drop_keys_outside_node(struct btree *); 132 133 void bch2_btree_build_aux_trees(struct btree *); 134 void bch2_btree_init_next(struct btree_trans *, struct btree *); 135 136 int bch2_btree_node_read_done(struct bch_fs *, struct bch_dev *, 137 struct btree *, 138 struct bch_io_failures *, 139 struct printbuf *); 140 void bch2_btree_node_read(struct btree_trans *, struct btree *, bool); 141 int bch2_btree_root_read(struct bch_fs *, enum btree_id, 142 const struct bkey_i *, unsigned); 143 144 void bch2_btree_read_bio_to_text(struct printbuf *, struct btree_read_bio *); 145 146 int bch2_btree_node_scrub(struct btree_trans *, enum btree_id, unsigned, 147 struct bkey_s_c, unsigned); 148 149 bool bch2_btree_post_write_cleanup(struct bch_fs *, struct btree *); 150 151 enum btree_write_flags { 152 __BTREE_WRITE_ONLY_IF_NEED = BTREE_WRITE_TYPE_BITS, 153 __BTREE_WRITE_ALREADY_STARTED, 154 }; 155 #define BTREE_WRITE_ONLY_IF_NEED BIT(__BTREE_WRITE_ONLY_IF_NEED) 156 #define BTREE_WRITE_ALREADY_STARTED BIT(__BTREE_WRITE_ALREADY_STARTED) 157 158 void __bch2_btree_node_write(struct bch_fs *, struct btree *, unsigned); 159 void bch2_btree_node_write(struct bch_fs *, struct btree *, 160 enum six_lock_type, unsigned); 161 void bch2_btree_node_write_trans(struct btree_trans *, struct btree *, 162 enum six_lock_type, unsigned); 163 164 static inline void btree_node_write_if_need(struct btree_trans *trans, struct btree *b, 165 enum six_lock_type lock_held) 166 { 167 bch2_btree_node_write_trans(trans, b, lock_held, BTREE_WRITE_ONLY_IF_NEED); 168 } 169 170 bool bch2_btree_flush_all_reads(struct bch_fs *); 171 bool bch2_btree_flush_all_writes(struct bch_fs *); 172 173 static inline void compat_bformat(unsigned level, enum btree_id btree_id, 174 unsigned version, unsigned big_endian, 175 int write, struct bkey_format *f) 176 { 177 if (version < bcachefs_metadata_version_inode_btree_change && 178 btree_id == BTREE_ID_inodes) { 179 swap(f->bits_per_field[BKEY_FIELD_INODE], 180 f->bits_per_field[BKEY_FIELD_OFFSET]); 181 swap(f->field_offset[BKEY_FIELD_INODE], 182 f->field_offset[BKEY_FIELD_OFFSET]); 183 } 184 185 if (version < bcachefs_metadata_version_snapshot && 186 (level || btree_type_has_snapshots(btree_id))) { 187 u64 max_packed = 188 ~(~0ULL << f->bits_per_field[BKEY_FIELD_SNAPSHOT]); 189 190 f->field_offset[BKEY_FIELD_SNAPSHOT] = write 191 ? 0 192 : cpu_to_le64(U32_MAX - max_packed); 193 } 194 } 195 196 static inline void compat_bpos(unsigned level, enum btree_id btree_id, 197 unsigned version, unsigned big_endian, 198 int write, struct bpos *p) 199 { 200 if (big_endian != CPU_BIG_ENDIAN) 201 bch2_bpos_swab(p); 202 203 if (version < bcachefs_metadata_version_inode_btree_change && 204 btree_id == BTREE_ID_inodes) 205 swap(p->inode, p->offset); 206 } 207 208 static inline void compat_btree_node(unsigned level, enum btree_id btree_id, 209 unsigned version, unsigned big_endian, 210 int write, 211 struct btree_node *bn) 212 { 213 if (version < bcachefs_metadata_version_inode_btree_change && 214 btree_id_is_extents(btree_id) && 215 !bpos_eq(bn->min_key, POS_MIN) && 216 write) 217 bn->min_key = bpos_nosnap_predecessor(bn->min_key); 218 219 if (version < bcachefs_metadata_version_snapshot && 220 write) 221 bn->max_key.snapshot = 0; 222 223 compat_bpos(level, btree_id, version, big_endian, write, &bn->min_key); 224 compat_bpos(level, btree_id, version, big_endian, write, &bn->max_key); 225 226 if (version < bcachefs_metadata_version_snapshot && 227 !write) 228 bn->max_key.snapshot = U32_MAX; 229 230 if (version < bcachefs_metadata_version_inode_btree_change && 231 btree_id_is_extents(btree_id) && 232 !bpos_eq(bn->min_key, POS_MIN) && 233 !write) 234 bn->min_key = bpos_nosnap_successor(bn->min_key); 235 } 236 237 void bch2_btree_write_stats_to_text(struct printbuf *, struct bch_fs *); 238 239 #endif /* _BCACHEFS_BTREE_IO_H */ 240