bcachefs; Don't use btree write buffer until journal replay is finished

The keys being replayed by journal replay have to be synchronized with
updates by other threads that overwrite them. We rely on btree node
locks for synchronizing - but since btree write buffer updates take no
btree locks, that won't work.

Instead, simply disable using the btree write buffer until journal
replay is finished.

This fixes a rare backpointers error in the merge_torture_flakey test.

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

1 files changed, 16 insertions, 0 deletions

diff --git a/fs/bcachefs/btree_update.c b/fs/bcachefs/btree_update.c
index 324767c0ddcc..25fdca00bf7b 100644
--- a/fs/bcachefs/btree_update.c
+++ b/fs/bcachefs/btree_update.c
@@ -554,6 +554,19 @@ int __must_check bch2_trans_update_seq(struct btree_trans *trans, u64 seq,
 						 BTREE_UPDATE_PREJOURNAL);
 }
 
+static noinline int bch2_btree_insert_clone_trans(struct btree_trans *trans,
+						  enum btree_id btree,
+						  struct bkey_i *k)
+{
+	struct bkey_i *n = bch2_trans_kmalloc(trans, bkey_bytes(&k->k));
+	int ret = PTR_ERR_OR_ZERO(n);
+	if (ret)
+		return ret;
+
+	bkey_copy(n, k);
+	return bch2_btree_insert_trans(trans, btree, n, 0);
+}
+
 int __must_check bch2_trans_update_buffered(struct btree_trans *trans,
 					    enum btree_id btree,
 					    struct bkey_i *k)
@@ -564,6 +577,9 @@ int __must_check bch2_trans_update_buffered(struct btree_trans *trans,
 	EBUG_ON(trans->nr_wb_updates > trans->wb_updates_size);
 	EBUG_ON(k->k.u64s > BTREE_WRITE_BUFERED_U64s_MAX);
 
+	if (unlikely(trans->journal_replay_not_finished))
+		return bch2_btree_insert_clone_trans(trans, btree, k);
+
 	trans_for_each_wb_update(trans, i) {
 		if (i->btree == btree && bpos_eq(i->k.k.p, k->k.p)) {
 			bkey_copy(&i->k, k);