On Wed, 27 Nov 2024 23:21:33 +0000 Jordan Rife wrote:
The second command fails with "Invalid argument" (EINVAL) on the unpatched kernel. This simplifies things, as there's no need for clients to explicitly probe to see if this attribute is supported. I will do the following:
Revert WG_GENL_VERSION back to 1.
Add a check for new flags similar to the one you mentioned for WGPEER_A_FLAGS.
if (attrs[WGPEER_A_FLAGS]) flags = nla_get_u32(attrs[WGPEER_A_FLAGS]); ret = -EOPNOTSUPP; if (flags & ~__WGPEER_F_ALL) goto out;This should be sufficient. We might want to consider how best to bubble this error up to users. In the case of wg, "Invalid argument" may not be very helpful in determining where you went wrong. We could always detect when EINVAL is returned in response to an operation that sets WGALLOWEDIP_A_FLAGS and print something more helpful like "Operation not supported on this kernel". However, these are details that can be worked out.
Better still use NLA_POLICY_MASK() so that nla_parse_nested() can perform the validation and attach a machine readable info about the failure.