bcachefs: Fix a race with b->write_type

b->write_type needs to be set atomically with setting the
btree_node_need_write flag, so move it into b->flags.

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

6 files changed, 40 insertions, 28 deletions

diff --git a/fs/bcachefs/bcachefs.h b/fs/bcachefs/bcachefs.h
index c8ab0e2029df..e61dc1e6da06 100644
--- a/fs/bcachefs/bcachefs.h
+++ b/fs/bcachefs/bcachefs.h
@@ -627,23 +627,6 @@ typedef struct {
 #define BCACHEFS_ROOT_SUBVOL_INUM					\
 	((subvol_inum) { BCACHEFS_ROOT_SUBVOL,	BCACHEFS_ROOT_INO })
 
-#define BCH_BTREE_WRITE_TYPES()						\
-	x(initial,		0)					\
-	x(init_next_bset,	1)					\
-	x(cache_reclaim,	2)					\
-	x(journal_reclaim,	3)					\
-	x(interior,		4)
-
-enum btree_write_type {
-#define x(t, n) BTREE_WRITE_##t,
-	BCH_BTREE_WRITE_TYPES()
-#undef x
-	BTREE_WRITE_TYPE_NR,
-};
-
-#define BTREE_WRITE_TYPE_MASK	(roundup_pow_of_two(BTREE_WRITE_TYPE_NR) - 1)
-#define BTREE_WRITE_TYPE_BITS	ilog2(BTREE_WRITE_TYPE_MASK)
-
 struct bch_fs {
 	struct closure		cl;
 
diff --git a/fs/bcachefs/btree_cache.c b/fs/bcachefs/btree_cache.c
index 5adfdc5afbea..c9d287f38d63 100644
--- a/fs/bcachefs/btree_cache.c
+++ b/fs/bcachefs/btree_cache.c
@@ -631,7 +631,6 @@ out:
 	b->flags		= 0;
 	b->written		= 0;
 	b->nsets		= 0;
-	b->write_type		= 0;
 	b->sib_u64s[0]		= 0;
 	b->sib_u64s[1]		= 0;
 	b->whiteout_u64s	= 0;
diff --git a/fs/bcachefs/btree_io.c b/fs/bcachefs/btree_io.c
index 5d750f447241..8dbe930c1eb2 100644
--- a/fs/bcachefs/btree_io.c
+++ b/fs/bcachefs/btree_io.c
@@ -1636,6 +1636,7 @@ static void __btree_node_write_done(struct bch_fs *c, struct btree *b)
 {
 	struct btree_write *w = btree_prev_write(b);
 	unsigned long old, new, v;
+	unsigned type = 0;
 
 	bch2_btree_complete_write(c, b, w);
 
@@ -1654,6 +1655,9 @@ static void __btree_node_write_done(struct bch_fs *c, struct btree *b)
 			new |=  (1U << BTREE_NODE_write_in_flight_inner);
 			new |=  (1U << BTREE_NODE_just_written);
 			new ^=  (1U << BTREE_NODE_write_idx);
+
+			type = new & BTREE_WRITE_TYPE_MASK;
+			new &= ~BTREE_WRITE_TYPE_MASK;
 		} else {
 			new &= ~(1U << BTREE_NODE_write_in_flight);
 			new &= ~(1U << BTREE_NODE_write_in_flight_inner);
@@ -1661,7 +1665,7 @@ static void __btree_node_write_done(struct bch_fs *c, struct btree *b)
 	} while ((v = cmpxchg(&b->flags, old, new)) != old);
 
 	if (new & (1U << BTREE_NODE_write_in_flight))
-		__bch2_btree_node_write(c, b, BTREE_WRITE_ALREADY_STARTED|b->write_type);
+		__bch2_btree_node_write(c, b, BTREE_WRITE_ALREADY_STARTED|type);
 	else
 		wake_up_bit(&b->flags, BTREE_NODE_write_in_flight);
 }
@@ -1846,6 +1850,10 @@ void __bch2_btree_node_write(struct bch_fs *c, struct btree *b, unsigned flags)
 		if (old & (1 << BTREE_NODE_write_in_flight))
 			return;
 
+		if (flags & BTREE_WRITE_ONLY_IF_NEED)
+			type = new & BTREE_WRITE_TYPE_MASK;
+		new &= ~BTREE_WRITE_TYPE_MASK;
+
 		new &= ~(1 << BTREE_NODE_dirty);
 		new &= ~(1 << BTREE_NODE_need_write);
 		new |=  (1 << BTREE_NODE_write_in_flight);
@@ -1857,10 +1865,6 @@ void __bch2_btree_node_write(struct bch_fs *c, struct btree *b, unsigned flags)
 	if (new & (1U << BTREE_NODE_need_write))
 		return;
 do_write:
-	if ((flags & BTREE_WRITE_ONLY_IF_NEED))
-		type = b->write_type;
-	b->write_type = 0;
-
 	BUG_ON((type == BTREE_WRITE_initial) != (b->written == 0));
 
 	atomic_dec(&c->btree_cache.dirty);
diff --git a/fs/bcachefs/btree_types.h b/fs/bcachefs/btree_types.h
index 38c4754dbd7e..72e6a214b89a 100644
--- a/fs/bcachefs/btree_types.h
+++ b/fs/bcachefs/btree_types.h
@@ -77,7 +77,6 @@ struct btree {
 	u8			nsets;
 	u8			nr_key_bits;
 	u16			version_ondisk;
-	u8			write_type;
 
 	struct bkey_format	format;
 
@@ -445,6 +444,23 @@ struct btree_trans {
 	struct replicas_delta_list *fs_usage_deltas;
 };
 
+#define BCH_BTREE_WRITE_TYPES()						\
+	x(initial,		0)					\
+	x(init_next_bset,	1)					\
+	x(cache_reclaim,	2)					\
+	x(journal_reclaim,	3)					\
+	x(interior,		4)
+
+enum btree_write_type {
+#define x(t, n) BTREE_WRITE_##t,
+	BCH_BTREE_WRITE_TYPES()
+#undef x
+	BTREE_WRITE_TYPE_NR,
+};
+
+#define BTREE_WRITE_TYPE_MASK	(roundup_pow_of_two(BTREE_WRITE_TYPE_NR) - 1)
+#define BTREE_WRITE_TYPE_BITS	ilog2(roundup_pow_of_two(BTREE_WRITE_TYPE_NR))
+
 #define BTREE_FLAGS()							\
 	x(read_in_flight)						\
 	x(read_error)							\
@@ -464,6 +480,8 @@ struct btree_trans {
 	x(never_write)
 
 enum btree_flags {
+	/* First bits for btree node write type */
+	BTREE_NODE_FLAGS_START = BTREE_WRITE_TYPE_BITS - 1,
 #define x(flag)	BTREE_NODE_##flag,
 	BTREE_FLAGS()
 #undef x
diff --git a/fs/bcachefs/btree_update_interior.c b/fs/bcachefs/btree_update_interior.c
index ac3a5ef1b1af..03e016758af3 100644
--- a/fs/bcachefs/btree_update_interior.c
+++ b/fs/bcachefs/btree_update_interior.c
@@ -1257,6 +1257,7 @@ static void bch2_insert_fixup_btree_ptr(struct btree_update *as,
 	struct bch_fs *c = as->c;
 	struct bkey_packed *k;
 	struct printbuf buf = PRINTBUF;
+	unsigned long old, new, v;
 
 	BUG_ON(insert->k.type == KEY_TYPE_btree_ptr_v2 &&
 	       !btree_ptr_sectors_written(insert));
@@ -1294,8 +1295,15 @@ static void bch2_insert_fixup_btree_ptr(struct btree_update *as,
 
 	bch2_btree_bset_insert_key(trans, path, b, node_iter, insert);
 	set_btree_node_dirty_acct(c, b);
-	set_btree_node_need_write(b);
-	b->write_type = BTREE_WRITE_interior;
+
+	v = READ_ONCE(b->flags);
+	do {
+		old = new = v;
+
+		new &= ~BTREE_WRITE_TYPE_MASK;
+		new |= BTREE_WRITE_interior;
+		new |= 1 << BTREE_NODE_need_write;
+	} while ((v = cmpxchg(&b->flags, old, new)) != old);
 
 	printbuf_exit(&buf);
 }
diff --git a/fs/bcachefs/btree_update_leaf.c b/fs/bcachefs/btree_update_leaf.c
index 1405ad4eda02..445f8f57ef0c 100644
--- a/fs/bcachefs/btree_update_leaf.c
+++ b/fs/bcachefs/btree_update_leaf.c
@@ -178,11 +178,11 @@ static int __btree_node_flush(struct journal *j, struct journal_entry_pin *pin,
 		    w->journal.seq != seq)
 			break;
 
+		new &= ~BTREE_WRITE_TYPE_MASK;
+		new |= BTREE_WRITE_journal_reclaim;
 		new |= 1 << BTREE_NODE_need_write;
 	} while ((v = cmpxchg(&b->flags, old, new)) != old);
 
-	b->write_type = BTREE_WRITE_journal_reclaim;
-
 	btree_node_write_if_need(c, b, SIX_LOCK_read);
 	six_unlock_read(&b->c.lock);