bcache: Add struct btree_keys

Soon, bset.c won't need to depend on struct btree.
Signed-off-by: Kent Overstreet <kmo@daterainc.com>

drivers/md/bcache/bcache.h

@@ -679,9 +679,9 @@ struct cache_set {
 	unsigned		error_decay;
 
 	unsigned short		journal_delay_ms;
+	bool			expensive_debug_checks;
 	unsigned		verify:1;
 	unsigned		key_merging_disabled:1;
-	unsigned		expensive_debug_checks:1;
 	unsigned		gc_always_rewrite:1;
 	unsigned		shrinker_disabled:1;
 	unsigned		copy_gc_enabled:1;

drivers/md/bcache/bset.h

@@ -145,6 +145,9 @@
  */
 
 struct btree;
+struct btree_keys;
+struct btree_iter;
+struct btree_iter_set;
 struct bkey_float;
 
 #define MAX_BSETS		4U
@@ -181,6 +184,74 @@ struct bset_tree {
 	struct bset		*data;
 };
 
+struct btree_keys_ops {
+	bool		(*sort_cmp)(struct btree_iter_set,
+				    struct btree_iter_set);
+	struct bkey	*(*sort_fixup)(struct btree_iter *, struct bkey *);
+	bool		(*key_invalid)(struct btree_keys *,
+				       const struct bkey *);
+	bool		(*key_bad)(struct btree_keys *, const struct bkey *);
+	bool		(*key_merge)(struct btree_keys *,
+				     struct bkey *, struct bkey *);
+
+	/*
+	 * Only used for deciding whether to use START_KEY(k) or just the key
+	 * itself in a couple places
+	 */
+	bool		is_extents;
+};
+
+struct btree_keys {
+	const struct btree_keys_ops	*ops;
+	uint8_t			page_order;
+	uint8_t			nsets;
+	unsigned		last_set_unwritten:1;
+	bool			*expensive_debug_checks;
+
+	/*
+	 * Sets of sorted keys - the real btree node - plus a binary search tree
+	 *
+	 * set[0] is special; set[0]->tree, set[0]->prev and set[0]->data point
+	 * to the memory we have allocated for this btree node. Additionally,
+	 * set[0]->data points to the entire btree node as it exists on disk.
+	 */
+	struct bset_tree	set[MAX_BSETS];
+};
+
+static inline struct bset_tree *bset_tree_last(struct btree_keys *b)
+{
+	return b->set + b->nsets;
+}
+
+static inline bool bset_written(struct btree_keys *b, struct bset_tree *t)
+{
+	return t <= b->set + b->nsets - b->last_set_unwritten;
+}
+
+static inline bool bkey_written(struct btree_keys *b, struct bkey *k)
+{
+	return !b->last_set_unwritten || k < b->set[b->nsets].data->start;
+}
+
+static inline unsigned bset_byte_offset(struct btree_keys *b, struct bset *i)
+{
+	return ((size_t) i) - ((size_t) b->set->data);
+}
+
+static inline unsigned bset_sector_offset(struct btree_keys *b, struct bset *i)
+{
+	return bset_byte_offset(b, i) >> 9;
+}
+
+static inline bool btree_keys_expensive_checks(struct btree_keys *b)
+{
+#ifdef CONFIG_BCACHE_DEBUG
+	return *b->expensive_debug_checks;
+#else
+	return false;
+#endif
+}
+
 #define __set_bytes(i, k)	(sizeof(*(i)) + (k) * sizeof(uint64_t))
 #define set_bytes(i)		__set_bytes(i, i->keys)
 
@@ -189,12 +260,34 @@ struct bset_tree {
 #define set_blocks(i, block_bytes)				\
 	__set_blocks(i, (i)->keys, block_bytes)
 
-void bch_btree_keys_free(struct btree *);
-int bch_btree_keys_alloc(struct btree *, unsigned, gfp_t);
+static inline struct bset *bset_next_set(struct btree_keys *b,
+					 unsigned block_bytes)
+{
+	struct bset *i = bset_tree_last(b)->data;
+
+	return ((void *) i) + roundup(set_bytes(i), block_bytes);
+}
+
+void bch_btree_keys_free(struct btree_keys *);
+int bch_btree_keys_alloc(struct btree_keys *, unsigned, gfp_t);
+void bch_btree_keys_init(struct btree_keys *, const struct btree_keys_ops *,
+			 bool *);
 
-void bch_bset_fix_invalidated_key(struct btree *, struct bkey *);
-void bch_bset_init_next(struct btree *, struct bset *, uint64_t);
-void bch_bset_insert(struct btree *, struct bkey *, struct bkey *);
+void bch_bset_init_next(struct btree_keys *, struct bset *, uint64_t);
+void bch_bset_build_written_tree(struct btree_keys *);
+void bch_bset_fix_invalidated_key(struct btree_keys *, struct bkey *);
+void bch_bset_insert(struct btree_keys *, struct bkey *, struct bkey *);
+
+/*
+ * Tries to merge l and r: l should be lower than r
+ * Returns true if we were able to merge. If we did merge, l will be the merged
+ * key, r will be untouched.
+ */
+static inline bool bch_bkey_try_merge(struct btree_keys *b,
+				      struct bkey *l, struct bkey *r)
+{
+	return b->ops->key_merge ?  b->ops->key_merge(b, l, r) : false;
+}
 
 /* Btree key iteration */
 
@@ -208,11 +301,11 @@ struct btree_iter {
 	} data[MAX_BSETS];
 };
 
-typedef bool (*ptr_filter_fn)(struct btree *, const struct bkey *);
+typedef bool (*ptr_filter_fn)(struct btree_keys *, const struct bkey *);
 
 struct bkey *bch_btree_iter_next(struct btree_iter *);
 struct bkey *bch_btree_iter_next_filter(struct btree_iter *,
-					struct btree *, ptr_filter_fn);
+					struct btree_keys *, ptr_filter_fn);
 
 void bch_btree_iter_push(struct btree_iter *, struct bkey *, struct bkey *);
 struct bkey *bch_btree_iter_init(struct btree *, struct btree_iter *,
@@ -246,7 +339,7 @@ int bch_bset_sort_state_init(struct bset_sort_state *, unsigned);
 void bch_btree_sort_lazy(struct btree *, struct bset_sort_state *);
 void bch_btree_sort_into(struct btree *, struct btree *,
 			 struct bset_sort_state *);
-void bch_btree_sort_and_fix_extents(struct btree *, struct btree_iter *,
+void bch_btree_sort_and_fix_extents(struct btree_keys *, struct btree_iter *,
 				    struct bset_sort_state *);
 void bch_btree_sort_partial(struct btree *, unsigned,
 			    struct bset_sort_state *);
@@ -311,6 +404,16 @@ static inline bool bch_cut_back(const struct bkey *where, struct bkey *k)
 	_ret;							\
 })
 
+static inline bool bch_ptr_invalid(struct btree_keys *b, const struct bkey *k)
+{
+	return b->ops->key_invalid(b, k);
+}
+
+static inline bool bch_ptr_bad(struct btree_keys *b, const struct bkey *k)
+{
+	return b->ops->key_bad(b, k);
+}
+
 /* Keylists */
 
 struct keylist {

drivers/md/bcache/btree.h

@@ -113,28 +113,7 @@ struct btree_write {
 	int			prio_blocked;
 };
 
-struct btree_keys_ops {
-	bool			(*sort_cmp)(struct btree_iter_set,
-					    struct btree_iter_set);
-	struct bkey		*(*sort_fixup)(struct btree_iter *,
-					       struct bkey *);
-	bool			(*key_invalid)(struct btree *,
-					       const struct bkey *);
-	bool			(*key_bad)(struct btree *,
-					   const struct bkey *);
-	bool			(*key_merge)(struct btree *,
-					     struct bkey *, struct bkey *);
-
-
-	/*
-	 * Only used for deciding whether to use START_KEY(k) or just the key
-	 * itself in a couple places
-	 */
-	bool		is_extents;
-};
-
 struct btree {
-	const struct btree_keys_ops	*ops;
 	/* Hottest entries first */
 	struct hlist_node	hash;
 
@@ -151,17 +130,8 @@ struct btree {
 	unsigned long		flags;
 	uint16_t		written;	/* would be nice to kill */
 	uint8_t			level;
-	uint8_t			nsets;
-	uint8_t			page_order;
 
-	/*
-	 * Set of sorted keys - the real btree node - plus a binary search tree
-	 *
-	 * sets[0] is special; set[0]->tree, set[0]->prev and set[0]->data point
-	 * to the memory we have allocated for this btree node. Additionally,
-	 * set[0]->data points to the entire btree node as it exists on disk.
-	 */
-	struct bset_tree	sets[MAX_BSETS];
+	struct btree_keys	keys;
 
 	/* For outstanding btree writes, used as a lock - protects write_idx */
 	struct closure		io;
@@ -201,49 +171,19 @@ static inline struct btree_write *btree_prev_write(struct btree *b)
 	return b->writes + (btree_node_write_idx(b) ^ 1);
 }
 
-static inline struct bset_tree *bset_tree_last(struct btree *b)
-{
-	return b->sets + b->nsets;
-}
-
 static inline struct bset *btree_bset_first(struct btree *b)
 {
-	return b->sets->data;
+	return b->keys.set->data;
 }
 
 static inline struct bset *btree_bset_last(struct btree *b)
 {
-	return bset_tree_last(b)->data;
-}
-
-static inline unsigned bset_byte_offset(struct btree *b, struct bset *i)
-{
-	return ((size_t) i) - ((size_t) b->sets->data);
-}
-
-static inline unsigned bset_sector_offset(struct btree *b, struct bset *i)
-{
-	return (((void *) i) - ((void *) btree_bset_first(b))) >> 9;
+	return bset_tree_last(&b->keys)->data;
 }
 
 static inline unsigned bset_block_offset(struct btree *b, struct bset *i)
 {
-	return bset_sector_offset(b, i) >> b->c->block_bits;
-}
-
-static inline struct bset *write_block(struct btree *b)
-{
-	return ((void *) b->sets[0].data) + b->written * block_bytes(b->c);
-}
-
-static inline bool bset_written(struct btree *b, struct bset_tree *t)
-{
-	return t->data < write_block(b);
-}
-
-static inline bool bkey_written(struct btree *b, struct bkey *k)
-{
-	return k < write_block(b)->start;
+	return bset_sector_offset(&b->keys, i) >> b->c->block_bits;
 }
 
 static inline void set_gc_sectors(struct cache_set *c)
@@ -251,27 +191,6 @@ static inline void set_gc_sectors(struct cache_set *c)
 	atomic_set(&c->sectors_to_gc, c->sb.bucket_size * c->nbuckets / 16);
 }
 
-static inline bool bch_ptr_invalid(struct btree *b, const struct bkey *k)
-{
-	return b->ops->key_invalid(b, k);
-}
-
-static inline bool bch_ptr_bad(struct btree *b, const struct bkey *k)
-{
-	return b->ops->key_bad(b, k);
-}
-
-/*
- * Tries to merge l and r: l should be lower than r
- * Returns true if we were able to merge. If we did merge, l will be the merged
- * key, r will be untouched.
- */
-static inline bool bch_bkey_try_merge(struct btree *b,
-				      struct bkey *l, struct bkey *r)
-{
-	return b->ops->key_merge ?  b->ops->key_merge(b, l, r) : false;
-}
-
 void bkey_put(struct cache_set *c, struct bkey *k);
 
 /* Looping macros */
@@ -284,7 +203,7 @@ void bkey_put(struct cache_set *c, struct bkey *k);
 
 #define for_each_key_filter(b, k, iter, filter)				\
 	for (bch_btree_iter_init((b), (iter), NULL);			\
-	     ((k) = bch_btree_iter_next_filter((iter), b, filter));)
+	     ((k) = bch_btree_iter_next_filter((iter), &(b)->keys, filter));)
 
 #define for_each_key(b, k, iter)					\
 	for (bch_btree_iter_init((b), (iter), NULL);			\

drivers/md/bcache/debug.c

@@ -113,9 +113,9 @@ static void bch_dump_bucket(struct btree *b)
 	unsigned i;
 
 	console_lock();
-	for (i = 0; i <= b->nsets; i++)
-		dump_bset(b, b->sets[i].data,
-			  bset_block_offset(b, b->sets[i].data));
+	for (i = 0; i <= b->keys.nsets; i++)
+		dump_bset(b, b->keys.set[i].data,
+			  bset_block_offset(b, b->keys.set[i].data));
 	console_unlock();
 }
 
@@ -139,13 +139,13 @@ void bch_btree_verify(struct btree *b)
 	mutex_lock(&b->c->verify_lock);
 
 	ondisk = b->c->verify_ondisk;
-	sorted = b->c->verify_data->sets->data;
-	inmemory = b->sets->data;
+	sorted = b->c->verify_data->keys.set->data;
+	inmemory = b->keys.set->data;
 
 	bkey_copy(&v->key, &b->key);
 	v->written = 0;
 	v->level = b->level;
-	v->ops = b->ops;
+	v->keys.ops = b->keys.ops;
 
 	bio = bch_bbio_alloc(b->c);
 	bio->bi_bdev		= PTR_CACHE(b->c, &b->key, 0)->bdev;
@@ -159,7 +159,7 @@ void bch_btree_verify(struct btree *b)
 	memcpy(ondisk, sorted, KEY_SIZE(&v->key) << 9);
 
 	bch_btree_node_read_done(v);
-	sorted = v->sets->data;
+	sorted = v->keys.set->data;
 
 	if (inmemory->keys != sorted->keys ||
 	    memcmp(inmemory->start,
@@ -264,14 +264,14 @@ void __bch_check_keys(struct btree *b, const char *fmt, ...)
 			if (p && bkey_cmp(&START_KEY(p), &START_KEY(k)) > 0)
 				goto bug;
 
-			if (bch_ptr_invalid(b, k))
+			if (bch_ptr_invalid(&b->keys, k))
 				continue;
 
 			err =  "Overlapping keys";
 			if (p && bkey_cmp(p, &START_KEY(k)) > 0)
 				goto bug;
 		} else {
-			if (bch_ptr_bad(b, k))
+			if (bch_ptr_bad(&b->keys, k))
 				continue;
 
 			err = "Duplicate keys";

drivers/md/bcache/extents.c

@@ -81,8 +81,9 @@ bool __bch_btree_ptr_invalid(struct cache_set *c, const struct bkey *k)
 	return true;
 }
 
-static bool bch_btree_ptr_invalid(struct btree *b, const struct bkey *k)
+static bool bch_btree_ptr_invalid(struct btree_keys *bk, const struct bkey *k)
 {
+	struct btree *b = container_of(bk, struct btree, keys);
 	return __bch_btree_ptr_invalid(b->c, k);
 }
 
@@ -118,13 +119,14 @@ static bool btree_ptr_bad_expensive(struct btree *b, const struct bkey *k)
 	return true;
 }
 
-static bool bch_btree_ptr_bad(struct btree *b, const struct bkey *k)
+static bool bch_btree_ptr_bad(struct btree_keys *bk, const struct bkey *k)
 {
+	struct btree *b = container_of(bk, struct btree, keys);
 	unsigned i;
 
 	if (!bkey_cmp(k, &ZERO_KEY) ||
 	    !KEY_PTRS(k) ||
-	    bch_ptr_invalid(b, k))
+	    bch_ptr_invalid(bk, k))
 		return true;
 
 	for (i = 0; i < KEY_PTRS(k); i++)
@@ -209,8 +211,9 @@ static struct bkey *bch_extent_sort_fixup(struct btree_iter *iter,
 	return NULL;
 }
 
-static bool bch_extent_invalid(struct btree *b, const struct bkey *k)
+static bool bch_extent_invalid(struct btree_keys *bk, const struct bkey *k)
 {
+	struct btree *b = container_of(bk, struct btree, keys);
 	char buf[80];
 
 	if (!KEY_SIZE(k))
@@ -259,13 +262,14 @@ static bool bch_extent_bad_expensive(struct btree *b, const struct bkey *k,
 	return true;
 }
 
-static bool bch_extent_bad(struct btree *b, const struct bkey *k)
+static bool bch_extent_bad(struct btree_keys *bk, const struct bkey *k)
 {
+	struct btree *b = container_of(bk, struct btree, keys);
 	struct bucket *g;
 	unsigned i, stale;
 
 	if (!KEY_PTRS(k) ||
-	    bch_extent_invalid(b, k))
+	    bch_extent_invalid(bk, k))
 		return true;
 
 	for (i = 0; i < KEY_PTRS(k); i++)
@@ -303,8 +307,9 @@ static uint64_t merge_chksums(struct bkey *l, struct bkey *r)
 		~((uint64_t)1 << 63);
 }
 
-static bool bch_extent_merge(struct btree *b, struct bkey *l, struct bkey *r)
+static bool bch_extent_merge(struct btree_keys *bk, struct bkey *l, struct bkey *r)
 {
+	struct btree *b = container_of(bk, struct btree, keys);
 	unsigned i;
 
 	if (key_merging_disabled(b->c))

drivers/md/bcache/sysfs.c

@@ -433,7 +433,7 @@ SHOW(__bch_cache_set)
 
 		mutex_lock(&c->bucket_lock);
 		list_for_each_entry(b, &c->btree_cache, list)
-			ret += 1 << (b->page_order + PAGE_SHIFT);
+			ret += 1 << (b->keys.page_order + PAGE_SHIFT);
 
 		mutex_unlock(&c->bucket_lock);
 		return ret;

include/trace/events/bcache.h

@@ -247,7 +247,7 @@ TRACE_EVENT(bcache_btree_write,
 	TP_fast_assign(
 		__entry->bucket	= PTR_BUCKET_NR(b->c, &b->key, 0);
 		__entry->block	= b->written;
-		__entry->keys	= b->sets[b->nsets].data->keys;
+		__entry->keys	= b->keys.set[b->keys.nsets].data->keys;
 	),
 
 	TP_printk("bucket %zu", __entry->bucket)