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 "extents.h"
   9 #include "io_types.h"
  10
  11 struct bch_fs;
  12 struct btree_write;
  13 struct btree;
  14 struct btree_iter;
  15
  16 struct btree_read_bio {
  17         struct bch_fs           *c;
  18         u64                     start_time;
  19         unsigned                have_ioref:1;
  20         struct extent_ptr_decoded       pick;
  21         struct work_struct      work;
  22         struct bio              bio;
  23 };
  24
  25 struct btree_write_bio {
  26         void                    *data;
  27         struct work_struct      work;
  28         struct bch_write_bio    wbio;
  29 };
  30
  31 static inline void btree_node_io_unlock(struct btree *b)
  32 {
  33         EBUG_ON(!btree_node_write_in_flight(b));
  34         clear_btree_node_write_in_flight(b);
  35         wake_up_bit(&b->flags, BTREE_NODE_write_in_flight);
  36 }
  37
  38 static inline void btree_node_io_lock(struct btree *b)
  39 {
  40         wait_on_bit_lock_io(&b->flags, BTREE_NODE_write_in_flight,
  41                             TASK_UNINTERRUPTIBLE);
  42 }
  43
  44 static inline void btree_node_wait_on_io(struct btree *b)
  45 {
  46         wait_on_bit_io(&b->flags, BTREE_NODE_write_in_flight,
  47                        TASK_UNINTERRUPTIBLE);
  48 }
  49
  50 static inline bool btree_node_may_write(struct btree *b)
  51 {
  52         return list_empty_careful(&b->write_blocked) &&
  53                 (!b->written || !b->will_make_reachable);
  54 }
  55
  56 enum compact_mode {
  57         COMPACT_LAZY,
  58         COMPACT_ALL,
  59 };
  60
  61 bool bch2_compact_whiteouts(struct bch_fs *, struct btree *,
  62                             enum compact_mode);
  63
  64 static inline bool should_compact_bset_lazy(struct btree *b,
  65                                             struct bset_tree *t)
  66 {
  67         unsigned total_u64s = bset_u64s(t);
  68         unsigned dead_u64s = bset_dead_u64s(b, t);
  69
  70         return dead_u64s > 64 && dead_u64s * 3 > total_u64s;
  71 }
  72
  73 static inline bool bch2_maybe_compact_whiteouts(struct bch_fs *c, struct btree *b)
  74 {
  75         struct bset_tree *t;
  76
  77         for_each_bset(b, t)
  78                 if (should_compact_bset_lazy(b, t))
  79                         return bch2_compact_whiteouts(c, b, COMPACT_LAZY);
  80
  81         return false;
  82 }
  83
  84 void bch2_btree_sort_into(struct bch_fs *, struct btree *, struct btree *);
  85
  86 void bch2_btree_build_aux_trees(struct btree *);
  87 void bch2_btree_init_next(struct bch_fs *, struct btree *,
  88                          struct btree_iter *);
  89
  90 int bch2_btree_node_read_done(struct bch_fs *, struct btree *, bool);
  91 void bch2_btree_node_read(struct bch_fs *, struct btree *, bool);
  92 int bch2_btree_root_read(struct bch_fs *, enum btree_id,
  93                          const struct bkey_i *, unsigned);
  94
  95 void bch2_btree_complete_write(struct bch_fs *, struct btree *,
  96                               struct btree_write *);
  97 void bch2_btree_write_error_work(struct work_struct *);
  98
  99 void __bch2_btree_node_write(struct bch_fs *, struct btree *,
 100                             enum six_lock_type);
 101 bool bch2_btree_post_write_cleanup(struct bch_fs *, struct btree *);
 102
 103 void bch2_btree_node_write(struct bch_fs *, struct btree *,
 104                           enum six_lock_type);
 105
 106 static inline void btree_node_write_if_need(struct bch_fs *c, struct btree *b,
 107                                             enum six_lock_type lock_held)
 108 {
 109         while (b->written &&
 110                btree_node_need_write(b) &&
 111                btree_node_may_write(b)) {
 112                 if (!btree_node_write_in_flight(b)) {
 113                         bch2_btree_node_write(c, b, lock_held);
 114                         break;
 115                 }
 116
 117                 six_unlock_type(&b->c.lock, lock_held);
 118                 btree_node_wait_on_io(b);
 119                 btree_node_lock_type(c, b, lock_held);
 120         }
 121 }
 122
 123 #define bch2_btree_node_write_cond(_c, _b, cond)                        \
 124 do {                                                                    \
 125         unsigned long old, new, v = READ_ONCE((_b)->flags);             \
 126                                                                         \
 127         do {                                                            \
 128                 old = new = v;                                          \
 129                                                                         \
 130                 if (!(old & (1 << BTREE_NODE_dirty)) || !(cond))        \
 131                         break;                                          \
 132                                                                         \
 133                 new |= (1 << BTREE_NODE_need_write);                    \
 134         } while ((v = cmpxchg(&(_b)->flags, old, new)) != old);         \
 135                                                                         \
 136         btree_node_write_if_need(_c, _b, SIX_LOCK_read);                \
 137 } while (0)
 138
 139 void bch2_btree_flush_all_reads(struct bch_fs *);
 140 void bch2_btree_flush_all_writes(struct bch_fs *);
 141 void bch2_btree_verify_flushed(struct bch_fs *);
 142 ssize_t bch2_dirty_btree_nodes_print(struct bch_fs *, char *);
 143
 144 static inline void compat_bformat(unsigned level, enum btree_id btree_id,
 145                                  unsigned version, unsigned big_endian,
 146                                  int write, struct bkey_format *f)
 147 {
 148         if (version < bcachefs_metadata_version_inode_btree_change &&
 149             btree_id == BTREE_ID_INODES) {
 150                 swap(f->bits_per_field[BKEY_FIELD_INODE],
 151                      f->bits_per_field[BKEY_FIELD_OFFSET]);
 152                 swap(f->field_offset[BKEY_FIELD_INODE],
 153                      f->field_offset[BKEY_FIELD_OFFSET]);
 154         }
 155 }
 156
 157 static inline void compat_bpos(unsigned level, enum btree_id btree_id,
 158                                unsigned version, unsigned big_endian,
 159                                int write, struct bpos *p)
 160 {
 161         if (big_endian != CPU_BIG_ENDIAN)
 162                 bch2_bpos_swab(p);
 163
 164         if (version < bcachefs_metadata_version_inode_btree_change &&
 165             btree_id == BTREE_ID_INODES)
 166                 swap(p->inode, p->offset);
 167 }
 168
 169 static inline void compat_btree_node(unsigned level, enum btree_id btree_id,
 170                                      unsigned version, unsigned big_endian,
 171                                      int write,
 172                                      struct btree_node *bn)
 173 {
 174         if (version < bcachefs_metadata_version_inode_btree_change &&
 175             btree_node_type_is_extents(btree_id) &&
 176             bkey_cmp(bn->min_key, POS_MIN) &&
 177             write)
 178                 bn->min_key = bkey_predecessor(bn->min_key);
 179
 180         compat_bpos(level, btree_id, version, big_endian, write, &bn->min_key);
 181         compat_bpos(level, btree_id, version, big_endian, write, &bn->max_key);
 182
 183         if (version < bcachefs_metadata_version_inode_btree_change &&
 184             btree_node_type_is_extents(btree_id) &&
 185             bkey_cmp(bn->min_key, POS_MIN) &&
 186             !write)
 187                 bn->min_key = bkey_successor(bn->min_key);
 188 }
 189
 190 #endif /* _BCACHEFS_BTREE_IO_H */