1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _BCACHEFS_BTREE_TYPES_H
3 #define _BCACHEFS_BTREE_TYPES_H
5 #include <linux/list.h>
6 #include <linux/rhashtable.h>
9 #include "bkey_methods.h"
10 #include "buckets_types.h"
12 #include "journal_types.h"
20 struct btree_nr_keys {
23 * Amount of live metadata (i.e. size of node after a compaction) in
27 u16 bset_u64s[MAX_BSETS];
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
42 /* size of the binary tree and prev array */
45 /* function of size - precalculated for to_inorder() */
54 struct journal_entry_pin journal;
58 struct open_buckets ob;
59 __BKEY_PADDED(k, BKEY_BTREE_PTR_VAL_U64s_MAX);
62 struct btree_bkey_cached_common {
69 struct btree_bkey_cached_common c;
71 struct rhash_head hash;
80 struct bkey_format format;
82 struct btree_node *data;
86 * Sets of sorted keys - the real btree node - plus a binary search tree
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.
92 struct bset_tree set[MAX_BSETS];
94 struct btree_nr_keys nr;
100 struct btree_write writes[2];
102 /* Key/pointer for this btree node */
103 __BKEY_PADDED(key, BKEY_BTREE_PTR_VAL_U64s_MAX);
106 * XXX: add a delete sequence number, so when bch2_btree_node_relock()
107 * fails because the lock sequence number has changed - i.e. the
108 * contents were modified - we can still relock the node if it's still
109 * the one we want, without redoing the traversal
113 * For asynchronous splits/interior node updates:
114 * When we do a split, we allocate new child nodes and update the parent
115 * node to point to them: we update the parent in memory immediately,
116 * but then we must wait until the children have been written out before
117 * the update to the parent can be written - this is a list of the
118 * btree_updates that are blocking this node from being
121 struct list_head write_blocked;
124 * Also for asynchronous splits/interior node updates:
125 * If a btree node isn't reachable yet, we don't want to kick off
126 * another write - because that write also won't yet be reachable and
127 * marking it as completed before it's reachable would be incorrect:
129 unsigned long will_make_reachable;
131 struct open_buckets ob;
134 struct list_head list;
138 struct rhashtable table;
139 bool table_init_done;
141 * We never free a struct btree, except on shutdown - we just put it on
142 * the btree_cache_freed list and reuse it later. This simplifies the
143 * code, and it doesn't cost us much memory as the memory usage is
144 * dominated by buffers that hold the actual btree node data and those
145 * can be freed - and the number of struct btrees allocated is
146 * effectively bounded.
148 * btree_cache_freeable effectively is a small cache - we use it because
149 * high order page allocations can be rather expensive, and it's quite
150 * common to delete and allocate btree nodes in quick succession. It
151 * should never grow past ~2-3 nodes in practice.
154 struct list_head live;
155 struct list_head freeable;
156 struct list_head freed_pcpu;
157 struct list_head freed_nonpcpu;
159 /* Number of elements in live + freeable lists */
163 struct shrinker shrink;
166 * If we need to allocate memory for a new btree node and that
167 * allocation fails, we can cannibalize another node in the btree cache
168 * to satisfy the allocation - lock to guarantee only one thread does
171 struct task_struct *alloc_lock;
172 struct closure_waitlist alloc_wait;
175 struct btree_node_iter {
176 struct btree_node_iter_set {
182 * Iterate over all possible positions, synthesizing deleted keys for holes:
184 #define BTREE_ITER_SLOTS (1 << 0)
185 #define BTREE_ITER_ALL_LEVELS (1 << 1)
187 * Indicates that intent locks should be taken on leaf nodes, because we expect
188 * to be doing updates:
190 #define BTREE_ITER_INTENT (1 << 2)
192 * Causes the btree iterator code to prefetch additional btree nodes from disk:
194 #define BTREE_ITER_PREFETCH (1 << 3)
196 * Used in bch2_btree_iter_traverse(), to indicate whether we're searching for
197 * @pos or the first key strictly greater than @pos
199 #define BTREE_ITER_IS_EXTENTS (1 << 4)
200 #define BTREE_ITER_NOT_EXTENTS (1 << 5)
201 #define BTREE_ITER_CACHED (1 << 6)
202 #define BTREE_ITER_CACHED_NOFILL (1 << 7)
203 #define BTREE_ITER_CACHED_NOCREATE (1 << 8)
204 #define BTREE_ITER_WITH_KEY_CACHE (1 << 9)
205 #define BTREE_ITER_WITH_UPDATES (1 << 10)
206 #define BTREE_ITER_WITH_JOURNAL (1 << 11)
207 #define __BTREE_ITER_ALL_SNAPSHOTS (1 << 12)
208 #define BTREE_ITER_ALL_SNAPSHOTS (1 << 13)
209 #define BTREE_ITER_FILTER_SNAPSHOTS (1 << 14)
210 #define BTREE_ITER_NOPRESERVE (1 << 15)
212 enum btree_path_uptodate {
213 BTREE_ITER_UPTODATE = 0,
214 BTREE_ITER_NEED_RELOCK = 1,
215 BTREE_ITER_NEED_TRAVERSE = 2,
218 #define BTREE_ITER_NO_NODE_GET_LOCKS ((struct btree *) 1)
219 #define BTREE_ITER_NO_NODE_DROP ((struct btree *) 2)
220 #define BTREE_ITER_NO_NODE_LOCK_ROOT ((struct btree *) 3)
221 #define BTREE_ITER_NO_NODE_UP ((struct btree *) 4)
222 #define BTREE_ITER_NO_NODE_DOWN ((struct btree *) 5)
223 #define BTREE_ITER_NO_NODE_INIT ((struct btree *) 6)
224 #define BTREE_ITER_NO_NODE_ERROR ((struct btree *) 7)
225 #define BTREE_ITER_NO_NODE_CACHED ((struct btree *) 8)
233 /* btree_iter_copy starts here: */
236 enum btree_id btree_id:4;
239 enum btree_path_uptodate uptodate:2;
241 * When true, failing to relock this path will cause the transaction to
244 bool should_be_locked:1;
248 nodes_intent_locked:4;
250 struct btree_path_level {
252 struct btree_node_iter iter;
254 } l[BTREE_MAX_DEPTH];
255 #ifdef CONFIG_BCACHEFS_DEBUG
256 unsigned long ip_allocated;
260 static inline struct btree_path_level *path_l(struct btree_path *path)
262 return path->l + path->level;
266 * @pos - iterator's current position
267 * @level - current btree depth
268 * @locks_want - btree level below which we start taking intent locks
269 * @nodes_locked - bitmask indicating which nodes in @nodes are locked
270 * @nodes_intent_locked - bitmask indicating which locks are intent locks
273 struct btree_trans *trans;
274 struct btree_path *path;
275 struct btree_path *update_path;
276 struct btree_path *key_cache_path;
278 enum btree_id btree_id:4;
279 unsigned min_depth:3;
282 /* btree_iter_copy starts here: */
285 /* When we're filtering by snapshot, the snapshot ID we're looking for: */
289 struct bpos pos_after_commit;
291 * Current unpacked key - so that bch2_btree_iter_next()/
292 * bch2_btree_iter_next_slot() can correctly advance pos.
296 /* BTREE_ITER_WITH_JOURNAL: */
298 struct bpos journal_pos;
299 #ifdef CONFIG_BCACHEFS_DEBUG
300 unsigned long ip_allocated;
304 struct btree_key_cache_freelist {
305 struct bkey_cached *objs[16];
309 struct btree_key_cache {
311 struct rhashtable table;
312 bool table_init_done;
313 struct list_head freed;
314 struct shrinker shrink;
315 unsigned shrink_iter;
316 struct btree_key_cache_freelist __percpu *pcpu_freed;
318 atomic_long_t nr_freed;
319 atomic_long_t nr_keys;
320 atomic_long_t nr_dirty;
323 struct bkey_cached_key {
326 } __attribute__((packed, aligned(4)));
328 #define BKEY_CACHED_ACCESSED 0
329 #define BKEY_CACHED_DIRTY 1
332 struct btree_bkey_cached_common c;
337 u32 btree_trans_barrier_seq;
338 struct bkey_cached_key key;
340 struct rhash_head hash;
341 struct list_head list;
343 struct journal_preres res;
344 struct journal_entry_pin journal;
349 struct btree_insert_entry {
352 enum btree_id btree_id:8;
355 bool insert_trigger_run:1;
356 bool overwrite_trigger_run:1;
357 bool key_cache_already_flushed:1;
359 * @old_k may be a key from the journal; @old_btree_u64s always refers
360 * to the size of the key being overwritten in the btree:
364 struct btree_path *path;
365 /* key being overwritten: */
367 const struct bch_val *old_v;
368 unsigned long ip_allocated;
371 #ifndef CONFIG_LOCKDEP
372 #define BTREE_ITER_MAX 64
374 #define BTREE_ITER_MAX 32
377 struct btree_trans_commit_hook;
378 typedef int (btree_trans_commit_hook_fn)(struct btree_trans *, struct btree_trans_commit_hook *);
380 struct btree_trans_commit_hook {
381 btree_trans_commit_hook_fn *fn;
382 struct btree_trans_commit_hook *next;
385 #define BTREE_TRANS_MEM_MAX (1U << 16)
387 #define BTREE_TRANS_MAX_LOCK_HOLD_TIME_NS 10000
392 struct list_head list;
394 struct btree *locking;
395 unsigned locking_path_idx;
396 struct bpos locking_pos;
399 u8 locking_lock_type;
400 struct task_struct *task;
407 bool in_traverse_all:1;
409 bool memory_allocation_failure:1;
410 bool is_initial_gc:1;
412 * For when bch2_trans_update notices we'll be splitting a compressed
415 unsigned extra_journal_res;
423 u8 sorted[BTREE_ITER_MAX];
424 struct btree_path *paths;
425 struct btree_insert_entry *updates;
428 struct btree_trans_commit_hook *hooks;
429 DARRAY(u64) extra_journal_entries;
430 struct journal_entry_pin *journal_pin;
432 struct journal_res journal_res;
433 struct journal_preres journal_preres;
435 struct disk_reservation *disk_res;
437 unsigned journal_u64s;
438 unsigned journal_preres_u64s;
439 struct replicas_delta_list *fs_usage_deltas;
442 #define BTREE_FLAGS() \
448 x(will_make_reachable) \
453 x(write_in_flight_inner) \
461 #define x(flag) BTREE_NODE_##flag,
467 static inline bool btree_node_ ## flag(struct btree *b) \
468 { return test_bit(BTREE_NODE_ ## flag, &b->flags); } \
470 static inline void set_btree_node_ ## flag(struct btree *b) \
471 { set_bit(BTREE_NODE_ ## flag, &b->flags); } \
473 static inline void clear_btree_node_ ## flag(struct btree *b) \
474 { clear_bit(BTREE_NODE_ ## flag, &b->flags); }
479 static inline struct btree_write *btree_current_write(struct btree *b)
481 return b->writes + btree_node_write_idx(b);
484 static inline struct btree_write *btree_prev_write(struct btree *b)
486 return b->writes + (btree_node_write_idx(b) ^ 1);
489 static inline struct bset_tree *bset_tree_last(struct btree *b)
492 return b->set + b->nsets - 1;
496 __btree_node_offset_to_ptr(const struct btree *b, u16 offset)
498 return (void *) ((u64 *) b->data + 1 + offset);
502 __btree_node_ptr_to_offset(const struct btree *b, const void *p)
504 u16 ret = (u64 *) p - 1 - (u64 *) b->data;
506 EBUG_ON(__btree_node_offset_to_ptr(b, ret) != p);
510 static inline struct bset *bset(const struct btree *b,
511 const struct bset_tree *t)
513 return __btree_node_offset_to_ptr(b, t->data_offset);
516 static inline void set_btree_bset_end(struct btree *b, struct bset_tree *t)
519 __btree_node_ptr_to_offset(b, vstruct_last(bset(b, t)));
522 static inline void set_btree_bset(struct btree *b, struct bset_tree *t,
523 const struct bset *i)
525 t->data_offset = __btree_node_ptr_to_offset(b, i);
526 set_btree_bset_end(b, t);
529 static inline struct bset *btree_bset_first(struct btree *b)
531 return bset(b, b->set);
534 static inline struct bset *btree_bset_last(struct btree *b)
536 return bset(b, bset_tree_last(b));
540 __btree_node_key_to_offset(const struct btree *b, const struct bkey_packed *k)
542 return __btree_node_ptr_to_offset(b, k);
545 static inline struct bkey_packed *
546 __btree_node_offset_to_key(const struct btree *b, u16 k)
548 return __btree_node_offset_to_ptr(b, k);
551 static inline unsigned btree_bkey_first_offset(const struct bset_tree *t)
553 return t->data_offset + offsetof(struct bset, _data) / sizeof(u64);
556 #define btree_bkey_first(_b, _t) \
558 EBUG_ON(bset(_b, _t)->start != \
559 __btree_node_offset_to_key(_b, btree_bkey_first_offset(_t)));\
561 bset(_b, _t)->start; \
564 #define btree_bkey_last(_b, _t) \
566 EBUG_ON(__btree_node_offset_to_key(_b, (_t)->end_offset) != \
567 vstruct_last(bset(_b, _t))); \
569 __btree_node_offset_to_key(_b, (_t)->end_offset); \
572 static inline unsigned bset_u64s(struct bset_tree *t)
574 return t->end_offset - t->data_offset -
575 sizeof(struct bset) / sizeof(u64);
578 static inline unsigned bset_dead_u64s(struct btree *b, struct bset_tree *t)
580 return bset_u64s(t) - b->nr.bset_u64s[t - b->set];
583 static inline unsigned bset_byte_offset(struct btree *b, void *i)
585 return i - (void *) b->data;
588 enum btree_node_type {
589 #define x(kwd, val) BKEY_TYPE_##kwd = val,
595 /* Type of a key in btree @id at level @level: */
596 static inline enum btree_node_type __btree_node_type(unsigned level, enum btree_id id)
598 return level ? BKEY_TYPE_btree : (enum btree_node_type) id;
601 /* Type of keys @b contains: */
602 static inline enum btree_node_type btree_node_type(struct btree *b)
604 return __btree_node_type(b->c.level, b->c.btree_id);
607 #define BTREE_NODE_TYPE_HAS_TRANS_TRIGGERS \
608 ((1U << BKEY_TYPE_extents)| \
609 (1U << BKEY_TYPE_alloc)| \
610 (1U << BKEY_TYPE_inodes)| \
611 (1U << BKEY_TYPE_stripes)| \
612 (1U << BKEY_TYPE_reflink)| \
613 (1U << BKEY_TYPE_btree))
615 #define BTREE_NODE_TYPE_HAS_MEM_TRIGGERS \
616 ((1U << BKEY_TYPE_alloc)| \
617 (1U << BKEY_TYPE_inodes)| \
618 (1U << BKEY_TYPE_stripes)| \
619 (1U << BKEY_TYPE_snapshots))
621 #define BTREE_NODE_TYPE_HAS_TRIGGERS \
622 (BTREE_NODE_TYPE_HAS_TRANS_TRIGGERS| \
623 BTREE_NODE_TYPE_HAS_MEM_TRIGGERS)
625 #define BTREE_ID_IS_EXTENTS \
626 ((1U << BTREE_ID_extents)| \
627 (1U << BTREE_ID_reflink)| \
628 (1U << BTREE_ID_freespace))
630 static inline bool btree_node_type_is_extents(enum btree_node_type type)
632 return (1U << type) & BTREE_ID_IS_EXTENTS;
635 #define BTREE_ID_HAS_SNAPSHOTS \
636 ((1U << BTREE_ID_extents)| \
637 (1U << BTREE_ID_inodes)| \
638 (1U << BTREE_ID_dirents)| \
639 (1U << BTREE_ID_xattrs))
641 #define BTREE_ID_HAS_PTRS \
642 ((1U << BTREE_ID_extents)| \
643 (1U << BTREE_ID_reflink))
645 static inline bool btree_type_has_snapshots(enum btree_id id)
647 return (1 << id) & BTREE_ID_HAS_SNAPSHOTS;
650 static inline bool btree_type_has_ptrs(enum btree_id id)
652 return (1 << id) & BTREE_ID_HAS_PTRS;
655 static inline bool btree_node_type_needs_gc(enum btree_node_type type)
657 return BTREE_NODE_TYPE_HAS_TRIGGERS & (1U << type);
663 /* On disk root - see async splits: */
664 __BKEY_PADDED(key, BKEY_BTREE_PTR_VAL_U64s_MAX);
670 enum btree_insert_ret {
672 /* leaf node needs to be split */
673 BTREE_INSERT_BTREE_NODE_FULL,
674 BTREE_INSERT_NEED_MARK_REPLICAS,
675 BTREE_INSERT_NEED_JOURNAL_RES,
676 BTREE_INSERT_NEED_JOURNAL_RECLAIM,
679 enum btree_gc_coalesce_fail_reason {
680 BTREE_GC_COALESCE_FAIL_RESERVE_GET,
681 BTREE_GC_COALESCE_FAIL_KEYLIST_REALLOC,
682 BTREE_GC_COALESCE_FAIL_FORMAT_FITS,
685 enum btree_node_sibling {
690 #endif /* _BCACHEFS_BTREE_TYPES_H */