Lines Matching refs:b
99 #define insert_lock(s, b) ((b)->level <= (s)->lock) argument
119 #define btree(fn, key, b, op, ...) \ argument
121 int _r, l = (b)->level - 1; \
123 struct btree *_child = bch_btree_node_get((b)->c, op, key, l, \
124 _w, b); \
160 static inline struct bset *write_block(struct btree *b) in write_block() argument
162 return ((void *) btree_bset_first(b)) + b->written * block_bytes(b->c); in write_block()
165 static void bch_btree_init_next(struct btree *b) in bch_btree_init_next() argument
168 if (b->level && b->keys.nsets) in bch_btree_init_next()
169 bch_btree_sort(&b->keys, &b->c->sort); in bch_btree_init_next()
171 bch_btree_sort_lazy(&b->keys, &b->c->sort); in bch_btree_init_next()
173 if (b->written < btree_blocks(b)) in bch_btree_init_next()
174 bch_bset_init_next(&b->keys, write_block(b), in bch_btree_init_next()
175 bset_magic(&b->c->sb)); in bch_btree_init_next()
192 static uint64_t btree_csum_set(struct btree *b, struct bset *i) in btree_csum_set() argument
194 uint64_t crc = b->key.ptr[0]; in btree_csum_set()
201 void bch_btree_node_read_done(struct btree *b) in bch_btree_node_read_done() argument
204 struct bset *i = btree_bset_first(b); in bch_btree_node_read_done()
207 iter = mempool_alloc(b->c->fill_iter, GFP_NOIO); in bch_btree_node_read_done()
208 iter->size = b->c->sb.bucket_size / b->c->sb.block_size; in bch_btree_node_read_done()
212 iter->b = &b->keys; in bch_btree_node_read_done()
219 b->written < btree_blocks(b) && i->seq == b->keys.set[0].data->seq; in bch_btree_node_read_done()
220 i = write_block(b)) { in bch_btree_node_read_done()
226 if (b->written + set_blocks(i, block_bytes(b->c)) > in bch_btree_node_read_done()
227 btree_blocks(b)) in bch_btree_node_read_done()
231 if (i->magic != bset_magic(&b->c->sb)) in bch_btree_node_read_done()
241 if (i->csum != btree_csum_set(b, i)) in bch_btree_node_read_done()
247 if (i != b->keys.set[0].data && !i->keys) in bch_btree_node_read_done()
252 b->written += set_blocks(i, block_bytes(b->c)); in bch_btree_node_read_done()
256 for (i = write_block(b); in bch_btree_node_read_done()
257 bset_sector_offset(&b->keys, i) < KEY_SIZE(&b->key); in bch_btree_node_read_done()
258 i = ((void *) i) + block_bytes(b->c)) in bch_btree_node_read_done()
259 if (i->seq == b->keys.set[0].data->seq) in bch_btree_node_read_done()
262 bch_btree_sort_and_fix_extents(&b->keys, iter, &b->c->sort); in bch_btree_node_read_done()
264 i = b->keys.set[0].data; in bch_btree_node_read_done()
266 if (b->keys.set[0].size && in bch_btree_node_read_done()
267 bkey_cmp(&b->key, &b->keys.set[0].end) < 0) in bch_btree_node_read_done()
270 if (b->written < btree_blocks(b)) in bch_btree_node_read_done()
271 bch_bset_init_next(&b->keys, write_block(b), in bch_btree_node_read_done()
272 bset_magic(&b->c->sb)); in bch_btree_node_read_done()
274 mempool_free(iter, b->c->fill_iter); in bch_btree_node_read_done()
277 set_btree_node_io_error(b); in bch_btree_node_read_done()
278 bch_cache_set_error(b->c, "%s at bucket %zu, block %u, %u keys", in bch_btree_node_read_done()
279 err, PTR_BUCKET_NR(b->c, &b->key, 0), in bch_btree_node_read_done()
280 bset_block_offset(b, i), i->keys); in bch_btree_node_read_done()
290 static void bch_btree_node_read(struct btree *b) in bch_btree_node_read() argument
296 trace_bcache_btree_read(b); in bch_btree_node_read()
300 bio = bch_bbio_alloc(b->c); in bch_btree_node_read()
301 bio->bi_iter.bi_size = KEY_SIZE(&b->key) << 9; in bch_btree_node_read()
306 bch_bio_map(bio, b->keys.set[0].data); in bch_btree_node_read()
308 bch_submit_bbio(bio, b->c, &b->key, 0); in bch_btree_node_read()
312 set_btree_node_io_error(b); in bch_btree_node_read()
314 bch_bbio_free(bio, b->c); in bch_btree_node_read()
316 if (btree_node_io_error(b)) in bch_btree_node_read()
319 bch_btree_node_read_done(b); in bch_btree_node_read()
320 bch_time_stats_update(&b->c->btree_read_time, start_time); in bch_btree_node_read()
324 bch_cache_set_error(b->c, "io error reading bucket %zu", in bch_btree_node_read()
325 PTR_BUCKET_NR(b->c, &b->key, 0)); in bch_btree_node_read()
328 static void btree_complete_write(struct btree *b, struct btree_write *w) in btree_complete_write() argument
331 !atomic_sub_return(w->prio_blocked, &b->c->prio_blocked)) in btree_complete_write()
332 wake_up_allocators(b->c); in btree_complete_write()
336 __closure_wake_up(&b->c->journal.wait); in btree_complete_write()
345 struct btree *b = container_of(cl, struct btree, io); in btree_node_write_unlock() local
347 up(&b->io_mutex); in btree_node_write_unlock()
352 struct btree *b = container_of(cl, struct btree, io); in __btree_node_write_done() local
353 struct btree_write *w = btree_prev_write(b); in __btree_node_write_done()
355 bch_bbio_free(b->bio, b->c); in __btree_node_write_done()
356 b->bio = NULL; in __btree_node_write_done()
357 btree_complete_write(b, w); in __btree_node_write_done()
359 if (btree_node_dirty(b)) in __btree_node_write_done()
360 schedule_delayed_work(&b->work, 30 * HZ); in __btree_node_write_done()
367 struct btree *b = container_of(cl, struct btree, io); in btree_node_write_done() local
369 bio_free_pages(b->bio); in btree_node_write_done()
376 struct btree *b = container_of(cl, struct btree, io); in btree_node_write_endio() local
379 set_btree_node_io_error(b); in btree_node_write_endio()
381 bch_bbio_count_io_errors(b->c, bio, bio->bi_status, "writing btree"); in btree_node_write_endio()
385 static void do_btree_node_write(struct btree *b) in do_btree_node_write() argument
387 struct closure *cl = &b->io; in do_btree_node_write()
388 struct bset *i = btree_bset_last(b); in do_btree_node_write()
392 i->csum = btree_csum_set(b, i); in do_btree_node_write()
394 BUG_ON(b->bio); in do_btree_node_write()
395 b->bio = bch_bbio_alloc(b->c); in do_btree_node_write()
397 b->bio->bi_end_io = btree_node_write_endio; in do_btree_node_write()
398 b->bio->bi_private = cl; in do_btree_node_write()
399 b->bio->bi_iter.bi_size = roundup(set_bytes(i), block_bytes(b->c)); in do_btree_node_write()
400 b->bio->bi_opf = REQ_OP_WRITE | REQ_META | REQ_FUA; in do_btree_node_write()
401 bch_bio_map(b->bio, i); in do_btree_node_write()
418 bkey_copy(&k.key, &b->key); in do_btree_node_write()
420 bset_sector_offset(&b->keys, i)); in do_btree_node_write()
422 if (!bio_alloc_pages(b->bio, __GFP_NOWARN|GFP_NOWAIT)) { in do_btree_node_write()
427 bio_for_each_segment_all(bv, b->bio, j) in do_btree_node_write()
431 bch_submit_bbio(b->bio, b->c, &k.key, 0); in do_btree_node_write()
435 b->bio->bi_vcnt = 0; in do_btree_node_write()
436 bch_bio_map(b->bio, i); in do_btree_node_write()
438 bch_submit_bbio(b->bio, b->c, &k.key, 0); in do_btree_node_write()
445 void __bch_btree_node_write(struct btree *b, struct closure *parent) in __bch_btree_node_write() argument
447 struct bset *i = btree_bset_last(b); in __bch_btree_node_write()
449 lockdep_assert_held(&b->write_lock); in __bch_btree_node_write()
451 trace_bcache_btree_write(b); in __bch_btree_node_write()
454 BUG_ON(b->written >= btree_blocks(b)); in __bch_btree_node_write()
455 BUG_ON(b->written && !i->keys); in __bch_btree_node_write()
456 BUG_ON(btree_bset_first(b)->seq != i->seq); in __bch_btree_node_write()
457 bch_check_keys(&b->keys, "writing"); in __bch_btree_node_write()
459 cancel_delayed_work(&b->work); in __bch_btree_node_write()
462 down(&b->io_mutex); in __bch_btree_node_write()
463 closure_init(&b->io, parent ?: &b->c->cl); in __bch_btree_node_write()
465 clear_bit(BTREE_NODE_dirty, &b->flags); in __bch_btree_node_write()
466 change_bit(BTREE_NODE_write_idx, &b->flags); in __bch_btree_node_write()
468 do_btree_node_write(b); in __bch_btree_node_write()
470 atomic_long_add(set_blocks(i, block_bytes(b->c)) * b->c->sb.block_size, in __bch_btree_node_write()
471 &PTR_CACHE(b->c, &b->key, 0)->btree_sectors_written); in __bch_btree_node_write()
473 b->written += set_blocks(i, block_bytes(b->c)); in __bch_btree_node_write()
476 void bch_btree_node_write(struct btree *b, struct closure *parent) in bch_btree_node_write() argument
478 unsigned nsets = b->keys.nsets; in bch_btree_node_write()
480 lockdep_assert_held(&b->lock); in bch_btree_node_write()
482 __bch_btree_node_write(b, parent); in bch_btree_node_write()
488 if (nsets && !b->keys.nsets) in bch_btree_node_write()
489 bch_btree_verify(b); in bch_btree_node_write()
491 bch_btree_init_next(b); in bch_btree_node_write()
494 static void bch_btree_node_write_sync(struct btree *b) in bch_btree_node_write_sync() argument
500 mutex_lock(&b->write_lock); in bch_btree_node_write_sync()
501 bch_btree_node_write(b, &cl); in bch_btree_node_write_sync()
502 mutex_unlock(&b->write_lock); in bch_btree_node_write_sync()
509 struct btree *b = container_of(to_delayed_work(w), struct btree, work); in btree_node_write_work() local
511 mutex_lock(&b->write_lock); in btree_node_write_work()
512 if (btree_node_dirty(b)) in btree_node_write_work()
513 __bch_btree_node_write(b, NULL); in btree_node_write_work()
514 mutex_unlock(&b->write_lock); in btree_node_write_work()
517 static void bch_btree_leaf_dirty(struct btree *b, atomic_t *journal_ref) in bch_btree_leaf_dirty() argument
519 struct bset *i = btree_bset_last(b); in bch_btree_leaf_dirty()
520 struct btree_write *w = btree_current_write(b); in bch_btree_leaf_dirty()
522 lockdep_assert_held(&b->write_lock); in bch_btree_leaf_dirty()
524 BUG_ON(!b->written); in bch_btree_leaf_dirty()
527 if (!btree_node_dirty(b)) in bch_btree_leaf_dirty()
528 schedule_delayed_work(&b->work, 30 * HZ); in bch_btree_leaf_dirty()
530 set_btree_node_dirty(b); in bch_btree_leaf_dirty()
534 journal_pin_cmp(b->c, w->journal, journal_ref)) { in bch_btree_leaf_dirty()
548 bch_btree_node_write(b, NULL); in bch_btree_leaf_dirty()
561 static void mca_data_free(struct btree *b) in mca_data_free() argument
563 BUG_ON(b->io_mutex.count != 1); in mca_data_free()
565 bch_btree_keys_free(&b->keys); in mca_data_free()
567 b->c->btree_cache_used--; in mca_data_free()
568 list_move(&b->list, &b->c->btree_cache_freed); in mca_data_free()
571 static void mca_bucket_free(struct btree *b) in mca_bucket_free() argument
573 BUG_ON(btree_node_dirty(b)); in mca_bucket_free()
575 b->key.ptr[0] = 0; in mca_bucket_free()
576 hlist_del_init_rcu(&b->hash); in mca_bucket_free()
577 list_move(&b->list, &b->c->btree_cache_freeable); in mca_bucket_free()
585 static void mca_data_alloc(struct btree *b, struct bkey *k, gfp_t gfp) in mca_data_alloc() argument
587 if (!bch_btree_keys_alloc(&b->keys, in mca_data_alloc()
589 ilog2(b->c->btree_pages), in mca_data_alloc()
592 b->c->btree_cache_used++; in mca_data_alloc()
593 list_move(&b->list, &b->c->btree_cache); in mca_data_alloc()
595 list_move(&b->list, &b->c->btree_cache_freed); in mca_data_alloc()
602 struct btree *b = kzalloc(sizeof(struct btree), gfp); in mca_bucket_alloc() local
603 if (!b) in mca_bucket_alloc()
606 init_rwsem(&b->lock); in mca_bucket_alloc()
607 lockdep_set_novalidate_class(&b->lock); in mca_bucket_alloc()
608 mutex_init(&b->write_lock); in mca_bucket_alloc()
609 lockdep_set_novalidate_class(&b->write_lock); in mca_bucket_alloc()
610 INIT_LIST_HEAD(&b->list); in mca_bucket_alloc()
611 INIT_DELAYED_WORK(&b->work, btree_node_write_work); in mca_bucket_alloc()
612 b->c = c; in mca_bucket_alloc()
613 sema_init(&b->io_mutex, 1); in mca_bucket_alloc()
615 mca_data_alloc(b, k, gfp); in mca_bucket_alloc()
616 return b; in mca_bucket_alloc()
619 static int mca_reap(struct btree *b, unsigned min_order, bool flush) in mca_reap() argument
624 lockdep_assert_held(&b->c->bucket_lock); in mca_reap()
626 if (!down_write_trylock(&b->lock)) in mca_reap()
629 BUG_ON(btree_node_dirty(b) && !b->keys.set[0].data); in mca_reap()
631 if (b->keys.page_order < min_order) in mca_reap()
635 if (btree_node_dirty(b)) in mca_reap()
638 if (down_trylock(&b->io_mutex)) in mca_reap()
640 up(&b->io_mutex); in mca_reap()
643 mutex_lock(&b->write_lock); in mca_reap()
644 if (btree_node_dirty(b)) in mca_reap()
645 __bch_btree_node_write(b, &cl); in mca_reap()
646 mutex_unlock(&b->write_lock); in mca_reap()
651 down(&b->io_mutex); in mca_reap()
652 up(&b->io_mutex); in mca_reap()
656 rw_unlock(true, b); in mca_reap()
664 struct btree *b, *t; in bch_mca_scan() local
693 list_for_each_entry_safe(b, t, &c->btree_cache_freeable, list) { in bch_mca_scan()
698 !mca_reap(b, 0, false)) { in bch_mca_scan()
699 mca_data_free(b); in bch_mca_scan()
700 rw_unlock(true, b); in bch_mca_scan()
709 b = list_first_entry(&c->btree_cache, struct btree, list); in bch_mca_scan()
712 if (!b->accessed && in bch_mca_scan()
713 !mca_reap(b, 0, false)) { in bch_mca_scan()
714 mca_bucket_free(b); in bch_mca_scan()
715 mca_data_free(b); in bch_mca_scan()
716 rw_unlock(true, b); in bch_mca_scan()
719 b->accessed = 0; in bch_mca_scan()
742 struct btree *b; in bch_btree_cache_free() local
762 b = list_first_entry(&c->btree_cache, struct btree, list); in bch_btree_cache_free()
764 if (btree_node_dirty(b)) in bch_btree_cache_free()
765 btree_complete_write(b, btree_current_write(b)); in bch_btree_cache_free()
766 clear_bit(BTREE_NODE_dirty, &b->flags); in bch_btree_cache_free()
768 mca_data_free(b); in bch_btree_cache_free()
772 b = list_first_entry(&c->btree_cache_freed, in bch_btree_cache_free()
774 list_del(&b->list); in bch_btree_cache_free()
775 cancel_delayed_work_sync(&b->work); in bch_btree_cache_free()
776 kfree(b); in bch_btree_cache_free()
829 struct btree *b; in mca_find() local
832 hlist_for_each_entry_rcu(b, mca_hash(c, k), hash) in mca_find()
833 if (PTR_HASH(c, &b->key) == PTR_HASH(c, k)) in mca_find()
835 b = NULL; in mca_find()
838 return b; in mca_find()
859 struct btree *b; in mca_cannibalize() local
866 list_for_each_entry_reverse(b, &c->btree_cache, list) in mca_cannibalize()
867 if (!mca_reap(b, btree_order(k), false)) in mca_cannibalize()
868 return b; in mca_cannibalize()
870 list_for_each_entry_reverse(b, &c->btree_cache, list) in mca_cannibalize()
871 if (!mca_reap(b, btree_order(k), true)) in mca_cannibalize()
872 return b; in mca_cannibalize()
895 struct btree *b; in mca_alloc() local
907 list_for_each_entry(b, &c->btree_cache_freeable, list) in mca_alloc()
908 if (!mca_reap(b, btree_order(k), false)) in mca_alloc()
914 list_for_each_entry(b, &c->btree_cache_freed, list) in mca_alloc()
915 if (!mca_reap(b, 0, false)) { in mca_alloc()
916 mca_data_alloc(b, k, __GFP_NOWARN|GFP_NOIO); in mca_alloc()
917 if (!b->keys.set[0].data) in mca_alloc()
923 b = mca_bucket_alloc(c, k, __GFP_NOWARN|GFP_NOIO); in mca_alloc()
924 if (!b) in mca_alloc()
927 BUG_ON(!down_write_trylock(&b->lock)); in mca_alloc()
928 if (!b->keys.set->data) in mca_alloc()
931 BUG_ON(b->io_mutex.count != 1); in mca_alloc()
933 bkey_copy(&b->key, k); in mca_alloc()
934 list_move(&b->list, &c->btree_cache); in mca_alloc()
935 hlist_del_init_rcu(&b->hash); in mca_alloc()
936 hlist_add_head_rcu(&b->hash, mca_hash(c, k)); in mca_alloc()
938 lock_set_subclass(&b->lock.dep_map, level + 1, _THIS_IP_); in mca_alloc()
939 b->parent = (void *) ~0UL; in mca_alloc()
940 b->flags = 0; in mca_alloc()
941 b->written = 0; in mca_alloc()
942 b->level = level; in mca_alloc()
944 if (!b->level) in mca_alloc()
945 bch_btree_keys_init(&b->keys, &bch_extent_keys_ops, in mca_alloc()
946 &b->c->expensive_debug_checks); in mca_alloc()
948 bch_btree_keys_init(&b->keys, &bch_btree_keys_ops, in mca_alloc()
949 &b->c->expensive_debug_checks); in mca_alloc()
951 return b; in mca_alloc()
953 if (b) in mca_alloc()
954 rw_unlock(true, b); in mca_alloc()
956 b = mca_cannibalize(c, op, k); in mca_alloc()
957 if (!IS_ERR(b)) in mca_alloc()
960 return b; in mca_alloc()
977 struct btree *b; in bch_btree_node_get() local
981 b = mca_find(c, k); in bch_btree_node_get()
983 if (!b) { in bch_btree_node_get()
988 b = mca_alloc(c, op, k, level); in bch_btree_node_get()
991 if (!b) in bch_btree_node_get()
993 if (IS_ERR(b)) in bch_btree_node_get()
994 return b; in bch_btree_node_get()
996 bch_btree_node_read(b); in bch_btree_node_get()
999 downgrade_write(&b->lock); in bch_btree_node_get()
1001 rw_lock(write, b, level); in bch_btree_node_get()
1002 if (PTR_HASH(c, &b->key) != PTR_HASH(c, k)) { in bch_btree_node_get()
1003 rw_unlock(write, b); in bch_btree_node_get()
1006 BUG_ON(b->level != level); in bch_btree_node_get()
1009 b->parent = parent; in bch_btree_node_get()
1010 b->accessed = 1; in bch_btree_node_get()
1012 for (; i <= b->keys.nsets && b->keys.set[i].size; i++) { in bch_btree_node_get()
1013 prefetch(b->keys.set[i].tree); in bch_btree_node_get()
1014 prefetch(b->keys.set[i].data); in bch_btree_node_get()
1017 for (; i <= b->keys.nsets; i++) in bch_btree_node_get()
1018 prefetch(b->keys.set[i].data); in bch_btree_node_get()
1020 if (btree_node_io_error(b)) { in bch_btree_node_get()
1021 rw_unlock(write, b); in bch_btree_node_get()
1025 BUG_ON(!b->written); in bch_btree_node_get()
1027 return b; in bch_btree_node_get()
1032 struct btree *b; in btree_node_prefetch() local
1035 b = mca_alloc(parent->c, NULL, k, parent->level - 1); in btree_node_prefetch()
1038 if (!IS_ERR_OR_NULL(b)) { in btree_node_prefetch()
1039 b->parent = parent; in btree_node_prefetch()
1040 bch_btree_node_read(b); in btree_node_prefetch()
1041 rw_unlock(true, b); in btree_node_prefetch()
1047 static void btree_node_free(struct btree *b) in btree_node_free() argument
1049 trace_bcache_btree_node_free(b); in btree_node_free()
1051 BUG_ON(b == b->c->root); in btree_node_free()
1053 mutex_lock(&b->write_lock); in btree_node_free()
1055 if (btree_node_dirty(b)) in btree_node_free()
1056 btree_complete_write(b, btree_current_write(b)); in btree_node_free()
1057 clear_bit(BTREE_NODE_dirty, &b->flags); in btree_node_free()
1059 mutex_unlock(&b->write_lock); in btree_node_free()
1061 cancel_delayed_work(&b->work); in btree_node_free()
1063 mutex_lock(&b->c->bucket_lock); in btree_node_free()
1064 bch_bucket_free(b->c, &b->key); in btree_node_free()
1065 mca_bucket_free(b); in btree_node_free()
1066 mutex_unlock(&b->c->bucket_lock); in btree_node_free()
1074 struct btree *b = ERR_PTR(-EAGAIN); in __bch_btree_node_alloc() local
1084 b = mca_alloc(c, op, &k.key, level); in __bch_btree_node_alloc()
1085 if (IS_ERR(b)) in __bch_btree_node_alloc()
1088 if (!b) { in __bch_btree_node_alloc()
1094 b->accessed = 1; in __bch_btree_node_alloc()
1095 b->parent = parent; in __bch_btree_node_alloc()
1096 bch_bset_init_next(&b->keys, b->keys.set->data, bset_magic(&b->c->sb)); in __bch_btree_node_alloc()
1100 trace_bcache_btree_node_alloc(b); in __bch_btree_node_alloc()
1101 return b; in __bch_btree_node_alloc()
1108 return b; in __bch_btree_node_alloc()
1118 static struct btree *btree_node_alloc_replacement(struct btree *b, in btree_node_alloc_replacement() argument
1121 struct btree *n = bch_btree_node_alloc(b->c, op, b->level, b->parent); in btree_node_alloc_replacement()
1124 bch_btree_sort_into(&b->keys, &n->keys, &b->c->sort); in btree_node_alloc_replacement()
1125 bkey_copy_key(&n->key, &b->key); in btree_node_alloc_replacement()
1132 static void make_btree_freeing_key(struct btree *b, struct bkey *k) in make_btree_freeing_key() argument
1136 mutex_lock(&b->c->bucket_lock); in make_btree_freeing_key()
1138 atomic_inc(&b->c->prio_blocked); in make_btree_freeing_key()
1140 bkey_copy(k, &b->key); in make_btree_freeing_key()
1145 bch_inc_gen(PTR_CACHE(b->c, &b->key, i), in make_btree_freeing_key()
1146 PTR_BUCKET(b->c, &b->key, i))); in make_btree_freeing_key()
1148 mutex_unlock(&b->c->bucket_lock); in make_btree_freeing_key()
1151 static int btree_check_reserve(struct btree *b, struct btree_op *op) in btree_check_reserve() argument
1153 struct cache_set *c = b->c; in btree_check_reserve()
1155 unsigned i, reserve = (c->root->level - b->level) * 2 + 1; in btree_check_reserve()
1170 return mca_cannibalize_lock(b->c, op); in btree_check_reserve()
1227 #define btree_mark_key(b, k) __bch_btree_mark_key(b->c, b->level, k) argument
1236 struct bucket *b = PTR_BUCKET(c, k, i); in bch_initial_mark_key() local
1238 b->gen = PTR_GEN(k, i); in bch_initial_mark_key()
1241 b->prio = BTREE_PRIO; in bch_initial_mark_key()
1242 else if (!level && b->prio == BTREE_PRIO) in bch_initial_mark_key()
1243 b->prio = INITIAL_PRIO; in bch_initial_mark_key()
1249 static bool btree_gc_mark_node(struct btree *b, struct gc_stat *gc) in btree_gc_mark_node() argument
1259 for_each_key_filter(&b->keys, k, &iter, bch_ptr_invalid) { in btree_gc_mark_node()
1260 stale = max(stale, btree_mark_key(b, k)); in btree_gc_mark_node()
1263 if (bch_ptr_bad(&b->keys, k)) in btree_gc_mark_node()
1273 for (t = b->keys.set; t <= &b->keys.set[b->keys.nsets]; t++) in btree_gc_mark_node()
1275 bset_written(&b->keys, t) && in btree_gc_mark_node()
1276 bkey_cmp(&b->key, &t->end) < 0, in btree_gc_mark_node()
1277 b, "found short btree key in gc"); in btree_gc_mark_node()
1279 if (b->c->gc_always_rewrite) in btree_gc_mark_node()
1294 struct btree *b; member
1301 static int btree_gc_coalesce(struct btree *b, struct btree_op *op, in btree_gc_coalesce() argument
1312 if (btree_check_reserve(b, NULL)) in btree_gc_coalesce()
1318 while (nodes < GC_MERGE_NODES && !IS_ERR_OR_NULL(r[nodes].b)) in btree_gc_coalesce()
1321 blocks = btree_default_blocks(b->c) * 2 / 3; in btree_gc_coalesce()
1324 __set_blocks(b->keys.set[0].data, keys, in btree_gc_coalesce()
1325 block_bytes(b->c)) > blocks * (nodes - 1)) in btree_gc_coalesce()
1329 new_nodes[i] = btree_node_alloc_replacement(r[i].b, NULL); in btree_gc_coalesce()
1340 if (btree_check_reserve(b, NULL)) in btree_gc_coalesce()
1359 block_bytes(b->c)) > blocks) in btree_gc_coalesce()
1375 block_bytes(b->c)) > in btree_gc_coalesce()
1381 last = &r->b->key; in btree_gc_coalesce()
1384 BUG_ON(__set_blocks(n1, n1->keys + keys, block_bytes(b->c)) > in btree_gc_coalesce()
1424 if (__bch_keylist_realloc(&keylist, bkey_u64s(&r[i].b->key))) in btree_gc_coalesce()
1427 make_btree_freeing_key(r[i].b, keylist.top); in btree_gc_coalesce()
1431 bch_btree_insert_node(b, op, &keylist, NULL, NULL); in btree_gc_coalesce()
1435 btree_node_free(r[i].b); in btree_gc_coalesce()
1436 rw_unlock(true, r[i].b); in btree_gc_coalesce()
1438 r[i].b = new_nodes[i]; in btree_gc_coalesce()
1442 r[nodes - 1].b = ERR_PTR(-EINTR); in btree_gc_coalesce()
1458 atomic_dec(&b->c->prio_blocked); in btree_gc_coalesce()
1468 static int btree_gc_rewrite_node(struct btree *b, struct btree_op *op, in btree_gc_rewrite_node() argument
1474 if (btree_check_reserve(b, NULL)) in btree_gc_rewrite_node()
1480 if (btree_check_reserve(b, NULL)) { in btree_gc_rewrite_node()
1494 bch_btree_insert_node(b, op, &keys, NULL, NULL); in btree_gc_rewrite_node()
1504 static unsigned btree_gc_count_keys(struct btree *b) in btree_gc_count_keys() argument
1510 for_each_key_filter(&b->keys, k, &iter, bch_ptr_bad) in btree_gc_count_keys()
1516 static int btree_gc_recurse(struct btree *b, struct btree_op *op, in btree_gc_recurse() argument
1526 bch_btree_iter_init(&b->keys, &iter, &b->c->gc_done); in btree_gc_recurse()
1529 i->b = ERR_PTR(-EINTR); in btree_gc_recurse()
1532 k = bch_btree_iter_next_filter(&iter, &b->keys, bch_ptr_bad); in btree_gc_recurse()
1534 r->b = bch_btree_node_get(b->c, op, k, b->level - 1, in btree_gc_recurse()
1535 true, b); in btree_gc_recurse()
1536 if (IS_ERR(r->b)) { in btree_gc_recurse()
1537 ret = PTR_ERR(r->b); in btree_gc_recurse()
1541 r->keys = btree_gc_count_keys(r->b); in btree_gc_recurse()
1543 ret = btree_gc_coalesce(b, op, gc, r); in btree_gc_recurse()
1548 if (!last->b) in btree_gc_recurse()
1551 if (!IS_ERR(last->b)) { in btree_gc_recurse()
1552 should_rewrite = btree_gc_mark_node(last->b, gc); in btree_gc_recurse()
1554 ret = btree_gc_rewrite_node(b, op, last->b); in btree_gc_recurse()
1559 if (last->b->level) { in btree_gc_recurse()
1560 ret = btree_gc_recurse(last->b, op, writes, gc); in btree_gc_recurse()
1565 bkey_copy_key(&b->c->gc_done, &last->b->key); in btree_gc_recurse()
1571 mutex_lock(&last->b->write_lock); in btree_gc_recurse()
1572 if (btree_node_dirty(last->b)) in btree_gc_recurse()
1573 bch_btree_node_write(last->b, writes); in btree_gc_recurse()
1574 mutex_unlock(&last->b->write_lock); in btree_gc_recurse()
1575 rw_unlock(true, last->b); in btree_gc_recurse()
1579 r->b = NULL; in btree_gc_recurse()
1588 if (!IS_ERR_OR_NULL(i->b)) { in btree_gc_recurse()
1589 mutex_lock(&i->b->write_lock); in btree_gc_recurse()
1590 if (btree_node_dirty(i->b)) in btree_gc_recurse()
1591 bch_btree_node_write(i->b, writes); in btree_gc_recurse()
1592 mutex_unlock(&i->b->write_lock); in btree_gc_recurse()
1593 rw_unlock(true, i->b); in btree_gc_recurse()
1599 static int bch_btree_gc_root(struct btree *b, struct btree_op *op, in bch_btree_gc_root() argument
1606 should_rewrite = btree_gc_mark_node(b, gc); in bch_btree_gc_root()
1608 n = btree_node_alloc_replacement(b, NULL); in bch_btree_gc_root()
1614 btree_node_free(b); in bch_btree_gc_root()
1621 __bch_btree_mark_key(b->c, b->level + 1, &b->key); in bch_btree_gc_root()
1623 if (b->level) { in bch_btree_gc_root()
1624 ret = btree_gc_recurse(b, op, writes, gc); in bch_btree_gc_root()
1629 bkey_copy_key(&b->c->gc_done, &b->key); in bch_btree_gc_root()
1637 struct bucket *b; in btree_gc_start() local
1649 for_each_bucket(b, ca) { in btree_gc_start()
1650 b->last_gc = b->gen; in btree_gc_start()
1651 if (!atomic_read(&b->pin)) { in btree_gc_start()
1652 SET_GC_MARK(b, 0); in btree_gc_start()
1653 SET_GC_SECTORS_USED(b, 0); in btree_gc_start()
1663 struct bucket *b; in bch_btree_gc_finish() local
1711 for_each_bucket(b, ca) { in bch_btree_gc_finish()
1712 c->need_gc = max(c->need_gc, bucket_gc_gen(b)); in bch_btree_gc_finish()
1714 if (atomic_read(&b->pin)) in bch_btree_gc_finish()
1717 BUG_ON(!GC_MARK(b) && GC_SECTORS_USED(b)); in bch_btree_gc_finish()
1719 if (!GC_MARK(b) || GC_MARK(b) == GC_MARK_RECLAIMABLE) in bch_btree_gc_finish()
1813 static int bch_btree_check_recurse(struct btree *b, struct btree_op *op) in bch_btree_check_recurse() argument
1819 for_each_key_filter(&b->keys, k, &iter, bch_ptr_invalid) in bch_btree_check_recurse()
1820 bch_initial_mark_key(b->c, b->level, k); in bch_btree_check_recurse()
1822 bch_initial_mark_key(b->c, b->level + 1, &b->key); in bch_btree_check_recurse()
1824 if (b->level) { in bch_btree_check_recurse()
1825 bch_btree_iter_init(&b->keys, &iter, NULL); in bch_btree_check_recurse()
1828 k = bch_btree_iter_next_filter(&iter, &b->keys, in bch_btree_check_recurse()
1831 btree_node_prefetch(b, k); in bch_btree_check_recurse()
1834 ret = btree(check_recurse, p, b, op); in bch_btree_check_recurse()
1855 struct bucket *b; in bch_initial_gc_finish() local
1872 for_each_bucket(b, ca) { in bch_initial_gc_finish()
1877 if (bch_can_invalidate_bucket(ca, b) && in bch_initial_gc_finish()
1878 !GC_MARK(b)) { in bch_initial_gc_finish()
1879 __bch_invalidate_one_bucket(ca, b); in bch_initial_gc_finish()
1881 b - ca->buckets)) in bch_initial_gc_finish()
1883 b - ca->buckets); in bch_initial_gc_finish()
1893 static bool btree_insert_key(struct btree *b, struct bkey *k, in btree_insert_key() argument
1898 BUG_ON(bkey_cmp(k, &b->key) > 0); in btree_insert_key()
1900 status = bch_btree_insert_key(&b->keys, k, replace_key); in btree_insert_key()
1902 bch_check_keys(&b->keys, "%u for %s", status, in btree_insert_key()
1905 trace_bcache_btree_insert_key(b, k, replace_key != NULL, in btree_insert_key()
1912 static size_t insert_u64s_remaining(struct btree *b) in insert_u64s_remaining() argument
1914 long ret = bch_btree_keys_u64s_remaining(&b->keys); in insert_u64s_remaining()
1919 if (b->keys.ops->is_extents) in insert_u64s_remaining()
1925 static bool bch_btree_insert_keys(struct btree *b, struct btree_op *op, in bch_btree_insert_keys() argument
1930 int oldsize = bch_count_data(&b->keys); in bch_btree_insert_keys()
1935 if (bkey_u64s(k) > insert_u64s_remaining(b)) in bch_btree_insert_keys()
1938 if (bkey_cmp(k, &b->key) <= 0) { in bch_btree_insert_keys()
1939 if (!b->level) in bch_btree_insert_keys()
1940 bkey_put(b->c, k); in bch_btree_insert_keys()
1942 ret |= btree_insert_key(b, k, replace_key); in bch_btree_insert_keys()
1944 } else if (bkey_cmp(&START_KEY(k), &b->key) < 0) { in bch_btree_insert_keys()
1948 bch_cut_back(&b->key, &temp.key); in bch_btree_insert_keys()
1949 bch_cut_front(&b->key, insert_keys->keys); in bch_btree_insert_keys()
1951 ret |= btree_insert_key(b, &temp.key, replace_key); in bch_btree_insert_keys()
1961 BUG_ON(!bch_keylist_empty(insert_keys) && b->level); in bch_btree_insert_keys()
1963 BUG_ON(bch_count_data(&b->keys) < oldsize); in bch_btree_insert_keys()
1967 static int btree_split(struct btree *b, struct btree_op *op, in btree_split() argument
1980 if (btree_check_reserve(b, op)) { in btree_split()
1981 if (!b->level) in btree_split()
1987 n1 = btree_node_alloc_replacement(b, op); in btree_split()
1992 block_bytes(n1->c)) > (btree_blocks(b) * 4) / 5; in btree_split()
1997 trace_bcache_btree_node_split(b, btree_bset_first(n1)->keys); in btree_split()
1999 n2 = bch_btree_node_alloc(b->c, op, b->level, b->parent); in btree_split()
2003 if (!b->parent) { in btree_split()
2004 n3 = bch_btree_node_alloc(b->c, op, b->level + 1, NULL); in btree_split()
2034 bkey_copy_key(&n2->key, &b->key); in btree_split()
2041 trace_bcache_btree_node_compact(b, btree_bset_first(n1)->keys); in btree_split()
2062 } else if (!b->parent) { in btree_split()
2069 make_btree_freeing_key(b, parent_keys.top); in btree_split()
2072 bch_btree_insert_node(b->parent, op, &parent_keys, NULL, NULL); in btree_split()
2076 btree_node_free(b); in btree_split()
2079 bch_time_stats_update(&b->c->btree_split_time, start_time); in btree_split()
2083 bkey_put(b->c, &n2->key); in btree_split()
2087 bkey_put(b->c, &n1->key); in btree_split()
2091 WARN(1, "bcache: btree split failed (level %u)", b->level); in btree_split()
2101 static int bch_btree_insert_node(struct btree *b, struct btree_op *op, in bch_btree_insert_node() argument
2108 BUG_ON(b->level && replace_key); in bch_btree_insert_node()
2112 mutex_lock(&b->write_lock); in bch_btree_insert_node()
2114 if (write_block(b) != btree_bset_last(b) && in bch_btree_insert_node()
2115 b->keys.last_set_unwritten) in bch_btree_insert_node()
2116 bch_btree_init_next(b); /* just wrote a set */ in bch_btree_insert_node()
2118 if (bch_keylist_nkeys(insert_keys) > insert_u64s_remaining(b)) { in bch_btree_insert_node()
2119 mutex_unlock(&b->write_lock); in bch_btree_insert_node()
2123 BUG_ON(write_block(b) != btree_bset_last(b)); in bch_btree_insert_node()
2125 if (bch_btree_insert_keys(b, op, insert_keys, replace_key)) { in bch_btree_insert_node()
2126 if (!b->level) in bch_btree_insert_node()
2127 bch_btree_leaf_dirty(b, journal_ref); in bch_btree_insert_node()
2129 bch_btree_node_write(b, &cl); in bch_btree_insert_node()
2132 mutex_unlock(&b->write_lock); in bch_btree_insert_node()
2140 op->lock = b->c->root->level + 1; in bch_btree_insert_node()
2142 } else if (op->lock <= b->c->root->level) { in bch_btree_insert_node()
2143 op->lock = b->c->root->level + 1; in bch_btree_insert_node()
2147 int ret = btree_split(b, op, insert_keys, replace_key); in bch_btree_insert_node()
2157 int bch_btree_insert_check_key(struct btree *b, struct btree_op *op, in bch_btree_insert_check_key() argument
2161 uint64_t btree_ptr = b->key.ptr[0]; in bch_btree_insert_check_key()
2162 unsigned long seq = b->seq; in bch_btree_insert_check_key()
2169 rw_unlock(false, b); in bch_btree_insert_check_key()
2170 rw_lock(true, b, b->level); in bch_btree_insert_check_key()
2172 if (b->key.ptr[0] != btree_ptr || in bch_btree_insert_check_key()
2173 b->seq != seq + 1) { in bch_btree_insert_check_key()
2174 op->lock = b->level; in bch_btree_insert_check_key()
2186 ret = bch_btree_insert_node(b, op, &insert, NULL, NULL); in bch_btree_insert_check_key()
2191 downgrade_write(&b->lock); in bch_btree_insert_check_key()
2202 static int btree_insert_fn(struct btree_op *b_op, struct btree *b) in btree_insert_fn() argument
2207 int ret = bch_btree_insert_node(b, &op->op, op->keys, in btree_insert_fn()
2249 void bch_btree_set_root(struct btree *b) in bch_btree_set_root() argument
2256 trace_bcache_btree_set_root(b); in bch_btree_set_root()
2258 BUG_ON(!b->written); in bch_btree_set_root()
2260 for (i = 0; i < KEY_PTRS(&b->key); i++) in bch_btree_set_root()
2261 BUG_ON(PTR_BUCKET(b->c, &b->key, i)->prio != BTREE_PRIO); in bch_btree_set_root()
2263 mutex_lock(&b->c->bucket_lock); in bch_btree_set_root()
2264 list_del_init(&b->list); in bch_btree_set_root()
2265 mutex_unlock(&b->c->bucket_lock); in bch_btree_set_root()
2267 b->c->root = b; in bch_btree_set_root()
2269 bch_journal_meta(b->c, &cl); in bch_btree_set_root()
2275 static int bch_btree_map_nodes_recurse(struct btree *b, struct btree_op *op, in bch_btree_map_nodes_recurse() argument
2281 if (b->level) { in bch_btree_map_nodes_recurse()
2285 bch_btree_iter_init(&b->keys, &iter, from); in bch_btree_map_nodes_recurse()
2287 while ((k = bch_btree_iter_next_filter(&iter, &b->keys, in bch_btree_map_nodes_recurse()
2289 ret = btree(map_nodes_recurse, k, b, in bch_btree_map_nodes_recurse()
2298 if (!b->level || flags == MAP_ALL_NODES) in bch_btree_map_nodes_recurse()
2299 ret = fn(op, b); in bch_btree_map_nodes_recurse()
2310 static int bch_btree_map_keys_recurse(struct btree *b, struct btree_op *op, in bch_btree_map_keys_recurse() argument
2318 bch_btree_iter_init(&b->keys, &iter, from); in bch_btree_map_keys_recurse()
2320 while ((k = bch_btree_iter_next_filter(&iter, &b->keys, bch_ptr_bad))) { in bch_btree_map_keys_recurse()
2321 ret = !b->level in bch_btree_map_keys_recurse()
2322 ? fn(op, b, k) in bch_btree_map_keys_recurse()
2323 : btree(map_keys_recurse, k, b, op, from, fn, flags); in bch_btree_map_keys_recurse()
2330 if (!b->level && (flags & MAP_END_KEY)) in bch_btree_map_keys_recurse()
2331 ret = fn(op, b, &KEY(KEY_INODE(&b->key), in bch_btree_map_keys_recurse()
2332 KEY_OFFSET(&b->key), 0)); in bch_btree_map_keys_recurse()
2369 static int refill_keybuf_fn(struct btree_op *op, struct btree *b, in refill_keybuf_fn() argument