At the moment, ethtool falls back to ioctl implementation whenever either in response to request type not implemented in kernel or netlink interface is unavailable or netlink request fails with EOPNOTSUPP error code. This is not perfect as EOPNOTSUPP can have different meanings and we only want to fall back if it is caused by kernel lacking netlink implementation of the request. In other cases, we would needlessly repeat the same failure trying both netlink and ioctl.
These two patches improve the logic to avoid such duplicate failures and improve handling of cases where fallback to ioctl is impossible for other reasons (e.g. wildcard device name or no ioctl handler). Michal Kubecek (2): refactor interface between ioctl and netlink code netlink: use genetlink ops information to decide about fallback ethtool.c | 51 +++--------- netlink/extapi.h | 14 ++-- netlink/monitor.c | 15 +++- netlink/netlink.c | 193 ++++++++++++++++++++++++++++++++++++++------- netlink/netlink.h | 6 ++ netlink/parser.c | 7 ++ netlink/settings.c | 7 ++ 7 files changed, 220 insertions(+), 73 deletions(-) -- 2.26.2
