Hi Gyan/Ted,

> On Nov 20, 2019, at 9:12 AM, Gyan Mishra <[email protected]> wrote:
> 
> 
> 
> On Nov 19, 2019, at 2:30 AM, Ted Lemon <[email protected] 
> <mailto:[email protected]>> wrote:
> 
>> On Nov 18, 2019, at 1:50 AM, Gyan Mishra <[email protected] 
>> <mailto:[email protected]>> wrote:
>>> ULA mentioned in section 2.1 mentions use case described in RFC 4864 but 
>>> should mention section 3.2 of that document.  I think we could add some 
>>> verbiage that ULA is used for local scope communications only. Also maybe a 
>>> case of SHIM6 multi homing or where 6to6  nat - nat64 is used where ULA is 
>>> inside local Nat to outside global address use case. Maybe also mention 
>>> default address selection RFC 6724 where a host is configured with both ULA 
>>> and Global and how ULA is used for internal local communications and how 
>>> the global address is used for external internet communications.
>> 
>> I don’t think you should mention anything that’s not ready to deploy.  SHIM6 
>> is a neat idea that nobody has deployed, as far as I know, and it’s not 
>> clear that it can be deployed.  I may be wrong about that—that’s just my 
>> impression—but the point is that if you’re going to suggest something, it 
>> should be readily actionable.
> 
> Agreed.
>> 
>>> 2.1.4 static does mention obfuscation of static address but 2.1.6 does not 
>>> since DHCPv6 ISC or other DHCPV6 server implementations the pool address 
>>> picked let’s say within the scope is randomly picked each time the server 
>>> doles ourt an IPv6 lease by default and not picked sequentially by the 
>>> server which is why obfuscation is not mentioned.   We could mention that 
>>> is how the DHCPv6 sever picks the address so that is not confused.  We 
>>> don’t mention split scope is required for redundancy which should be 
>>> mentioned and how that works in comparison to IPv4 where state sharing is 
>>> done between active and backup server where with DHCPV6 no active leases 
>>> state sharing.  We will get that added as well.  Also a caveat with DHCPv6 
>>> that when the primary server goes down and the lease has to be renewed 
>>> rebind time that the IPv6 address has to change at which time any active 
>>> session will have to reset.  I have actually deployed this with BT Diamond 
>>> which used ISC and to get around this issue we ended up setting a long 7 
>>> day lease time with larger block size for the scope being a /112 versus our 
>>> initial /116 scope.  Also good to mention that since the 128 bits is 
>>> managed to use smaller scope size for split scope then 2 /65 since their is 
>>> not any privacy extension which stateful that for security it’s better to 
>>> have a smaller DHCPv6 scope for managed address.  In section 2.1.6 the DUID 
>>> has the mac embedded in the address embedded in the DUID as the 48 LSB 
>>> bits.  Can you provide the use case where the DUID does not have embedded 
>>> mac.  I agree  that 2.1.7 needs to some extra verbiage as to the use case 
>>> of when and why addressing would be done as /64 per host and refer back to 
>>> 2.1.4 static addressing use case.
>> 
>> Why do you need a split scope?  If the prefix is managed, just give out 
>> random addresses.   Use a good random number generator.  Use failover to 
>> keep the servers in sync.
> 
> See RFC 6853.
> 
> https://tools.ietf.org/html/rfc6853#section-6.1 
> <https://tools.ietf.org/html/rfc6853#section-6.1>
> 
> With DHCPV6 all servers are active and that is why there is not any state 
> sharing since the pool has to be different and there a a preference option as 
> to which server is preferred.  This does go deeper into host configuration 
> which is out of scope for this document so will leave out.
>> 
>> DUID is opaque.  We recommended generating it using the MAC address because 
>> we hadn’t thought about privacy.   RFC7824 makes it clear that this is a 
>> problem.  Any DHCP implementation on a device where privacy is a likely 
>> issue should already be randomizing the DUID.  DUID-LLT with a random MAC 
>> address should be safe to use and unique.  It can be regenerated as 
>> frequently as the client decides.
>> 
> DUID-LLT which is most commonly used still uses an embedded Mac but now along 
> with time source to generate the DUID versus DUID-LL.

There are specific recommendations for this (generation of DUIDs) in Section 
4.3 of RFC7844, essentially describing randomizing contents (and usage of “old” 
random timestamps for DUID-LLT). 

Regards
Suresh

_______________________________________________
OPSEC mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/opsec

Reply via email to