On Mon, 2020-10-05 at 12:25 -0700, Jakub Kicinski wrote: > On Mon, 05 Oct 2020 20:58:57 +0200 Johannes Berg wrote: > > On Mon, 2020-10-05 at 08:57 -0700, Jakub Kicinski wrote: > > > @@ -47,19 +61,16 @@ int ethnl_parse_header_dev_get(struct ethnl_req_info > > > *req_info, > > > NL_SET_ERR_MSG(extack, "request header missing"); > > > return -EINVAL; > > > } > > > + /* Use most permissive header policy here, ops should specify their > > > + * actual header policy via NLA_POLICY_NESTED(), and the real > > > + * validation will happen in genetlink code. > > > + */ > > > ret = nla_parse_nested(tb, ETHTOOL_A_HEADER_MAX, header, > > > - ethnl_header_policy, extack); > > > + ethnl_header_policy_stats, extack); > > > > Would it make sense to just remove the validation here? It's already > > done, so it just costs extra cycles and can't really fail, and if there > > are more diverse policies in the future this might also very quickly get > > out of hand? > > I was slightly worried I missed a command and need last line of defence,
Ah. I was just about to suggest to put it into the family policy/maxattr but that won't work of course since this is nested. But actually what you _could_ put there is a dummy policy (non-NULL pointer) with a maxattr of 0, and then all attrs will be completely rejected for a command where the policy was missed. Not if you missed the NLA_POLICY_NESTED() link, though. johannes
