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