bcache: FUA fixes
Journal writes need to be marked FUA, not just REQ_FLUSH. And btree node
writes have... weird ordering requirements.
Signed-off-by: Kent Overstreet <koverstreet@google.com>
diff --git a/drivers/md/bcache/btree.c b/drivers/md/bcache/btree.c
index 09fb8a2..a6ad49a 100644
--- a/drivers/md/bcache/btree.c
+++ b/drivers/md/bcache/btree.c
@@ -328,10 +328,25 @@
b->bio->bi_end_io = btree_node_write_endio;
b->bio->bi_private = &b->io.cl;
- b->bio->bi_rw = REQ_META|WRITE_SYNC;
- b->bio->bi_size = set_blocks(i, b->c) * block_bytes(b->c);
+ b->bio->bi_rw = REQ_META|WRITE_SYNC|REQ_FUA;
+ b->bio->bi_size = set_blocks(i, b->c) * block_bytes(b->c);
bch_bio_map(b->bio, i);
+ /*
+ * If we're appending to a leaf node, we don't technically need FUA -
+ * this write just needs to be persisted before the next journal write,
+ * which will be marked FLUSH|FUA.
+ *
+ * Similarly if we're writing a new btree root - the pointer is going to
+ * be in the next journal entry.
+ *
+ * But if we're writing a new btree node (that isn't a root) or
+ * appending to a non leaf btree node, we need either FUA or a flush
+ * when we write the parent with the new pointer. FUA is cheaper than a
+ * flush, and writes appending to leaf nodes aren't blocking anything so
+ * just make all btree node writes FUA to keep things sane.
+ */
+
bkey_copy(&k.key, &b->key);
SET_PTR_OFFSET(&k.key, 0, PTR_OFFSET(&k.key, 0) + bset_offset(b, i));
@@ -2092,6 +2107,9 @@
void bch_btree_set_root(struct btree *b)
{
unsigned i;
+ struct closure cl;
+
+ closure_init_stack(&cl);
trace_bcache_btree_set_root(b);
@@ -2107,7 +2125,8 @@
b->c->root = b;
__bkey_put(b->c, &b->key);
- bch_journal_meta(b->c, NULL);
+ bch_journal_meta(b->c, &cl);
+ closure_sync(&cl);
}
/* Cache lookup */