Weblog entry #21 for lters
See all weblog entries made by lters.
#21
IPv6 Prefix length
Posted by lters on Fri 28 Jan 2011 at 16:16
When working with routers and using linux to do routing, is there a way to stop the kernel warning "kernel: IPv6 addrconf: prefix with wrong length 112" ?
Seems the kernel thinks its never except-able to have a prefix other than /64?
Seems the kernel thinks its never except-able to have a prefix other than /64?
Comments on this Entry
Posted by Anonymous (84.153.xx.xx) on Sat 29 Jan 2011 at 12:09
PS: As to "why a /64"... You can fit the whole MAC address space in there with room to spare. The short answer is that you will never ever have problems autoconfiguration any more.
Unless you want/need fixed non-MAC addresses somewhere, but even then, there is no need to use anything longer than a /64.
Unless you want/need fixed non-MAC addresses somewhere, but even then, there is no need to use anything longer than a /64.
[ Parent | Reply to this comment ]
Posted by lters (12.162.xx.xx) on Sat 29 Jan 2011 at 15:41
[ Send Message | View lters's Scratchpad | View Weblogs ]
[ Send Message | View lters's Scratchpad | View Weblogs ]
It still seems there should be a way to get the kernel to not complain when it sees it.
And on further research, it seems I had a router that was making advertisements about a /112 which was causing the kernels to complain that had visibility. Even though they were setup statically.
If every takes the idea that you never go smaller than /64, IPv6 will run out sooner than you think, and we will be looking at IPv8.
That does not appeal to me, as difficult to get IPv6 off the ground as it seems.
regards, lters
And on further research, it seems I had a router that was making advertisements about a /112 which was causing the kernels to complain that had visibility. Even though they were setup statically.
If every takes the idea that you never go smaller than /64, IPv6 will run out sooner than you think, and we will be looking at IPv8.
That does not appeal to me, as difficult to get IPv6 off the ground as it seems.
regards, lters
[ Parent | Reply to this comment ]
Posted by AJxn (2001:0xx:0xx:0xxx:0xxx:0xxx:xx) on Wed 2 Feb 2011 at 14:57
[ Send Message | View Weblogs ]
[ Send Message | View Weblogs ]
If you don't follow how IPv6 is suppose to work, you have to get used to get this warning, as it is a usefull warning to all others.
Let see now. With a /64 IPv6 net to all Internet users, you still have plenty of addresses compared to IPv4.
Ponder a total use of all addresses in IPv4, you get at most 2^32 addresses/users (minus a lot of adresses, like net, broadcast and at least a router for each net).
Same calculation for IPv6. With /64 nets to each users, you get aprox. 2^64 nets! If we say we need one additional net for each users router, we still have 2^63 nets/users.
We will lose those standard nets for link locals and broadcasts etc, but still. For each address in IPv4, we get 2^63 (≃ 9 * 10^18) nets (minus those nets mentioned). That is 2^31 (≃ 2 * 10^9) nets for each IPv4 address.
Yes, It isn't a propper calculations, but it would give us an idea that IPv6 addresses will not run out very soon. So there will take time until we get IPv7. At least not because we have runned out of addresses.
Notice the length of your and mine address in our posts. ;)
But I am interested in a more realistic and correct calculation, if you want to do that. But you must do that on both IPv4 and IPv6 :)
Let see now. With a /64 IPv6 net to all Internet users, you still have plenty of addresses compared to IPv4.
Ponder a total use of all addresses in IPv4, you get at most 2^32 addresses/users (minus a lot of adresses, like net, broadcast and at least a router for each net).
Same calculation for IPv6. With /64 nets to each users, you get aprox. 2^64 nets! If we say we need one additional net for each users router, we still have 2^63 nets/users.
We will lose those standard nets for link locals and broadcasts etc, but still. For each address in IPv4, we get 2^63 (≃ 9 * 10^18) nets (minus those nets mentioned). That is 2^31 (≃ 2 * 10^9) nets for each IPv4 address.
Yes, It isn't a propper calculations, but it would give us an idea that IPv6 addresses will not run out very soon. So there will take time until we get IPv7. At least not because we have runned out of addresses.
Notice the length of your and mine address in our posts. ;)
But I am interested in a more realistic and correct calculation, if you want to do that. But you must do that on both IPv4 and IPv6 :)
[ Parent | Reply to this comment ]
One thing more. If you are routing, you should only need link local automatic IPv6-addresses (fe80::/10) between routers. If you have to set up a tunnel, you need a global IPv6-net in the tunnel, but othervise you should be ok without any need to use global IPv6-addresses between routers.
[ Parent | Reply to this comment ]
If your ISP hands you a /56, you have 256 /64. A /48 is 65k /64.
Stop thinking in IPv4, IPv6 is simply different.
[ Parent | Reply to this comment ]