bcachefs: split up btree cache counters for live, freeable

this is prep for introducing a second live list and shrinker for pinned
nodes

Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

6 files changed, 47 insertions, 32 deletions

diff --git a/fs/bcachefs/btree_cache.c b/fs/bcachefs/btree_cache.c
index a7eb07d6e7f9..e66853b98857 100644
--- a/fs/bcachefs/btree_cache.c
+++ b/fs/bcachefs/btree_cache.c
@@ -49,7 +49,7 @@ void bch2_recalc_btree_reserve(struct bch_fs *c)
 
 static inline size_t btree_cache_can_free(struct btree_cache *bc)
 {
-	return max_t(int, 0, bc->nr_used - bc->nr_reserve);
+	return max_t(int, 0, bc->nr_live + bc->nr_freeable - bc->nr_reserve);
 }
 
 static void btree_node_to_freedlist(struct btree_cache *bc, struct btree *b)
@@ -64,6 +64,8 @@ static void btree_node_data_free(struct bch_fs *c, struct btree *b)
 {
 	struct btree_cache *bc = &c->btree_cache;
 
+	BUG_ON(btree_node_hashed(b));
+
 	/*
 	 * This should really be done in slub/vmalloc, but we're using the
 	 * kmalloc_large() path, so we're working around a slub bug by doing
@@ -87,7 +89,7 @@ static void btree_node_data_free(struct bch_fs *c, struct btree *b)
 #endif
 	b->aux_data = NULL;
 
-	bc->nr_used--;
+	bc->nr_freeable--;
 
 	btree_node_to_freedlist(bc, b);
 }
@@ -167,7 +169,7 @@ struct btree *__bch2_btree_node_mem_alloc(struct bch_fs *c)
 
 	bch2_btree_lock_init(&b->c, 0);
 
-	bc->nr_used++;
+	bc->nr_freeable++;
 	list_add(&b->list, &bc->freeable);
 	return b;
 }
@@ -186,6 +188,7 @@ void bch2_btree_node_to_freelist(struct bch_fs *c, struct btree *b)
 
 void bch2_btree_node_hash_remove(struct btree_cache *bc, struct btree *b)
 {
+	lockdep_assert_held(&bc->lock);
 	int ret = rhashtable_remove_fast(&bc->table, &b->hash, bch_btree_cache_params);
 
 	BUG_ON(ret);
@@ -195,6 +198,10 @@ void bch2_btree_node_hash_remove(struct btree_cache *bc, struct btree *b)
 
 	if (b->c.btree_id < BTREE_ID_NR)
 		--bc->nr_by_btree[b->c.btree_id];
+
+	bc->nr_live--;
+	bc->nr_freeable++;
+	list_move(&b->list, &bc->freeable);
 }
 
 int __bch2_btree_node_hash_insert(struct btree_cache *bc, struct btree *b)
@@ -204,23 +211,25 @@ int __bch2_btree_node_hash_insert(struct btree_cache *bc, struct btree *b)
 
 	int ret = rhashtable_lookup_insert_fast(&bc->table, &b->hash,
 						bch_btree_cache_params);
-	if (!ret && b->c.btree_id < BTREE_ID_NR)
+	if (ret)
+		return ret;
+
+	if (b->c.btree_id < BTREE_ID_NR)
 		bc->nr_by_btree[b->c.btree_id]++;
-	return ret;
+	bc->nr_live++;
+	bc->nr_freeable--;
+	list_move_tail(&b->list, &bc->live);
+	return 0;
 }
 
 int bch2_btree_node_hash_insert(struct btree_cache *bc, struct btree *b,
 				unsigned level, enum btree_id id)
 {
-	int ret;
-
 	b->c.level	= level;
 	b->c.btree_id	= id;
 
 	mutex_lock(&bc->lock);
-	ret = __bch2_btree_node_hash_insert(bc, b);
-	if (!ret)
-		list_add_tail(&b->list, &bc->live);
+	int ret = __bch2_btree_node_hash_insert(bc, b);
 	mutex_unlock(&bc->lock);
 
 	return ret;
@@ -402,7 +411,7 @@ static unsigned long bch2_btree_cache_scan(struct shrinker *shrink,
 	unsigned i, flags;
 	unsigned long ret = SHRINK_STOP;
 	bool trigger_writes = atomic_long_read(&bc->nr_dirty) + nr >=
-		bc->nr_used * 3 / 4;
+		(bc->nr_live + bc->nr_freeable) * 3 / 4;
 
 	if (bch2_btree_shrinker_disabled)
 		return SHRINK_STOP;
@@ -451,11 +460,12 @@ restart:
 			bc->not_freed[BCH_BTREE_CACHE_NOT_FREED_access_bit]++;
 			--touched;;
 		} else if (!btree_node_reclaim(c, b, true)) {
+			bch2_btree_node_hash_remove(bc, b);
+
 			freed++;
 			btree_node_data_free(c, b);
 			bc->nr_freed++;
 
-			bch2_btree_node_hash_remove(bc, b);
 			six_unlock_write(&b->c.lock);
 			six_unlock_intent(&b->c.lock);
 
@@ -506,7 +516,7 @@ static unsigned long bch2_btree_cache_count(struct shrinker *shrink,
 void bch2_fs_btree_cache_exit(struct bch_fs *c)
 {
 	struct btree_cache *bc = &c->btree_cache;
-	struct btree *b;
+	struct btree *b, *t;
 	unsigned i, flags;
 
 	shrinker_free(bc->shrink);
@@ -527,11 +537,10 @@ void bch2_fs_btree_cache_exit(struct bch_fs *c)
 			list_add(&r->b->list, &bc->live);
 	}
 
-	list_splice(&bc->freeable, &bc->live);
-
-	while (!list_empty(&bc->live)) {
-		b = list_first_entry(&bc->live, struct btree, list);
+	list_for_each_entry_safe(b, t, &bc->live, list)
+		bch2_btree_node_hash_remove(bc, b);
 
+	list_for_each_entry_safe(b, t, &bc->freeable, list) {
 		BUG_ON(btree_node_read_in_flight(b) ||
 		       btree_node_write_in_flight(b));
 
@@ -543,8 +552,7 @@ void bch2_fs_btree_cache_exit(struct bch_fs *c)
 
 	list_splice(&bc->freed_pcpu, &bc->freed_nonpcpu);
 
-	while (!list_empty(&bc->freed_nonpcpu)) {
-		b = list_first_entry(&bc->freed_nonpcpu, struct btree, list);
+	list_for_each_entry_safe(b, t, &bc->freed_nonpcpu, list) {
 		list_del(&b->list);
 		six_lock_exit(&b->c.lock);
 		kfree(b);
@@ -553,6 +561,11 @@ void bch2_fs_btree_cache_exit(struct bch_fs *c)
 	mutex_unlock(&bc->lock);
 	memalloc_nofs_restore(flags);
 
+	for (unsigned i = 0; i < ARRAY_SIZE(bc->nr_by_btree); i++)
+		BUG_ON(bc->nr_by_btree[i]);
+	BUG_ON(bc->nr_live);
+	BUG_ON(bc->nr_freeable);
+
 	if (bc->table_init_done)
 		rhashtable_destroy(&bc->table);
 }
@@ -739,7 +752,7 @@ got_node:
 	}
 
 	mutex_lock(&bc->lock);
-	bc->nr_used++;
+	bc->nr_freeable++;
 got_mem:
 	mutex_unlock(&bc->lock);
 
@@ -1280,8 +1293,8 @@ wait_on_io:
 	BUG_ON(btree_node_dirty(b));
 
 	mutex_lock(&bc->lock);
-	btree_node_data_free(c, b);
 	bch2_btree_node_hash_remove(bc, b);
+	btree_node_data_free(c, b);
 	mutex_unlock(&bc->lock);
 out:
 	six_unlock_write(&b->c.lock);
@@ -1374,7 +1387,8 @@ void bch2_btree_cache_to_text(struct printbuf *out, const struct btree_cache *bc
 	if (!out->nr_tabstops)
 		printbuf_tabstop_push(out, 32);
 
-	prt_btree_cache_line(out, c, "total:",		bc->nr_used);
+	prt_btree_cache_line(out, c, "nr_live:",	bc->nr_live);
+	prt_btree_cache_line(out, c, "nr_freeable:",	bc->nr_freeable);
 	prt_btree_cache_line(out, c, "nr dirty:",	atomic_long_read(&bc->nr_dirty));
 	prt_printf(out, "cannibalize lock:\t%p\n",	bc->alloc_lock);
 	prt_newline(out);
diff --git a/fs/bcachefs/btree_gc.c b/fs/bcachefs/btree_gc.c
index 120ffd68ab0a..b5e0692f03c6 100644
--- a/fs/bcachefs/btree_gc.c
+++ b/fs/bcachefs/btree_gc.c
@@ -549,9 +549,8 @@ reconstruct_root:
 		six_unlock_read(&b->c.lock);
 
 		if (ret == DROP_THIS_NODE) {
-			bch2_btree_node_hash_remove(&c->btree_cache, b);
 			mutex_lock(&c->btree_cache.lock);
-			list_move(&b->list, &c->btree_cache.freeable);
+			bch2_btree_node_hash_remove(&c->btree_cache, b);
 			mutex_unlock(&c->btree_cache.lock);
 
 			r->b = NULL;
diff --git a/fs/bcachefs/btree_io.c b/fs/bcachefs/btree_io.c
index aad89ba16b9b..cb48a9477514 100644
--- a/fs/bcachefs/btree_io.c
+++ b/fs/bcachefs/btree_io.c
@@ -1749,10 +1749,8 @@ static int __bch2_btree_root_read(struct btree_trans *trans, enum btree_id id,
 	bch2_btree_node_read(trans, b, true);
 
 	if (btree_node_read_error(b)) {
-		bch2_btree_node_hash_remove(&c->btree_cache, b);
-
 		mutex_lock(&c->btree_cache.lock);
-		list_move(&b->list, &c->btree_cache.freeable);
+		bch2_btree_node_hash_remove(&c->btree_cache, b);
 		mutex_unlock(&c->btree_cache.lock);
 
 		ret = -BCH_ERR_btree_node_read_error;
diff --git a/fs/bcachefs/btree_types.h b/fs/bcachefs/btree_types.h
index 806d27b7f41b..ee3df2a486cc 100644
--- a/fs/bcachefs/btree_types.h
+++ b/fs/bcachefs/btree_types.h
@@ -179,8 +179,8 @@ struct btree_cache {
 	struct list_head	freed_pcpu;
 	struct list_head	freed_nonpcpu;
 
-	/* Number of elements in live + freeable lists */
-	size_t			nr_used;
+	size_t			nr_live;
+	size_t			nr_freeable;
 	size_t			nr_reserve;
 	size_t			nr_by_btree[BTREE_ID_NR];
 	atomic_long_t		nr_dirty;
diff --git a/fs/bcachefs/btree_update_interior.c b/fs/bcachefs/btree_update_interior.c
index 24e445574ab8..18494a662e0a 100644
--- a/fs/bcachefs/btree_update_interior.c
+++ b/fs/bcachefs/btree_update_interior.c
@@ -251,8 +251,13 @@ static void bch2_btree_node_free_inmem(struct btree_trans *trans,
 	unsigned i, level = b->c.level;
 
 	bch2_btree_node_lock_write_nofail(trans, path, &b->c);
+
+	mutex_lock(&c->btree_cache.lock);
 	bch2_btree_node_hash_remove(&c->btree_cache, b);
+	mutex_unlock(&c->btree_cache.lock);
+
 	__btree_node_free(trans, b);
+
 	six_unlock_write(&b->c.lock);
 	mark_btree_node_locked_noreset(path, level, BTREE_NODE_INTENT_LOCKED);
 
@@ -284,7 +289,6 @@ static void bch2_btree_node_free_never_used(struct btree_update *as,
 	clear_btree_node_need_write(b);
 
 	mutex_lock(&c->btree_cache.lock);
-	list_del_init(&b->list);
 	bch2_btree_node_hash_remove(&c->btree_cache, b);
 	mutex_unlock(&c->btree_cache.lock);
 
diff --git a/fs/bcachefs/journal_reclaim.c b/fs/bcachefs/journal_reclaim.c
index 9794b6d214cd..f8e045982753 100644
--- a/fs/bcachefs/journal_reclaim.c
+++ b/fs/bcachefs/journal_reclaim.c
@@ -681,7 +681,7 @@ static int __bch2_journal_reclaim(struct journal *j, bool direct, bool kicked)
 		if (j->watermark != BCH_WATERMARK_stripe)
 			min_nr = 1;
 
-		if (atomic_long_read(&c->btree_cache.nr_dirty) * 2 > c->btree_cache.nr_used)
+		if (atomic_long_read(&c->btree_cache.nr_dirty) * 2 > c->btree_cache.nr_live)
 			min_nr = 1;
 
 		min_key_cache = min(bch2_nr_btree_keys_need_flush(c), (size_t) 128);
@@ -690,7 +690,7 @@ static int __bch2_journal_reclaim(struct journal *j, bool direct, bool kicked)
 				direct, kicked,
 				min_nr, min_key_cache,
 				atomic_long_read(&c->btree_cache.nr_dirty),
-				c->btree_cache.nr_used,
+				c->btree_cache.nr_live,
 				atomic_long_read(&c->btree_key_cache.nr_dirty),
 				atomic_long_read(&c->btree_key_cache.nr_keys));