Re: [homenet] Fwd: I-D Action: draft-ietf-homenet-naming-architecture-dhc-options-01.txt

[Ray Hunter <[e-mail]>]

> Daniel Migault <mailto:[e-mail]>
> 16 March 2015 02:48
> Hi,
>
> Thank you for the feed back. Here is an update of the renumbering 
> section. I considered the two cases make-before-break and 
> break-before-make.
>
> Feel free to make any comment.
>
> BR,
> Daniel
>
> 9 Renumbering
>
> This section details how the CPE is expected to handle renumbering. 
> Renumbering has been extensively described in RFC4192 and analyzed in 
> RFC7010. This section is largely inspired by these document, and the 
> reader is expected to become familiar with them.
>
> This section considers two ways to renumber the home network:
>     - make-before-break: In this scenario, the new prefix is 
> advertised, the network is configured to prepare the transition to the 
> new prefix. During a period of time, the two prefixes old and new 
> coexist, before the old prefix is completely removed.
>     - break-before-make: In this scenario, the new prefix is 
> advertised making the old prefix obsolete.
>
> The make-before-break scenario is recommended, but this section 
> details both.
>
> In both scenarios, this section is focused on how renumbering affects 
> the DNS(SEC) Homenet Zone and how renumbering affects the 
> communication between the Hidden Master and the slave hosted on the 
> Public Authoritative Name Server Set. This section designates by 
> OLD_PREFIX and OLD_IP the IP prefix and IP addresses that are to be 
> replaced and NEW_PREFIX and the NEW_IP the replacing IP prefix and IP 
> addresses.
>
> The initial situation is as follows: the CPE has configured its naming 
> architecture as depicted in Figure 1 (Section 4.1). More specifically, 
> the CPE has set the DNS(SEC) Homenet Zone as detailed in Section 4.2, 
> and it is assumed that some hosts are configured with OLD_PREFIX. In 
> addition the Hidden Master and the Public Authoritative Name Server 
> Set are configured as detailed in Section 5. The Public Authoritative 
> Name Server Set is configured as a slave for the Homenet Domain Name. 
> The communication between the Hidden Master and the slave uses the 
> OLD_IP of the Hidden Master. Usually, the slave is configured with the 
> master's IP address. In our case the slave is configured with OLD_IP.
>
>     9.1 make-before-break
>
>     In the make-before-break renumbering scenario, the CPE is informed 
> that renumbering will happen in the future. The CPE waits that routers 
> ad switches are ready to route both OLD_PREFIX and OLD_PREFIX on the 
> home network. The necessary time for the CPE to be aware of the 
> bindings between the hosts of the home network and the NEW_IPs may 
> vary, depending on how binding between the NEW_IPs and the various 
> hosts in the home network is performed as well as how the CPE is 
> informed of the newly assigned IP addresses. After some time, either 
> that all hosts have provided their corresponding NEW_IPs or a timer 
> has expired, the CPE provides an updated version of the DNS(SEC) 
> Homenet Zone to the Hidden Master. As explained in Section 9.1.1, the 
> TTLs may also be adjusted.
>
>     The updated DNS(SEC) Homenet Zone is then updated on the slave 
> using the same configuration as used until now. In fact, there is no 
> need to update the master/slave configuration as currently both IP 
> addresses of the Hidden Master are reachable (IP_OLD and IP_NEW). This 
> process may have been proceed earlier, or may be processed later. Even 
> though updating the DNS(SEC) Homenet Zone and updating the IP address 
> used by the Hidden Master are completely different processes, they may 
> be combined as explained in section 9.1.2. At that time, the DNS(SEC) 
> Homenet Zone has been updated with both NEW_IPs and OLD_IPs, the 
> DNS(SEC) Homenet Zone is being publicly published on the Public 
> Masters, and the master/slave mechanism uses the NEW_IP of the Hidden 
> Master.
>
>     The final step consists in removing the OLD_IPs of the DNS(SEC) 
> Homenet Zone file and updating the zone on the Public Masters. The 
> removal of the OLD_IPs RRsets SHOULD be performed such that there is 
> no more RRSets contains OLD_IPs in any cache when OLD_PREFIX is not 
> reachable anymore. TTLs should be chosen appropriately to meet timing 
> constraints as explained in section 9.1.1.
>
>     9.1.1 Adjusting TTLs
>
>     As TTL designates how long the RRsets are cached, they should be 
> adjusted regarding the time both OLD_IP and NEW_IP are reachable and 
> the time OLD_IPs are removed from the DNS(SEC) Homenet Zone. Let 
> T_OLD_NEW designate the time the DNS(SEC) Homenet Zone contains both 
> OLD_IPs and NEW_IPs, and T_NEW the time when OLD_IPs are removed from 
> the DNS(SEC) Homenet Zone. Let OLD_TTL the TTL value before T_OLD_NEW, 
> OLD_NEW_TTL the TTL value after T_OLD_NEW and NEW_TTL the TTL value 
> after T_NEW.
>
>     When a RRset is modified, it takes 2*TTL seconds for the previous 
> value to be removed from all caches. As a result, it will take 
> 2*OLD_TTL before the DNS(SEC) Homenet Zone becomes stable have have 
> its RRsets containing both OLD_IPs and NEW_IP in all caches. Going to 
> a stable state where both OLD_IP and NEW_IP are stored in caches is 
> not mandatory. On the other hand, it is of primary importance that 
> OLD_IP is not anymore in caches when OLD_PREFIX becomes unreachable, 
> i.e. T_OLD_NEW + 2 OLD_TTL < T_OLD_PREFIX_UNREACHABLE. This equation 
> provides indication on how to chose T_OLD_NEW  as other parameters may 
> be fixed. Similarly, T_NEW should be chosen so that T_NEW + 
> 2*OLD_NEW_TTL < T_OLD_PREFIX_UNREACHABLE. This equation provides 
> indications on how OLD_NEW_TTL and T_NEW may be chosen.
>
>     Figure below illustrates a timing example.
>
>
>
>        Updating Zone    Last RRsets with      Updating Zone   Last 
> RRsets with     T_OLD_PREFIX_UNREACHABLE
>        with OLD_IP      OLD_IP only expires   with NEW_IP     OLD_IP 
> and NEW_IP    OLD_PREFIX becomes
>        and NEW_IPs      from cache            only            expires 
> from cache   unreachable
>              |               |                       |            
> |                 |
>              |               |                       |            
> |                 |
>    OLD_TTL   v  OLD_NEW_TTL  v                       v  NEW_TTL   
> v                 v
>    
> ----------+---------------+-----------------------+------------+----------------------
>           T_OLD_NEW       T_OLD_NEW + 2 OLD_TTL    T_NEW      T_NEW + 
> 2*OLD_NEW_TTL
>
>
>
>
>     9.1.2 Updating the IP address of the Hidden Master
>
>
>     This section details how the slave may be advertised the Hidden 
> Master has changed its IP address, and consequently update its 
> configuration. Note also that the IP address of the Hidden Master does 
> not appear in the DNS(SEC) Homenet Zone as the Hidden Master is kept 
> "hidden".
>
>     The Hidden Master sends a NOTIFY DNS query to the masters. When 
> the slave receives the NOTIFY DNS query and have authenticated the 
> NOTIFY DNS query, it sends a response back to the master.  The slave 
> may notices that the IP address is not associated to the Hidden Master 
> IP address(es), however, the slave is not expected to update its 
> master / slave configuration as the IP address is most likely not 
> protected. Instead the slave is expected to perform a reachability 
> check before setting the master/slave configuration. The reachability 
> check may be combined with the regular master/slave synchronization, 
> that is to say a DNS query for the SOA is sent by the slave to the 
> master. The difference with the regular master/slave synchronization 
> is that the IP used to reach the slave is not provided by the 
> master/slave setting, but is instead provided by the NOTIFY DNS query. 
> If an authenticated response is received the slave MAY update its 
> master/slave settings with the new IP address.
>
> The reason a change of master does not necessarily results in updating 
> the master/slave configuration is that the CPE may be multi-homed, in 
> which case multiple IP addresses may be used in parallel. In such 
> cases, it is not necessary to update the master/slave configuration. 
> Instead, the update should be performed only when the IP is not in the 
> pool of the IP addresses used by the CPE.  The edns-client-subnet 
> option described in [draft-vandergaast-edns-client-subnet-02] may be 
> used to advertise the valid IP addresses of the Hidden Master. 
> However, for sake of simplicity, we recommend that unless the slave is 
> aware that the Hidden Master uses multiple IP addresses, the Hidden 
> Master uses a single IP address at a time, meaning that a change of IP 
> address would result in a modification of the master/slave setting on 
> the slave.
> Note that the NOTIFY DNS query can be authenticated even though the IP 
> address is unknown to the slave. As described in RFC1996, the NOTIFY 
> payload carries the Homenet Domain Name. Similarly, if SIG(0) or TSIG 
> is used to secure the transaction between the Hidden Master and the 
> slave, the Homenet Domain Name is also indicated in the RRSIG RRset. 
> Unless the transport layer can handle with an IP update, if DTLS/TLS 
> is used, than a new socket may be opened, which results results in a 
> new authentication exchange before the NOTIFY DNS query is sent. If 
> IPsec is used, authentication is based on the IP address. Unless 
> MOBIKE is supported by IKEv2, a new IPsec Security Association should 
> be negotiated. If MOBIKE is supported, than the Hidden Master can 
> inform the slave that its IP address has been updated.
>
> Note also that the IP address is not authenticated, unless IPsec with 
> the AH mode is used. The reliability of this IP address is left to the 
> reachability check. As a results, the IP address may not be the one 
> associated to the Hidden Master, but instead the one of a proxy or any 
> middle box. For this reason it matters to authenticate and protect the 
> IP payload with appropriated security protocols listed in Section 
> 5.2.  One way DNS could be used to provide the IP addresses associated 
> to the CPE would be to use the edns-client-subnet option described in 
> [draft-vandergaast-edns-client-subnet-02], however, this may be used 
> cautiously in case any kind of translation address mechanism occurs 
> between the master and the slave.
>
>     9.2 break-before-make
>
>     In the break-before-make, the CPE is being assigned a new prefix 
> at the time the old prefix is not reachable. This situation 
> necessarily results in making the hosts unreachable, as the DNS cannot 
> be updated instantly. More specifically, RRset with OLD_IP will remain 
> in cache for 2*TTL.
>
>     In this case, the DNS(SEC) Homenet Zone is rebuilt with the 
> NEW_IPs only, and eventually resigned. In fact OLD_IPs are already 
> unreachable, thus they are not mentioned in the zone. The Hidden 
> Master then performs sends a NOTIFY using the newly assigned IP 
> address and performs both a zone synchronization and a configuration 
> update at the same time as described in section 9.1.2.
>
>
>
>
>
>
>
>
> -- 
> Daniel Migault
> Ericsson
> ------------------------------------------------------------------------

I think the description of the requirements and overall chain of events 
is excellent. Thanks for that.

However I'm not particularly convinced by the security and 
authentication mechanisms, as the proposed text just seems to punt these 
to another transport protocol.

I also don't see the recovery mechanism/ how to distinguish between a 
break before make event, and a make before break where the ISP's DNS 
servers may not be reachable at the moment of renumbering, or that a 
Homenet is simply not reachable for some time, and comes up "unnumbered" 
with no old IP, or that the Homenet never comes back at all. It seems to 
me that you could also end up hosting lots of cruft in the ISP servers, 
and there's a need for some garbage collection.

I also expect we need some more limitations on when/why a slave would 
poll an unknown hidden master, otherwise it seems to me that we have a 
nice DoS vector/ amplification attack. A number of attack machines could 
spoof notification messages pointing to various ISP slave systems, all 
pointing to a common target victim "new" hidden master.

It feels to me like
1) we need a more reliable update mechanism between the hidden master 
and the slaves (to cover longer-term unreachability events, or a Homenet 
hidden master going off air permanently)
2) we need a more effective authentication mechanism for triggering 
updates, that is source IP address agnostic.

-- 
Regards,
RayH