----- Original Message ----- 
From: "Jim Reid" <[email protected]>
To: "George Barwood" <[email protected]>
Cc: <[email protected]>
Sent: Sunday, March 07, 2010 10:20 AM
Subject: Re: [DNSOP] Should root-servers.net be signed


> On 7 Mar 2010, at 08:06, George Barwood wrote:
> 
>> If root-servers.net is unsigned, it's not possible for the resolver  
>> to validate
>> the set of root IP addresses
> 
> So what? If the served zones are signed, it simply doesn't matter if  
> the address of a name server is spoofed or hijacked. The Bad Guy won't  
> have the private keys, so will be unable to return answers which  
> validate. In the context of a referral from the root, what matters is  
> the signature over the TLD's RRset (and its KSKs), not the IP address  
> of the root server or any signature that might or might not exist over  
> its name.
> 
>> (a) An attacker can control every unsigned zone.
>>
>> (b) An attacker can monitor every request to a signed zone ( no  
>> privacy ).
>>
>> (c) An attacker can deny service to any zone, on a selective basis.
> 
> It's not clear what point you're making or what your concerns are.  
> None of these things listed above are remotely relevant. Apart from  
> (a) which is hardly news: zones can be spoofed if they're not signed.  
> [What next? Can we expect revelations about what bears do in the  
> woods?] 

Prevention of spoofing is a goal of DNSSEC.
However, if a zone ( such as root-servers.net ) is not signed, spoofing is not 
prevented.
This has negative consequences, and weakens the protection that DNSSEC offers.

> Privacy -- whatever that might mean -- has never been a design  
> goal of DNS. Or Secure DNS for that matter. An eavesdropper can  
> monitor *any* DNS request (signed or not) if they're close enough to  
> the client or server. 

Yes, but I'm talking about blind spoofing attacks, where a single successful 
spoof
results in the attacker gaining a lot, and which can be stopped by signing 
root-server.net

> DoS attacks can and are mounted on any zone,  
> whether or not they're signed. Meanwhile, in other news, water is  
> discovered to be wet and fire is proven to be hot.

>> Apparently there are currently no plans to sign root-servers.net
> 
> There's no point doing that IMO until .net is signed and there's a  
> single chain of trust from root-servers.net to the One True Trust  
> Anchor, the signed root. 

I hadn't considered that. The dependency on .net is perhaps a little odd.
A more logical set of name server names might be simply

a.root-servers
b.root-servers
..
k.root-servers

so that the dependency is removed. There may be objections to this.

> If the zone was to be self-signed, that would  
> mean yet another TA would need to be embedded and maintained in  
> validator configurations. Which creates more failure modes and scope  
> for errors. And since validating the answers for root-servers.net will  
> rarely if ever matter, adding that TA would be a lot of risk for  
> almost no reward.

I agree that there is little point in signing root-servers if it can not
be validated without configuring a special trust anchor.
Of course another solution to that is to sign .net

-- George
_______________________________________________
DNSOP mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/dnsop

Reply via email to