tor/doc/spec/proposals/162-consensus-flavors.txt

Filename: 162-consensus-flavors.txt
Title: Publish the consensus in multiple flavors
Author: Nick Mathewson
Created: 14-May-2009
Target: 0.2.2
Status: Open


Overview:

   This proposal describes a way to publish each consensus in
   multiple simultaneous formats, or "flavors".  This will reduce the
   amount of time needed to deploy new consensus-like documents, and
   reduce the size of consensus documents in the long term.

Motivation:

   In the future, we will almost surely want different fields and
   data in the network-status document.  Examples include:
      - Publishing hashes of microdescriptors instead of hashes of
        full descriptors (Proposal 158).
      - Including different digests of descriptors, instead of the
        perhaps-soon-to-be-totally-broken SHA1.

   Note that in both cases, from the client's point of view, this
   information _replaces_ older information.  If we're using a
   SHA256 hash, we don't need to see the SHA1.  If clients only want
   microdescriptors, they don't (necessarily) need to see hashes of
   other things.

   Our past approach to cases like this has been to shovel all of
   the data into the consensus document.  But this is rather poor
   for bandwidth.  Adding a single SHA256 hash to a consensus for
   each router increases the compressed consensus size by 47%.  In
   comparison, replacing a single SHA1 hash with a SHA256 hash for
   each listed router increases the consensus size by only 18%.

Design in brief:

   Let the voting process will remain as it is, until a consensus is
   generated.  With future versions of the voting algorithm, instead
   of just a single consensus being generated, multiple consensus
   "flavors" are produced.

   Consensuses (all of them) include a list of which flavors are
   being generated.  Caches fetch and serve all flavors of consensus
   that are listed, regardless of whether they can parse or validate
   them, and serve them to clients.  Thus, once this design is in
   place, we won't need to deploy more cache changes in order to get
   new flavors of consensus to be cached.

   Clients download only the consensus flavor they want.

A note on hashes:

   Everything in this document is specified to use SHA256, and to be
   upgradeable to use better hashes in the future.

Spec modifications:

   1. URLs and changes to the current consensus format.

   Every consensus flavor has a name consisting of a sequence of one
   or more alphanumeric characters and dashes.  For compatibility
   current descriptor flavor is called "ns".

   The supported consensus flavors are defined as part of the
   authorities' consensus method.

   For each supported flavors, every authority calculates another
   consensus document of as-yet-unspecified format, and exchange
   detached signatures for these documents as in the current consensus
   design.

   In addition to the consensus currently served at
   /tor/status-vote/(current|next)/consensus.z , authorities serve
   another consensus of each flavor "F" from the location
   /tor/status-vote/(current|next)/F/consensus.z.

   When caches serve these documents, they do so from the same
   locations.

   2. Document format: generic consensus.

   The format of a flavored consensus is as-yet-unspecified, except
   that the first line is:
      "network-status-version" SP version SP flavor NL

   where version is 3 or higher, and the flavor is a string
   consisting of alphanumeric characters and dashes, matching the
   corresponding flavor listed in the unflavored consensus.

   3. Document format: detached signatures.

   In addition to the current detached signature format, we allow
   the first line to take the form,
      "consensus-digest" SP flavor SP 1*(Algname "=" Digest) NL

   The consensus-signatures URL should contain the signatures
   for _all_ flavors of consensus.

   4. The consensus index:

   Authorities additionally generate and serve a consensus-index
   document.  Its format is:

       Header ValidAfter ValidUntil Documents Signatures

       Header = "consensus-index" SP version NL
       ValidAfter = as in a consensus
       ValidUntil = as in a consensus
       Documents = Document*
       Document = "document" SP flavor SP SignedLength
                                    1*(SP AlgorithmName "=" Digest) NL
       Signatures = Signature *
       Signature = "directory-signature" SP algname SP identity
                           SP signing-key-digest NL signature

    There must be one Document line for each generated consensus flavor
    Each Document line describes the length of the signed portion of
    a consensus (the signatures themselves are not included), along
    with one or more digests of that signed portion.  Digests are
    given in hex.  The algorithm "sha256" MUST be included; others
    are allowed.

    The algname part of a signature describes what algorithm was
    used to hash the identity and signing keys, and to compute the
    signature.  The algorithm "sha256" MUST be recognized;
    signatures with unrecognized algorithms MUST be ignored.
    (See below).

    The consensus index is made available at
       /tor/status-vote/(current|next)/consensus-index.z.

    Caches should fetch this document so they can check the
    correctness of the different consensus documents they fetch.
    They do not need to check anything about an unrecognized
    consensus document beyond its digest.

    4.1. The "sha256" signature format.

    The 'SHA256' signature format for directory objects is defined as
    the RSA signature of the OAEP+-padded SHA256 digest of the SHA256
    digest of the the item to be signed.  When checking signatures,
    the signature MUST be treated as valid if the signed material
    begins with SHA256(SHA256(document)); this allows us to add other
    data later.

Considerations:

    - We should not create a new flavor of consensus when adding a
      field wouldn't be too onerous.

    - We should not proliferate flavors lightly: clients will be
      distinguishable based on which flavor they download.

Migration:

    - Stage one: authorities begin generating and serving
      consensus-index files.

    - Stage two: Caches begin downloading consenusus-index files,
      validating them, and using them to decide what flavors of
      consensus documents to cache.  They download all listed
      documents, and compare them to the digests given in the
      consensus.

    - Stage three: Once we want to make a significant change to the
      consensus format, we deploy another flavor of consensus at the
      authorities.  This will immediately start getting cached by the
      caches, and clients can start fetching the new flavor without
      waiting a version or two for enough caches to begin supporting
      it.

Acknowledgements:

    Aspects of this design and its applications to hash migration were
    heavily influenced by IRC conversations with Marian.