EIP-1459: Node Discovery via DNS
Scheme for authenticated updateable Ethereum node lists via DNS.
Abstract
This document describes a scheme for authenticated, updateable Ethereum node lists retrievable via DNS.
Motivation
Many Ethereum clients contain hard-coded bootstrap node lists. Updating those lists requires a software update. The current lists are small, giving the client little choice of initial entry point into the Ethereum network. We would like to maintain larger node lists containing hundreds of nodes, and update them regularly.
The scheme described here is a replacement for client bootstrap node lists with equivalent security and many additional benefits. Large lists populated by traversing the node discovery DHT can serve as a fallback option for nodes which can't join the DHT due to restrictive network policy. DNS-based node lists may also be useful to Ethereum peering providers because their customers can configure the client to use the provider's list.
Specification
A 'node list' is a list of 'node records' as defined by EIP-778 of arbitrary length. Lists may refer to other lists using links. The entire list is signed using a secp256k1 private key. The corresponding public key must be known to the client in order to verify the list.
To refer to a DNS node list, clients use a URL with 'enrtree' scheme. The URL contains the DNS name on which the list can be found as well as the public key that signed the list. The public key is contained in the username part of the URL and is the base32 encoding (RFC-4648) of the compressed 32-byte binary public key.
Example:
This URL refers to a node list at the DNS name 'nodes.example.org' and is signed
by the public key
0x049f88229042fef9200246f49f94d9b77c4e954721442714e85850cb6d9e5daf2d880ea0e53cb3ac1a75f9923c2726a4f941f7d326781baa6380754a360de5c2b6
DNS Record Structure
The nodes in a list are encoded as a merkle tree for distribution via the DNS protocol. Entries of the merkle tree are contained in DNS TXT records. The root of the tree is a TXT record with the following content:
where
enr-root
andlink-root
refer to the root hashes of subtrees containing nodes and links subtrees.sequence-number
is the tree's update sequence number, a decimal integer.signature
is a 65-byte secp256k1 EC signature over the keccak256 hash of the record content, excluding thesig=
part, encoded as URL-safe base64 (RFC-4648).
Further TXT records on subdomains map hashes to one of three entry types. The subdomain name of any entry is the base32 encoding of the (abbreviated) keccak256 hash of its text content.
enrtree-branch:<h₁>,<h₂>,...,<hₙ>
is an intermediate tree entry containing hashes of subtree entries.enrtree://<key>@<fqdn>
is a leaf pointing to a different list located at another fully qualified domain name. Note that this format matches the URL encoding. This type of entry may only appear in the subtree pointed to bylink-root
.enr:<node-record>
is a leaf containing a node record. The node record is encoded as a URL-safe base64 string. Note that this type of entry matches the canonical ENR text encoding. It may only appear in theenr-root
subtree.
No particular ordering or structure is defined for the tree. Whenever the tree
is updated, its sequence number should increase. The content of any TXT record
should be small enough to fit into the 512 byte limit imposed on UDP DNS
packets. This limits the number of hashes that can be placed into an
enrtree-branch
entry.
Example in zone file format:
Client Protocol
To find nodes at a given DNS name, say "mynodes.org":
- Resolve the TXT record of the name and check whether it contains a valid
"enrtree-root=v1" entry. Let's say the
enr-root
hash contained in the entry is "CFZUWDU7JNQR4VTCZVOJZ5ROV4". - Verify the signature on the root against the known public key and check whether the sequence number is larger than or equal to any previous number seen for that name.
- Resolve the TXT record of the hash subdomain, e.g. "CFZUWDU7JNQR4VTCZVOJZ5ROV4.mynodes.org" and verify whether the content matches the hash.
- The next step depends on the entry type found:
- for
enrtree-branch
: parse the list of hashes and continue resolving them (step 3). - for
enr
: decode, verify the node record and import it to local node storage.
- for
During traversal, the client must track hashes and domains which are already resolved to avoid going into an infinite loop. It's in the client's best interest to traverse the tree in random order.
Client implementations should avoid downloading the entire tree at once during normal operation. It's much better to request entries via DNS when-needed, i.e. at the time when the client is looking for peers.
Rationale
Why DNS?
We have chosen DNS as the distribution medium because it is always available, even under restrictive network conditions. The protocol provides low latency and answers to DNS queries can be cached by intermediate resolvers. No custom server software is needed. Node lists can be deployed to any DNS provider such as CloudFlare DNS, dnsimple, Amazon Route 53 using their respective client libraries.
Why is this a merkle tree?
Being a merkle tree, any node list can be authenticated by a single signature on the root. Hash subdomains protect the integrity of the list. At worst intermediate resolvers can block access to the list or disallow updates to it, but cannot corrupt its content. The sequence number prevents replacing the root with an older version.
Synchronizing updates on the client side can be done incrementally, which matters for large lists. Individual entries of the tree are small enough to fit into a single UDP packet, ensuring compatibility with environments where only basic UDP DNS is available. The tree format also works well with caching resolvers: only the root of the tree needs a short TTL. Intermediate entries and leaves can be cached for days.
Why does the link subtree exist?
Links between lists enable federation and web-of-trust functionality. The operator of a large list can delegate maintenance to other list providers. If two node lists link to each other, users can use either list and get nodes from both.
The link subtree is separate from the tree containing ENRs. This is done to enable client implementations to sync these trees independently. A client wanting to get as many nodes as possible will sync the link tree first and add all linked names to the sync horizon.
Security Considerations
Discovery via DNS is less secure than via DHT, because it relies on a trusted party to publish the records regularly. The actor could easily eclipse bootstrapping nodes by only publishing node records that it controls.
Copyright
Copyright and related rights waived via CC0.