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

   1 /* SPDX-License-Identifier: GPL-2.0 */
   2 #ifndef _BCACHEFS_BTREE_TYPES_H
   3 #define _BCACHEFS_BTREE_TYPES_H
   4
   5 #include <linux/list.h>
   6 #include <linux/rhashtable.h>
   7 #include <linux/six.h>
   8
   9 #include "bkey_methods.h"
  10 #include "buckets_types.h"
  11 #include "journal_types.h"
  12
  13 struct open_bucket;
  14 struct btree_update;
  15 struct btree_trans;
  16
  17 #define MAX_BSETS               3U
  18
  19 struct btree_nr_keys {
  20
  21         /*
  22          * Amount of live metadata (i.e. size of node after a compaction) in
  23          * units of u64s
  24          */
  25         u16                     live_u64s;
  26         u16                     bset_u64s[MAX_BSETS];
  27
  28         /* live keys only: */
  29         u16                     packed_keys;
  30         u16                     unpacked_keys;
  31 };
  32
  33 struct bset_tree {
  34         /*
  35          * We construct a binary tree in an array as if the array
  36          * started at 1, so that things line up on the same cachelines
  37          * better: see comments in bset.c at cacheline_to_bkey() for
  38          * details
  39          */
  40
  41         /* size of the binary tree and prev array */
  42         u16                     size;
  43
  44         /* function of size - precalculated for to_inorder() */
  45         u16                     extra;
  46
  47         u16                     data_offset;
  48         u16                     aux_data_offset;
  49         u16                     end_offset;
  50
  51         struct bpos             max_key;
  52 };
  53
  54 struct btree_write {
  55         struct journal_entry_pin        journal;
  56 };
  57
  58 struct btree_alloc {
  59         struct open_buckets     ob;
  60         BKEY_PADDED(k);
  61 };
  62
  63 struct btree_bkey_cached_common {
  64         struct six_lock         lock;
  65         u8                      level;
  66         u8                      btree_id;
  67 };
  68
  69 struct btree {
  70         struct btree_bkey_cached_common c;
  71
  72         struct rhash_head       hash;
  73         u64                     hash_val;
  74
  75         unsigned long           flags;
  76         u16                     written;
  77         u8                      nsets;
  78         u8                      nr_key_bits;
  79
  80         struct bkey_format      format;
  81
  82         struct btree_node       *data;
  83         void                    *aux_data;
  84
  85         /*
  86          * Sets of sorted keys - the real btree node - plus a binary search tree
  87          *
  88          * set[0] is special; set[0]->tree, set[0]->prev and set[0]->data point
  89          * to the memory we have allocated for this btree node. Additionally,
  90          * set[0]->data points to the entire btree node as it exists on disk.
  91          */
  92         struct bset_tree        set[MAX_BSETS];
  93
  94         struct btree_nr_keys    nr;
  95         u16                     sib_u64s[2];
  96         u16                     whiteout_u64s;
  97         u8                      page_order;
  98         u8                      unpack_fn_len;
  99
 100         /*
 101          * XXX: add a delete sequence number, so when bch2_btree_node_relock()
 102          * fails because the lock sequence number has changed - i.e. the
 103          * contents were modified - we can still relock the node if it's still
 104          * the one we want, without redoing the traversal
 105          */
 106
 107         /*
 108          * For asynchronous splits/interior node updates:
 109          * When we do a split, we allocate new child nodes and update the parent
 110          * node to point to them: we update the parent in memory immediately,
 111          * but then we must wait until the children have been written out before
 112          * the update to the parent can be written - this is a list of the
 113          * btree_updates that are blocking this node from being
 114          * written:
 115          */
 116         struct list_head        write_blocked;
 117
 118         /*
 119          * Also for asynchronous splits/interior node updates:
 120          * If a btree node isn't reachable yet, we don't want to kick off
 121          * another write - because that write also won't yet be reachable and
 122          * marking it as completed before it's reachable would be incorrect:
 123          */
 124         unsigned long           will_make_reachable;
 125
 126         struct open_buckets     ob;
 127
 128         /* lru list */
 129         struct list_head        list;
 130
 131         struct btree_write      writes[2];
 132
 133 #ifdef CONFIG_BCACHEFS_DEBUG
 134         bool                    *expensive_debug_checks;
 135 #endif
 136
 137         /* Key/pointer for this btree node */
 138         __BKEY_PADDED(key, BKEY_BTREE_PTR_VAL_U64s_MAX);
 139 };
 140
 141 struct btree_cache {
 142         struct rhashtable       table;
 143         bool                    table_init_done;
 144         /*
 145          * We never free a struct btree, except on shutdown - we just put it on
 146          * the btree_cache_freed list and reuse it later. This simplifies the
 147          * code, and it doesn't cost us much memory as the memory usage is
 148          * dominated by buffers that hold the actual btree node data and those
 149          * can be freed - and the number of struct btrees allocated is
 150          * effectively bounded.
 151          *
 152          * btree_cache_freeable effectively is a small cache - we use it because
 153          * high order page allocations can be rather expensive, and it's quite
 154          * common to delete and allocate btree nodes in quick succession. It
 155          * should never grow past ~2-3 nodes in practice.
 156          */
 157         struct mutex            lock;
 158         struct list_head        live;
 159         struct list_head        freeable;
 160         struct list_head        freed;
 161
 162         /* Number of elements in live + freeable lists */
 163         unsigned                used;
 164         unsigned                reserve;
 165         struct shrinker         shrink;
 166
 167         /*
 168          * If we need to allocate memory for a new btree node and that
 169          * allocation fails, we can cannibalize another node in the btree cache
 170          * to satisfy the allocation - lock to guarantee only one thread does
 171          * this at a time:
 172          */
 173         struct task_struct      *alloc_lock;
 174         struct closure_waitlist alloc_wait;
 175 };
 176
 177 struct btree_node_iter {
 178         struct btree_node_iter_set {
 179                 u16     k, end;
 180         } data[MAX_BSETS];
 181 };
 182
 183 enum btree_iter_type {
 184         BTREE_ITER_KEYS,
 185         BTREE_ITER_NODES,
 186         BTREE_ITER_CACHED,
 187 };
 188
 189 #define BTREE_ITER_TYPE                 ((1 << 2) - 1)
 190
 191 /*
 192  * Iterate over all possible positions, synthesizing deleted keys for holes:
 193  */
 194 #define BTREE_ITER_SLOTS                (1 << 2)
 195 /*
 196  * Indicates that intent locks should be taken on leaf nodes, because we expect
 197  * to be doing updates:
 198  */
 199 #define BTREE_ITER_INTENT               (1 << 3)
 200 /*
 201  * Causes the btree iterator code to prefetch additional btree nodes from disk:
 202  */
 203 #define BTREE_ITER_PREFETCH             (1 << 4)
 204 /*
 205  * Indicates that this iterator should not be reused until transaction commit,
 206  * either because a pending update references it or because the update depends
 207  * on that particular key being locked (e.g. by the str_hash code, for hash
 208  * table consistency)
 209  */
 210 #define BTREE_ITER_KEEP_UNTIL_COMMIT    (1 << 5)
 211 /*
 212  * Used in bch2_btree_iter_traverse(), to indicate whether we're searching for
 213  * @pos or the first key strictly greater than @pos
 214  */
 215 #define BTREE_ITER_IS_EXTENTS           (1 << 6)
 216 #define BTREE_ITER_ERROR                (1 << 7)
 217 #define BTREE_ITER_SET_POS_AFTER_COMMIT (1 << 8)
 218 #define BTREE_ITER_CACHED_NOFILL        (1 << 9)
 219 #define BTREE_ITER_CACHED_NOCREATE      (1 << 10)
 220
 221 #define BTREE_ITER_USER_FLAGS                           \
 222         (BTREE_ITER_SLOTS                               \
 223         |BTREE_ITER_INTENT                              \
 224         |BTREE_ITER_PREFETCH                            \
 225         |BTREE_ITER_CACHED_NOFILL                       \
 226         |BTREE_ITER_CACHED_NOCREATE)
 227
 228 enum btree_iter_uptodate {
 229         BTREE_ITER_UPTODATE             = 0,
 230         BTREE_ITER_NEED_PEEK            = 1,
 231         BTREE_ITER_NEED_RELOCK          = 2,
 232         BTREE_ITER_NEED_TRAVERSE        = 3,
 233 };
 234
 235 #define BTREE_ITER_NO_NODE_GET_LOCKS    ((struct btree *) 1)
 236 #define BTREE_ITER_NO_NODE_DROP         ((struct btree *) 2)
 237 #define BTREE_ITER_NO_NODE_LOCK_ROOT    ((struct btree *) 3)
 238 #define BTREE_ITER_NO_NODE_UP           ((struct btree *) 4)
 239 #define BTREE_ITER_NO_NODE_DOWN         ((struct btree *) 5)
 240 #define BTREE_ITER_NO_NODE_INIT         ((struct btree *) 6)
 241 #define BTREE_ITER_NO_NODE_ERROR        ((struct btree *) 7)
 242
 243 /*
 244  * @pos                 - iterator's current position
 245  * @level               - current btree depth
 246  * @locks_want          - btree level below which we start taking intent locks
 247  * @nodes_locked        - bitmask indicating which nodes in @nodes are locked
 248  * @nodes_intent_locked - bitmask indicating which locks are intent locks
 249  */
 250 struct btree_iter {
 251         struct btree_trans      *trans;
 252         struct bpos             pos;
 253         struct bpos             pos_after_commit;
 254
 255         u16                     flags;
 256         u8                      idx;
 257
 258         enum btree_id           btree_id:4;
 259         enum btree_iter_uptodate uptodate:4;
 260         unsigned                level:4,
 261                                 min_depth:4,
 262                                 locks_want:4,
 263                                 nodes_locked:4,
 264                                 nodes_intent_locked:4;
 265
 266         struct btree_iter_level {
 267                 struct btree    *b;
 268                 struct btree_node_iter iter;
 269                 u32             lock_seq;
 270         }                       l[BTREE_MAX_DEPTH];
 271
 272         /*
 273          * Current unpacked key - so that bch2_btree_iter_next()/
 274          * bch2_btree_iter_next_slot() can correctly advance pos.
 275          */
 276         struct bkey             k;
 277         unsigned long           ip_allocated;
 278 };
 279
 280 static inline enum btree_iter_type
 281 btree_iter_type(const struct btree_iter *iter)
 282 {
 283         return iter->flags & BTREE_ITER_TYPE;
 284 }
 285
 286 static inline struct btree_iter_level *iter_l(struct btree_iter *iter)
 287 {
 288         return iter->l + iter->level;
 289 }
 290
 291 struct btree_key_cache {
 292         struct mutex            lock;
 293         struct rhashtable       table;
 294         struct list_head        freed;
 295         struct list_head        clean;
 296 };
 297
 298 struct bkey_cached_key {
 299         u32                     btree_id;
 300         struct bpos             pos;
 301 } __packed;
 302
 303 #define BKEY_CACHED_DIRTY               0
 304
 305 struct bkey_cached {
 306         struct btree_bkey_cached_common c;
 307
 308         unsigned long           flags;
 309         u8                      u64s;
 310         bool                    valid;
 311         struct bkey_cached_key  key;
 312
 313         struct rhash_head       hash;
 314         struct list_head        list;
 315
 316         struct journal_preres   res;
 317         struct journal_entry_pin journal;
 318
 319         struct bkey_i           *k;
 320 };
 321
 322 struct btree_insert_entry {
 323         unsigned                trigger_flags;
 324         unsigned                trans_triggers_run:1;
 325         struct bkey_i           *k;
 326         struct btree_iter       *iter;
 327 };
 328
 329 #ifndef CONFIG_LOCKDEP
 330 #define BTREE_ITER_MAX          64
 331 #else
 332 #define BTREE_ITER_MAX          32
 333 #endif
 334
 335 struct btree_trans {
 336         struct bch_fs           *c;
 337 #ifdef CONFIG_BCACHEFS_DEBUG
 338         struct list_head        list;
 339         struct btree            *locking;
 340         unsigned                locking_iter_idx;
 341         struct bpos             locking_pos;
 342         u8                      locking_btree_id;
 343         u8                      locking_level;
 344         pid_t                   pid;
 345 #endif
 346         unsigned long           ip;
 347
 348         u64                     iters_linked;
 349         u64                     iters_live;
 350         u64                     iters_touched;
 351
 352         u8                      nr_iters;
 353         u8                      nr_updates;
 354         u8                      nr_updates2;
 355         u8                      size;
 356         unsigned                used_mempool:1;
 357         unsigned                error:1;
 358         unsigned                nounlock:1;
 359         unsigned                need_reset:1;
 360         unsigned                in_traverse_all:1;
 361
 362         unsigned                mem_top;
 363         unsigned                mem_bytes;
 364         void                    *mem;
 365
 366         struct btree_iter       *iters;
 367         struct btree_insert_entry *updates;
 368         struct btree_insert_entry *updates2;
 369
 370         /* update path: */
 371         struct jset_entry       *extra_journal_entries;
 372         unsigned                extra_journal_entry_u64s;
 373         struct journal_entry_pin *journal_pin;
 374
 375         struct journal_res      journal_res;
 376         struct journal_preres   journal_preres;
 377         u64                     *journal_seq;
 378         struct disk_reservation *disk_res;
 379         unsigned                flags;
 380         unsigned                journal_u64s;
 381         unsigned                journal_preres_u64s;
 382         struct replicas_delta_list *fs_usage_deltas;
 383
 384         struct btree_iter       iters_onstack[2];
 385         struct btree_insert_entry updates_onstack[2];
 386         struct btree_insert_entry updates2_onstack[2];
 387 };
 388
 389 #define BTREE_FLAG(flag)                                                \
 390 static inline bool btree_node_ ## flag(struct btree *b)                 \
 391 {       return test_bit(BTREE_NODE_ ## flag, &b->flags); }              \
 392                                                                         \
 393 static inline void set_btree_node_ ## flag(struct btree *b)             \
 394 {       set_bit(BTREE_NODE_ ## flag, &b->flags); }                      \
 395                                                                         \
 396 static inline void clear_btree_node_ ## flag(struct btree *b)           \
 397 {       clear_bit(BTREE_NODE_ ## flag, &b->flags); }
 398
 399 enum btree_flags {
 400         BTREE_NODE_read_in_flight,
 401         BTREE_NODE_read_error,
 402         BTREE_NODE_dirty,
 403         BTREE_NODE_need_write,
 404         BTREE_NODE_noevict,
 405         BTREE_NODE_write_idx,
 406         BTREE_NODE_accessed,
 407         BTREE_NODE_write_in_flight,
 408         BTREE_NODE_just_written,
 409         BTREE_NODE_dying,
 410         BTREE_NODE_fake,
 411         BTREE_NODE_old_extent_overwrite,
 412 };
 413
 414 BTREE_FLAG(read_in_flight);
 415 BTREE_FLAG(read_error);
 416 BTREE_FLAG(dirty);
 417 BTREE_FLAG(need_write);
 418 BTREE_FLAG(noevict);
 419 BTREE_FLAG(write_idx);
 420 BTREE_FLAG(accessed);
 421 BTREE_FLAG(write_in_flight);
 422 BTREE_FLAG(just_written);
 423 BTREE_FLAG(dying);
 424 BTREE_FLAG(fake);
 425 BTREE_FLAG(old_extent_overwrite);
 426
 427 static inline struct btree_write *btree_current_write(struct btree *b)
 428 {
 429         return b->writes + btree_node_write_idx(b);
 430 }
 431
 432 static inline struct btree_write *btree_prev_write(struct btree *b)
 433 {
 434         return b->writes + (btree_node_write_idx(b) ^ 1);
 435 }
 436
 437 static inline struct bset_tree *bset_tree_last(struct btree *b)
 438 {
 439         EBUG_ON(!b->nsets);
 440         return b->set + b->nsets - 1;
 441 }
 442
 443 static inline void *
 444 __btree_node_offset_to_ptr(const struct btree *b, u16 offset)
 445 {
 446         return (void *) ((u64 *) b->data + 1 + offset);
 447 }
 448
 449 static inline u16
 450 __btree_node_ptr_to_offset(const struct btree *b, const void *p)
 451 {
 452         u16 ret = (u64 *) p - 1 - (u64 *) b->data;
 453
 454         EBUG_ON(__btree_node_offset_to_ptr(b, ret) != p);
 455         return ret;
 456 }
 457
 458 static inline struct bset *bset(const struct btree *b,
 459                                 const struct bset_tree *t)
 460 {
 461         return __btree_node_offset_to_ptr(b, t->data_offset);
 462 }
 463
 464 static inline void set_btree_bset_end(struct btree *b, struct bset_tree *t)
 465 {
 466         t->end_offset =
 467                 __btree_node_ptr_to_offset(b, vstruct_last(bset(b, t)));
 468 }
 469
 470 static inline void set_btree_bset(struct btree *b, struct bset_tree *t,
 471                                   const struct bset *i)
 472 {
 473         t->data_offset = __btree_node_ptr_to_offset(b, i);
 474         set_btree_bset_end(b, t);
 475 }
 476
 477 static inline struct bset *btree_bset_first(struct btree *b)
 478 {
 479         return bset(b, b->set);
 480 }
 481
 482 static inline struct bset *btree_bset_last(struct btree *b)
 483 {
 484         return bset(b, bset_tree_last(b));
 485 }
 486
 487 static inline u16
 488 __btree_node_key_to_offset(const struct btree *b, const struct bkey_packed *k)
 489 {
 490         return __btree_node_ptr_to_offset(b, k);
 491 }
 492
 493 static inline struct bkey_packed *
 494 __btree_node_offset_to_key(const struct btree *b, u16 k)
 495 {
 496         return __btree_node_offset_to_ptr(b, k);
 497 }
 498
 499 static inline unsigned btree_bkey_first_offset(const struct bset_tree *t)
 500 {
 501         return t->data_offset + offsetof(struct bset, _data) / sizeof(u64);
 502 }
 503
 504 #define btree_bkey_first(_b, _t)                                        \
 505 ({                                                                      \
 506         EBUG_ON(bset(_b, _t)->start !=                                  \
 507                 __btree_node_offset_to_key(_b, btree_bkey_first_offset(_t)));\
 508                                                                         \
 509         bset(_b, _t)->start;                                            \
 510 })
 511
 512 #define btree_bkey_last(_b, _t)                                         \
 513 ({                                                                      \
 514         EBUG_ON(__btree_node_offset_to_key(_b, (_t)->end_offset) !=     \
 515                 vstruct_last(bset(_b, _t)));                            \
 516                                                                         \
 517         __btree_node_offset_to_key(_b, (_t)->end_offset);               \
 518 })
 519
 520 static inline unsigned bset_u64s(struct bset_tree *t)
 521 {
 522         return t->end_offset - t->data_offset -
 523                 sizeof(struct bset) / sizeof(u64);
 524 }
 525
 526 static inline unsigned bset_dead_u64s(struct btree *b, struct bset_tree *t)
 527 {
 528         return bset_u64s(t) - b->nr.bset_u64s[t - b->set];
 529 }
 530
 531 static inline unsigned bset_byte_offset(struct btree *b, void *i)
 532 {
 533         return i - (void *) b->data;
 534 }
 535
 536 enum btree_node_type {
 537 #define x(kwd, val, name) BKEY_TYPE_##kwd = val,
 538         BCH_BTREE_IDS()
 539 #undef x
 540         BKEY_TYPE_BTREE,
 541 };
 542
 543 /* Type of a key in btree @id at level @level: */
 544 static inline enum btree_node_type __btree_node_type(unsigned level, enum btree_id id)
 545 {
 546         return level ? BKEY_TYPE_BTREE : (enum btree_node_type) id;
 547 }
 548
 549 /* Type of keys @b contains: */
 550 static inline enum btree_node_type btree_node_type(struct btree *b)
 551 {
 552         return __btree_node_type(b->c.level, b->c.btree_id);
 553 }
 554
 555 static inline bool btree_node_type_is_extents(enum btree_node_type type)
 556 {
 557         switch (type) {
 558         case BKEY_TYPE_EXTENTS:
 559         case BKEY_TYPE_REFLINK:
 560                 return true;
 561         default:
 562                 return false;
 563         }
 564 }
 565
 566 static inline bool btree_node_is_extents(struct btree *b)
 567 {
 568         return btree_node_type_is_extents(btree_node_type(b));
 569 }
 570
 571 #define BTREE_NODE_TYPE_HAS_TRIGGERS                    \
 572         ((1U << BKEY_TYPE_EXTENTS)|                     \
 573          (1U << BKEY_TYPE_ALLOC)|                       \
 574          (1U << BKEY_TYPE_INODES)|                      \
 575          (1U << BKEY_TYPE_REFLINK)|                     \
 576          (1U << BKEY_TYPE_EC)|                          \
 577          (1U << BKEY_TYPE_BTREE))
 578
 579 #define BTREE_NODE_TYPE_HAS_TRANS_TRIGGERS              \
 580         ((1U << BKEY_TYPE_EXTENTS)|                     \
 581          (1U << BKEY_TYPE_INODES)|                      \
 582          (1U << BKEY_TYPE_REFLINK))
 583
 584 enum btree_trigger_flags {
 585         __BTREE_TRIGGER_NORUN,          /* Don't run triggers at all */
 586         __BTREE_TRIGGER_NOOVERWRITES,   /* Don't run triggers on overwrites */
 587
 588         __BTREE_TRIGGER_INSERT,
 589         __BTREE_TRIGGER_OVERWRITE,
 590         __BTREE_TRIGGER_OVERWRITE_SPLIT,
 591
 592         __BTREE_TRIGGER_GC,
 593         __BTREE_TRIGGER_BUCKET_INVALIDATE,
 594         __BTREE_TRIGGER_ALLOC_READ,
 595         __BTREE_TRIGGER_NOATOMIC,
 596 };
 597
 598 #define BTREE_TRIGGER_NORUN             (1U << __BTREE_TRIGGER_NORUN)
 599 #define BTREE_TRIGGER_NOOVERWRITES      (1U << __BTREE_TRIGGER_NOOVERWRITES)
 600
 601 #define BTREE_TRIGGER_INSERT            (1U << __BTREE_TRIGGER_INSERT)
 602 #define BTREE_TRIGGER_OVERWRITE         (1U << __BTREE_TRIGGER_OVERWRITE)
 603 #define BTREE_TRIGGER_OVERWRITE_SPLIT   (1U << __BTREE_TRIGGER_OVERWRITE_SPLIT)
 604
 605 #define BTREE_TRIGGER_GC                (1U << __BTREE_TRIGGER_GC)
 606 #define BTREE_TRIGGER_BUCKET_INVALIDATE (1U << __BTREE_TRIGGER_BUCKET_INVALIDATE)
 607 #define BTREE_TRIGGER_ALLOC_READ        (1U << __BTREE_TRIGGER_ALLOC_READ)
 608 #define BTREE_TRIGGER_NOATOMIC          (1U << __BTREE_TRIGGER_NOATOMIC)
 609
 610 static inline bool btree_node_type_needs_gc(enum btree_node_type type)
 611 {
 612         return BTREE_NODE_TYPE_HAS_TRIGGERS & (1U << type);
 613 }
 614
 615 struct btree_root {
 616         struct btree            *b;
 617
 618         /* On disk root - see async splits: */
 619         __BKEY_PADDED(key, BKEY_BTREE_PTR_VAL_U64s_MAX);
 620         u8                      level;
 621         u8                      alive;
 622         s8                      error;
 623 };
 624
 625 /*
 626  * Optional hook that will be called just prior to a btree node update, when
 627  * we're holding the write lock and we know what key is about to be overwritten:
 628  */
 629
 630 enum btree_insert_ret {
 631         BTREE_INSERT_OK,
 632         /* leaf node needs to be split */
 633         BTREE_INSERT_BTREE_NODE_FULL,
 634         BTREE_INSERT_ENOSPC,
 635         BTREE_INSERT_NEED_MARK_REPLICAS,
 636         BTREE_INSERT_NEED_JOURNAL_RES,
 637 };
 638
 639 enum btree_gc_coalesce_fail_reason {
 640         BTREE_GC_COALESCE_FAIL_RESERVE_GET,
 641         BTREE_GC_COALESCE_FAIL_KEYLIST_REALLOC,
 642         BTREE_GC_COALESCE_FAIL_FORMAT_FITS,
 643 };
 644
 645 enum btree_node_sibling {
 646         btree_prev_sib,
 647         btree_next_sib,
 648 };
 649
 650 typedef struct btree_nr_keys (*sort_fix_overlapping_fn)(struct bset *,
 651                                                         struct btree *,
 652                                                         struct btree_node_iter *);
 653
 654 #endif /* _BCACHEFS_BTREE_TYPES_H */