[Ndn-interest] [EXT]Re: KeyLocator name vs Certificate name
Lan Wang (lanwang)
lanwang at memphis.edu
Mon May 4 10:41:45 PDT 2020
I see. I think (1) can be discussed further at the weekly call on Friday. For (2), you can discuss with John.
On May 4, 2020, at 12:32 PM, Junxiao Shi <shijunxiao at email.arizona.edu<mailto:shijunxiao at email.arizona.edu>> wrote:
1. I think the KeyLocator field could be extended to contain multiple locators if the situation you mentioned below is to be supported.
For a scenario with multiple unrelated trust anchors, having more than one Name in the KeyLocator still leaves the question of "which certificate to fetch". However, the consumer could always pursue every certificate chain, and find the one trust anchor.
2. For the current testbed, why is there a need to use exact name of the certificate given there is only one trust anchor?
There's no guarantee that "NDN Testbed Root" is the only trust anchor. Application can create an "overlay" that uses a different trust anchor.
If I publish a certificate following a different trust anchor and it somehow gets into the cache, future validations that expect "NDN Testbed Root" trust anchor will fail.
3. I’m just curious why you are encountering a problem right now (maybe this is related to 2).
I published a self-signed certificate in a repo. This isn't invalid action, because a self-signed certificate is a valid Data packet.
However, this breaks validation, because the validator's certificate retrieval Interest is being satisfied by the self-signed certificate, and then it cannot trace back to the trust anchor.
On May 4, 2020, at 11:03 AM, Junxiao Shi <shijunxiao at email.arizona.edu<mailto:shijunxiao at email.arizona.edu>> wrote:
Putting the exact or full name of the certificate in KeyLocator is certainly an option.
It works well in the common case, where each key is used under only one trust model.
It becomes suboptimal in an uncommon use case:
* There are two distinct groups of consumers that want to fetch my Data.
* Both groups use the hierarchical trust model, but with different trust anchors.
* Currently, I could:
* Obtain a certificate under each trust anchor, with different IssuerId.
* Publish the Data only once with KeyLocator ending at the KeyId component.
* Suppose the consumers can magically figure out the IssuerId for their group, they can verify the Data successfully.
* If the KeyLocator should be exact/full names, I'll have to publish the Data twice.
However, given today's testbed has a single trust anchor, should the KeyChain implementations default to putting the exact name of the certificate into KeyLocator?
To start, the certificates issued by testbed infrastructure should have exact names in KeyLocator.
On Mon, May 4, 2020 at 11:46 AM Lan Wang (lanwang) <lanwang at memphis.edu<mailto:lanwang at memphis.edu>> wrote:
If the KeyLocator uses the full name of the certificate, would there still be a problem? If not, then the application just needs to use the certificate name.
There is also a KeyDigest option defined in the spec. That may be another solution.
A KeyLocator specifies either Name that points to another Data packet containing certificate or public key or KeyDigest to identify the public key within a specific trust model (the trust model definition is outside the scope of the current specification). Note that although KeyLocator is defined as an optional field in SignatureInfo block, some signature types may require presence of it and some require KeyLocator absence.
KeyLocator = KEY-LOCATOR-TYPE TLV-LENGTH (Name / KeyDigest)
KeyDigest = KEY-DIGEST-TYPE TLV-LENGTH *OCTET
See Name specification<https://named-data.net/doc/NDN-packet-spec/current/name.html#name> for the definition of Name field.
The specific definition of the usage of Name and KeyDigest options in KeyLocator field is outside the scope of this specification. Generally, Name names the Data packet with the corresponding certificate. However, it is up to the specific trust model to define whether this name is a full name of the Data packet or a prefix that can match multiple Data packets. For example, the hierarchical trust model [BZA+13]<https://named-data.net/doc/NDN-packet-spec/current/signature.html#testbed-key-management> uses the latter approach, requiring clients to fetch the latest version of the Data packet pointed by the KeyLocator (the latest version of the public key certificate) in order to ensure that the public key was not yet revoked.
On Fri, Oct 4, 2019 at 11:33 AM Junxiao Shi <shijunxiao at email.arizona.edu<mailto:shijunxiao at email.arizona.edu>> wrote:
Whether "both certificates are valid for the same key" depends on the trust schema of the validator. I'm not dealing with multiple trust anchors here.
A real world example: when Let's Encrypt started, its root certificate "ISRG Root X1" was not recognized by browsers, so they had their "Authority X3" key cross-signed by IdenTrust.
The equivalent in NDN would be "Authority X3" key has two certificates:
On a packet signed by "Authority X3" key, the KeyLocator name would be /lets-encrypt/KEY/X3.
Suppose we have a validator that ONLY knows IdenTrust as a trust anchor, but does not recognize "ISRG Root X1" as a trust anchor.
When it sends an Interest for /lets-encrypt/KEY/X3 , it could receive the certificate /lets-encrypt/KEY/X3/ISRG/35=%01 that have a KeyLocator /isrg-root/KEY.X1 .
The validator will then retrieve the "ISRG Root X1" certificate, and see it's a self-signed certificate but not a trust anchor (remember this validator ONLY recognizes IdenTrust).
Eventually, validator rejects the original packet.
What SHOULD happen is that, the validator needs to retrieve the /lets-encrypt/KEY/X3/IdenTrust/35=%01 certificate instead of (or in addition to) /lets-encrypt/KEY/X3/ISRG/35=%01 certificate.
This would allow it to continue validation toward a known trust anchor.
The question here is: how could the validator know to retrieve the /lets-encrypt/KEY/X3/IdenTrust/35=%01 certificate?
On Thu, Oct 3, 2019 at 8:19 AM Lan Wang (lanwang) <lanwang at memphis.edu<mailto:lanwang at memphis.edu>> wrote:
If both certificates are valid (for the same key), why would the trust model expect only one of the certificates, not the other one? The trust model should be updated to allow either one. My recent ICN paper talks about supporting multiple trust anchors and schemas for this case. This is something the application can do, although maybe library support will help. For more information, you can read:
A. Gawande, J. Clark, D. Coomes, L. Wang, "Decentralized and Secure Multimedia Sharing Application over Named Data Networking<http://web0.cs.memphis.edu/~lanwang/paper/ICN19-npchat.pdf>," in Proceedings of ACM Conference on Information Centric Networking, Sept. 2019
On Oct 2, 2019, at 8:35 AM, Junxiao Shi <shijunxiao at email.arizona.edu<mailto:shijunxiao at email.arizona.edu>> wrote:
Finding the latest certificate name ("latest" as in "latest version") is a solved problem: Realtime Data Retrieval (RDR) protocol allows a validator to identify which version is the latest.
Suppose there are two certificate versions from the same issuer:
Using RDR protocol, the validator could ask: /ndn/web/KEY/%29%964%F6%11%12%1C%9A/NA/32=metadata CanBePrefix=1 MustBeFresh=1, and get a reply that %FD%02 is the latest version.
However, knowing the latest version does not help when there are multiple issuers.
In the example given in my last post<https://www.lists.cs.ucla.edu/pipermail/ndn-interest/2019-October/002600.html>, there are certificates issued by two separate issuers:
RDR cannot currently operate on this level, but can easily be extended to do so. Then, the latest version among these two certificates would be /ndn/web/KEY/%29%964%F6%11%12%1C%9A/X/%FD%00%00%01m%88%1E%0A%07 as its version component has a greater number. However, if the validator is expecting ndn-testbed-root-v2 trust anchor, the validation will fail.
Enumerating the namespace using namespace catalog protocol<https://redmine.named-data.net/issues/4666> might help. The validator can:
1. Enumerate /ndn/web/KEY/%29%964%F6%11%12%1C%9A namespace, and find there are two issuerIds "NA" and "X".
2. Fetch the latest certificate under each issuerId.
3. Determine which certificate(s) to follow up according to trust schema.
However, isn't it better to make issuerId part of the trust schema itself, so that the validator could skip the namespace enumeration step.
A related problem: both RDR and namespace catalog protocol need some way to verify the response Data packets come from the namespace owner. If we don't yet have the certificate of the namespace owner, how can we trust the responses?
On Wed, Oct 2, 2019 at 8:52 AM Lan Wang (lanwang) <lanwang at memphis.edu<mailto:lanwang at memphis.edu>> wrote:
It seems to me that a separate system is needed to find the latest certificate name (e.g., through NDNS). That’s an orthogonal issue to the KeyLocator format which seems to be fine the way it is.
On Oct 1, 2019, at 12:51 PM, Junxiao Shi <shijunxiao at email.arizona.edu<mailto:shijunxiao at email.arizona.edu>> wrote:
I wonder what's the relation between KeyLocator name and Certificate name?
Packet format spec<https://named-data.net/doc/NDN-packet-spec/0.3/signature.html#keylocator> says this on KeyLocator name:
The specific definition of the usage of Name and KeyDigest options in KeyLocator field is outside the scope of this specification. Generally, Name names the Data packet with the corresponding certificate. However, it is up to the specific trust model to define whether this name is a full name of the Data packet or a prefix that can match multiple Data packets. For example, the hierarchical trust model<https://named-data.net/publications/techreports/trpublishkey-rev2/> uses the latter approach, requiring clients to fetch the latest version of the Data packet pointed by the KeyLocator (the latest version of the public key certificate) in order to ensure that the public key was not yet revoked.
Certificate format spec<https://named-data.net/doc/ndn-cxx/0.6.6/specs/certificate-format.html#name> defines certificate name to be:
KeyId is an opaque name component to identify an instance of the public key for the certificate namespace. The value of Key ID is controlled by the namespace owner and can be an 8-byte random number, SHA-256 digest of the public key, timestamp, or a simple numerical identifier.
Issuer Id is an opaque name component to identify issuer of the certificate. The value is controlled by the certificate issuer and, similar to KeyId, can be an 8-byte random number, SHA-256 digest of the issuer’s public key, or a simple numerical identifier.
In today's testbed, most applications are setting the KeyLocator name to be /<SubjectName>/KEY/[KeyId].
For example, ivoosh server's Data packet /ndn/web/video/research_questions/hls/playlist.m3u8/%FD01/%00%00 uses KeyLocator name /ndn/web/KEY/%29%964%F6%11%12%1C%9A .
After expressing an Interest for that name, I receive a certificate named /ndn/web/KEY/%29%964%F6%11%12%1C%9A/NA/%FD%00%00%01i%FE%FD~%BF .
Everything seems to be working fine. However, what if there are more than one certificates associated with the key?
Suppose the namespace owner requests me to issue a new certificate to his existing public key, I could execute:
ndnpeek /ndn/web/KEY/%29%964%F6%11%12%1C%9A/NA/%FD%00%00%01i%FE%FD~%BF | base64 | ndnsec cert-gen -i X - | base64 -d > X.data
and then let him publish the new certificate: /ndn/web/KEY/%29%964%F6%11%12%1C%9A/X/%FD%00%00%01m%88%1E%0A%07
Now, anyone who asks for a certificate using the KeyLocator name /ndn/web/KEY/%29%964%F6%11%12%1C%9A could receive either of the two certificates.
The validator, depending on the trust model, could be expecting one of these certificates.
If it happens to receive the other certificate, validation would fail because the signer does not match policy.
So, what's the solution space?
Trying to ensure there is only one issuer per key defeats the purpose of having IssuerId name component in the first place.
The reason to have IssuerId is that, a namespace owner can obtain certificates from multiple issuers on the same public key, so that the same content can be made available under multiple trust models.
Adding IssuerId to KeyLocator name seems to fix the problem of validator receiving the wrong certificate.
However, this would require the namespace owner to generate separate Data packets for each trust model.
In this case, she may as well use separate keys.
Providing all certificates in a bundle could solve the problem.
The producer would include certificates from multiple distinct trust models in the certificate bundle<https://redmine.named-data.net/issues/2766#note-30>, and then the validator should attempt to validate the original packet using each certificate matching the KeyLocator name, and accept the original packet if any certificate chain leads to a trust anchor configured in the validator.
A drawback of this solution is inflation of certificate bundle size.
Appending an IssuerId component in the validator appears to be the best solution. The validator needs to know, through static configuration, what IssuerId to append to each KeyLocator name.
For example, a validator configured to work with ndncert-legacy's certificate chain could be configured to always append 'NA' as IssuerId. Given the KeyLocator name shown above, the validator would fetch certificate using /ndn/web/KEY/%29%964%F6%11%12%1C%9A/NA and it would receive the expected certificate.
Please REPLY-ALL with your thoughts on this topic.
Ndn-interest mailing list
Ndn-interest at lists.cs.ucla.edu<mailto:Ndn-interest at lists.cs.ucla.edu>
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the Ndn-interest