1 // SPDX-License-Identifier: GPL-2.0
5 #include "btree_cache.h"
7 #include "btree_iter.h"
8 #include "btree_locking.h"
13 #include <linux/prefetch.h>
14 #include <linux/sched/mm.h>
15 #include <trace/events/bcachefs.h>
17 #define BTREE_CACHE_NOT_FREED_INCREMENT(counter) \
19 if (shrinker_counter) \
20 bc->not_freed_##counter++; \
23 const char * const bch2_btree_node_flags[] = {
30 void bch2_recalc_btree_reserve(struct bch_fs *c)
32 unsigned i, reserve = 16;
34 if (!c->btree_roots[0].b)
37 for (i = 0; i < BTREE_ID_NR; i++)
38 if (c->btree_roots[i].b)
39 reserve += min_t(unsigned, 1,
40 c->btree_roots[i].b->c.level) * 8;
42 c->btree_cache.reserve = reserve;
45 static inline unsigned btree_cache_can_free(struct btree_cache *bc)
47 return max_t(int, 0, bc->used - bc->reserve);
50 static void btree_node_to_freedlist(struct btree_cache *bc, struct btree *b)
52 if (b->c.lock.readers)
53 list_move(&b->list, &bc->freed_pcpu);
55 list_move(&b->list, &bc->freed_nonpcpu);
58 static void btree_node_data_free(struct bch_fs *c, struct btree *b)
60 struct btree_cache *bc = &c->btree_cache;
62 EBUG_ON(btree_node_write_in_flight(b));
64 kvpfree(b->data, btree_bytes(c));
69 munmap(b->aux_data, btree_aux_data_bytes(b));
75 btree_node_to_freedlist(bc, b);
78 static int bch2_btree_cache_cmp_fn(struct rhashtable_compare_arg *arg,
81 const struct btree *b = obj;
82 const u64 *v = arg->key;
84 return b->hash_val == *v ? 0 : 1;
87 static const struct rhashtable_params bch_btree_cache_params = {
88 .head_offset = offsetof(struct btree, hash),
89 .key_offset = offsetof(struct btree, hash_val),
90 .key_len = sizeof(u64),
91 .obj_cmpfn = bch2_btree_cache_cmp_fn,
94 static int btree_node_data_alloc(struct bch_fs *c, struct btree *b, gfp_t gfp)
96 BUG_ON(b->data || b->aux_data);
98 b->data = kvpmalloc(btree_bytes(c), gfp);
102 b->aux_data = vmalloc_exec(btree_aux_data_bytes(b), gfp);
104 b->aux_data = mmap(NULL, btree_aux_data_bytes(b),
105 PROT_READ|PROT_WRITE|PROT_EXEC,
106 MAP_PRIVATE|MAP_ANONYMOUS, 0, 0);
107 if (b->aux_data == MAP_FAILED)
111 kvpfree(b->data, btree_bytes(c));
119 static struct btree *__btree_node_mem_alloc(struct bch_fs *c, gfp_t gfp)
121 struct btree *b = kzalloc(sizeof(struct btree), gfp);
125 bkey_btree_ptr_init(&b->key);
126 __six_lock_init(&b->c.lock, "b->c.lock", &bch2_btree_node_lock_key);
127 #ifdef CONFIG_DEBUG_LOCK_ALLOC
128 lockdep_set_no_check_recursion(&b->c.lock.dep_map);
130 INIT_LIST_HEAD(&b->list);
131 INIT_LIST_HEAD(&b->write_blocked);
132 b->byte_order = ilog2(btree_bytes(c));
136 struct btree *__bch2_btree_node_mem_alloc(struct bch_fs *c)
138 struct btree_cache *bc = &c->btree_cache;
139 struct btree *b = __btree_node_mem_alloc(c, GFP_KERNEL);
143 if (btree_node_data_alloc(c, b, GFP_KERNEL)) {
149 list_add(&b->list, &bc->freeable);
153 /* Btree in memory cache - hash table */
155 void bch2_btree_node_hash_remove(struct btree_cache *bc, struct btree *b)
157 int ret = rhashtable_remove_fast(&bc->table, &b->hash, bch_btree_cache_params);
160 /* Cause future lookups for this node to fail: */
164 int __bch2_btree_node_hash_insert(struct btree_cache *bc, struct btree *b)
167 b->hash_val = btree_ptr_hash_val(&b->key);
169 return rhashtable_lookup_insert_fast(&bc->table, &b->hash,
170 bch_btree_cache_params);
173 int bch2_btree_node_hash_insert(struct btree_cache *bc, struct btree *b,
174 unsigned level, enum btree_id id)
181 mutex_lock(&bc->lock);
182 ret = __bch2_btree_node_hash_insert(bc, b);
184 list_add_tail(&b->list, &bc->live);
185 mutex_unlock(&bc->lock);
191 static inline struct btree *btree_cache_find(struct btree_cache *bc,
192 const struct bkey_i *k)
194 u64 v = btree_ptr_hash_val(k);
196 return rhashtable_lookup_fast(&bc->table, &v, bch_btree_cache_params);
200 * this version is for btree nodes that have already been freed (we're not
201 * reaping a real btree node)
203 static int __btree_node_reclaim(struct bch_fs *c, struct btree *b, bool flush, bool shrinker_counter)
205 struct btree_cache *bc = &c->btree_cache;
208 lockdep_assert_held(&bc->lock);
210 if (b->flags & ((1U << BTREE_NODE_dirty)|
211 (1U << BTREE_NODE_read_in_flight)|
212 (1U << BTREE_NODE_write_in_flight))) {
214 if (btree_node_dirty(b))
215 BTREE_CACHE_NOT_FREED_INCREMENT(dirty);
216 else if (btree_node_read_in_flight(b))
217 BTREE_CACHE_NOT_FREED_INCREMENT(read_in_flight);
218 else if (btree_node_write_in_flight(b))
219 BTREE_CACHE_NOT_FREED_INCREMENT(write_in_flight);
223 /* XXX: waiting on IO with btree cache lock held */
224 bch2_btree_node_wait_on_read(b);
225 bch2_btree_node_wait_on_write(b);
228 if (!six_trylock_intent(&b->c.lock)) {
229 BTREE_CACHE_NOT_FREED_INCREMENT(lock_intent);
233 if (!six_trylock_write(&b->c.lock)) {
234 BTREE_CACHE_NOT_FREED_INCREMENT(lock_write);
235 goto out_unlock_intent;
238 /* recheck under lock */
239 if (b->flags & ((1U << BTREE_NODE_read_in_flight)|
240 (1U << BTREE_NODE_write_in_flight))) {
242 if (btree_node_read_in_flight(b))
243 BTREE_CACHE_NOT_FREED_INCREMENT(read_in_flight);
244 else if (btree_node_write_in_flight(b))
245 BTREE_CACHE_NOT_FREED_INCREMENT(write_in_flight);
248 six_unlock_write(&b->c.lock);
249 six_unlock_intent(&b->c.lock);
253 if (btree_node_noevict(b)) {
254 BTREE_CACHE_NOT_FREED_INCREMENT(noevict);
257 if (btree_node_write_blocked(b)) {
258 BTREE_CACHE_NOT_FREED_INCREMENT(write_blocked);
261 if (btree_node_will_make_reachable(b)) {
262 BTREE_CACHE_NOT_FREED_INCREMENT(will_make_reachable);
266 if (btree_node_dirty(b)) {
268 BTREE_CACHE_NOT_FREED_INCREMENT(dirty);
272 * Using the underscore version because we don't want to compact
273 * bsets after the write, since this node is about to be evicted
274 * - unless btree verify mode is enabled, since it runs out of
275 * the post write cleanup:
277 if (bch2_verify_btree_ondisk)
278 bch2_btree_node_write(c, b, SIX_LOCK_intent, 0);
280 __bch2_btree_node_write(c, b, 0);
282 six_unlock_write(&b->c.lock);
283 six_unlock_intent(&b->c.lock);
287 if (b->hash_val && !ret)
288 trace_and_count(c, btree_cache_reap, c, b);
291 six_unlock_write(&b->c.lock);
293 six_unlock_intent(&b->c.lock);
298 static int btree_node_reclaim(struct bch_fs *c, struct btree *b, bool shrinker_counter)
300 return __btree_node_reclaim(c, b, false, shrinker_counter);
303 static int btree_node_write_and_reclaim(struct bch_fs *c, struct btree *b)
305 return __btree_node_reclaim(c, b, true, false);
308 static unsigned long bch2_btree_cache_scan(struct shrinker *shrink,
309 struct shrink_control *sc)
311 struct bch_fs *c = container_of(shrink, struct bch_fs,
313 struct btree_cache *bc = &c->btree_cache;
315 unsigned long nr = sc->nr_to_scan;
316 unsigned long can_free = 0;
317 unsigned long freed = 0;
318 unsigned long touched = 0;
320 unsigned long ret = SHRINK_STOP;
321 bool trigger_writes = atomic_read(&bc->dirty) + nr >=
324 if (bch2_btree_shrinker_disabled)
327 mutex_lock(&bc->lock);
328 flags = memalloc_nofs_save();
331 * It's _really_ critical that we don't free too many btree nodes - we
332 * have to always leave ourselves a reserve. The reserve is how we
333 * guarantee that allocating memory for a new btree node can always
334 * succeed, so that inserting keys into the btree can always succeed and
335 * IO can always make forward progress:
337 can_free = btree_cache_can_free(bc);
338 nr = min_t(unsigned long, nr, can_free);
341 list_for_each_entry_safe(b, t, &bc->freeable, list) {
343 * Leave a few nodes on the freeable list, so that a btree split
344 * won't have to hit the system allocator:
354 if (!btree_node_reclaim(c, b, true)) {
355 btree_node_data_free(c, b);
356 six_unlock_write(&b->c.lock);
357 six_unlock_intent(&b->c.lock);
363 list_for_each_entry_safe(b, t, &bc->live, list) {
366 if (btree_node_accessed(b)) {
367 clear_btree_node_accessed(b);
368 bc->not_freed_access_bit++;
369 } else if (!btree_node_reclaim(c, b, true)) {
371 btree_node_data_free(c, b);
374 bch2_btree_node_hash_remove(bc, b);
375 six_unlock_write(&b->c.lock);
376 six_unlock_intent(&b->c.lock);
380 } else if (trigger_writes &&
381 btree_node_dirty(b) &&
382 !btree_node_will_make_reachable(b) &&
383 !btree_node_write_blocked(b) &&
384 six_trylock_read(&b->c.lock)) {
385 list_move(&bc->live, &b->list);
386 mutex_unlock(&bc->lock);
387 __bch2_btree_node_write(c, b, 0);
388 six_unlock_read(&b->c.lock);
391 mutex_lock(&bc->lock);
399 if (&t->list != &bc->live)
400 list_move_tail(&bc->live, &t->list);
402 mutex_unlock(&bc->lock);
405 memalloc_nofs_restore(flags);
406 trace_and_count(c, btree_cache_scan, sc->nr_to_scan, can_free, ret);
410 static unsigned long bch2_btree_cache_count(struct shrinker *shrink,
411 struct shrink_control *sc)
413 struct bch_fs *c = container_of(shrink, struct bch_fs,
415 struct btree_cache *bc = &c->btree_cache;
417 if (bch2_btree_shrinker_disabled)
420 return btree_cache_can_free(bc);
423 static void bch2_btree_cache_shrinker_to_text(struct printbuf *out, struct shrinker *shrink)
425 struct bch_fs *c = container_of(shrink, struct bch_fs,
428 bch2_btree_cache_to_text(out, &c->btree_cache);
431 void bch2_fs_btree_cache_exit(struct bch_fs *c)
433 struct btree_cache *bc = &c->btree_cache;
437 if (bc->shrink.list.next)
438 unregister_shrinker(&bc->shrink);
440 /* vfree() can allocate memory: */
441 flags = memalloc_nofs_save();
442 mutex_lock(&bc->lock);
445 list_move(&c->verify_data->list, &bc->live);
447 kvpfree(c->verify_ondisk, btree_bytes(c));
449 for (i = 0; i < BTREE_ID_NR; i++)
450 if (c->btree_roots[i].b)
451 list_add(&c->btree_roots[i].b->list, &bc->live);
453 list_splice(&bc->freeable, &bc->live);
455 while (!list_empty(&bc->live)) {
456 b = list_first_entry(&bc->live, struct btree, list);
458 BUG_ON(btree_node_read_in_flight(b) ||
459 btree_node_write_in_flight(b));
461 if (btree_node_dirty(b))
462 bch2_btree_complete_write(c, b, btree_current_write(b));
463 clear_btree_node_dirty_acct(c, b);
465 btree_node_data_free(c, b);
468 BUG_ON(atomic_read(&c->btree_cache.dirty));
470 list_splice(&bc->freed_pcpu, &bc->freed_nonpcpu);
472 while (!list_empty(&bc->freed_nonpcpu)) {
473 b = list_first_entry(&bc->freed_nonpcpu, struct btree, list);
475 six_lock_pcpu_free(&b->c.lock);
479 mutex_unlock(&bc->lock);
480 memalloc_nofs_restore(flags);
482 if (bc->table_init_done)
483 rhashtable_destroy(&bc->table);
486 int bch2_fs_btree_cache_init(struct bch_fs *c)
488 struct btree_cache *bc = &c->btree_cache;
492 pr_verbose_init(c->opts, "");
494 ret = rhashtable_init(&bc->table, &bch_btree_cache_params);
498 bc->table_init_done = true;
500 bch2_recalc_btree_reserve(c);
502 for (i = 0; i < bc->reserve; i++)
503 if (!__bch2_btree_node_mem_alloc(c)) {
508 list_splice_init(&bc->live, &bc->freeable);
510 mutex_init(&c->verify_lock);
512 bc->shrink.count_objects = bch2_btree_cache_count;
513 bc->shrink.scan_objects = bch2_btree_cache_scan;
514 bc->shrink.to_text = bch2_btree_cache_shrinker_to_text;
515 bc->shrink.seeks = 4;
516 ret = register_shrinker(&bc->shrink, "%s/btree_cache", c->name);
518 pr_verbose_init(c->opts, "ret %i", ret);
522 void bch2_fs_btree_cache_init_early(struct btree_cache *bc)
524 mutex_init(&bc->lock);
525 INIT_LIST_HEAD(&bc->live);
526 INIT_LIST_HEAD(&bc->freeable);
527 INIT_LIST_HEAD(&bc->freed_pcpu);
528 INIT_LIST_HEAD(&bc->freed_nonpcpu);
532 * We can only have one thread cannibalizing other cached btree nodes at a time,
533 * or we'll deadlock. We use an open coded mutex to ensure that, which a
534 * cannibalize_bucket() will take. This means every time we unlock the root of
535 * the btree, we need to release this lock if we have it held.
537 void bch2_btree_cache_cannibalize_unlock(struct bch_fs *c)
539 struct btree_cache *bc = &c->btree_cache;
541 if (bc->alloc_lock == current) {
542 trace_and_count(c, btree_cache_cannibalize_unlock, c);
543 bc->alloc_lock = NULL;
544 closure_wake_up(&bc->alloc_wait);
548 int bch2_btree_cache_cannibalize_lock(struct bch_fs *c, struct closure *cl)
550 struct btree_cache *bc = &c->btree_cache;
551 struct task_struct *old;
553 old = cmpxchg(&bc->alloc_lock, NULL, current);
554 if (old == NULL || old == current)
558 trace_and_count(c, btree_cache_cannibalize_lock_fail, c);
562 closure_wait(&bc->alloc_wait, cl);
564 /* Try again, after adding ourselves to waitlist */
565 old = cmpxchg(&bc->alloc_lock, NULL, current);
566 if (old == NULL || old == current) {
568 closure_wake_up(&bc->alloc_wait);
572 trace_and_count(c, btree_cache_cannibalize_lock_fail, c);
576 trace_and_count(c, btree_cache_cannibalize_lock, c);
580 static struct btree *btree_node_cannibalize(struct bch_fs *c)
582 struct btree_cache *bc = &c->btree_cache;
585 list_for_each_entry_reverse(b, &bc->live, list)
586 if (!btree_node_reclaim(c, b, false))
590 list_for_each_entry_reverse(b, &bc->live, list)
591 if (!btree_node_write_and_reclaim(c, b))
595 * Rare case: all nodes were intent-locked.
598 WARN_ONCE(1, "btree cache cannibalize failed\n");
603 struct btree *bch2_btree_node_mem_alloc(struct bch_fs *c, bool pcpu_read_locks)
605 struct btree_cache *bc = &c->btree_cache;
606 struct list_head *freed = pcpu_read_locks
608 : &bc->freed_nonpcpu;
609 struct btree *b, *b2;
610 u64 start_time = local_clock();
613 flags = memalloc_nofs_save();
614 mutex_lock(&bc->lock);
617 * We never free struct btree itself, just the memory that holds the on
618 * disk node. Check the freed list before allocating a new one:
620 list_for_each_entry(b, freed, list)
621 if (!btree_node_reclaim(c, b, false)) {
622 list_del_init(&b->list);
626 b = __btree_node_mem_alloc(c, __GFP_NOWARN);
628 mutex_unlock(&bc->lock);
629 b = __btree_node_mem_alloc(c, GFP_KERNEL);
632 mutex_lock(&bc->lock);
636 six_lock_pcpu_alloc(&b->c.lock);
638 BUG_ON(!six_trylock_intent(&b->c.lock));
639 BUG_ON(!six_trylock_write(&b->c.lock));
643 * btree_free() doesn't free memory; it sticks the node on the end of
644 * the list. Check if there's any freed nodes there:
646 list_for_each_entry(b2, &bc->freeable, list)
647 if (!btree_node_reclaim(c, b2, false)) {
648 swap(b->data, b2->data);
649 swap(b->aux_data, b2->aux_data);
650 btree_node_to_freedlist(bc, b2);
651 six_unlock_write(&b2->c.lock);
652 six_unlock_intent(&b2->c.lock);
656 mutex_unlock(&bc->lock);
658 if (btree_node_data_alloc(c, b, __GFP_NOWARN|GFP_KERNEL))
661 mutex_lock(&bc->lock);
664 mutex_unlock(&bc->lock);
666 BUG_ON(btree_node_hashed(b));
667 BUG_ON(btree_node_dirty(b));
668 BUG_ON(btree_node_write_in_flight(b));
675 b->whiteout_u64s = 0;
676 bch2_btree_keys_init(b);
677 set_btree_node_accessed(b);
679 bch2_time_stats_update(&c->times[BCH_TIME_btree_node_mem_alloc],
682 memalloc_nofs_restore(flags);
685 mutex_lock(&bc->lock);
687 /* Try to cannibalize another cached btree node: */
688 if (bc->alloc_lock == current) {
689 b2 = btree_node_cannibalize(c);
690 bch2_btree_node_hash_remove(bc, b2);
693 swap(b->data, b2->data);
694 swap(b->aux_data, b2->aux_data);
695 btree_node_to_freedlist(bc, b2);
696 six_unlock_write(&b2->c.lock);
697 six_unlock_intent(&b2->c.lock);
700 list_del_init(&b->list);
703 mutex_unlock(&bc->lock);
705 trace_and_count(c, btree_cache_cannibalize, c);
709 mutex_unlock(&bc->lock);
710 memalloc_nofs_restore(flags);
711 return ERR_PTR(-ENOMEM);
714 /* Slowpath, don't want it inlined into btree_iter_traverse() */
715 static noinline struct btree *bch2_btree_node_fill(struct bch_fs *c,
716 struct btree_trans *trans,
717 struct btree_path *path,
718 const struct bkey_i *k,
719 enum btree_id btree_id,
721 enum six_lock_type lock_type,
724 struct btree_cache *bc = &c->btree_cache;
728 BUG_ON(level + 1 >= BTREE_MAX_DEPTH);
730 * Parent node must be locked, else we could read in a btree node that's
733 if (trans && !bch2_btree_node_relock(trans, path, level + 1)) {
734 trace_and_count(c, trans_restart_relock_parent_for_fill, trans, _THIS_IP_, path);
735 return ERR_PTR(btree_trans_restart(trans, BCH_ERR_transaction_restart_fill_relock));
738 b = bch2_btree_node_mem_alloc(c, level != 0);
740 if (trans && b == ERR_PTR(-ENOMEM)) {
741 trans->memory_allocation_failure = true;
742 trace_and_count(c, trans_restart_memory_allocation_failure, trans, _THIS_IP_, path);
743 return ERR_PTR(btree_trans_restart(trans, BCH_ERR_transaction_restart_fill_mem_alloc_fail));
749 bkey_copy(&b->key, k);
750 if (bch2_btree_node_hash_insert(bc, b, level, btree_id)) {
751 /* raced with another fill: */
753 /* mark as unhashed... */
756 mutex_lock(&bc->lock);
757 list_add(&b->list, &bc->freeable);
758 mutex_unlock(&bc->lock);
760 six_unlock_write(&b->c.lock);
761 six_unlock_intent(&b->c.lock);
765 set_btree_node_read_in_flight(b);
767 six_unlock_write(&b->c.lock);
768 seq = b->c.lock.state.seq;
769 six_unlock_intent(&b->c.lock);
771 /* Unlock before doing IO: */
773 bch2_trans_unlock(trans);
775 bch2_btree_node_read(c, b, sync);
781 int ret = bch2_trans_relock(trans) ?:
782 bch2_btree_path_relock_intent(trans, path);
784 BUG_ON(!trans->restarted);
789 if (!six_relock_type(&b->c.lock, lock_type, seq)) {
791 trace_and_count(c, trans_restart_relock_after_fill, trans, _THIS_IP_, path);
792 return ERR_PTR(btree_trans_restart(trans, BCH_ERR_transaction_restart_relock_after_fill));
798 static noinline void btree_bad_header(struct bch_fs *c, struct btree *b)
800 struct printbuf buf = PRINTBUF;
802 if (!test_bit(BCH_FS_INITIAL_GC_DONE, &c->flags))
806 "btree node header doesn't match ptr\n"
807 "btree %s level %u\n"
809 bch2_btree_ids[b->c.btree_id], b->c.level);
810 bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(&b->key));
812 prt_printf(&buf, "\nheader: btree %s level %llu\n"
814 bch2_btree_ids[BTREE_NODE_ID(b->data)],
815 BTREE_NODE_LEVEL(b->data));
816 bch2_bpos_to_text(&buf, b->data->min_key);
818 prt_printf(&buf, "\nmax ");
819 bch2_bpos_to_text(&buf, b->data->max_key);
821 bch2_fs_inconsistent(c, "%s", buf.buf);
825 static inline void btree_check_header(struct bch_fs *c, struct btree *b)
827 if (b->c.btree_id != BTREE_NODE_ID(b->data) ||
828 b->c.level != BTREE_NODE_LEVEL(b->data) ||
829 bpos_cmp(b->data->max_key, b->key.k.p) ||
830 (b->key.k.type == KEY_TYPE_btree_ptr_v2 &&
831 bpos_cmp(b->data->min_key,
832 bkey_i_to_btree_ptr_v2(&b->key)->v.min_key)))
833 btree_bad_header(c, b);
837 * bch_btree_node_get - find a btree node in the cache and lock it, reading it
838 * in from disk if necessary.
840 * If IO is necessary and running under generic_make_request, returns -EAGAIN.
842 * The btree node will have either a read or a write lock held, depending on
843 * the @write parameter.
845 struct btree *bch2_btree_node_get(struct btree_trans *trans, struct btree_path *path,
846 const struct bkey_i *k, unsigned level,
847 enum six_lock_type lock_type,
848 unsigned long trace_ip)
850 struct bch_fs *c = trans->c;
851 struct btree_cache *bc = &c->btree_cache;
856 EBUG_ON(level >= BTREE_MAX_DEPTH);
858 b = btree_node_mem_ptr(k);
861 * Check b->hash_val _before_ calling btree_node_lock() - this might not
862 * be the node we want anymore, and trying to lock the wrong node could
863 * cause an unneccessary transaction restart:
865 if (likely(c->opts.btree_node_mem_ptr_optimization &&
867 b->hash_val == btree_ptr_hash_val(k)))
870 b = btree_cache_find(bc, k);
873 * We must have the parent locked to call bch2_btree_node_fill(),
874 * else we could read in a btree node from disk that's been
877 b = bch2_btree_node_fill(c, trans, path, k, path->btree_id,
878 level, lock_type, true);
880 /* We raced and found the btree node in the cache */
889 * There's a potential deadlock with splits and insertions into
890 * interior nodes we have to avoid:
892 * The other thread might be holding an intent lock on the node
893 * we want, and they want to update its parent node so they're
894 * going to upgrade their intent lock on the parent node to a
897 * But if we're holding a read lock on the parent, and we're
898 * trying to get the intent lock they're holding, we deadlock.
900 * So to avoid this we drop the read locks on parent nodes when
901 * we're starting to take intent locks - and handle the race.
903 * The race is that they might be about to free the node we
904 * want, and dropping our read lock on the parent node lets them
905 * update the parent marking the node we want as freed, and then
908 * To guard against this, btree nodes are evicted from the cache
909 * when they're freed - and b->hash_val is zeroed out, which we
910 * check for after we lock the node.
912 * Then, bch2_btree_node_relock() on the parent will fail - because
913 * the parent was modified, when the pointer to the node we want
914 * was removed - and we'll bail out:
916 if (btree_node_read_locked(path, level + 1))
917 btree_node_unlock(trans, path, level + 1);
919 ret = btree_node_lock(trans, path, &b->c, level, lock_type, trace_ip);
920 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
925 if (unlikely(b->hash_val != btree_ptr_hash_val(k) ||
926 b->c.level != level ||
928 six_unlock_type(&b->c.lock, lock_type);
929 if (bch2_btree_node_relock(trans, path, level + 1))
932 trace_and_count(c, trans_restart_btree_node_reused, trans, trace_ip, path);
933 return ERR_PTR(btree_trans_restart(trans, BCH_ERR_transaction_restart_lock_node_reused));
937 if (unlikely(btree_node_read_in_flight(b))) {
938 u32 seq = b->c.lock.state.seq;
940 six_unlock_type(&b->c.lock, lock_type);
941 bch2_trans_unlock(trans);
943 bch2_btree_node_wait_on_read(b);
946 * should_be_locked is not set on this path yet, so we need to
947 * relock it specifically:
950 int ret = bch2_trans_relock(trans) ?:
951 bch2_btree_path_relock_intent(trans, path);
953 BUG_ON(!trans->restarted);
958 if (!six_relock_type(&b->c.lock, lock_type, seq))
962 prefetch(b->aux_data);
964 for_each_bset(b, t) {
965 void *p = (u64 *) b->aux_data + t->aux_data_offset;
967 prefetch(p + L1_CACHE_BYTES * 0);
968 prefetch(p + L1_CACHE_BYTES * 1);
969 prefetch(p + L1_CACHE_BYTES * 2);
972 /* avoid atomic set bit if it's not needed: */
973 if (!btree_node_accessed(b))
974 set_btree_node_accessed(b);
976 if (unlikely(btree_node_read_error(b))) {
977 six_unlock_type(&b->c.lock, lock_type);
978 return ERR_PTR(-EIO);
981 EBUG_ON(b->c.btree_id != path->btree_id);
982 EBUG_ON(BTREE_NODE_LEVEL(b->data) != level);
983 btree_check_header(c, b);
988 struct btree *bch2_btree_node_get_noiter(struct btree_trans *trans,
989 const struct bkey_i *k,
990 enum btree_id btree_id,
994 struct bch_fs *c = trans->c;
995 struct btree_cache *bc = &c->btree_cache;
1000 EBUG_ON(level >= BTREE_MAX_DEPTH);
1002 if (c->opts.btree_node_mem_ptr_optimization) {
1003 b = btree_node_mem_ptr(k);
1008 b = btree_cache_find(bc, k);
1013 b = bch2_btree_node_fill(c, NULL, NULL, k, btree_id,
1014 level, SIX_LOCK_read, true);
1016 /* We raced and found the btree node in the cache */
1021 !bch2_btree_cache_cannibalize_lock(c, NULL))
1028 ret = btree_node_lock_nopath(trans, &b->c, SIX_LOCK_read);
1029 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
1030 return ERR_PTR(ret);
1034 if (unlikely(b->hash_val != btree_ptr_hash_val(k) ||
1035 b->c.btree_id != btree_id ||
1036 b->c.level != level)) {
1037 six_unlock_read(&b->c.lock);
1042 /* XXX: waiting on IO with btree locks held: */
1043 __bch2_btree_node_wait_on_read(b);
1045 prefetch(b->aux_data);
1047 for_each_bset(b, t) {
1048 void *p = (u64 *) b->aux_data + t->aux_data_offset;
1050 prefetch(p + L1_CACHE_BYTES * 0);
1051 prefetch(p + L1_CACHE_BYTES * 1);
1052 prefetch(p + L1_CACHE_BYTES * 2);
1055 /* avoid atomic set bit if it's not needed: */
1056 if (!btree_node_accessed(b))
1057 set_btree_node_accessed(b);
1059 if (unlikely(btree_node_read_error(b))) {
1060 six_unlock_read(&b->c.lock);
1065 EBUG_ON(b->c.btree_id != btree_id);
1066 EBUG_ON(BTREE_NODE_LEVEL(b->data) != level);
1067 btree_check_header(c, b);
1069 bch2_btree_cache_cannibalize_unlock(c);
1073 int bch2_btree_node_prefetch(struct bch_fs *c,
1074 struct btree_trans *trans,
1075 struct btree_path *path,
1076 const struct bkey_i *k,
1077 enum btree_id btree_id, unsigned level)
1079 struct btree_cache *bc = &c->btree_cache;
1082 BUG_ON(trans && !btree_node_locked(path, level + 1));
1083 BUG_ON(level >= BTREE_MAX_DEPTH);
1085 b = btree_cache_find(bc, k);
1089 b = bch2_btree_node_fill(c, trans, path, k, btree_id,
1090 level, SIX_LOCK_read, false);
1091 return PTR_ERR_OR_ZERO(b);
1094 void bch2_btree_node_evict(struct btree_trans *trans, const struct bkey_i *k)
1096 struct bch_fs *c = trans->c;
1097 struct btree_cache *bc = &c->btree_cache;
1100 b = btree_cache_find(bc, k);
1104 /* not allowed to wait on io with btree locks held: */
1106 /* XXX we're called from btree_gc which will be holding other btree
1109 __bch2_btree_node_wait_on_read(b);
1110 __bch2_btree_node_wait_on_write(b);
1112 btree_node_lock_nopath_nofail(trans, &b->c, SIX_LOCK_intent);
1113 btree_node_lock_nopath_nofail(trans, &b->c, SIX_LOCK_write);
1115 if (btree_node_dirty(b)) {
1116 __bch2_btree_node_write(c, b, 0);
1117 six_unlock_write(&b->c.lock);
1118 six_unlock_intent(&b->c.lock);
1122 BUG_ON(btree_node_dirty(b));
1124 mutex_lock(&bc->lock);
1125 btree_node_data_free(c, b);
1126 bch2_btree_node_hash_remove(bc, b);
1127 mutex_unlock(&bc->lock);
1129 six_unlock_write(&b->c.lock);
1130 six_unlock_intent(&b->c.lock);
1133 void bch2_btree_node_to_text(struct printbuf *out, struct bch_fs *c,
1136 const struct bkey_format *f = &b->format;
1137 struct bset_stats stats;
1139 memset(&stats, 0, sizeof(stats));
1141 bch2_btree_keys_stats(b, &stats);
1143 prt_printf(out, "l %u ", b->c.level);
1144 bch2_bpos_to_text(out, b->data->min_key);
1145 prt_printf(out, " - ");
1146 bch2_bpos_to_text(out, b->data->max_key);
1147 prt_printf(out, ":\n"
1149 bch2_val_to_text(out, c, bkey_i_to_s_c(&b->key));
1151 prt_printf(out, "\n"
1152 " format: u64s %u fields %u %u %u %u %u\n"
1153 " unpack fn len: %u\n"
1154 " bytes used %zu/%zu (%zu%% full)\n"
1155 " sib u64s: %u, %u (merge threshold %u)\n"
1156 " nr packed keys %u\n"
1157 " nr unpacked keys %u\n"
1159 " failed unpacked %zu\n",
1161 f->bits_per_field[0],
1162 f->bits_per_field[1],
1163 f->bits_per_field[2],
1164 f->bits_per_field[3],
1165 f->bits_per_field[4],
1167 b->nr.live_u64s * sizeof(u64),
1168 btree_bytes(c) - sizeof(struct btree_node),
1169 b->nr.live_u64s * 100 / btree_max_u64s(c),
1172 c->btree_foreground_merge_threshold,
1174 b->nr.unpacked_keys,
1179 void bch2_btree_cache_to_text(struct printbuf *out, struct btree_cache *bc)
1181 prt_printf(out, "nr nodes:\t\t%u\n", bc->used);
1182 prt_printf(out, "nr dirty:\t\t%u\n", atomic_read(&bc->dirty));
1183 prt_printf(out, "cannibalize lock:\t%p\n", bc->alloc_lock);
1185 prt_printf(out, "freed:\t\t\t\t%u\n", bc->freed);
1186 prt_printf(out, "not freed, dirty:\t\t%u\n", bc->not_freed_dirty);
1187 prt_printf(out, "not freed, write in flight:\t%u\n", bc->not_freed_write_in_flight);
1188 prt_printf(out, "not freed, read in flight:\t%u\n", bc->not_freed_read_in_flight);
1189 prt_printf(out, "not freed, lock intent failed:\t%u\n", bc->not_freed_lock_intent);
1190 prt_printf(out, "not freed, lock write failed:\t%u\n", bc->not_freed_lock_write);
1191 prt_printf(out, "not freed, access bit:\t\t%u\n", bc->not_freed_access_bit);
1192 prt_printf(out, "not freed, no evict failed:\t%u\n", bc->not_freed_noevict);
1193 prt_printf(out, "not freed, write blocked:\t%u\n", bc->not_freed_write_blocked);
1194 prt_printf(out, "not freed, will make reachable:\t%u\n", bc->not_freed_will_make_reachable);