RPL lwtunnel encapsulation
Hello linux-wpan devs,
I have been experimenting with the RPL SRH support implemented by Alexander Aring. I hope related questions fit into this mailing list, otherwise feel free to redirect me elsewhere.
In the original merge commit (96376cad3508) for this feature, the log indicates that IPv6-in-IPv6 encapsulation is implemented:
In receive handling I add handling for IPIP encapsulation as RFC6554 describes it as possible.
However, my observations suggest that it is actually not the case when forwarding packets. Instead, the IPv6 header of the packet is modified in a way which violates the IPv6 specification (RFC 8200 section 4):
Extension headers (except for the Hop-by-Hop Options header) are not processed, inserted, or deleted by any node along a packet's delivery path, until the packet reaches the node (or each of the set of nodes, in the case of multicast) identified in the Destination Address field of the IPv6 header.
In the appendix I describe more precisely the steps I took to come to this conclusion. A couple questions follow from my observations:
- First of all, is my analysis flawed?
- Then, was this behavior implemented knowingly?
- Finally, should it be changed?
########################################################################
Appendix: my experiment
I have 2 TUN devices setup and a couple simple routes to do testing.
$ ip -6 route
[...]
2001:db8::1 dev tun1 proto kernel metric 256 pref medium
2001:db8::2 dev tun2 proto kernel metric 256 pref medium
2001:db8::3 dev tun2 metric 1024 pref medium
2001:db8::4 encap rpl segs 1 [ 2001:db8::3 ] dev tun2 metric 1024
pref medium
[...]I am then generating a packet at tun1, destined to 2001:db8::4, and observing what goes through both interfaces using tshark.
$ tshark -i tun1 -i tun2 -V
Capturing on 'tun1' and 'tun2'
Frame 1: 48 bytes on wire (384 bits), 48 bytes captured (384 bits)
on interface tun1, id 0
[...]
Internet Protocol Version 6, Src: 2001:db8::1, Dst: 2001:db8::4
0110 .... = Version: 6
.... 0000 0000 .... .... .... .... .... = Traffic Class: 0x00
(DSCP: CS0, ECN: Not-ECT)
.... 0000 00.. .... .... .... .... .... = Differentiated
Services Codepoint: Default (0)
.... .... ..00 .... .... .... .... .... = Explicit
Congestion Notification: Not ECN-Capable Transport (0)
.... .... .... 0000 0000 0000 0000 0000 = Flow Label: 0x00000
Payload Length: 8
Next Header: UDP (17)
Hop Limit: 255
Source Address: 2001:db8::1
Destination Address: 2001:db8::4
User Datagram Protocol, Src Port: 5000, Dst Port: 5000
Source Port: 5000
Destination Port: 5000
Length: 8
Checksum: 0x7d57 [unverified]
[Checksum Status: Unverified]
[Stream index: 0] Frame 2: 64 bytes on wire (512 bits), 64 bytes captured (512 bits)
on interface tun2, id 1
[...]
Internet Protocol Version 6, Src: 2001:db8::1, Dst: 2001:db8::3
0110 .... = Version: 6
.... 0000 0000 .... .... .... .... .... = Traffic Class: 0x00
(DSCP: CS0, ECN: Not-ECT)
.... 0000 00.. .... .... .... .... .... = Differentiated
Services Codepoint: Default (0)
.... .... ..00 .... .... .... .... .... = Explicit Congestion Notification: Not ECN-Capable Transport (0)
.... .... .... 0000 0000 0000 0000 0000 = Flow Label: 0x00000 Payload Length: 24
Next Header: Routing Header for IPv6 (43)
Hop Limit: 254
Source Address: 2001:db8::1
Destination Address: 2001:db8::3
Routing Header for IPv6 (RPL Source Route)
Next Header: UDP (17)
Length: 1
[Length: 16 bytes]
Type: RPL Source Route (3)
Segments Left: 1
1111 .... .... .... .... .... .... .... = Compressed
Internal Octets (CmprI): 15
.... 1111 .... .... .... .... .... .... = Compressed Final
Octets (CmprE): 15
.... .... 0111 .... .... .... .... .... = Padding Bytes: 7
.... .... .... 0000 0000 0000 0000 0000 = Reserved: 0
[Total Address Count: 1]
Address: 04
[Address[1]: 2001:db8::4]
User Datagram Protocol, Src Port: 5000, Dst Port: 5000
Source Port: 5000
Destination Port: 5000
Length: 8
Checksum: 0x7d57 [unverified]
[Checksum Status: Unverified]
[Stream index: 0]We can then see that the original packet header was modified, with destination address changed, and a routing extension header added. If IPv6-in-IPv6 encapsulation was at play, the original header would have been copied unchanged (except the hop limit I suppose), and an additional IPv6 header accompanied with the SRH would have been added as an outer header.
Quoted 3 lines · Mathis Marion
However, my observations suggest that it is actually not the case when forwarding packets. Instead, the IPv6 header of the packet is modified in a way which violates the IPv6 specification (RFC 8200 section 4):
I have not sat down to read the code to understand what it actually does, so I can't really comment at this point. I salute you for having gotten into whether the code is compliant.
But, I did write spend way too much of my life writing
https://datatracker.ietf.org/doc/rfc9008/
to deal with the perception that RPL networks had to violate 8200.
I know that Linux does not (yet) deal with all the minutia in 9008. I wish that I had time to fix that.
Hi Michael,
Quoted 11 lines · Michael Richardson
Quoted 3 lines · Mathis Marion
However, my observations suggest that it is actually not the case when forwarding packets. Instead, the IPv6 header of the packet is modified in a way which violates the IPv6 specification (RFC 8200 section 4):
I have not sat down to read the code to understand what it actually does, so I can't really comment at this point. I salute you for having gotten into whether the code is compliant.
Quoted 6 lines
But, I did write spend way too much of my life writing
https://datatracker.ietf.org/doc/rfc9008/to deal with the perception that RPL networks had to violate 8200.
I know that Linux does not (yet) deal with all the minutia in 9008. I wish that I had time to fix that.
To put everything into IPIP and back is not a question of doing a iptunnel ip6tnl [0] and doing the right configuration... just do get everything over "the internet" which I think is the whole reason why putting everything into IPIP?
- Alex
[0] https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/net/ipv6/ip6_tunnel.c
Hi,
Quoted 19 lines · Alexander Aring
Hi Michael,
Quoted 15 lines · Michael Richardson
Quoted 3 lines · Mathis Marion
However, my observations suggest that it is actually not the case when forwarding packets. Instead, the IPv6 header of the packet is modified in a way which violates the IPv6 specification (RFC 8200 section 4):
I have not sat down to read the code to understand what it actually does, so I can't really comment at this point. I salute you for having gotten into whether the code is compliant.
But, I did write spend way too much of my life writing
https://datatracker.ietf.org/doc/rfc9008/to deal with the perception that RPL networks had to violate 8200.
Quoted 8 lines
Quoted 2 lines
I know that Linux does not (yet) deal with all the minutia in 9008. I wish that I had time to fix that.
To put everything into IPIP and back is not a question of doing a iptunnel ip6tnl [0] and doing the right configuration... just do get everything over "the internet" which I think is the whole reason why putting everything into IPIP?
what I mean is, that encap rpl will never do any IPIP itself, you need to use existing iptunnel implementation [0] to make it working like you want to work and this is in my opinion "if we have luck" only a matter of configure everything "right".
- Alex
Quoted 13 lines · Alexander Aring
Quoted 13 lines
Mathis Marion <[email protected]> wrote: > However, my observations suggest that it is actually not the case when > forwarding packets. Instead, the IPv6 header of the packet is modified
Quoted 1 line
in a way which violates the IPv6 specification (RFC 8200 section 4):
I have not sat down to read the code to understand what it actually does, so I can't really comment at this point. I salute you for having gotten into whether the code is compliant.
But, I did write spend way too much of my life writing
https://datatracker.ietf.org/doc/rfc9008/ to deal with the perception that RPL networks had to violate 8200.
Quoted 7 lines
Quoted 2 lines
I know that Linux does not (yet) deal with all the minutia in 9008. I wish that I had time to fix that.
To put everything into IPIP and back is not a question of doing a iptunnel ip6tnl [0] and doing the right configuration... just do get everything over "the internet" which I think is the whole reason why putting everything into IPIP?
I agree that modelling it an infinite series of iptunnel/ip6tnl is the wrong approach. I would model it akin to how ND and ARP work: something that happens which then resolves into some bytes that get prefixed and/or removed.
--
] Never tell me the odds! | ipv6 mesh networks [ ] Michael Richardson, Sandelman Software Works | network architect [ ] [email protected] http://www.sandelman.ca/ | ruby on rails [
Hi,
Quoted 19 lines · Michael Richardson
Quoted 15 lines · Alexander Aring
Quoted 15 lines
Mathis Marion <[email protected]> wrote: > However, my observations suggest that it is actually not the case when > forwarding packets. Instead, the IPv6 header of the packet is modified
Quoted 1 line
in a way which violates the IPv6 specification (RFC 8200 section 4):
I have not sat down to read the code to understand what it actually does, so I can't really comment at this point. I salute you for having gotten into whether the code is compliant.
But, I did write spend way too much of my life writing
https://datatracker.ietf.org/doc/rfc9008/ to deal with the perception that RPL networks had to violate 8200.I know that Linux does not (yet) deal with all the minutia in 9008. I wish that I had time to fix that.
Quoted 8 lines
Quoted 4 lines
To put everything into IPIP and back is not a question of doing a iptunnel ip6tnl [0] and doing the right configuration... just do get everything over "the internet" which I think is the whole reason why putting everything into IPIP?
I agree that modelling it an infinite series of iptunnel/ip6tnl is the wrong approach. I would model it akin to how ND and ARP work: something that happens which then resolves into some bytes that get prefixed and/or removed.
then it is currently possible, but not in a nice way (you will configure yourself to death)... there might be new config options of iptunnel/ip6tnl to do your ND approach (maybe with accessing ND cache?).
It is not that the rpl tunnel for source routing header needs to deal with IP6IP6, it already exists in the kernel with ip6tnl implementation... it's just terrible to configure it.
- Alex