Hi Thomas,
thanks for showing up here on the Linux-CAN mailing list!
On 02.12.20 13:28, Thomas Wagner wrote:
as already discussed with Oliver Hartkopp via Github (https://github.com/hartkopp/can-isotp/issues/36), I suggest a new option/flag to enable ISO-TP sockets with just a TX address, while the RX address is being ignored/discarded. This can be useful for OBD2/UDS usage, where functional requests are used and sent in a 1-to-N scenario, while still having an ISO-TP style PCI.
Oliver:
It would be something like a new flag, e.g. CAN_ISOTP_FADDR_TX_ONLY where only the tp.tx_id is used at bind() time (no registering of a tp.rx_id) and where you only can send PDUs that are max 7 bytes => SF.
Yes. I would be fine with it.
Btw. I would like to ask about the naming of the flag.
CAN_ISOTP_FADDR_TX_ONLY
CAN_ISOTP_FUNCADDR_TX
CAN_ISOTP_FUNC_ADDR
CAN_ISOTP_FUNC_ADDR_TX
(?? other ideas ??)
In depth Use-Case explanation:
With OBD2 and UDS requests on CAN, ISO-TP is used as a transport protocol. Requests can be physically addressed by a tester to a specific ECU using the CAN-IDs 0x7E0 to 0x7E7 (11-bit IDs).
The ECUs respond to the tester using the CAN-IDs 0x7E8 to 0x7EF. A tester can open 8 ISO-TP sockets, using the specific RX and TX addresses, to communicate with each ECU.
In some cases requests aren't sent physically addressed to a specific ECU, but broadcasted/functionally addressed to all ECUs who might care/can respond. This is done using the special CAN-ID 0x7DF.
Any ECU that might respond does so, using it's to-tester address (0x7E8 to 0x7EF). This way a 1-to-N broadcast/request can be done, after which multiple 1-to-1 connections can be established to transfer large PDUs.
Btw. If you already have established the 8 1-to-1 sockets as described
above the ECU can also send PDUs with length >7 bytes as your socket
would do the protocol handshake.
In the current ISO-TP/SocketCAN implementation, an additional 9th socket needs to be used for those functional requests. Using an ISO-TP socket as 9th socket forces the user to choose an RX address that is never actually used. Using a RAW CAN socket forces the user to calculate/build a PCI manually and have the socket discard/filter all RX traffic.
You can simply remove the single default CAN_RAW filter which receives
all CAN-IDs:
(see in cansend.c code)
setsockopt(s, SOL_CAN_RAW, CAN_RAW_FILTER, NULL, 0);
AFAIK functional requests are always <7B data, so ISO-TP would be fine without a RX address for flow control on the 7DF request socket.
Now I seed what you meant with "<7B" .. "<7 bytes"
:-D
Any feedback/thoughts on that? :)
I'm currently working on the sja1000 arbitration lost topic but I can
send a RFC patch for this suggestion afterwards.
Are you currently working on a Linux 5.10-rc kernel with ISO-TP included
in the tree - or would you like the testing based on my GitHub isotp repo?
Best regards,
Oliver
