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 struct btree_read_bio {
  18         struct bch_fs           *c;
  19         u64                     start_time;
  20         unsigned                have_ioref:1;
  21         struct extent_ptr_decoded       pick;
  22         struct work_struct      work;
  23         struct bio              bio;
  24 };
  25
  26 struct btree_write_bio {
  27         struct work_struct      work;
  28         void                    *data;
  29         unsigned                bytes;
  30         struct bch_write_bio    wbio;
  31 };
  32
  33 static inline void btree_node_io_unlock(struct btree *b)
  34 {
  35         EBUG_ON(!btree_node_write_in_flight(b));
  36         clear_btree_node_write_in_flight(b);
  37         wake_up_bit(&b->flags, BTREE_NODE_write_in_flight);
  38 }
  39
  40 static inline void btree_node_io_lock(struct btree *b)
  41 {
  42         wait_on_bit_lock_io(&b->flags, BTREE_NODE_write_in_flight,
  43                             TASK_UNINTERRUPTIBLE);
  44 }
  45
  46 static inline void btree_node_wait_on_io(struct btree *b)
  47 {
  48         wait_on_bit_io(&b->flags, BTREE_NODE_write_in_flight,
  49                        TASK_UNINTERRUPTIBLE);
  50 }
  51
  52 static inline bool btree_node_may_write(struct btree *b)
  53 {
  54         return list_empty_careful(&b->write_blocked) &&
  55                 (!b->written || !b->will_make_reachable);
  56 }
  57
  58 enum compact_mode {
  59         COMPACT_LAZY,
  60         COMPACT_ALL,
  61 };
  62
  63 bool bch2_compact_whiteouts(struct bch_fs *, struct btree *,
  64                             enum compact_mode);
  65
  66 static inline bool should_compact_bset_lazy(struct btree *b,
  67                                             struct bset_tree *t)
  68 {
  69         unsigned total_u64s = bset_u64s(t);
  70         unsigned dead_u64s = bset_dead_u64s(b, t);
  71
  72         return dead_u64s > 64 && dead_u64s * 3 > total_u64s;
  73 }
  74
  75 static inline bool bch2_maybe_compact_whiteouts(struct bch_fs *c, struct btree *b)
  76 {
  77         struct bset_tree *t;
  78
  79         for_each_bset(b, t)
  80                 if (should_compact_bset_lazy(b, t))
  81                         return bch2_compact_whiteouts(c, b, COMPACT_LAZY);
  82
  83         return false;
  84 }
  85
  86 static inline struct nonce btree_nonce(struct bset *i, unsigned offset)
  87 {
  88         return (struct nonce) {{
  89                 [0] = cpu_to_le32(offset),
  90                 [1] = ((__le32 *) &i->seq)[0],
  91                 [2] = ((__le32 *) &i->seq)[1],
  92                 [3] = ((__le32 *) &i->journal_seq)[0]^BCH_NONCE_BTREE,
  93         }};
  94 }
  95
  96 static inline void bset_encrypt(struct bch_fs *c, struct bset *i, unsigned offset)
  97 {
  98         struct nonce nonce = btree_nonce(i, offset);
  99
 100         if (!offset) {
 101                 struct btree_node *bn = container_of(i, struct btree_node, keys);
 102                 unsigned bytes = (void *) &bn->keys - (void *) &bn->flags;
 103
 104                 bch2_encrypt(c, BSET_CSUM_TYPE(i), nonce, &bn->flags,
 105                              bytes);
 106
 107                 nonce = nonce_add(nonce, round_up(bytes, CHACHA_BLOCK_SIZE));
 108         }
 109
 110         bch2_encrypt(c, BSET_CSUM_TYPE(i), nonce, i->_data,
 111                      vstruct_end(i) - (void *) i->_data);
 112 }
 113
 114 void bch2_btree_sort_into(struct bch_fs *, struct btree *, struct btree *);
 115
 116 void bch2_btree_build_aux_trees(struct btree *);
 117 void bch2_btree_init_next(struct bch_fs *, struct btree *,
 118                          struct btree_iter *);
 119
 120 int bch2_btree_node_read_done(struct bch_fs *, struct btree *, bool);
 121 void bch2_btree_node_read(struct bch_fs *, struct btree *, bool);
 122 int bch2_btree_root_read(struct bch_fs *, enum btree_id,
 123                          const struct bkey_i *, unsigned);
 124
 125 void bch2_btree_complete_write(struct bch_fs *, struct btree *,
 126                               struct btree_write *);
 127 void bch2_btree_write_error_work(struct work_struct *);
 128
 129 void __bch2_btree_node_write(struct bch_fs *, struct btree *,
 130                             enum six_lock_type);
 131 bool bch2_btree_post_write_cleanup(struct bch_fs *, struct btree *);
 132
 133 void bch2_btree_node_write(struct bch_fs *, struct btree *,
 134                           enum six_lock_type);
 135
 136 static inline void btree_node_write_if_need(struct bch_fs *c, struct btree *b,
 137                                             enum six_lock_type lock_held)
 138 {
 139         while (b->written &&
 140                btree_node_need_write(b) &&
 141                btree_node_may_write(b)) {
 142                 if (!btree_node_write_in_flight(b)) {
 143                         bch2_btree_node_write(c, b, lock_held);
 144                         break;
 145                 }
 146
 147                 six_unlock_type(&b->c.lock, lock_held);
 148                 btree_node_wait_on_io(b);
 149                 btree_node_lock_type(c, b, lock_held);
 150         }
 151 }
 152
 153 #define bch2_btree_node_write_cond(_c, _b, cond)                        \
 154 do {                                                                    \
 155         unsigned long old, new, v = READ_ONCE((_b)->flags);             \
 156                                                                         \
 157         do {                                                            \
 158                 old = new = v;                                          \
 159                                                                         \
 160                 if (!(old & (1 << BTREE_NODE_dirty)) || !(cond))        \
 161                         break;                                          \
 162                                                                         \
 163                 new |= (1 << BTREE_NODE_need_write);                    \
 164         } while ((v = cmpxchg(&(_b)->flags, old, new)) != old);         \
 165                                                                         \
 166         btree_node_write_if_need(_c, _b, SIX_LOCK_read);                \
 167 } while (0)
 168
 169 void bch2_btree_flush_all_reads(struct bch_fs *);
 170 void bch2_btree_flush_all_writes(struct bch_fs *);
 171 void bch2_btree_verify_flushed(struct bch_fs *);
 172 void bch2_dirty_btree_nodes_to_text(struct printbuf *, struct bch_fs *);
 173
 174 static inline void compat_bformat(unsigned level, enum btree_id btree_id,
 175                                  unsigned version, unsigned big_endian,
 176                                  int write, struct bkey_format *f)
 177 {
 178         if (version < bcachefs_metadata_version_inode_btree_change &&
 179             btree_id == BTREE_ID_INODES) {
 180                 swap(f->bits_per_field[BKEY_FIELD_INODE],
 181                      f->bits_per_field[BKEY_FIELD_OFFSET]);
 182                 swap(f->field_offset[BKEY_FIELD_INODE],
 183                      f->field_offset[BKEY_FIELD_OFFSET]);
 184         }
 185 }
 186
 187 static inline void compat_bpos(unsigned level, enum btree_id btree_id,
 188                                unsigned version, unsigned big_endian,
 189                                int write, struct bpos *p)
 190 {
 191         if (big_endian != CPU_BIG_ENDIAN)
 192                 bch2_bpos_swab(p);
 193
 194         if (version < bcachefs_metadata_version_inode_btree_change &&
 195             btree_id == BTREE_ID_INODES)
 196                 swap(p->inode, p->offset);
 197 }
 198
 199 static inline void compat_btree_node(unsigned level, enum btree_id btree_id,
 200                                      unsigned version, unsigned big_endian,
 201                                      int write,
 202                                      struct btree_node *bn)
 203 {
 204         if (version < bcachefs_metadata_version_inode_btree_change &&
 205             btree_node_type_is_extents(btree_id) &&
 206             bkey_cmp(bn->min_key, POS_MIN) &&
 207             write)
 208                 bn->min_key = bkey_predecessor(bn->min_key);
 209
 210         compat_bpos(level, btree_id, version, big_endian, write, &bn->min_key);
 211         compat_bpos(level, btree_id, version, big_endian, write, &bn->max_key);
 212
 213         if (version < bcachefs_metadata_version_inode_btree_change &&
 214             btree_node_type_is_extents(btree_id) &&
 215             bkey_cmp(bn->min_key, POS_MIN) &&
 216             !write)
 217                 bn->min_key = bkey_successor(bn->min_key);
 218 }
 219
 220 #endif /* _BCACHEFS_BTREE_IO_H */