Tracking: potential incompatibility with Select calls

[szszszsz]

While I was implementing support for KeepassXC' challenge-response over CCID, I've noticed that the SELECT call used there had not contain the Le byte. This caused the response to SELECT to be 61xx, which needed additional implementation for getting the remaining data. That sprinkled the discussion, why it was not needed before for the other supported there smart card readers. Apparently the SELECT call for them always result with the data in the direct response to the request.

I am documenting this behavior here for the future reference. Perhaps the SELECT command should be exempted from testing the Le byte, and following the usual rules (e.g. to make it work with improper implementations).

Yubikey 4 responds immediately as well AFAICS. Some YK4 traffic excerpt containing the SELECT (00 A4 04 00) calls:

07477498 APDU: 00 A4 04 00 07 A0 00 00 05 27 20 01
00001034 SW: 04 03 05 03 0B 00 06 0F 00 00 90 00
00000136 APDU: 00 A4 04 00 07 A0 00 00 05 27 20 01
00001019 SW: 04 03 05 03 0B 00 06 0F 00 00 90 00
00000062 APDU: 00 03 00 00 06
00000854 SW: 04 03 05 03 0B 00 90 00
00000137 APDU: 00 A4 04 00 07 A0 00 00 05 27 20 01
00001023 SW: 04 03 05 03 0B 00 06 0F 00 00 90 00
00000055 APDU: 00 01 10 00 06
00000849 SW: 00 56 7F B0 90 00
00000130 APDU: 00 A4 04 00 07 A0 00 00 05 27 20 01
00001000 SW: 04 03 05 03 0B 00 06 0F 00 00 90 00

The call used in KeepassXC for SELECT is:

        uint8_t pbSendBuffer_head[5] = {
            CLA_ISO, INS_SELECT, SEL_APP_AID, 0, static_cast<uint8_t>(handle.second.size())};
        auto pbSendBuffer = new uint8_t[5 + handle.second.size()];
        memcpy(pbSendBuffer, pbSendBuffer_head, 5);
        memcpy(pbSendBuffer + 5, handle.second.constData(), handle.second.size());
       // Give it more space in case custom implementations have longer answer to select
        uint8_t pbRecvBuffer[64] = {0}; 
       // 3 bytes version, 1 byte program counter, other stuff for various implementations, 2 bytes status
        SCUINT dwRecvLength = sizeof pbRecvBuffer;
        auto rv = transmit(handle.first, pbSendBuffer, 5 + handle.second.size(), pbRecvBuffer, dwRecvLength);

PR: keepassxreboot/keepassxc#9397

Code in question:

apdu-dispatch/src/dispatch.rs

Lines 280 to 334 in a6f0011

	fn handle_reply(&mut self) {
	// Consider if we need to reply via chaining method.
	// If the reader is using chaining, we will simply
	// reply 61XX, and put the response in a buffer.
	// It is up to the reader to then send GetResponse
	// requests, to which we will return up to `Le` bytes at a time.
	let (new_state, response) = match &mut self.buffer.raw {
	RawApduBuffer::Request(_) | RawApduBuffer::None => {
	info!("Unexpected GetResponse request.");
	(RawApduBuffer::None, Status::UnspecifiedCheckingError.into())
	}
	RawApduBuffer::Response(res) => {
	let max_response_len = self.response_len_expected.min(MAX_INTERCHANGE_DATA);
	if self.was_request_chained || res.len() > max_response_len {
	// Do not send more than the expected bytes
	let boundary = max_response_len.min(res.len());
	let to_send = &res[..boundary];
	let remaining = &res[boundary..];
	let mut message = interchanges::Data::from_slice(to_send).unwrap();
	let return_code = if remaining.len() > 255 {
	// XX = 00 indicates more than 255 bytes of data
	0x6100u16
	} else if !remaining.is_empty() {
	0x6100 + (remaining.len() as u16)
	} else {
	// Last chunk has success code
	0x9000
	};
	message
	.extend_from_slice(&return_code.to_be_bytes())
	.expect("Failed add to status bytes");
	if return_code == 0x9000 {
	(RawApduBuffer::None, message)
	} else {
	info!("Still {} bytes in response buffer", remaining.len());
	(
	RawApduBuffer::Response(response::Data::from_slice(remaining).unwrap()),
	message,
	)
	}
	} else {
	// Add success code
	res.extend_from_slice(&[0x90, 00])
	.expect("Failed to add the status bytes");
	(
	RawApduBuffer::None,
	interchanges::Data::from_slice(res.as_slice()).unwrap(),
	)
	}
	}
	};
	self.buffer.raw = new_state;
	self.respond(response);
	}

[sosthene-nitrokey]

Relevant parts of the OpenPGP specification is in paragraph 7:

Le field Absent for encoding Ne = 0, present for encoding Ne > 0

• Ne denotes the maximum number of bytes expected in the response data field.
  • If the Le field is absent, then Ne is zero.
  • A short Le field consists of one byte with any value.
  • If the byte is set to '00', then Ne is 256 dec.
  • An extended Le field consists of either three bytes (one byte set to '00' followed
    by two bytes with any value) if the Lc field is absent, or two bytes (with any
    value) if an extended Lc field is present.
  • If the two bytes are set to '0000', then Ne is 65536 dec.
• Nr denotes the number of bytes in the response data field. Nr shall be less than or
  equal to Ne

The same is present in paragraph 5.1 of ISO7816-4