Re: [dnsext] Historical root keys: The Large Router Vendor Speaks

[Olafur Gudmundsson <[e-mail]>]

On 01/02/2011 11:50 AM, George Barwood wrote:
>
> ----- Original Message -----
> From: "Tony Finch"<[e-mail]>
> To: "George Barwood"<[e-mail]>
> Cc:<[e-mail]>
> Sent: Tuesday, February 01, 2011 4:25 PM
> Subject: Re: [dnsext] Historical root keys: The Large Router Vendor Speaks
>
>
>> On Fri, 28 Jan 2011, George Barwood wrote:
>>>
>>> I think it's necessary to roll the key eventually because DNSSEC
>>> signature dates wrap, (and signatures can therefore be replayed) but
>>> only after 136 years.
>>
>> There are no dates on DNS keys so I don't understand the relevance of this
>> point.
>
> Ok, we are talking about replay attacks.
>
> The date fields in RRSIG records wrap around after 136 years,  so an attacker
> can replay responses from 136 years ago and they will be accepted as current,
> if no KSK rollover has been performed.
>
> Thus in DNSSEC, you must roll any KSK at least once every 136 years to remain secure.
>
> ( 136 years = 2^32 seconds )
>

Actually there is no explicit requirement to roll, you need signatures 
to be included that show the key has not changed every 34 years or so, 
as validators are supposed to only accept timer values that are not that 
far off each other (there are times I wish we had gone with at least 40 
bit timer values).

	Olafur