xen/blkfront: Only check REQ_FUA for writes

[ Upstream commit b6ebaa8100090092aa602530d7e8316816d0c98d ]

The existing code silently converts read operations with the
REQ_FUA bit set into write-barrier operations. This results in data
loss as the backend scribbles zeroes over the data instead of returning
it.

While the REQ_FUA bit doesn't make sense on a read operation, at least
one well-known out-of-tree kernel module does set it and since it
results in data loss, let's be safe here and only look at REQ_FUA for
writes.

Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com>
Acked-by: Juergen Gross <jgross@suse.com>
Link: https://lore.kernel.org/r/20230426164005.2213139-1-ross.lagerwall@citrix.com
Signed-off-by: Juergen Gross <jgross@suse.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>

diff --git a/drivers/block/xen-blkfront.c b/drivers/block/xen-blkfront.c
index cd58f582c50c16d4ccf664ba450270073ef5a42d..b649f1a68b417edd4ebb84973b94aa5e82ca8763 100644 (file)
--- a/drivers/block/xen-blkfront.c
+++ b/drivers/block/xen-blkfront.c
@@ -780,7 +780,8 @@ static int blkif_queue_rw_req(struct request *req, struct blkfront_ring_info *ri
                ring_req->u.rw.handle = info->handle;
                ring_req->operation = rq_data_dir(req) ?
                        BLKIF_OP_WRITE : BLKIF_OP_READ;
-               if (req_op(req) == REQ_OP_FLUSH || req->cmd_flags & REQ_FUA) {
+               if (req_op(req) == REQ_OP_FLUSH ||
+                   (req_op(req) == REQ_OP_WRITE && (req->cmd_flags & REQ_FUA))) {
                        /*
                         * Ideally we can do an unordered flush-to-disk.
                         * In case the backend onlysupports barriers, use that.