git.sesse.net Git - bcachefs-tools-debian/blob - libbcache/btree_types.h

   1 #ifndef _BCACHE_BTREE_TYPES_H
   2 #define _BCACHE_BTREE_TYPES_H
   3
   4 #include <linux/bcache.h>
   5 #include <linux/list.h>
   6 #include <linux/rhashtable.h>
   7 #include <linux/semaphore.h>
   8 #include <linux/workqueue.h>
   9
  10 #include "bkey_methods.h"
  11 #include "journal_types.h"
  12 #include "six.h"
  13
  14 struct cache_set;
  15 struct open_bucket;
  16 struct btree_interior_update;
  17
  18 #define MAX_BSETS               3U
  19
  20 struct btree_nr_keys {
  21
  22         /*
  23          * Amount of live metadata (i.e. size of node after a compaction) in
  24          * units of u64s
  25          */
  26         u16                     live_u64s;
  27         u16                     bset_u64s[MAX_BSETS];
  28
  29         /* live keys only: */
  30         u16                     packed_keys;
  31         u16                     unpacked_keys;
  32 };
  33
  34 struct bset_tree {
  35         /*
  36          * We construct a binary tree in an array as if the array
  37          * started at 1, so that things line up on the same cachelines
  38          * better: see comments in bset.c at cacheline_to_bkey() for
  39          * details
  40          */
  41
  42         /* size of the binary tree and prev array */
  43         u16                     size;
  44
  45         /* function of size - precalculated for to_inorder() */
  46         u16                     extra;
  47
  48         u16                     data_offset;
  49         u16                     aux_data_offset;
  50         u16                     end_offset;
  51
  52         struct bpos             max_key;
  53 };
  54
  55 struct btree_write {
  56         struct journal_entry_pin        journal;
  57         struct closure_waitlist         wait;
  58 };
  59
  60 struct btree {
  61         /* Hottest entries first */
  62         struct rhash_head       hash;
  63
  64         /* Key/pointer for this btree node */
  65         __BKEY_PADDED(key, BKEY_BTREE_PTR_VAL_U64s_MAX);
  66
  67         struct six_lock         lock;
  68
  69         unsigned long           flags;
  70         u16                     written;
  71         u8                      level;
  72         u8                      btree_id;
  73         u8                      nsets;
  74         u8                      nr_key_bits;
  75
  76         struct bkey_format      format;
  77
  78         struct btree_node       *data;
  79         void                    *aux_data;
  80
  81         /*
  82          * Sets of sorted keys - the real btree node - plus a binary search tree
  83          *
  84          * set[0] is special; set[0]->tree, set[0]->prev and set[0]->data point
  85          * to the memory we have allocated for this btree node. Additionally,
  86          * set[0]->data points to the entire btree node as it exists on disk.
  87          */
  88         struct bset_tree        set[MAX_BSETS];
  89
  90         struct btree_nr_keys    nr;
  91         u16                     sib_u64s[2];
  92         u16                     whiteout_u64s;
  93         u16                     uncompacted_whiteout_u64s;
  94         u8                      page_order;
  95         u8                      unpack_fn_len;
  96
  97         /*
  98          * XXX: add a delete sequence number, so when btree_node_relock() fails
  99          * because the lock sequence number has changed - i.e. the contents were
 100          * modified - we can still relock the node if it's still the one we
 101          * want, without redoing the traversal
 102          */
 103
 104         /*
 105          * For asynchronous splits/interior node updates:
 106          * When we do a split, we allocate new child nodes and update the parent
 107          * node to point to them: we update the parent in memory immediately,
 108          * but then we must wait until the children have been written out before
 109          * the update to the parent can be written - this is a list of the
 110          * btree_interior_updates that are blocking this node from being
 111          * written:
 112          */
 113         struct list_head        write_blocked;
 114
 115         struct open_bucket      *ob;
 116
 117         /* lru list */
 118         struct list_head        list;
 119
 120         struct btree_write      writes[2];
 121
 122 #ifdef CONFIG_BCACHE_DEBUG
 123         bool                    *expensive_debug_checks;
 124 #endif
 125 };
 126
 127 #define BTREE_FLAG(flag)                                                \
 128 static inline bool btree_node_ ## flag(struct btree *b)                 \
 129 {       return test_bit(BTREE_NODE_ ## flag, &b->flags); }              \
 130                                                                         \
 131 static inline void set_btree_node_ ## flag(struct btree *b)             \
 132 {       set_bit(BTREE_NODE_ ## flag, &b->flags); }                      \
 133                                                                         \
 134 static inline void clear_btree_node_ ## flag(struct btree *b)           \
 135 {       clear_bit(BTREE_NODE_ ## flag, &b->flags); }
 136
 137 enum btree_flags {
 138         BTREE_NODE_read_error,
 139         BTREE_NODE_write_error,
 140         BTREE_NODE_dirty,
 141         BTREE_NODE_write_idx,
 142         BTREE_NODE_accessed,
 143         BTREE_NODE_write_in_flight,
 144         BTREE_NODE_just_written,
 145 };
 146
 147 BTREE_FLAG(read_error);
 148 BTREE_FLAG(write_error);
 149 BTREE_FLAG(dirty);
 150 BTREE_FLAG(write_idx);
 151 BTREE_FLAG(accessed);
 152 BTREE_FLAG(write_in_flight);
 153 BTREE_FLAG(just_written);
 154
 155 static inline struct btree_write *btree_current_write(struct btree *b)
 156 {
 157         return b->writes + btree_node_write_idx(b);
 158 }
 159
 160 static inline struct btree_write *btree_prev_write(struct btree *b)
 161 {
 162         return b->writes + (btree_node_write_idx(b) ^ 1);
 163 }
 164
 165 static inline struct bset_tree *bset_tree_last(struct btree *b)
 166 {
 167         EBUG_ON(!b->nsets);
 168         return b->set + b->nsets - 1;
 169 }
 170
 171 static inline struct bset *bset(const struct btree *b,
 172                                 const struct bset_tree *t)
 173 {
 174         return (void *) b->data + t->data_offset * sizeof(u64);
 175 }
 176
 177 static inline struct bset *btree_bset_first(struct btree *b)
 178 {
 179         return bset(b, b->set);
 180 }
 181
 182 static inline struct bset *btree_bset_last(struct btree *b)
 183 {
 184         return bset(b, bset_tree_last(b));
 185 }
 186
 187 static inline u16
 188 __btree_node_key_to_offset(const struct btree *b, const struct bkey_packed *k)
 189 {
 190         size_t ret = (u64 *) k - (u64 *) b->data - 1;
 191
 192         EBUG_ON(ret > U16_MAX);
 193         return ret;
 194 }
 195
 196 static inline struct bkey_packed *
 197 __btree_node_offset_to_key(const struct btree *b, u16 k)
 198 {
 199         return (void *) ((u64 *) b->data + k + 1);
 200 }
 201
 202 #define __bkey_idx(_set, _offset)                               \
 203         ((_set)->_data + (_offset))
 204
 205 #define bkey_idx(_set, _offset)                                 \
 206         ((typeof(&(_set)->start[0])) __bkey_idx((_set), (_offset)))
 207
 208 #define __bset_bkey_last(_set)                                  \
 209          __bkey_idx((_set), (_set)->u64s)
 210
 211 #define bset_bkey_last(_set)                                    \
 212          bkey_idx((_set), le16_to_cpu((_set)->u64s))
 213
 214 #define btree_bkey_first(_b, _t)        (bset(_b, _t)->start)
 215
 216 #define btree_bkey_last(_b, _t)                                         \
 217 ({                                                                      \
 218         EBUG_ON(__btree_node_offset_to_key(_b, (_t)->end_offset) !=     \
 219                 bset_bkey_last(bset(_b, _t)));                          \
 220                                                                         \
 221         __btree_node_offset_to_key(_b, (_t)->end_offset);               \
 222 })
 223
 224 static inline void set_btree_bset_end(struct btree *b, struct bset_tree *t)
 225 {
 226         t->end_offset =
 227                 __btree_node_key_to_offset(b, bset_bkey_last(bset(b, t)));
 228         btree_bkey_last(b, t);
 229 }
 230
 231 static inline void set_btree_bset(struct btree *b, struct bset_tree *t,
 232                                   const struct bset *i)
 233 {
 234         t->data_offset = (u64 *) i - (u64 *) b->data;
 235
 236         EBUG_ON(bset(b, t) != i);
 237
 238         set_btree_bset_end(b, t);
 239 }
 240
 241 static inline unsigned bset_byte_offset(struct btree *b, void *i)
 242 {
 243         return i - (void *) b->data;
 244 }
 245
 246 /* Type of keys @b contains: */
 247 static inline enum bkey_type btree_node_type(struct btree *b)
 248 {
 249         return b->level ? BKEY_TYPE_BTREE : b->btree_id;
 250 }
 251
 252 static inline const struct bkey_ops *btree_node_ops(struct btree *b)
 253 {
 254         return bch_bkey_ops[btree_node_type(b)];
 255 }
 256
 257 static inline bool btree_node_has_ptrs(struct btree *b)
 258 {
 259         return btree_type_has_ptrs(btree_node_type(b));
 260 }
 261
 262 static inline bool btree_node_is_extents(struct btree *b)
 263 {
 264         return btree_node_type(b) == BKEY_TYPE_EXTENTS;
 265 }
 266
 267 struct btree_root {
 268         struct btree            *b;
 269
 270         struct btree_interior_update *as;
 271
 272         /* On disk root - see async splits: */
 273         __BKEY_PADDED(key, BKEY_BTREE_PTR_VAL_U64s_MAX);
 274         u8                      level;
 275         u8                      alive;
 276 };
 277
 278 /*
 279  * Optional hook that will be called just prior to a btree node update, when
 280  * we're holding the write lock and we know what key is about to be overwritten:
 281  */
 282
 283 struct btree_iter;
 284 struct bucket_stats_cache_set;
 285 struct btree_node_iter;
 286
 287 enum extent_insert_hook_ret {
 288         BTREE_HOOK_DO_INSERT,
 289         BTREE_HOOK_NO_INSERT,
 290         BTREE_HOOK_RESTART_TRANS,
 291 };
 292
 293 struct extent_insert_hook {
 294         enum extent_insert_hook_ret
 295         (*fn)(struct extent_insert_hook *, struct bpos, struct bpos,
 296               struct bkey_s_c, const struct bkey_i *);
 297 };
 298
 299 enum btree_insert_ret {
 300         BTREE_INSERT_OK,
 301         /* extent spanned multiple leaf nodes: have to traverse to next node: */
 302         BTREE_INSERT_NEED_TRAVERSE,
 303         /* write lock held for too long */
 304         BTREE_INSERT_NEED_RESCHED,
 305         /* leaf node needs to be split */
 306         BTREE_INSERT_BTREE_NODE_FULL,
 307         BTREE_INSERT_JOURNAL_RES_FULL,
 308         BTREE_INSERT_ENOSPC,
 309         BTREE_INSERT_NEED_GC_LOCK,
 310 };
 311
 312 enum btree_gc_coalesce_fail_reason {
 313         BTREE_GC_COALESCE_FAIL_RESERVE_GET,
 314         BTREE_GC_COALESCE_FAIL_KEYLIST_REALLOC,
 315         BTREE_GC_COALESCE_FAIL_FORMAT_FITS,
 316 };
 317
 318 typedef struct btree_nr_keys (*sort_fix_overlapping_fn)(struct bset *,
 319                                                         struct btree *,
 320                                                         struct btree_node_iter *);
 321
 322 #endif /* _BCACHE_BTREE_TYPES_H */