For sig types in the experimental range (65280-65534), floodfills should accept netdb stores without checking the signature.
This will support testing of new sig types.
The GOST proposal 134 has revealed two issues with the previously-unused experimental sig type range.
First, since sig types in the experimental range cannot be reserved, they may be used for multiple sig types at once.
Second, unless a router info or lease set with an experimental sig type can be stored at a floodfill, the new sig type is difficult to fully test or use on a trial basis.
Floodfills should accept, and flood, LS stores with sig types in the experimental range, without checking the signature. Support for RI stores is TBD, and may have more security implications.
For sig types in the experimental range, a floodfill should accept and flood netdb stores without checking the signature.
To prevent spoofing of non-experimental routers and destinations, a floodfill should never accept a store of an experimental sig type that has a hash collision with an existing netdb entry of a different sig type. This prevents hijacking of a previous netdb entry.
Additionally, a floodfill should overwrite an experimental netdb entry with a store of a non-experimental sig type that is a hash collision, to prevent hijacking of a previously-absent hash.
Floodfills should assume the signing public key length is 128, or derive it from the key certificate length, if longer. Some implementations may not support longer lengths unless the sig type is informally reserved.
Once this feature is supported, in a known router version, experimental sig type netdb entries may be stored to floodfills of that version or higher.
If some router implementations do not support this feature, the netdb store will fail, but that's the same as it is now.
There may be additional security implications, to be researched (see proposal 137)
Some implementations may not support key lengths greater than 128, as described above. Additionally, it may be necessary to enforce a maximum of 128 (in other words, there is no excess key data in the key cert), to reduce the ability of attackers to generate hash collisions.
Similar issues will need to be addressed with non-zero encryption types, which has not yet been formally proposed.
NetDB stores of unknown sig types that are not in the experimental range will continue to be rejected by floodfills, as the signature cannot be verified.
Proposal 134 Proposal 137