Comments in line.
On Tue, 6 Apr 2010, Sriram, Kotikalapudi wrote:
....
So, for a wild example, suppose the validation check looked to see if one
single AS in the origin AS_SET matched a valid ROA for the prefix, and
passed the BGP route as valid. That would give a malicious person a way
to ensure that their bogus route would be accepted. If on the other hand,
the validation code looked to see if all ASs in the AS_SET matched a valid
ROA, that might break legitimate uses.
....
--Sandy
Sandy's hypothetical attack example as well John's original message made me
think about
enumerating all the use-case scenarios for {Prefix, Origin} validation with
AS-sets.
<snip>
There are two essential things that the address-resource owner needs to be made
aware of about validation:
1. In the long run, when deployment is deemed complete, an update with an AS
set will be considered Invalid except in cases A3, B3, and C3.
2. NO concession is made during validation by way of aggregation of more
specific prefixes in ROAs to “effectively” create a ROA for a less specific
prefix.
Example: Existence of the following four ROAs, namely, ROA-1: {10.1.0.0/18,
AS64496}, ROA-2: {10.1.64.0/18, AS64497}, ROA-3: {10.1.128.0/18, AS64498},
ROA-4: {10.1.192.0/18, AS64499} WILL NOT by itself result in positive
validation of an update that has the aggregate prefix 10.1.0.0/16 with the AS
set {AS64496, AS64497, AS64498, AS64499} as origin.
You seem to be presuming specific validation outcomes here. I'm not
saying you are wrong, but I'm not certain the use cases document should be
saying ANYTHING about the validation algorithm.
I also don't know if we have a consensus of the security concerns with
AS_SETS to be making presumptions about the validation outcomes. It
appears that John, for one, does not care for the roa-validation draft's
suggested outcome.
As wg chair, I'd appreciate some more discussion on what to do with
AS_SETS. Jump in any time folks. As I tried to say yesterday, the object
here is to avoid leaving a hole in the spec implementers would have to
imaginatively fill, creating inter-op problems and maybe a vulnerability.
About point 2: There is this thing called "proxy aggregation" where an
ISP discovers two external peer BGP routes that are aggregatable and
announces the aggregate. Long long ago, IIRC, we decided that proxy
aggregation was too hard and too rare and we would declare it out of
scope. Your point 2 would specifically disallow proxy aggregation. Not
saying that's a bad thing, for a feature that was declared out of scope.
Just warning about the side effect.
--Sandy
Note: Sandy’s hypothetical attack scenario is easily mitigated if the outcome
of validation in use case C2 is always “Invalid.” The algorithms should be
designed to carefully consider all of the nine use-case scenarios listed above
(assuming we are not missing any that are of any consequence).
Sriram
_______________________________________________
sidr mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/sidr