Re: [dhcwg] draft-pruss-dhcp-auth-dsl-03

["Templin, Fred L" <[e-mail]>]

 

>-----Original Message-----
>From: Yoshihiro Ohba [mailto:[e-mail]] 
>Sent: Tuesday, July 29, 2008 6:24 AM
>To: Templin, Fred L
>Cc: Richard Pruss; [e-mail]
>Subject: Re: [dhcwg] draft-pruss-dhcp-auth-dsl-03
>
>On Mon, Jul 28, 2008 at 09:28:27AM -0700, Templin, Fred L wrote:
>> I haven't touched 'draft-templin-dhcpmtu' since the
>> last time this subject came up, but AFAICT it will
>> still work fine even if relay agent information
>> options are inserted.
>> 
>> Per RFC3046:
>> 
>>    "A DHCP relay agent adding a Relay Agent Information 
>field SHALL add
>>    it as the last option (but before 'End Option' 255, if present) in
>>    the DHCP options field of any recognized BOOTP or DHCP packet
>>    forwarded from a client to a server."
>> 
>> So, as long as 'draft-templin-dhcpmtu' is updated to
>> say that the reassembly process ignores any relay agent
>> information options the DHCP fragmentation/reassembly
>> will still work fine. 
>
>One issue with 'draft-templin-dhcpmtu' is that essentially there is a
>possibility of collision of fragments generated by multiple DHCP
>clients using 0.0.0.0 address.  This is because IP source address, the
>'xid' field in the DHCP message header, and the Identification field
>in the DHCP fragment option are used for reassembly, and it is
>possible that multiple DHCP clients prior to address allocation happen
>to use the same set of values for them.

You are kidding, right? True, we don't have an IP source
address (the whole reason for considering this approach)
but we do have 2 32bit randomly-chosen fields. That gives
64bits of uniqueness, and the probability for collision
for two nodes selecting randomly-chosen 64bit IDs is
"vanishingly small" as analyzed in RFC4429, Appendix A.
The probability of collision is certainly negligible
when compared to that for IP fragmentation which uses
a fixed 32bit IP source address plus 16bit ip_id.

In fact, draft-templin-dhcpmtu as it stands is overkill
along several different vectors. First, the sprite-mtu
checksum is completely unnecessary when we have such
high assurance against misassociating fragments during
reassembly. Secondly, the 32bit Identification field
is overkill when we consider that we already have a
randomly-chosen 32bit xid and that ordinary use cases
would not have millions of nodes on the same link
trying to bootstrap themselves at the same time.
Also, the UDP checksum will catch most reassembly
errors if fragment misassociation somehow occurs.
Finally, how bad could it be if a reassembly error
actually slipped past the UDP checksums? The server
would get confused and discontinue processing of
the message, and the client would time out and
try again.

These factors (plus some easily-fixed errors such as
failure to observe the 255 byte length restriction
for DHCP options) seem like reason enough to revive
the draft.

Thanks - Fred
[e-mail]  


>
>Yoshihiro Ohba
>
>> 
>> Ric is right that the normal-case link MTU is 1500 bytes,
>> but there is a remote possibility of smaller link MTUs
>> and/or larger DHCP messages; draft-templin-dhcpmtu'
>> addresses both possibilities. The approach would be
>> for the DHCP AUTH client to try w/o fragmentation first,
>> then try again using fragmentation if the unfragmented
>> attempt fails.
>> 
>> Fred
>> [e-mail] 
>>  
>> 
>> >-----Original Message-----
>> >From: Richard Pruss [mailto:[e-mail]] 
>> >Sent: Sunday, July 27, 2008 4:47 AM
>> >To: Yoshihiro Ohba
>> >Cc: [e-mail]
>> >Subject: Re: [dhcwg] draft-pruss-dhcp-auth-dsl-03
>> >
>> >Thanks for your comments,
>> >
>> >On 27/07/2008, at 10:19 AM, Yoshihiro Ohba wrote:
>> >
>> >> I have a couple of comments on new dhcp-auth I-D.
>> >>
>> >> - It still does not seem to address the issue of the difference in
>> >> retransmission directions.  Especially I am not sure how dhcp-auth
>> >> works when EAP-Success/Failure gets lost.
>> >>
>> >> - Comment on fragmentation.  The current draft says that 
>> >there is over
>> >> 200-octet space available more than the EAP MTU of 1020 octets.
>> >> However, I am not sure that if over 200-octet space is really
>> >> sufficient for 1500-octet MTU considering that DHCP relay agent
>> >> information option can be inserted by DHCP relay agent as well as
>> >> there can be 'shim' layers below IP.
>> >
>> >Relay's typically add only port information so I think we can 
>> >be quiet  
>> >safe with our 200 bytes
>> >also considering the real world EAP packet sizes.
>> >
>> >- Ric
>> >
>> >
>> >>
>> >>
>> >> - DHCP EAP request response message can be more confusing, 
>> >considering
>> >> the new extension to EAP such as ERX 
>(draft-ietf-hokey-erx) where two
>> >> new messages are defined that are neither request nor response.
>> >> Considering ERX, I would strongly discourage combining 
>DHCP and EAP
>> >> because ERX can make integration of DHCP and EAP even 
>more difficult.
>> >> It is best if we separate IP address configuration from 
>> >network access
>> >> authentication.
>> >>
>> >> Regards,
>> >> Yoshihiro Ohba
>> >>
>> >> On Fri, Jul 25, 2008 at 08:41:44AM +1000, Richard Pruss wrote:
>> >>> Hi,
>> >>>
>> >>> To help the discussion next week I was prompted to put out a  
>> >>> summary of
>> >>> changes.
>> >>> http://tools.ietf.org/html/draft-pruss-dhcp-auth-dsl-03
>> >>>
>> >>> We have tried in this version to address concerns raised 
>in IETF 70.
>> >>> Jari and Ralph's preso may remind you of those:
>> >>> http://www.ietf.org/proceedings/07dec/slides/intarea-2/sld1.htm
>> >>>
>> >>> We have added a first draft proposal for DHCPv6 messages for a  
>> >>> limited
>> >>> set of IPv6 deployments.
>> >>> 
>http://tools.ietf.org/html/draft-pruss-dhcp-auth-dsl-03#section-5.2
>> >>>
>> >>> We now have added a DHCP relay model to the DHCP 
>> >proxy/server model  
>> >>> that
>> >>> was the document model.
>> >>> (DHCP proxy is a term used in the DSL architectures, where 
>> >the BRAS  
>> >>> acts
>> >>> as a server to the client.)
>> >>>
>> >>> We have added a section on fragmentation.
>> >>> http://tools.ietf.org/html/draft-pruss-dhcp-auth-dsl-03#section-8
>> >>>
>> >>> The DHCP EAP request response messages are now separate 
>messages to
>> >>> possibly make the flow clearer
>> >>> and hopefully make the discussion around DHCP vs EAP 
>retransmission
>> >>> responsibility easier for people to understand.
>> >>>
>> >>> There is a section on backwards compatibility and a 
>number of cases
>> >>> considered, no updates to that but it addresses one of the 
>> >bullets on
>> >>> the slides in IETF-70.
>> >>>
>> >>> - Ric
>> >>>
>> >>> _______________________________________________
>> >>> dhcwg mailing list
>> >>> [e-mail]
>> >>> https://www.ietf.org/mailman/listinfo/dhcwg
>> >>>
>> >
>> >_______________________________________________
>> >dhcwg mailing list
>> >[e-mail]
>> >https://www.ietf.org/mailman/listinfo/dhcwg
>> >
>> 
>> 
>
_______________________________________________
dhcwg mailing list
[e-mail]
https://www.ietf.org/mailman/listinfo/dhcwg