sys::socket adding Linux packet filtering on ipv4/ipv6/loopback traffics

[devnexen]

Respectively sys::socket::SockProtocol::EthIp,
sys::socket::SockProtocol::EthIpv6 and
sys::socket::SockProtocol::EthLoop if we want more refined filtering as opposed to the existing sys::socket::SockProtocol::EthAll which captures everything.

[devnexen]

@SteveLauC let me know which direction you prefer me to take here. e.g. using this for all these Eth* flags to avoid future conflicts, not using it all ...

[SteveLauC]

The current approach looks good, i.e., add them via associated constants when we see conflicts

[SteveLauC]

Suggested change

	/// NOTE: placed here due to conflict (little endian arch) with SockProtocol::NetLinkISCI
	// NOTE: placed here due to conflict (little endian arch) with SockProtocol::NetLinkISCI

This should be a code comment, not a doc comment.

conflict (little endian arch)

BTW, wouldn't this conflict always occur regardless of the byte order since they have the same value

[SteveLauC]

BTW, wouldn't this conflict always occur regardless of the byte order since they have the same value

I was wrong, they only have the same value on Little-endian machines. With big-endian, EthIp is 2048

[SteveLauC]

Looks like we need to do conditional compilation to fix the UB issue that fails our CI:

    /// Packet filter on IPv4 traffic
    // NOTE: placed here due to conflict (little endian arch) with `SockProtocol::NetLinkISCI`
    #[cfg(linux_android)]
    #[allow(non_upper_case_globals)]
    #[cfg(target_endian = "little")]
    pub const EthIp: SockProtocol = unsafe { std::mem::transmute::<i32, SockProtocol>((libc::ETH_P_IP as u16).to_be() as i32) };
    
     /// Packet filter on IPv4 traffic
    #[cfg(linux_android)]
    #[cfg(target_endian = "big")]
    EthIp = (libc::ETH_P_IP as u16).to_be() as i32,

[SteveLauC]

Sorry for being unclear, this should be an enum variant

[SteveLauC]

On big-endian targets, no enum variant has value 2048, so that std::mem::transmute() is UB, which is the reason why I said we should make it an enum variant

[devnexen]

yes it should work now

[SteveLauC]

Thanks