bcachefs: bkey_written()

also cleanups of btree node offsets

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

6 files changed, 87 insertions, 73 deletions

diff --git a/fs/bcachefs/bset.c b/fs/bcachefs/bset.c
index a74e93a7215c..cf83911b3f5d 100644
--- a/fs/bcachefs/bset.c
+++ b/fs/bcachefs/bset.c
@@ -20,12 +20,14 @@
 
 struct bset_tree *bch2_bkey_to_bset(struct btree *b, struct bkey_packed *k)
 {
+	unsigned offset = __btree_node_key_to_offset(b, k);
 	struct bset_tree *t;
 
 	for_each_bset(b, t)
-		if (k >= btree_bkey_first(b, t) &&
-		    k < btree_bkey_last(b, t))
+		if (offset <= t->end_offset) {
+			EBUG_ON(offset < btree_bkey_first_offset(t));
 			return t;
+		}
 
 	BUG();
 }
@@ -172,34 +174,29 @@ static void bch2_btree_node_iter_next_check(struct btree_node_iter *iter,
 void bch2_btree_node_iter_verify(struct btree_node_iter *iter,
 				struct btree *b)
 {
-	struct btree_node_iter_set *set, *prev = NULL;
+	struct btree_node_iter_set *set, *s2;
 	struct bset_tree *t;
-	struct bkey_packed *k, *first;
 
-	if (bch2_btree_node_iter_end(iter))
-		return;
+	/* Verify no duplicates: */
+	btree_node_iter_for_each(iter, set)
+		btree_node_iter_for_each(iter, s2)
+			BUG_ON(set != s2 && set->end == s2->end);
 
+	/* Verify that set->end is correct: */
 	btree_node_iter_for_each(iter, set) {
-		k = __btree_node_offset_to_key(b, set->k);
-		t = bch2_bkey_to_bset(b, k);
-
-		BUG_ON(__btree_node_offset_to_key(b, set->end) !=
-		       btree_bkey_last(b, t));
-
-		BUG_ON(prev &&
-		       btree_node_iter_cmp(iter, b, *prev, *set) > 0);
-
-		prev = set;
+		for_each_bset(b, t)
+			if (set->end == t->end_offset)
+				goto found;
+		BUG();
+found:
+		BUG_ON(set->k < btree_bkey_first_offset(t) ||
+		       set->k >= t->end_offset);
 	}
 
-	first = __btree_node_offset_to_key(b, iter->data[0].k);
-
-	for_each_bset(b, t)
-		if (bch2_btree_node_iter_bset_pos(iter, b, t) ==
-		    btree_bkey_last(b, t) &&
-		    (k = bch2_bkey_prev_all(b, t, btree_bkey_last(b, t))))
-			BUG_ON(__btree_node_iter_cmp(iter->is_extents, b,
-						     k, first) > 0);
+	/* Verify iterator is sorted: */
+	btree_node_iter_for_each(iter, set)
+		BUG_ON(set != iter->data &&
+		       btree_node_iter_cmp(iter, b, set[-1], set[0]) > 0);
 }
 
 void bch2_verify_key_order(struct btree *b,
diff --git a/fs/bcachefs/btree_io.c b/fs/bcachefs/btree_io.c
index 2d004941c52e..5c36acef2b13 100644
--- a/fs/bcachefs/btree_io.c
+++ b/fs/bcachefs/btree_io.c
@@ -309,7 +309,7 @@ static unsigned should_compact_bset(struct btree *b, struct bset_tree *t,
 
 	if (mode == COMPACT_LAZY) {
 		if (should_compact_bset_lazy(b, t) ||
-		    (compacting && bset_unwritten(b, bset(b, t))))
+		    (compacting && !bset_written(b, bset(b, t))))
 			return dead_u64s;
 	} else {
 		if (bset_written(b, bset(b, t)))
@@ -356,7 +356,7 @@ bool __bch2_compact_whiteouts(struct bch_fs *c, struct btree *b,
 		struct bkey_packed *k, *n, *out, *start, *end;
 		struct btree_node_entry *src = NULL, *dst = NULL;
 
-		if (t != b->set && bset_unwritten(b, i)) {
+		if (t != b->set && !bset_written(b, i)) {
 			src = container_of(i, struct btree_node_entry, keys);
 			dst = max(write_block(b),
 				  (void *) btree_bkey_last(b, t -1));
@@ -396,7 +396,7 @@ bool __bch2_compact_whiteouts(struct bch_fs *c, struct btree *b,
 				continue;
 
 			if (bkey_whiteout(k)) {
-				unreserve_whiteout(b, t, k);
+				unreserve_whiteout(b, k);
 				memcpy_u64s(u_pos, k, bkeyp_key_u64s(f, k));
 				set_bkeyp_val_u64s(f, u_pos, 0);
 				u_pos = bkey_next(u_pos);
@@ -467,7 +467,7 @@ static bool bch2_drop_whiteouts(struct btree *b)
 		start	= btree_bkey_first(b, t);
 		end	= btree_bkey_last(b, t);
 
-		if (bset_unwritten(b, i) &&
+		if (!bset_written(b, i) &&
 		    t != b->set) {
 			struct bset *dst =
 			       max_t(struct bset *, write_block(b),
@@ -829,7 +829,7 @@ static bool btree_node_compact(struct bch_fs *c, struct btree *b,
 	for (unwritten_idx = 0;
 	     unwritten_idx < b->nsets;
 	     unwritten_idx++)
-		if (bset_unwritten(b, bset(b, &b->set[unwritten_idx])))
+		if (!bset_written(b, bset(b, &b->set[unwritten_idx])))
 			break;
 
 	if (b->nsets - unwritten_idx > 1) {
@@ -852,7 +852,7 @@ void bch2_btree_build_aux_trees(struct btree *b)
 
 	for_each_bset(b, t)
 		bch2_bset_build_aux_tree(b, t,
-				bset_unwritten(b, bset(b, t)) &&
+				!bset_written(b, bset(b, t)) &&
 				t == bset_tree_last(b));
 }
 
@@ -1949,9 +1949,9 @@ bool bch2_btree_post_write_cleanup(struct bch_fs *c, struct btree *b)
 	clear_btree_node_just_written(b);
 
 	/*
-	 * Note: immediately after write, bset_unwritten()/bset_written() don't
-	 * work - the amount of data we had to write after compaction might have
-	 * been smaller than the offset of the last bset.
+	 * Note: immediately after write, bset_written() doesn't work - the
+	 * amount of data we had to write after compaction might have been
+	 * smaller than the offset of the last bset.
 	 *
 	 * However, we know that all bsets have been written here, as long as
 	 * we're still holding the write lock:
diff --git a/fs/bcachefs/btree_types.h b/fs/bcachefs/btree_types.h
index fcd660470e52..5376388e91e6 100644
--- a/fs/bcachefs/btree_types.h
+++ b/fs/bcachefs/btree_types.h
@@ -340,10 +340,38 @@ static inline struct bset_tree *bset_tree_last(struct btree *b)
 	return b->set + b->nsets - 1;
 }
 
+static inline void *
+__btree_node_offset_to_ptr(const struct btree *b, u16 offset)
+{
+	return (void *) ((u64 *) b->data + 1 + offset);
+}
+
+static inline u16
+__btree_node_ptr_to_offset(const struct btree *b, const void *p)
+{
+	u16 ret = (u64 *) p - 1 - (u64 *) b->data;
+
+	EBUG_ON(__btree_node_offset_to_ptr(b, ret) != p);
+	return ret;
+}
+
 static inline struct bset *bset(const struct btree *b,
 				const struct bset_tree *t)
 {
-	return (void *) b->data + t->data_offset * sizeof(u64);
+	return __btree_node_offset_to_ptr(b, t->data_offset);
+}
+
+static inline void set_btree_bset_end(struct btree *b, struct bset_tree *t)
+{
+	t->end_offset =
+		__btree_node_ptr_to_offset(b, vstruct_last(bset(b, t)));
+}
+
+static inline void set_btree_bset(struct btree *b, struct bset_tree *t,
+				  const struct bset *i)
+{
+	t->data_offset = __btree_node_ptr_to_offset(b, i);
+	set_btree_bset_end(b, t);
 }
 
 static inline struct bset *btree_bset_first(struct btree *b)
@@ -359,16 +387,13 @@ static inline struct bset *btree_bset_last(struct btree *b)
 static inline u16
 __btree_node_key_to_offset(const struct btree *b, const struct bkey_packed *k)
 {
-	size_t ret = (u64 *) k - (u64 *) b->data - 1;
-
-	EBUG_ON(ret > U16_MAX);
-	return ret;
+	return __btree_node_ptr_to_offset(b, k);
 }
 
 static inline struct bkey_packed *
 __btree_node_offset_to_key(const struct btree *b, u16 k)
 {
-	return (void *) ((u64 *) b->data + k + 1);
+	return __btree_node_offset_to_ptr(b, k);
 }
 
 static inline unsigned btree_bkey_first_offset(const struct bset_tree *t)
@@ -376,7 +401,13 @@ static inline unsigned btree_bkey_first_offset(const struct bset_tree *t)
 	return t->data_offset + offsetof(struct bset, _data) / sizeof(u64);
 }
 
-#define btree_bkey_first(_b, _t)	(bset(_b, _t)->start)
+#define btree_bkey_first(_b, _t)					\
+({									\
+	EBUG_ON(bset(_b, _t)->start !=					\
+		__btree_node_offset_to_key(_b, btree_bkey_first_offset(_t)));\
+									\
+	bset(_b, _t)->start;						\
+})
 
 #define btree_bkey_last(_b, _t)						\
 ({									\
@@ -386,23 +417,6 @@ static inline unsigned btree_bkey_first_offset(const struct bset_tree *t)
 	__btree_node_offset_to_key(_b, (_t)->end_offset);		\
 })
 
-static inline void set_btree_bset_end(struct btree *b, struct bset_tree *t)
-{
-	t->end_offset =
-		__btree_node_key_to_offset(b, vstruct_last(bset(b, t)));
-	btree_bkey_last(b, t);
-}
-
-static inline void set_btree_bset(struct btree *b, struct bset_tree *t,
-				  const struct bset *i)
-{
-	t->data_offset = (u64 *) i - (u64 *) b->data;
-
-	EBUG_ON(bset(b, t) != i);
-
-	set_btree_bset_end(b, t);
-}
-
 static inline unsigned bset_byte_offset(struct btree *b, void *i)
 {
 	return i - (void *) b->data;
diff --git a/fs/bcachefs/btree_update_interior.h b/fs/bcachefs/btree_update_interior.h
index 7a19a52bbcff..711fbe63eb3a 100644
--- a/fs/bcachefs/btree_update_interior.h
+++ b/fs/bcachefs/btree_update_interior.h
@@ -241,14 +241,19 @@ static inline void *write_block(struct btree *b)
 	return (void *) b->data + (b->written << 9);
 }
 
+static inline bool __btree_addr_written(struct btree *b, void *p)
+{
+	return p < write_block(b);
+}
+
 static inline bool bset_written(struct btree *b, struct bset *i)
 {
-	return (void *) i < write_block(b);
+	return __btree_addr_written(b, i);
 }
 
-static inline bool bset_unwritten(struct btree *b, struct bset *i)
+static inline bool bkey_written(struct btree *b, struct bkey_packed *k)
 {
-	return (void *) i > write_block(b);
+	return __btree_addr_written(b, k);
 }
 
 static inline ssize_t __bch_btree_u64s_remaining(struct bch_fs *c,
@@ -307,10 +312,9 @@ static inline struct btree_node_entry *want_new_bset(struct bch_fs *c,
 	return NULL;
 }
 
-static inline void unreserve_whiteout(struct btree *b, struct bset_tree *t,
-				      struct bkey_packed *k)
+static inline void unreserve_whiteout(struct btree *b, struct bkey_packed *k)
 {
-	if (bset_written(b, bset(b, t))) {
+	if (bkey_written(b, k)) {
 		EBUG_ON(b->uncompacted_whiteout_u64s <
 			bkeyp_key_u64s(&b->format, k));
 		b->uncompacted_whiteout_u64s -=
@@ -318,10 +322,9 @@ static inline void unreserve_whiteout(struct btree *b, struct bset_tree *t,
 	}
 }
 
-static inline void reserve_whiteout(struct btree *b, struct bset_tree *t,
-				    struct bkey_packed *k)
+static inline void reserve_whiteout(struct btree *b, struct bkey_packed *k)
 {
-	if (bset_written(b, bset(b, t))) {
+	if (bkey_written(b, k)) {
 		BUG_ON(!k->needs_whiteout);
 		b->uncompacted_whiteout_u64s +=
 			bkeyp_key_u64s(&b->format, k);
diff --git a/fs/bcachefs/btree_update_leaf.c b/fs/bcachefs/btree_update_leaf.c
index 32126b02ce3a..ce0223bd52b5 100644
--- a/fs/bcachefs/btree_update_leaf.c
+++ b/fs/bcachefs/btree_update_leaf.c
@@ -40,7 +40,7 @@ bool bch2_btree_bset_insert_key(struct btree_iter *iter,
 
 		t = bch2_bkey_to_bset(b, k);
 
-		if (bset_unwritten(b, bset(b, t)) &&
+		if (!bkey_written(b, k) &&
 		    bkey_val_u64s(&insert->k) == bkeyp_val_u64s(f, k) &&
 		    !bkey_whiteout(&insert->k)) {
 			k->type = insert->k.type;
@@ -76,7 +76,7 @@ bool bch2_btree_bset_insert_key(struct btree_iter *iter,
 					k->u64s, k->u64s);
 
 		if (bkey_whiteout(&insert->k)) {
-			reserve_whiteout(b, t, k);
+			reserve_whiteout(b, k);
 			return true;
 		} else {
 			k->needs_whiteout = false;
diff --git a/fs/bcachefs/extents.c b/fs/bcachefs/extents.c
index 02a49d9845fb..803272b10e61 100644
--- a/fs/bcachefs/extents.c
+++ b/fs/bcachefs/extents.c
@@ -1429,7 +1429,7 @@ extent_squash(struct extent_insert_state *s, struct bkey_i *insert,
 		 * what k points to)
 		 */
 		bkey_reassemble(&split.k, k.s_c);
-		split.k.k.needs_whiteout |= bset_written(b, bset(b, t));
+		split.k.k.needs_whiteout |= bkey_written(b, _k);
 
 		bch2_cut_back(bkey_start_pos(&insert->k), &split.k.k);
 		BUG_ON(bkey_deleted(&split.k.k));
@@ -1499,9 +1499,9 @@ __bch2_delete_fixup_extent(struct extent_insert_state *s)
 			bch2_subtract_sectors(s, k.s_c,
 					     bkey_start_offset(k.k), k.k->size);
 			_k->type = KEY_TYPE_DISCARD;
-			reserve_whiteout(b, t, _k);
+			reserve_whiteout(b, _k);
 		} else if (k.k->needs_whiteout ||
-			   bset_written(b, bset(b, t))) {
+			   bkey_written(b, _k)) {
 			struct bkey_i discard = *insert;
 
 			discard.k.type = KEY_TYPE_DISCARD;
@@ -1573,13 +1573,13 @@ __bch2_insert_fixup_extent(struct extent_insert_state *s)
 			break;
 
 		if (k.k->size &&
-		    (k.k->needs_whiteout || bset_written(b, bset(b, t))))
+		    (k.k->needs_whiteout || bkey_written(b, _k)))
 			insert->k.needs_whiteout = true;
 
 		if (overlap == BCH_EXTENT_OVERLAP_ALL &&
 		    bkey_whiteout(k.k) &&
 		    k.k->needs_whiteout) {
-			unreserve_whiteout(b, t, _k);
+			unreserve_whiteout(b, _k);
 			_k->needs_whiteout = false;
 		}
 squash: