]> git.sesse.net Git - bcachefs-tools-debian/blob - libbcachefs/btree_types.h
Update bcachefs sources to 90a9c61e2b bcachefs: Switch bch2_btree_delete_range()...
[bcachefs-tools-debian] / 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 "darray.h"
12 #include "journal_types.h"
13
14 struct open_bucket;
15 struct btree_update;
16 struct btree_trans;
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
53 struct btree_write {
54         struct journal_entry_pin        journal;
55 };
56
57 struct btree_alloc {
58         struct open_buckets     ob;
59         __BKEY_PADDED(k, BKEY_BTREE_PTR_VAL_U64s_MAX);
60 };
61
62 struct btree_bkey_cached_common {
63         struct six_lock         lock;
64         u8                      level;
65         u8                      btree_id;
66 };
67
68 struct btree {
69         struct btree_bkey_cached_common c;
70
71         struct rhash_head       hash;
72         u64                     hash_val;
73
74         unsigned long           flags;
75         u16                     written;
76         u8                      nsets;
77         u8                      nr_key_bits;
78         u16                     version_ondisk;
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                      byte_order;
98         u8                      unpack_fn_len;
99
100         struct btree_write      writes[2];
101
102         /* Key/pointer for this btree node */
103         __BKEY_PADDED(key, BKEY_BTREE_PTR_VAL_U64s_MAX);
104
105         /*
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
110          */
111
112         /*
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
119          * written:
120          */
121         struct list_head        write_blocked;
122
123         /*
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:
128          */
129         unsigned long           will_make_reachable;
130
131         struct open_buckets     ob;
132
133         /* lru list */
134         struct list_head        list;
135 };
136
137 struct btree_cache {
138         struct rhashtable       table;
139         bool                    table_init_done;
140         /*
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.
147          *
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.
152          */
153         struct mutex            lock;
154         struct list_head        live;
155         struct list_head        freeable;
156         struct list_head        freed_pcpu;
157         struct list_head        freed_nonpcpu;
158
159         /* Number of elements in live + freeable lists */
160         unsigned                used;
161         unsigned                reserve;
162         atomic_t                dirty;
163         struct shrinker         shrink;
164
165         /*
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
169          * this at a time:
170          */
171         struct task_struct      *alloc_lock;
172         struct closure_waitlist alloc_wait;
173 };
174
175 struct btree_node_iter {
176         struct btree_node_iter_set {
177                 u16     k, end;
178         } data[MAX_BSETS];
179 };
180
181 /*
182  * Iterate over all possible positions, synthesizing deleted keys for holes:
183  */
184 #define BTREE_ITER_SLOTS                (1 << 0)
185 #define BTREE_ITER_ALL_LEVELS           (1 << 1)
186 /*
187  * Indicates that intent locks should be taken on leaf nodes, because we expect
188  * to be doing updates:
189  */
190 #define BTREE_ITER_INTENT               (1 << 2)
191 /*
192  * Causes the btree iterator code to prefetch additional btree nodes from disk:
193  */
194 #define BTREE_ITER_PREFETCH             (1 << 3)
195 /*
196  * Used in bch2_btree_iter_traverse(), to indicate whether we're searching for
197  * @pos or the first key strictly greater than @pos
198  */
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_WITH_KEY_CACHE       (1 << 7)
203 #define BTREE_ITER_WITH_UPDATES         (1 << 8)
204 #define BTREE_ITER_WITH_JOURNAL         (1 << 9)
205 #define __BTREE_ITER_ALL_SNAPSHOTS      (1 << 10)
206 #define BTREE_ITER_ALL_SNAPSHOTS        (1 << 11)
207 #define BTREE_ITER_FILTER_SNAPSHOTS     (1 << 12)
208 #define BTREE_ITER_NOPRESERVE           (1 << 13)
209
210 enum btree_path_uptodate {
211         BTREE_ITER_UPTODATE             = 0,
212         BTREE_ITER_NEED_RELOCK          = 1,
213         BTREE_ITER_NEED_TRAVERSE        = 2,
214 };
215
216 struct btree_path {
217         u8                      idx;
218         u8                      sorted_idx;
219         u8                      ref;
220         u8                      intent_ref;
221
222         /* btree_iter_copy starts here: */
223         struct bpos             pos;
224
225         enum btree_id           btree_id:4;
226         bool                    cached:1;
227         bool                    preserve:1;
228         enum btree_path_uptodate uptodate:2;
229         /*
230          * When true, failing to relock this path will cause the transaction to
231          * restart:
232          */
233         bool                    should_be_locked:1;
234         unsigned                level:3,
235                                 locks_want:4,
236                                 nodes_locked:4,
237                                 nodes_intent_locked:4;
238
239         struct btree_path_level {
240                 struct btree    *b;
241                 struct btree_node_iter iter;
242                 u32             lock_seq;
243 #ifdef CONFIG_BCACHEFS_LOCK_TIME_STATS
244                 u64             lock_taken_time;
245 #endif
246         }                       l[BTREE_MAX_DEPTH];
247 #ifdef CONFIG_BCACHEFS_DEBUG
248         unsigned long           ip_allocated;
249 #endif
250 };
251
252 static inline struct btree_path_level *path_l(struct btree_path *path)
253 {
254         return path->l + path->level;
255 }
256
257 /*
258  * @pos                 - iterator's current position
259  * @level               - current btree depth
260  * @locks_want          - btree level below which we start taking intent locks
261  * @nodes_locked        - bitmask indicating which nodes in @nodes are locked
262  * @nodes_intent_locked - bitmask indicating which locks are intent locks
263  */
264 struct btree_iter {
265         struct btree_trans      *trans;
266         struct btree_path       *path;
267         struct btree_path       *update_path;
268         struct btree_path       *key_cache_path;
269
270         enum btree_id           btree_id:4;
271         unsigned                min_depth:3;
272         unsigned                advanced:1;
273
274         /* btree_iter_copy starts here: */
275         u16                     flags;
276
277         /* When we're filtering by snapshot, the snapshot ID we're looking for: */
278         unsigned                snapshot;
279
280         struct bpos             pos;
281         struct bpos             pos_after_commit;
282         /*
283          * Current unpacked key - so that bch2_btree_iter_next()/
284          * bch2_btree_iter_next_slot() can correctly advance pos.
285          */
286         struct bkey             k;
287
288         /* BTREE_ITER_WITH_JOURNAL: */
289         size_t                  journal_idx;
290         struct bpos             journal_pos;
291 #ifdef CONFIG_BCACHEFS_DEBUG
292         unsigned long           ip_allocated;
293 #endif
294 };
295
296 struct btree_key_cache_freelist {
297         struct bkey_cached      *objs[16];
298         unsigned                nr;
299 };
300
301 struct btree_key_cache {
302         struct mutex            lock;
303         struct rhashtable       table;
304         bool                    table_init_done;
305         struct list_head        freed;
306         struct shrinker         shrink;
307         unsigned                shrink_iter;
308         struct btree_key_cache_freelist __percpu *pcpu_freed;
309
310         atomic_long_t           nr_freed;
311         atomic_long_t           nr_keys;
312         atomic_long_t           nr_dirty;
313 };
314
315 struct bkey_cached_key {
316         u32                     btree_id;
317         struct bpos             pos;
318 } __attribute__((packed, aligned(4)));
319
320 #define BKEY_CACHED_ACCESSED            0
321 #define BKEY_CACHED_DIRTY               1
322
323 struct bkey_cached {
324         struct btree_bkey_cached_common c;
325
326         unsigned long           flags;
327         u16                     u64s;
328         bool                    valid;
329         u32                     btree_trans_barrier_seq;
330         struct bkey_cached_key  key;
331
332         struct rhash_head       hash;
333         struct list_head        list;
334
335         struct journal_preres   res;
336         struct journal_entry_pin journal;
337
338         struct bkey_i           *k;
339 };
340
341 struct btree_insert_entry {
342         unsigned                flags;
343         u8                      bkey_type;
344         enum btree_id           btree_id:8;
345         u8                      level:4;
346         bool                    cached:1;
347         bool                    insert_trigger_run:1;
348         bool                    overwrite_trigger_run:1;
349         bool                    key_cache_already_flushed:1;
350         /*
351          * @old_k may be a key from the journal; @old_btree_u64s always refers
352          * to the size of the key being overwritten in the btree:
353          */
354         u8                      old_btree_u64s;
355         struct bkey_i           *k;
356         struct btree_path       *path;
357         /* key being overwritten: */
358         struct bkey             old_k;
359         const struct bch_val    *old_v;
360         unsigned long           ip_allocated;
361 };
362
363 #ifndef CONFIG_LOCKDEP
364 #define BTREE_ITER_MAX          64
365 #else
366 #define BTREE_ITER_MAX          32
367 #endif
368
369 struct btree_trans_commit_hook;
370 typedef int (btree_trans_commit_hook_fn)(struct btree_trans *, struct btree_trans_commit_hook *);
371
372 struct btree_trans_commit_hook {
373         btree_trans_commit_hook_fn      *fn;
374         struct btree_trans_commit_hook  *next;
375 };
376
377 #define BTREE_TRANS_MEM_MAX     (1U << 16)
378
379 #define BTREE_TRANS_MAX_LOCK_HOLD_TIME_NS       10000
380
381 struct btree_trans {
382         struct bch_fs           *c;
383         const char              *fn;
384         struct list_head        list;
385         u64                     last_begin_time;
386         struct btree_bkey_cached_common *locking;
387         unsigned                locking_path_idx;
388         struct bpos             locking_pos;
389         u8                      locking_btree_id;
390         u8                      locking_level;
391         u8                      locking_lock_type;
392         struct task_struct      *task;
393         int                     srcu_idx;
394
395         u8                      nr_sorted;
396         u8                      nr_updates;
397         u8                      traverse_all_idx;
398         bool                    used_mempool:1;
399         bool                    in_traverse_all:1;
400         bool                    memory_allocation_failure:1;
401         bool                    is_initial_gc:1;
402         enum bch_errcode        restarted:16;
403         u32                     restart_count;
404         unsigned long           last_restarted_ip;
405
406         /*
407          * For when bch2_trans_update notices we'll be splitting a compressed
408          * extent:
409          */
410         unsigned                extra_journal_res;
411
412         u64                     paths_allocated;
413
414         unsigned                mem_top;
415         unsigned                mem_bytes;
416         void                    *mem;
417
418         u8                      sorted[BTREE_ITER_MAX];
419         struct btree_path       *paths;
420         struct btree_insert_entry *updates;
421
422         /* update path: */
423         struct btree_trans_commit_hook *hooks;
424         DARRAY(u64)             extra_journal_entries;
425         struct journal_entry_pin *journal_pin;
426
427         struct journal_res      journal_res;
428         struct journal_preres   journal_preres;
429         u64                     *journal_seq;
430         struct disk_reservation *disk_res;
431         unsigned                flags;
432         unsigned                journal_u64s;
433         unsigned                journal_preres_u64s;
434         struct replicas_delta_list *fs_usage_deltas;
435         int                      lock_name_idx;
436 };
437
438 #define BTREE_FLAGS()                                                   \
439         x(read_in_flight)                                               \
440         x(read_error)                                                   \
441         x(dirty)                                                        \
442         x(need_write)                                                   \
443         x(write_blocked)                                                \
444         x(will_make_reachable)                                          \
445         x(noevict)                                                      \
446         x(write_idx)                                                    \
447         x(accessed)                                                     \
448         x(write_in_flight)                                              \
449         x(write_in_flight_inner)                                        \
450         x(just_written)                                                 \
451         x(dying)                                                        \
452         x(fake)                                                         \
453         x(need_rewrite)                                                 \
454         x(never_write)
455
456 enum btree_flags {
457 #define x(flag) BTREE_NODE_##flag,
458         BTREE_FLAGS()
459 #undef x
460 };
461
462 #define x(flag)                                                         \
463 static inline bool btree_node_ ## flag(struct btree *b)                 \
464 {       return test_bit(BTREE_NODE_ ## flag, &b->flags); }              \
465                                                                         \
466 static inline void set_btree_node_ ## flag(struct btree *b)             \
467 {       set_bit(BTREE_NODE_ ## flag, &b->flags); }                      \
468                                                                         \
469 static inline void clear_btree_node_ ## flag(struct btree *b)           \
470 {       clear_bit(BTREE_NODE_ ## flag, &b->flags); }
471
472 BTREE_FLAGS()
473 #undef x
474
475 static inline struct btree_write *btree_current_write(struct btree *b)
476 {
477         return b->writes + btree_node_write_idx(b);
478 }
479
480 static inline struct btree_write *btree_prev_write(struct btree *b)
481 {
482         return b->writes + (btree_node_write_idx(b) ^ 1);
483 }
484
485 static inline struct bset_tree *bset_tree_last(struct btree *b)
486 {
487         EBUG_ON(!b->nsets);
488         return b->set + b->nsets - 1;
489 }
490
491 static inline void *
492 __btree_node_offset_to_ptr(const struct btree *b, u16 offset)
493 {
494         return (void *) ((u64 *) b->data + 1 + offset);
495 }
496
497 static inline u16
498 __btree_node_ptr_to_offset(const struct btree *b, const void *p)
499 {
500         u16 ret = (u64 *) p - 1 - (u64 *) b->data;
501
502         EBUG_ON(__btree_node_offset_to_ptr(b, ret) != p);
503         return ret;
504 }
505
506 static inline struct bset *bset(const struct btree *b,
507                                 const struct bset_tree *t)
508 {
509         return __btree_node_offset_to_ptr(b, t->data_offset);
510 }
511
512 static inline void set_btree_bset_end(struct btree *b, struct bset_tree *t)
513 {
514         t->end_offset =
515                 __btree_node_ptr_to_offset(b, vstruct_last(bset(b, t)));
516 }
517
518 static inline void set_btree_bset(struct btree *b, struct bset_tree *t,
519                                   const struct bset *i)
520 {
521         t->data_offset = __btree_node_ptr_to_offset(b, i);
522         set_btree_bset_end(b, t);
523 }
524
525 static inline struct bset *btree_bset_first(struct btree *b)
526 {
527         return bset(b, b->set);
528 }
529
530 static inline struct bset *btree_bset_last(struct btree *b)
531 {
532         return bset(b, bset_tree_last(b));
533 }
534
535 static inline u16
536 __btree_node_key_to_offset(const struct btree *b, const struct bkey_packed *k)
537 {
538         return __btree_node_ptr_to_offset(b, k);
539 }
540
541 static inline struct bkey_packed *
542 __btree_node_offset_to_key(const struct btree *b, u16 k)
543 {
544         return __btree_node_offset_to_ptr(b, k);
545 }
546
547 static inline unsigned btree_bkey_first_offset(const struct bset_tree *t)
548 {
549         return t->data_offset + offsetof(struct bset, _data) / sizeof(u64);
550 }
551
552 #define btree_bkey_first(_b, _t)                                        \
553 ({                                                                      \
554         EBUG_ON(bset(_b, _t)->start !=                                  \
555                 __btree_node_offset_to_key(_b, btree_bkey_first_offset(_t)));\
556                                                                         \
557         bset(_b, _t)->start;                                            \
558 })
559
560 #define btree_bkey_last(_b, _t)                                         \
561 ({                                                                      \
562         EBUG_ON(__btree_node_offset_to_key(_b, (_t)->end_offset) !=     \
563                 vstruct_last(bset(_b, _t)));                            \
564                                                                         \
565         __btree_node_offset_to_key(_b, (_t)->end_offset);               \
566 })
567
568 static inline unsigned bset_u64s(struct bset_tree *t)
569 {
570         return t->end_offset - t->data_offset -
571                 sizeof(struct bset) / sizeof(u64);
572 }
573
574 static inline unsigned bset_dead_u64s(struct btree *b, struct bset_tree *t)
575 {
576         return bset_u64s(t) - b->nr.bset_u64s[t - b->set];
577 }
578
579 static inline unsigned bset_byte_offset(struct btree *b, void *i)
580 {
581         return i - (void *) b->data;
582 }
583
584 enum btree_node_type {
585 #define x(kwd, val) BKEY_TYPE_##kwd = val,
586         BCH_BTREE_IDS()
587 #undef x
588         BKEY_TYPE_btree,
589 };
590
591 /* Type of a key in btree @id at level @level: */
592 static inline enum btree_node_type __btree_node_type(unsigned level, enum btree_id id)
593 {
594         return level ? BKEY_TYPE_btree : (enum btree_node_type) id;
595 }
596
597 /* Type of keys @b contains: */
598 static inline enum btree_node_type btree_node_type(struct btree *b)
599 {
600         return __btree_node_type(b->c.level, b->c.btree_id);
601 }
602
603 #define BTREE_NODE_TYPE_HAS_TRANS_TRIGGERS              \
604         ((1U << BKEY_TYPE_extents)|                     \
605          (1U << BKEY_TYPE_alloc)|                       \
606          (1U << BKEY_TYPE_inodes)|                      \
607          (1U << BKEY_TYPE_stripes)|                     \
608          (1U << BKEY_TYPE_reflink)|                     \
609          (1U << BKEY_TYPE_btree))
610
611 #define BTREE_NODE_TYPE_HAS_MEM_TRIGGERS                \
612         ((1U << BKEY_TYPE_alloc)|                       \
613          (1U << BKEY_TYPE_inodes)|                      \
614          (1U << BKEY_TYPE_stripes)|                     \
615          (1U << BKEY_TYPE_snapshots))
616
617 #define BTREE_NODE_TYPE_HAS_TRIGGERS                    \
618         (BTREE_NODE_TYPE_HAS_TRANS_TRIGGERS|            \
619          BTREE_NODE_TYPE_HAS_MEM_TRIGGERS)
620
621 #define BTREE_ID_IS_EXTENTS                             \
622         ((1U << BTREE_ID_extents)|                      \
623          (1U << BTREE_ID_reflink)|                      \
624          (1U << BTREE_ID_freespace))
625
626 static inline bool btree_node_type_is_extents(enum btree_node_type type)
627 {
628         return (1U << type) & BTREE_ID_IS_EXTENTS;
629 }
630
631 #define BTREE_ID_HAS_SNAPSHOTS                          \
632         ((1U << BTREE_ID_extents)|                      \
633          (1U << BTREE_ID_inodes)|                       \
634          (1U << BTREE_ID_dirents)|                      \
635          (1U << BTREE_ID_xattrs))
636
637 #define BTREE_ID_HAS_PTRS                               \
638         ((1U << BTREE_ID_extents)|                      \
639          (1U << BTREE_ID_reflink))
640
641 static inline bool btree_type_has_snapshots(enum btree_id id)
642 {
643         return (1 << id) & BTREE_ID_HAS_SNAPSHOTS;
644 }
645
646 static inline bool btree_type_has_ptrs(enum btree_id id)
647 {
648         return (1 << id) & BTREE_ID_HAS_PTRS;
649 }
650
651 static inline bool btree_node_type_needs_gc(enum btree_node_type type)
652 {
653         return BTREE_NODE_TYPE_HAS_TRIGGERS & (1U << type);
654 }
655
656 struct btree_root {
657         struct btree            *b;
658
659         /* On disk root - see async splits: */
660         __BKEY_PADDED(key, BKEY_BTREE_PTR_VAL_U64s_MAX);
661         u8                      level;
662         u8                      alive;
663         s8                      error;
664 };
665
666 enum btree_insert_ret {
667         BTREE_INSERT_OK,
668         /* leaf node needs to be split */
669         BTREE_INSERT_BTREE_NODE_FULL,
670         BTREE_INSERT_NEED_MARK_REPLICAS,
671         BTREE_INSERT_NEED_JOURNAL_RES,
672         BTREE_INSERT_NEED_JOURNAL_RECLAIM,
673 };
674
675 enum btree_gc_coalesce_fail_reason {
676         BTREE_GC_COALESCE_FAIL_RESERVE_GET,
677         BTREE_GC_COALESCE_FAIL_KEYLIST_REALLOC,
678         BTREE_GC_COALESCE_FAIL_FORMAT_FITS,
679 };
680
681 enum btree_node_sibling {
682         btree_prev_sib,
683         btree_next_sib,
684 };
685
686 #endif /* _BCACHEFS_BTREE_TYPES_H */