Hi!
Xiaoqi Chen (Danny):
Hi Georg,
First of all, thanks to all of you for the effort writing up the proposal!
Sure. Thank you for your input.
I saw that email obfuscation was discussed but no solution was proposed. I want to throw in some ideas about obfuscating the email centrally:
- Let's not publish cleartext email in any public descriptor, and only
publish an obfuscated address, something like OperatorID@relay-operators.torproject.org or Fingerprint@relay-operators.torproject.org.
- Only those already operating a relay, using their ContactInfo as "from"
address, can send to these obfuscated addresses and get forwarded to the actual recipient email. Otherwise the email gets rejected or ignored. (Of course TorProject folks can also be added to the list of allowed senders.)
This will be very similar to how tor-relays@lists.torproject.org currently operates. This should reduce spamming as the cost for spamming is quite nontrivial (go through the hassle of setting up a new relay), while frictionlessly serving the existing and planned use cases, mainly 1) broadcast from tor project to all operators, 2) community discussion between relay operators, and 3) debugging and reaching individual operators or a small batch of them.
Of course, this solution might be unnecessary (and I agree that historically there's not much spam). We can balance the benefit and cost -- the marginal cost for adding forwarding logic as part of the future operator ID / email verification system.
This an interesting idea. However, I have my doubts that relay operators would be happy with it. I mean there are some folks that already have concerns that we require an email address to begin with (as this might leak things about themselves) but what you are suggesting is that Tor would centrally save those email addresses and keep a mapping to some public piece visible in the descriptor. Moreover and even worse, we would see whom of the relay operators would be talking to whom and when this would be happening (if I got the idea right), which sounds problematic to me and would probably not be in the interest of the operators.
I think the comparison to the tor-relays mailing list is a bit mis-matched as well given that there is no requirement to sign up to that mailing list.
So, setting up spam protections and filtering seems to be way less intrusive than what you have in mind. But I am fine thinking more about it if there is interest from the community.
Thanks, Georg
-- Danny
-- Yours sincerely, Xiaoqi Chen
On Sat, Oct 21, 2023 at 3:55 PM Georg Koppen gk@torproject.org wrote:
Hello everyone!
As indicated in our bug tracker a while ago[1] we have some strong incentives to redo our ContactInfo field. I've collected all the different use cases and combined them in a single proposal, discussing some potential concerns and future work we could get built upon it. The work is tracked on Gitlab as well[2] feel free to provide feedback there or here on the list as a follow-up to this announcement. For your convenience the proposal text is included below (if you like reading the .md file on Gitlab, see my personal repository for the latest draft[3]).
"""
Filename: 100-contactinfo-mandatory-email-address.md Title: Restricting ContactInfo to Mandatory Email Address Author: Georg Koppen Created: 2023-10-21 Status: Open
## Overview
This document proposes to change the ContactInfo field from a free text field to one that is only allowing an email address. Additionally, providing such an email address will be required after this proposal is implemented. This is a normative document.
## Motivation
Being able to reach out to and contact relay operators (bridge operators are included in that group) is important for our day-to-day work at Tor. While this has been brought up in the past as helpful in our fight against malicious relays, it goes well beyond that use case and is affecting general network health work, community-building efforts and quickly deploying anti-censorship related security fixes among other things.
Tor is providing a `torrc` configuration option, `ContactInfo`, which is supposed to contain an email address (potentially obfuscated) under which the relay operator may be reached. However, in practise this does not work very well for two reasons:
- `ContactInfo` is a free-form field which allows the operator to include not just email addresses but essentially any kind of text they want. This results in a lot of overhead when trying to contact all operators and failures to do so because not everything added to `ContactInfo` is a way to actually contact operators or undoing `ContactInfo` obfuscation turns out to be too hard.
- `ContactInfo` can be empty as it is not required for all operators and in cases where it is supposed to be required (e.g. for operators running more than one relay) the requirement is not enforced.
Over the years different teams at Tor developed different workarounds to the issues mentioned above without addressing them directly. The goal of this proposal is to make those workarounds obsolete.
## Fixing ContactInfo
As briefly mentioned in the [Overview section](#overview) the idea of this proposal is deceptively simple: `ContactInfo` will be turned into a field allowing just an email address, under which the operator can be reached. Moreover, providing such an address will be a requirement for all operators regardless whether they run one relay or many of them, or bridges.
Why will we require an email address and not, say, a domain set up somewhere or an account on our forum instead? Firstly, a lot of those other potential options require a valid email address themselves in the first place and would therefore just raise the costs for relay operators. Secondly, email still scales way better than other alternatives allowing us to reach more people in less time. And, thirdly, email addresses are required anyway in other network related infrastructure parts, e.g. the RIPE database for IP address space allocation comes to mind or abuse contacts for dealing with potentially malicious Tor traffic/Tor users.
Even though this proposal will not lay out the envisioned changes at a technical level (for that to happen we'll need a corresponding [tor-spec] proposal later on) there are a number of details to consider mainly stemming from the status quo existing for years and the accumulated usage patterns of the `ContactInfo` field "in the wild".
The remainder of this proposal will be devoted to explain those details and address potential concerns related to them.
## Implementation
The C-Tor codebase is in maintenance mode and not accepting any new features anymore (with a very narrow set of exceptions). We therefore plan to have this change included solely in the upcoming Arti relay work. Note: while we'll be using `ContactInfo` throughout this proposal it does no mean that the respective configuration option for Arti relay will be the same. It could just be [`contact`] or something else. The important point is that this proposal applies to whatever is providing the argument for the "contact" keyword in the server descriptor then.
Requiring a syntactically correct email address in an Arti relay configuration option is necessary but not sufficient for our purposes. Otherwise `ContactInfo` values like `nobody@example.com`, which is used by some relays and bridges in the Tor network at the time of writing this proposal, would be acceptable. However, in that case there would be no way to get into contact with respective relay operators. We intend to solve that problem by deploying an email verification service: relays without a verified `ContactInfo` value won't be allowed on the network.
## Concerns
There are some concerns related to this proposal which we should discuss and address where needed. The following sections will outline them and explain why (and how) we think they are non-issues or can be mitigated.
### Backward compatibility and breakage of external tooling
What do we do with `ContactInfo` strings that are not email addresses already? The main goal should be to make the transitioning process to the new `ContactInfo` string as smooth and easy for relay operators as possible. The good news is that we plan to have a transitioning tool for the overall change of C-Tor relays to Arti relays anyway and having the `ContactInfo` change be part of that should help. Running an awareness campaign and providing up-to-date documentation to reach operators that need to fix (or even set up) their `ContactInfo` manually should further minimize the risk of losing relays in the network and volunteers running them to begin with.
However, even if it were possible to extract an email address from the C-Tor relay configuration of any C-Tor relay or bridge without manual intervention by the operator there is another problem to consider: there is often additional information available in current `ContactInfo` strings, which sometimes external tools got built upon. One well-known example for that is [`OrNetStats`] which aims at displaying statistics related to relays and bridges in the Tor network. One important piece of it is grouping the displayed information around so-called Authenticated Relay Operator IDs (AROIs) were possible, which got developed in the [ContactInfo Information Sharing Specification] (CIISS) and is making use of overloading the `ContactInfo` field with additional information besides an email address. While we can't avoid breaking tools when evolving the `ContactInfo` semantics we can provide ways for those tools to get adapted to the new reality. For that we propose to create an additional configuration option that includes the non-email address information in the `ContactInfo` field and is made public in extra-info descriptors. Even though we believe validated email addresses as proposed in this proposal together with the planned [happy families proposal] and [proposal 347] provide similar benefits as the AROI mechanism, and hence make it superfluous, if someone would still like to center their relays statistics around AROIs they could do so by extracting the necessary information from extra-info descriptors after this proposal got implemented.
### Spam
Exposing an email address without obfuscation in the `ContactInfo` field is a potential spam risk which is why some operators have resorted to obfuscating it, often in very creative ways. However, we do think that those efforts are not effective if we want to retain a way of being able to reach operators by email. Moreover, our own and others' experience of running relays and providing an unobfuscated email address as `ContactInfo` shows that spam right now is actually not an issue. That might change in the future, though, which is why we strongly recommend to use a dedicated and long-term email address in the `ContactInfo` field. The long-term part is particularly important, as this proposal turns that email address into a validated relay operator ID, which is valuable for building trust over the longer run.
## Impact on relay operators and the growth of trust in the Tor network
As can be seen in the previous sections, the proposed changes in this document can come with a cost for relay operators. However, we do not expect them to have a lasting impact on our relay operator community and the health of the network as those changes are mainly accompanied by a one-time transition cost that puts a burden on our current relay operators. And that one-time cost we try to keep as low as possible. Additionally, we might lose or alienate some of our current and potential new operators, as they might not want to provide any (verifiable) email address at all as a requirement for running relays, but rather prefer participation anonymously. However, in order to fight malicious relays we need to strengthen our community and get to build connections and contacts with each other, which is hard if there is no way to get into contact to begin with. As outlined at the begin of this document, a lack of contact information makes other day-to-day work at Tor hard or impossible as well. Thus, overall we believe providing a working email address when running relays is a small cost compared to the improvements of the health of our network and the safety of our users.
There is another benefit which comes with this proposal: the validated email address can essentially work as a relay operator ID (similar to the above mentioned AROI). While we do deal with single relays or families of them when we are trying to determine whether they pose a danger to our users, what we are ultimately aiming at is the operator behind them. Having something like a relay operator ID does help with that part. But there is more: it allows us to think better about trust on the Tor network and how we might be able to implement ideas like "operator X is only allowed to run Y% of the Tor network" or "unknown operators are only allowed to run Z% of Tor's network capacity", to name just two. There are a number of proposals both [trust][torspec 103] [related][relay-search 40001] and [not trust-related][team 220] that would benefit from having relay operator IDs available.
## Acknowledgements
Big thanks go to @gus and @ahf for discussions and feedback around this proposal idea, which improved this document considerably.
[happy families proposal]:
https://gitlab.torproject.org/tpo/core/torspec/-/blob/82ba302230911cafd8683d... [proposal 347]:
https://gitlab.torproject.org/tpo/core/torspec/-/blob/523404500320ba9700fe5a... [torspec 103]: https://gitlab.torproject.org/tpo/core/torspec/-/issues/103 [relay-search https://gitlab.torproject.org/tpo/core/torspec/-/issues/103%5Brelay-search 40001]:
https://gitlab.torproject.org/tpo/network-health/metrics/relay-search/-/issu... [team 220]: https://gitlab.torproject.org/tpo/network-health/team/-/issues/220 """
Thanks, Georg
[1] https://gitlab.torproject.org/tpo/core/arti/-/issues/870 [2] https://gitlab.torproject.org/tpo/community/relays/-/issues/71 [3]
https://gitlab.torproject.org/gk/policies/-/blob/prop100/100-contactinfo-man... _______________________________________________ tor-relays mailing list tor-relays@lists.torproject.org https://lists.torproject.org/cgi-bin/mailman/listinfo/tor-relays
tor-relays mailing list tor-relays@lists.torproject.org https://lists.torproject.org/cgi-bin/mailman/listinfo/tor-relays