This is a work-in-progress draft, significant portions might change in a non-compatible way.
Tor users are facing persistent malicious actors repeatedly running large fractions of the tor network’s capacity to exploit them   . Detecting all malicious tor network capacity is not practically feasible using active scanners in many cases since attackers have moved from attacking all connections to more targeted approaches where only users of specific domains (that are not necessarily known to defenders) are exploited. Transport level encryption (HTTPS) can defeat many types of attacks by malicious exit relays and the global HTTPS availability has significantly increased over the past years but is still not ubiquitous yet, especially on the first connection. Therefore we propose to publish relay operator trust information to subsequently have the possibility to limit the fraction and impact of unknown tor network capacity by consuming trust information.
Trust in the context of this document is a proxy for a link between community members. For example a local hackerspace might trust its community members to run tor relays without malicious intent.
This document describes how trust information is published, can be retrieved and validated. It does not introduce any new requirements for tor relay operators.
The high level goal is to limit the impact of malicious tor relays on users and to increase the trustworthiness of the tor network for tor users. To achieve that goal we need information about whether a operator is known/trusted or completely unknown (no links to known entities).
On a more technical level the goal is to establish a simple protocol to publish and consume trust information about relay operator IDs. Relay operator IDs are the proven operator domain  . As of November 2021 over 50%  of the tor exit capacity has adopted the proven operator domain and it is the only available specified operator identifier that can not be arbitrarily spoofed and can automatically be verified.
Trust is issued to relay operator IDs and not individual relay fingerprints for scalability reasons and so new or changed relays for a known operator get the same level of trust without requiring any update to published trust information.
The trust scheme supports any relay type (guards, middle and exit relays) but it is expected that it will be more useful for relay types that impose a higher risk for tor users: guard and exit relays.
Adversaries want to maximize their network fraction of the overall tor network capacity and run their attacks without getting detected.
Trust information consumer want to learn about trusted operator IDs and to detect manipulated trust information (content of
A trust anchor is the initial starting point which is used to find trusted relay operator IDs. TAs publish relay operator IDs.
By publishing a relay operator ID a TA asserts that
TAs publish non-malicious operator IDs only. TAs do not publish any negative or “do not use” operator deny lists. Trust is binary. There is no notion of “some” trust. Consumers of trust information can use one or more trust anchors to find trusted operator IDs.
Trust anchors publish trusted relay operator IDs via a well-known URL for trust information consumers. Trust anchors must be able to serve a text file via HTTPS from the DNSSEC-enabled domain that trust information consumers have configured. Additionally TAs must publish integrity information (a hash of the text file) in DNSSEC-signed TXT records. This allows the TA to publish trust information via semi-trusted systems (i.e. CDNs) without giving them the power to modify trust information.
Relays operated by TAs are also considered trusted if their proven operator domain matches the one from the TA.
Optionally relay operators can also publish trust information and trust information consumers will use it when TAs assign the operator ID the necessary recursion flag. Trust information published by relay operators has the same meaning (asserts that the published operators are running relays without malicious intent) and requirements (HTTPS, hash published via DNSSEC-signed record, met physically at least once).
Relay operators that do not publish trust information do not have a DNSSEC requirement on their domain.
Consumers of trust information can choose what trust anchor they trust to find trusted operator IDs. In practice we assume that a initial default list of TAs is established. Using a common TA list also maximizes the anonymity set size of trust information consumers. In addition to specify the trust anchor, consumers can configure a global and per-TA max_depth value. The per-TA value overrides the global value. The max_depth describes the longest path, measured in edge counts that a consumer is willing to accept. A consumer does not fetch and validate operator IDs where max_depth is already exceeded.
|-||use the global max_depth configuration|
|-1||accept arbitrarily long trust paths (no restrictions).|
|0||The TA is a trusted operator ID, no dynamic discovery of additional trusted operator IDs. No DNSSEC check is performed.|
|1||Discover and trust operator IDs from the locally configured TA(s) but do not attempt to discover and trust any further.|
|2||Discover and accept trust information with a max edge count of up to 2. This is the recommended default value. In the following trust path: A(TA) -> B -> C -> D the last accepted ID would be “C”. “D” is ignored.|
|N||Trust up to N edges in the trust path. Ignore operator IDs where the edge count exceeds this value.|
The max_depth value is placed after the the trust anchors host.
Example of a consumer configuration file (ta.conf) using 3 distinct TAs:
global_max_depth:0 example.com:2 example.net:1 example.org:-
In addition to max_depth a consumer might also be interested to require multiple independend trust paths to a single operator ID. We leave this to a future iterations of the protocol to keep it simple for now. It will not require any change on how trust information is published, should this be added in the future.
A consumer can also specify a list of domains that a consumer never wants to trust for anything (no transitive trust and no relay operator trust) to ensure that dynamic discovery will never result in any trust in the listed entities and others that are only discoverable via these domains. If a certain operator-id is published by a negative trust entry and by a trusted TA, the operator-id is trusted.
negative-trust.conf configuration contains one entry per line. Lines starting with “#” are comments and ignored during parsing.
Trust anchors and optionally also operators publish trusted relay operator IDs under this well-known URL:
The URL MUST be HTTPS and use a valid certificate from a generally trusted root CA. Plain HTTP MUST not be used. The URL MUST be accessible by robots (no CAPTCHAs or similar). The URL MUST NOT redirect to another host.
operator-ids.txt contains one operator ID (a domain) per line followed by a “:” and the recursion flag: a value of 0 or 1.
This recursion flag tells the consumer whether the given operator ID is also trusted to publish (1)
operator-ids.txt files itself or not (0).
The recursion flag should only be enabled when the operator is known to publish
operator-ids.txt and have a DNSSEC-signed domain, to avoid unnecessary polling.
Publishers MUST only publish operator IDs where they are confident that the operator does NOT operate relays with malicious intent. It is generally expected that a publisher “knows” the operators for which it vouches.
Lines starting with “#” are comments and ignored.
# this is a comment line example.com:1 example.net:0
In addition to serving the file content via HTTPS the content MUST be authenticated using the following DNS TXT record - which MUST be DNSSEC-signed:
operator-ids-hash._tor.example.com. TXT sha512=ee2cdf110893ad62b2aff24e668e4f8d2...
operator-ids.txt files that can not be verified using the hash in the DNS TXT record MUST be ignored.
The DNS TTL value for the TXT record should not exceed 60 seconds to be able to update the
file withouth running out of sync for an extended amount of time due to DNS caching.
The only supported hash algorithm is SHA512.
Trust information consumers perform the following steps to find and validate trust information.
For this example walkthrough, the local configuration on the trust information consumer contains a single trust anchor:
A trust information consumer performs these steps:
operator-ids-hash._tor.example.comand ensure it is DNSSEC-signed.
operator-ids.txtfile matches the one provided in the DNS TXT record (from step 2) the
Local caches at trust information consumers should not exceed 7 days. Re-validation should happen after 4 days and MUST not occur more than once a day.