Hello, guys Well, I have a lot to say :)
> I load jool with an IPv6 pool (with jool -6 -a fg01::/96 ). When > I try to ping from A6 to C4, I have a correct ping (dmesg shows a > session was put up: Added session > fg01::3e52:10b1:6fc8:f0f1:94f8#1|fg01::c0a8:c0#1|192.168.0.100#62153|192.168.0.192#62153|ICMP). > This behavior is nice, but I'm not able to send back packets, > even if I start a UDP session from A6. I should be able to send > packets to 192.168.0.100:64065 for example, but they are not > received by A6. This is normal; it is a core nuance of the (Stateful) NAT64 design. Let me explain: When A6 sends a packet, G64 creates a mapping for it, and remembers it so it can handle the response: - A6 pings packet fg01::<blahblah>:94f8#1 -> fg01::c0a8:c0#1. - G64 looks at pool4 and decides 192.168.0.100#62153 is a suitable ICMP mask for fg01::<blahblah>:94f8#1. It then translates the packet into 192.168.0.100#62153 -> 192.168.0.192#62153 and (very important) remembers the [192.168.0.100#62153 fg01::<blahblah>:94f8#1] mask. - C4 answers using packet 192.168.0.192#62153 -> 192.168.0.100#62153. - G64 realizes that the destination address of the packet matches one of its masks, and looking at this mask, it realizes that C4's packet is intended for fg01::<blahblah>:94f8#1. So it translates the packet and the ping succeeds. On the other hand, this is what's happening to your UDP packet: - C4 writes (for example) packet 192.168.0.104:12345 -> 192.168.0.100#64065. - G64 does not have a mapping for 192.168.0.100#64065, so it doesn't know where it is headed. Therefore, it doesn't know how to translate the packet. Why can G64 infer a mapping for an IPv6 packet but not for an IPv4 one? because any source v6 address can be translated as any pool4 address, and any destination v6 address can be translated by removing the pool6 prefix. On the other hand, while a source v4 address can be translated by adding the pool6 prefix, there is no way to translate a v4 destination address without an already existing mapping. Think about it: You are masking 20-255 nodes using just G64's address (192.168.0.100). The fact that you already have a mapping that says that 192.168.0.100#62153 is fg01::<blahblah>:94f8#1, you can't automatically assume that any packet towards 192.168.0.100 should be headed towards fg01::<blahblah>:94f8. Otherwise A6 would be the only IPv4 node able to talk to the IPv4 network. This property of remembering mappings is what makes NAT64 "stateful". Its advantage is that you're using a few G64 addresses to grant a potentially large and dynamic number of IPv6 nodes access to IPv4 ones. The drawback is that IPv4 nodes cannot start communication with v6 ones because G64 doesn't initially know who they are talking to. (Unless you predefine mappings yourself. You can find more information at https://jool.mx/en/bib.html and https://jool.mx/en/usr-flags-bib.html) (Protip: Just for the sake of clarity, documentation labels these mappings "BIB entries". They are most of the time all you need to pay attention to; sessions are fairly useless from a user standpoint.) > Moreover, this is not the desired behavior as I want A6 to have > its own IPv4 address (to make it easier to talk to it). This is exactly what SIIT does; it lets you predetermine an IPv4 address for each of your IPv6 nodes. This means it frees you from the disadvantages of the BIB entries (and advantages, of course.) One of the downsides of SIIT is that this "predetermination" of address mappings is not going to be friendly with your "Every IPv6 device has a somewhat unpredictable IP where the prefix is fg01::/64" requirement. You can either find a way to impose more order to your address assignment policies (so they will be friendlier to SIIT's constraints) or you can try Taiga (http://www.litech.org/tayga/). Tayga is a SIIT/NAT64 hybrid in that it can assign v4 addresses to IPv6 nodes on the fly, and then the entire address (not just an address and port) are assigned to the IPv6 node. I can't vouch for Tayga because I don't use it and hasn't received updates in a long time, but perhaps you can squeeze something out of it. Alternatively, you might want to open this hybrid operation mode as a feature request for Jool: https://github.com/NICMx/Jool/issues > Here, I reckon that the gateway does not know that packets on the > network with the 192.168.0.54 address are in fact destined to > her, because when I manually add this IP to eth0 (with ip addr > add 192.168.0.54/24 dev eth0 ), suddenly, it works! A NAT64 is a device that masks IPv6 nodes behind it using its own IPv4 address (or addresses). An SIIT is a device that renames your IPv6 nodes, each using a separate IPv4 address, without them realizing it. Pay special attention to the part where I say "using its own IPv4 address". NAT64 does not rename v6 nodes like SIIT does; it pretends it itself is the IPv6 nodes it is masking. That's why they are called "masks" :) So yes; the v4 nodes need to be fooled into thinking that G64 *is* the v6 network. One way to do that is by adding the addresses to G64's interface. (So G64 answers ARP "neighbor" requests to these packets.) > Is it normal behavior that I cannot send back UDP packets to the NATed IPv4:port, even when the session has been correctly added? (I'm talking about a real UDP session, not an ICMP one) Yes. If v4-started traffic is one of your basic requirements, you need to either meddle with static BIB entries or switch over to SIIT. > How can I tell my eth0 interface that its should be concerned by packets adressed on the 192.168.0.54 (which it used to send packets beforehand since pinging worked), but without statically configuring it with this IP address? You need to realize that 192.168.0.54 belongs to C4's network (192.168.0.54/100), so C4 is expecting .54 to be one of its neighbors. For this reason, it doesn't intend to send the packet via any gateways. That said, you can always instead inform C4 that any traffic towards 192.168.0.54 should be routed via G64: romain@c4$ ip route add 192.168.0.54/32 via 192.168.0.100 But perhaps by this point you have realized that you should more likely redesign the setup :p > Can jool automatically and dynamically requests a new IPv4 from a DHCP server for an IPv6 device hidden behind? No, but why would you want to do that? (maybe I'm missing something) > Can jool automatically set up new address on the eth0 interface (when they are allocated by the DHCP server)? No, but can't a shell script do it? By the way: It seems that you're having trouble understanding these concepts from the documentation. Any tips on how to improve it? Alberto On Thu, Jul 28, 2016 at 3:33 PM, Michael Richardson <[email protected]> wrote: > > Hi, > What you are doing will become more and more common. > > Jool historically has needed an IP address different than the one on it's > primary interface. I've made this work (including getting the IP from > DHCPv4) using clone a macvlan interface, and some custom dhclient-up > scripts. > > That will take care of your needs: > > >* Can jool automatically and dynamically requests a new IPv4 from a > > DHCP server for an IPv6 device hidden behind? > >* Can jool automatically set up new address on the eth0 interface > > (when they are allocated by the DHCP server)? > > SIIT would let you embed IPv4 into your v6 network, which would also work. > > -- > ] Never tell me the odds! | ipv6 mesh > networks [ > ] Michael Richardson, Sandelman Software Works | network > architect [ > ] [email protected] http://www.sandelman.ca/ | ruby on > rails [ > > > _______________________________________________ > Jool-list mailing list > [email protected] > https://mail-lists.nic.mx/listas/listinfo/jool-list > >
_______________________________________________ Jool-list mailing list [email protected] https://mail-lists.nic.mx/listas/listinfo/jool-list
