bcachefs: Don't issue btree writes that weren't journalled

If we have an error in the btree interior update path that prevents us
from journalling the update, we can't issue the corresponding btree node
write - we didn't get a journal sequence number that would cause it to
be ignored in recovery.

Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>

3 files changed, 26 insertions, 6 deletions

diff --git a/fs/bcachefs/btree_io.c b/fs/bcachefs/btree_io.c
index 893ffe193479..9b19432ae7a5 100644
--- a/fs/bcachefs/btree_io.c
+++ b/fs/bcachefs/btree_io.c
@@ -1499,6 +1499,9 @@ void __bch2_btree_node_write(struct bch_fs *c, struct btree *b,
 		if (!btree_node_may_write(b))
 			return;
 
+		if (old & (1 << BTREE_NODE_never_write))
+			return;
+
 		if (old & (1 << BTREE_NODE_write_in_flight)) {
 			btree_node_wait_on_io(b);
 			continue;
@@ -1545,6 +1548,8 @@ void __bch2_btree_node_write(struct bch_fs *c, struct btree *b,
 		seq = max(seq, le64_to_cpu(i->journal_seq));
 	}
 
+	BUG_ON(b->written && !seq);
+
 	/* bch2_varint_decode may read up to 7 bytes past the end of the buffer: */
 	bytes += 8;
 
diff --git a/fs/bcachefs/btree_types.h b/fs/bcachefs/btree_types.h
index 15af60e92820..dc7de27112c6 100644
--- a/fs/bcachefs/btree_types.h
+++ b/fs/bcachefs/btree_types.h
@@ -415,6 +415,7 @@ enum btree_flags {
 	BTREE_NODE_fake,
 	BTREE_NODE_old_extent_overwrite,
 	BTREE_NODE_need_rewrite,
+	BTREE_NODE_never_write,
 };
 
 BTREE_FLAG(read_in_flight);
@@ -429,6 +430,7 @@ BTREE_FLAG(dying);
 BTREE_FLAG(fake);
 BTREE_FLAG(old_extent_overwrite);
 BTREE_FLAG(need_rewrite);
+BTREE_FLAG(never_write);
 
 static inline struct btree_write *btree_current_write(struct btree *b)
 {
diff --git a/fs/bcachefs/btree_update_interior.c b/fs/bcachefs/btree_update_interior.c
index edc11c22308c..4a169d366538 100644
--- a/fs/bcachefs/btree_update_interior.c
+++ b/fs/bcachefs/btree_update_interior.c
@@ -603,17 +603,30 @@ err:
 
 		list_del(&as->write_blocked_list);
 
-		if (!ret && as->b == b) {
+		/*
+		 * Node might have been freed, recheck under
+		 * btree_interior_update_lock:
+		 */
+		if (as->b == b) {
 			struct bset *i = btree_bset_last(b);
 
 			BUG_ON(!b->c.level);
 			BUG_ON(!btree_node_dirty(b));
 
-			i->journal_seq = cpu_to_le64(
-				max(journal_seq,
-				    le64_to_cpu(i->journal_seq)));
-
-			bch2_btree_add_journal_pin(c, b, journal_seq);
+			if (!ret) {
+				i->journal_seq = cpu_to_le64(
+					max(journal_seq,
+					    le64_to_cpu(i->journal_seq)));
+
+				bch2_btree_add_journal_pin(c, b, journal_seq);
+			} else {
+				/*
+				 * If we didn't get a journal sequence number we
+				 * can't write this btree node, because recovery
+				 * won't know to ignore this write:
+				 */
+				set_btree_node_never_write(b);
+			}
 		}
 
 		mutex_unlock(&c->btree_interior_update_lock);