Re: [Network-tokens] [TLS] Network Tokens I-D and TLS / ESNI

[Tom Herbert <[e-mail]>]

On Thu, Jul 9, 2020 at 5:33 PM Yiannis Yiakoumis
<[e-mail]> wrote:
>
>
> On Thu, Jul 09, 2020 at 4:51 PM, Tom Herbert <[e-mail]> wrote:
>>
>> On Thu, Jul 9, 2020 at 4:28 PM Yiannis Yiakoumis <[e-mail]> wrote:
>>>
>>> On Thu, Jul 09, 2020 at 3:46 PM, Tom Herbert <[e-mail]> wrote:
>>>>
>>>> On 6/26/2020 10:16 AM, Yiannis Yiakoumis wrote:
>>>>
>>>> On Fri, Jun 26, 2020 at 7:29 AM, Christian Huitema
>>>> <[e-mail] <mailto:[e-mail]>> wrote:
>>>>
>>>> On 6/25/2020 11:11 PM, Melinda Shore wrote:
>>>>
>>>> On 6/25/20 3:29 PM, Erik Nygren wrote:
>>>>
>>>> One quick comment is that binding tokens to IP addresses is strongly counter-recommended.
>>>> It doesn't survive NATs or proxies, mobility, and it is especially problematic in IPv6+IPv4 dual-stack environments.
>>>>
>>>> There's been a bunch of past work done developing similar sorts of protocols, and for what it's worth I wrote up a mechanism for using address tags and address rewrites, but unfortunately Cisco decided to patent it. Anyway, thre are ways of dealing with this problem that don't require binding the address to the token ("all technical problems can be solved by introducing a layer of indirection").
>>>>
>>>> There is also an interesting privacy issue. The token is meant to let a provider identify some properties of the connection. I suppose there are ways to do that without having it become a unique identifier that can be tracked by, well, pretty much everybody. But you have better spell out these ways.
>>>>
>>>> You are right that for the duration of a token, one could use it to identify an endpoint (either application or most likely a combination of user/application). Tokens expire and intermediary nodes cannot correlate tokens with each other as they are encrypted. So tracking cannot happen across different tokens (of the same user), or between token-enabled and non-token-enabled traffic. I guess similar type of tracking happens when users are not behind a NAT and their IP address can be used to track them. Would it make sense to have the user add a random value to a token, and then encrypt it with the network's public key, so that each token becomes unique and cannot be tracked. Would that address the privacy concerns better?
>>>>
>>>> That would certainly be better. The basic rule is that any such identifier should be used only once. Pretty much the same issue as the session resume tickets.
>>>>
>>>> Then, there are potential interactions with ESNI/ECH. The whole point of ECH is to keep private extensions private. The token extension would need to be placed in the outer envelope, which is public but does not expose seemingly important information like the SNI or the ALPN.
>>>>
>>>> Ah, I was not aware that ESNI can now include all CH extensions - thanks for the pointer. Yes, the token would have to stay on the outer envelope so the network can process it. The main idea is you can encrypt everything that is client-server specific, and just keep a token to explicitly exchange information with trusted networks.
>>>>
>>>> There are also implications for QUIC, in which the TLS data is part of an encrypted payload. The encryption key of the TLS carrying initial packets is not secret in V1, but it might well become so in a future version.
>>>>
>>>> Haven't looked into QUIC yet, but is on the list of things to do. If anyone is interested to help us explore this, please let me know.
>>>>
>>>> You may want to have that discussion in the QUIC WG. If you are building some kind of QoS service, you probably want it to work with QUIC too.
>>>>
>>>> I view the use of TLS to carry tokens as a temporary hack to make something work quickly. There are several obvious problems that prevent that from being the general solution: it's specific to TLS/TCP, it requires intermediate nodes to parse into TCP payloads, it requires tracking connection state in the network.
>>>>
>>>> IMO, the real solution is to encode tokens in HBH options. This addresses the aforementioned problems, works with any transport protocol (including QUIC, IPsec tunnels, arbitrary encapsulations), eliminates state and need for DPI in the network.
>>>>
>>>>
>>>
>>>
>>> + Christian as I think he posted at the tls mailing list. Lots of interesting points.
>>>
>>> Define "temporary" and "quickly" :) Tokens reflect a relationship between an end-user/app and a network, so you want apps/users to be able to insert these tokens to their traffic. Agreed that IPv6 can work with multiple transports and is more generic,  but you will then have to find ways to pass the token all the way from the user/app down to the IPv6 socket. TLS/QUIC is much closer to the end-user/developer, and most high-level frameworks (e.g., OkHTTP, web browsers) expose APIs to define certain properties. This is not the case with IPv4/IPv6 (e.g., I don't think it's possible to set DSCP bits from OkHTTP or Javascript).
>>
>>
>> Hi Yiannis,
>
>
>
> Hi Tom,
>
>> We don't need to find ways to pass the token, there are already sockets APIs to do that.
>
>
> How can an Android/iOS/web developer interact with these socket APIs? Are there side ways to do it while using the high-level frameworks available there, or do you have to skip them and build your code directly using sockets? Both STUN and TLS support custom extensions, and we've been able to use tokens with OpenSSL/BoringSSL (but we had to build and load our own version along with the app). If we make tokens a standard extension, then other frameworks will pick it up right away and they will become available for everyone to use (the same way it happened with SNI and ALPN).
>
>> The issue with network devices reading transport layer payload is that port numbers, which are used to distinguish TLS, QUIC payloads, etc. are not protocol numbers. They are only interpreted by the end points (RFC7605). The network never really knows if port 80 for TCP is HTTP, or for UDP is QUIC. It's an approximation that may be incorrect. Similarly, if the application uses some other port numbers for their protocols then they're out of luck unless the network knows about those. All of this was brought up in the SPUD discussions a while back, other than adding magic numbers into the payload, which is probabilistically correct, there really wasn't much of an answer to the problem.
>
>
> Yes, you don't have guarantees here. We have an implementation were tokens are carried over STUN messages during WebRTC calls. We assume that every UDP packet is STUN and parse it based on expected type-len fields as long as it gives us something valid. We haven't looked at the probability of error and performance penalty yet, these would be good things to understand. But given the trend towards programmable hardware/software packet parsers, I am optimistic.
>
>> There's other issues like segmentation of the TLS header, ossification of TLS headers, etc.
>
>
> Would DTLS help to assume that header is within a single packet?
>
>> Honestly, I doubt that network tokens in transport payload with the intent that intermediate devices parse them is something that could be standardized in IETF.
>
>
> Why is that? Isn't this how all firewalls/DPI/TDF etc work today? I think it's a good point to bring up into the TSV discussion and get feedback.
>
Yiannis,

It's because host-to-network signaling in UDP payload has been tried
before in IETF, tsvwg specifically, and failed to go anywhere. Please
take a look at SPUD, draft-hildebrand-spud-prototype-03 (there was
even a BOF or two on it with some pretty animated discussion!). It's
likely SPUD will be brought up in any TSV discussions :-).

Firewalls and DPI typically operate outside of the auspices of IETF
protocol specifications. Other than spin-bit in UDP, I don't believe
there are any other cases where protocols allow intermediate devices
to look at payload. In fact, DPI is actually one leading cause of
protocol ossification so there's a lot of motivation to eliminate it
(this is why transport protocols are moving to not just encrypting
payload but also the transport headers). This is another ongoing
discussion in TSV that will likely be brought up (see
draft-ietf-tsvwg-transport-encrypt-13 and all the animated discussion
around that :-) ). To that end moving network tokens in extension
headers eliminates the need for DPI and resolves the tussle around
transport layer header encryption.

Tom

>> A secondary issue that you might want to consider is security. A rogue application can easily spoof tokens in TLS or QUIC packets, the OS is out of the loop on what applications put in their payload. Setting an extension header on a socket requires going through the OS which can apply policy to see if the system configuration permits that, so it's going to be harder for random applications to spoof tokens.
>
>
> The role of the OS in describing and enforcing the desired policy is an important point and requires lots of discussion. Holding off for now though...
>
>> Tom
>>
>>>
>>>
>>> In terms of connection state and whether you want to do it per-flow or per packet. I think there is an overall trade-off here around how big is the token, how much state it introduces at the network, and what are the processing capabilities in terms of processing and decrypting tokens.
>>>
>>> For example, you can have a token which is a 128-bit random/opaque value that you use per-packet as a HBH option, but then the network needs to keep state for all the valid tokens, and accept some risk of a replay attack. The other alternative is to have a token that is encrypted and is therefore longer, but has enough information for the network to decrypt it and process it without requiring any state at all.
>>>
>>> I think both are viable, and it should be up to the network operator to decide what type of tokens it uses. if you look at how OATH2 and the IAM industry deals with it is that they support both.
>>>
>>>> Of course the naysayers will bring up the fact that HBH options aren't ubiquitously supported by routers. I think there are some ways around this:
>>>> 1) Network tokens really only make sense to the network that issued them, so their use could be restricted to that limited domain where they are consumed.
>>>> 2) Packets that are leaving the limited domain that carry network tokens in HBH options could be filtered and the HBH option is removed
>>>> (see proposal draft-herbert-6man-eh-attrib-01 to allow intermediate nodes to remove options).
>>>>
>>>> Limiting the visibility of tokens to the local network also helps their security.
>>>
>>>
>>> I am not a big fan of the assumption that tokens are relevant only to the local network, and can be issued only by it. In a zero-rating or firewall scenario, the token would be specific to the application provider (and maybe issued and signed by them), which is likely not co-located with the network. This is detailed on the application token in the I-D.
>>>
>>> Yiannis
>>>
>>>> Tom
>>>>
>>>> --
>>>> Network-tokens mailing list
>>>> [e-mail]
>>>> https://www.ietf.org/mailman/listinfo/network-tokens
>
>
>