Filename: 220-ecc-id-keys.txt
Title: Migrate server identity keys to Ed25519
Authors: Nick Mathewson
Created: 12 August 2013
Implemented-In: 0.3.0.1-alpha
Status: Closed

   [Note: This is a draft proposal; I've probably made some important
   mistakes, and there are parts that need more thinking.  I'm
   publishing it now so that we can do the thinking together.]

   (Sections 0-5 are currently implemented, except for section 2.3.  Sections
   6-8 are a work in progress, and may require revision.)

0. Introduction

   In current Tor designs, router identity keys are limited to
   1024-bit RSA keys.

   Clearly, that should change, because RSA doesn't represent a good
   performance-security tradeoff nowadays, and because 1024-bit RSA is
   just plain too short.

   We've already got an improved circuit extension handshake protocol
   that uses curve25519 in place of RSA1024, and we're using (where
   supported) P256 ECDHE in our TLS handshakes, but there are more uses
   of RSA1024 to replace, including:

      * Router identity keys
      * TLS link keys
      * Hidden service keys

   This proposal describes how we'll migrate away from using 1024-bit
   RSA in the first two, since they're tightly coupled.  Hidden service
   crypto changes will be complex, and will merit their own proposal.

   In this proposal, we'll also (incidentally) be extirpating a number
   of SHA1 usages.

1. Overview

   When this proposal is implemented, every router will have an Ed25519
   identity key in addition to its current RSA1024 public key.

   Ed25519 (specifically, Ed25519-SHA-512 as described and specified at
   http://ed25519.cr.yp.to/) is a desirable choice here: it's secure,
   fast, has small keys and small signatures, is bulletproof in several
   important ways, and supports fast batch verification. (It isn't quite
   as fast as RSA1024 when it comes to public key operations, since RSA
   gets to take advantage of small exponents when generating public
   keys.)

   (For reference: In Ed25519 public keys are 32 bytes long, private keys
   are 64 bytes long, and signatures are 64 bytes long.)

   To mirror the way that authority identity keys work, we'll fully
   support keeping Ed25519 identity keys offline; they'll be used to
   sign long-ish term signing keys, which in turn will do all of the
   heavy lifting.  A signing key will get used to sign the things that
   RSA1024 identity keys currently sign.

1.1. 'Personalized' signatures

   Each of the keys introduced here is used to sign more than one kind
   of document. While these documents should be unambiguous, I'm going
   to forward-proof the signatures by specifying each signature to be
   generated, not on the document itself, but on the document prefixed
   with some distinguishing string.

2. Certificates and Router descriptors.

2.1. Certificates

   When generating a signing key, we also generate a certificate for it.
   Unlike the certificates for authorities' signing keys, these
   certificates need to be sent around frequently, in significant
   numbers.  So we'll choose a compact representation.

         VERSION         [1 Byte]
         CERT_TYPE       [1 Byte]
         EXPIRATION_DATE [4 Bytes]
         CERT_KEY_TYPE   [1 byte]
         CERTIFIED_KEY   [32 Bytes]
         N_EXTENSIONS    [1 byte]
         EXTENSIONS      [N_EXTENSIONS times]
         SIGNATURE       [64 Bytes]

   The "VERSION" field holds the value [01].  The "CERT_TYPE" field
   holds a value depending on the type of certificate. (See appendix
   A.1.) The CERTIFIED_KEY field is an Ed25519 public key if
   CERT_KEY_TYPE is [01], or a SHA256 hash of some other key type
   depending on the value of CERT_KEY_TYPE. The EXPIRATION_DATE is a
   date, given in HOURS since the epoch, after which this
   certificate isn't valid. (A four-byte field here will work fine
   until 10136 A.D.)

   The EXTENSIONS field contains zero or more extensions, each of
   the format:

         ExtLength [2 bytes]
         ExtType   [1 byte]
         ExtFlags  [1 byte]
         ExtData   [Length bytes]

   The meaning of the ExtData field in an extension is type-dependent.

   The ExtFlags field holds flags; this flag is currently defined:

      1 -- AFFECTS_VALIDATION. If this flag is present, then the
           extension affects whether the certificate is valid; clients
           must not accept the certificate as valid unless they
           understand the extension.

   It is an error for an extension to be truncated; such a
   certificate is invalid.

   Before processing any certificate, parties MUST know which
   identity key it is supposed to be signed by, and then check the
   signature.  The signature is formed by signing the first N-64
   bytes of the certificate prefixed with the string "Tor node
   signing key certificate v1".

2.2. Basic extensions

2.2.1. Signed-with-ed25519-key extension [type 04]

   In several places, it's desirable to bundle the key signing a
   certificate along with the certificate.  We do so with this
   extension.

        ExtLength = 32
        ExtData =
           An ed25519 key    [32 bytes]

   When this extension is present, it MUST match the key used to
   sign the certificate.

2.3. Revoking keys.

   We also specify a revocation document for revoking a signing key or an
   identity key.  Its format is:
         FIXED_PREFIX    [8 Bytes]
         VERSION         [1 Byte]
         KEYTYPE         [1 Byte]
         IDENTITY_KEY    [32 Bytes]
         REVOKED_KEY     [32 Bytes]
         PUBLISHED       [8 Bytes]
         N_EXTENSIONS    [1 Byte]
           N_EXTENSIONS_TIMES:
           EXTENSIONS      [N_EXTENSIONS times]
         SIGNATURE       [64 Bytes]

   FIXED_PREFIX is "REVOKEID" or "REVOKESK". VERSION is [01]. KEYTYPE is
   [01] for revoking a signing key, [02] for revoking an identity key,
   or [03] for revoking an RSA identity key.
   REVOKED_KEY is the key being revoked or a SHA256 hash of the key if
   it is an RSA identity key; IDENTITY_KEY is the node's
   Ed25519 identity key. PUBLISHED is the time that the document was
   generated, in seconds since the epoch. REV_EXTENSIONS is left for a
   future version of this document.  The SIGNATURE is generated with
   the same key as in IDENTITY_KEY, and covers the entire revocation,
   prefixed with "Tor key revocation v1".

   Using these revocation documents is left for a later specification.

2.4. Managing keys

   By default, we can keep the easy-to-setup key management properties
   that Tor has now, so that node operators aren't required to have
   offline public keys:

        * When a Tor node starts up with no Ed25519 identity keys, it
          generates a new identity keypair.
        * When a Tor node has an Ed25519 identity keypair, and it has
          no signing key, or its signing key is going to expire within
          the next 48 hours, it generates a new signing key to last
          30 days.

   But we also support offline identity keys:

        * When a Tor node starts with an Ed25519 public identity key
          but no private identity key, it checks whether it has
          a currently valid certified signing keypair.  If it does,
          it starts.  Otherwise, it refuses to start.
        * If a Tor node's signing key is going to expire soon, it starts
          warning the user.  If it is expired, then the node shuts down.

2.5. Router descriptors

   We specify the following element that may appear at most once in
   each router descriptor:
      "identity-ed25519" NL "-----BEGIN ED25519 CERT-----" NL certificate
           "-----END ED25519 CERT-----" NL

   The certificate is base64-encoded with
   terminating =s removed.  When this element is present, it MUST appear
   as the first or second element in the router descriptor.
   [XXX The rationale here is to allow extracting the identity key and
   signing key and checking the signature before fully parsing the rest
   of the document. -NM]

   The certificate has CERT_TYPE of [04].  It must include a
   signed-with-ed25519-key extension (see section 2.2.1), so that we
   can extract the identity key.

   When an identity-ed25519 element is present, there must also be a
   "router-sig-ed25519" element.  It MUST be the next-to-last element in
   the descriptor, appearing immediately before the RSA signature.  (In
   future versions of the descriptor format that do not require an RSA
   identity key, it MUST be last.)  It MUST contain an ed25519 signature
   of a SHA256 digest of the entire document, from the first character
   up to and including the first space after the "router-sig-ed25519"
   string, prefixed
   with the string "Tor router descriptor signature v1".  Its format is:

      "router-sig-ed25519" SP signature NL

   Where 'signature' is encoded in base64 with terminating =s removed.

   The signing key in the certificate MUST
   be the one used to sign the document.

   Note that these keys cross-certify as follows: the ed25519 identity
   key signs the ed25519 signing key in the certificate.  The ed25519
   signing key signs itself and the ed25519 identity key and the RSA
   identity key as part of signing the descriptor.  And the RSA identity
   key also signs all three keys as part of signing the descriptor.


   When an ed25519 signature is present, there MAY be a "master-key-ed25519"
   element containing the base64 encoded ed25519 master key as a single
   argument.  If it is present, it MUST match the identity key in
   the certificate.

2.5.1. Checking descriptor signatures.

   Current versions of Tor will handle these new formats by ignoring the
   new fields, and not checking any ed25519 information.

   New versions of Tor will have a flag that tells them whether to check
   ed25519 information.  When it is set, they must check:

      * All RSA information and signatures that Tor implementations
        currently check.
      * If the identity-ed25519 line is present, it must be well-formed,
        and the certificate must be well-formed and correctly signed,
        and there must be a valid router-signature-ed25519 signature.
      * If we require an ed25519 key for this node (see 3.1 below), the
        ed25519 key must be present.

   Authorities and directory caches will have this flag always-on.  For
   clients, it will be controlled by a torrc option and consensus
   option, to be set to "always-on" in the future once enough clients
   support it.

2.5.2. Extra-info documents

   Extra-info documents now include "identity-ed25519" and
   "router-signature-ed25519" fields in the same positions in which they
   appear in router descriptors.

   Additionally, we add the base64-encoded, =-stripped SHA256 digest of
   a node's extra-info document field to the extra-info-digest line in
   the router descriptor. (All versions of Tor that recognize this line
   allow an extra field there.)

2.5.3. A note on signature verification

   Here and elsewhere, we're receiving a certificate and a document
   signed with the key certified by that certificate in the same step.
   This is a fine time to use the batch signature checking capability of
   Ed25519, so that we can check both signatures at once without (much)
   additional overhead over checking a single signature.

3. Consensus documents and authority operation

3.1. Handling router identity at the authority

   When receiving router descriptors, authorities must track mappings
   between RSA and Ed25519 keys.

   Rule 1: Once an authority has seen an Ed25519 identity key and an RSA
   identity key together on the same (valid) descriptor, it should no
   longer accept any descriptor signed by that RSA key with a different
   Ed25519 key, or that Ed25519 key with a different RSA key.

   Rule 2: Once an authority has seen an Ed25519 identity key and an RSA
   identity key on the same descriptor, it should no longer accept any
   descriptor signed by that RSA key unless it also has that Ed25519
   key.


   These rules together should enforce the property that, even if an
   attacker manages to steal or factor a node's RSA identity key, the
   attacker can't impersonate that node to the authorities, even when
   that node is identified by its RSA key.


   Enforcement of Rule 1 should be advisory-only for a little while (a
   release or two) while node operators get experience having Ed25519
   keys, in case there are any bugs that cause or force identity key
   replacement.  Enforcement of Rule 2 should be advisory-only for
   little while, so that node operators can try 0.2.5 but downgrade to
   0.2.4 without being de-listed from the consensus.


3.2. Formats

   Vote and microdescriptor documents now contain an optional "id"
   field for each routerstatus section.  Its format is:

       "id" SP "ed25519" SP ed25519-identity NL

   where ed25519-identity is base64-encoded, with trailing = characters
   omitted.  In vote documents, it may be replaced by the format:

       "id" SP "ed25519" SP "none" NL

   which indicates that the node does not have an ed25519 identity.  (In
   a microdescriptor, a lack of "id" line means that the node has no ed25519
   identity.)

   A vote or consensus document is ill-formed if it includes the same
   ed25519 identity key twice.

   A vote listing ed25519 identities must also include a new entry in its
   "r" lines, containing a base64-encoded SHA256 digest of the entire
   descriptor (including signature).  This kills off another place where
   we rely on sha1.  The format for 'r' lines is now:

    "r" SP nickname SP identity SP digest SP publication SP IP SP ORPort
        SP DirPort [ SP digest-sha256 ] NL

3.3. Generating votes

   An authority should pick which descriptor to choose for a node as
   before, and include the ed25519 identity key for the descriptor if
   it's present.

   As a transition, before Rule 1 and Rule 2 in 3.1 are fully enforced,
   authorities need a way to deal with the possibility that there might
   be two nodes with the same ed25519 key but different RSA keys.  In
   that case, it votes for the one with the most recent publication
   date.

   (The existing rules already prevent an authority from voting for two
   servers with the same RSA identity key.)

3.4. Generating a consensus from votes

   This proposal requires a new consensus vote method.  When we deploy
   it, we'll pick the next available vote method in sequence to use for
   this.

   When the new consensus method is in use, we must choose nodes first by ECC
   key, then by RSA key.  [This procedure is analogous to the current one,
   except that it is aware of multiple kinds of keys.]

3.4.1. Notation for voting

  We have a set of votes.  Each contains either 'old tuples' or 'new tuples'.

  Old tuples are:
    <id-RSA, descriptor-digest, published, nickname, IP, ports>

  New tuples are:
    <id-Ed, id-RSA, descriptor-digest, dd256, published, nickname, IP, ports>


3.4.2. Validating votes

  It is an error for a vote to have the same id-RSA or the same id-Ed listed
  twice.  Throw it away if it does.

3.4.3. Decide which ids to include.

  For each <id-Ed, id-RSA> that is listed by more than half of the total
    authorities (not just total votes), include it.  (No other <id-Ed, id-RSA'>
    can have as many votes.)

  Log any other id-RSA values corresponding to an id-Ed we included, and any
    other id-Ed values corresponding to an id-RSA we included.

  For each <id-RSA> that is not yet included, if it is listed by more than
    half of the total authorities, and we do not already have it listed with
    some <id-Ed>, include it without an id-Ed.

3.4.4. Decide which descriptors to include.

   A tuple belongs to an <id-RSA, id-Ed> identity if it is a new tuple that
   matches both ID parts, or if it is an old tuple that matches the RSA part.
   A tuple belongs to an <id-RSA> identity if its RSA identity matches.

   A tuple matches another tuple if all the fields that are present in both
   tuples are the same.

   For every included identity, consider the tuples belonging to that
   identity.  Group them into sets of matching tuples.  Include the tuple
   that matches the largest set, breaking ties in favor of the most recently
   published, and then in favor of the smaller server descriptor digest.

4. The link protocol

4.1. Overview of the status quo

   This section won't make much sense unless you grok the v3
   link protocol as described in tor-spec.txt, first proposed in
   proposal 195. So let's review.

   In the v3 link protocol, the client completes a TLS handshake
   with the server, in which the server uses an arbitrary
   certificate signed with an RSA key.  The client then sends a
   VERSIONS cell.  The server replies with a VERSIONS cell to
   negotiate version 3 or higher.  The server also sends a CERTS
   cell and an AUTH_CHALLENGE cell and a NETINFO cell.

   The CERTS cell from the server contains a set of one or more
   certificates that authenticate the RSA key used in the TLS
   handshake.  (Right now there's one self-signed RSA identity key
   certificate, and one certificate signing the RSA link key with
   the identity key.  These certificates are X509.)

   Having received a CERTS cell, the client has enough information
   to authenticate the server.  At this point, the client may send a
   NETINFO cell to finish the handshake.  But if the client wants to
   authenticate as well, it can send a CERTS cell and an AUTENTICATE
   cell.

   The client's CERTS cell also contains certs of the same general
   kinds as the server's key file: a self-signed identity
   certificate, and an authentication certificate signed with the
   identity key.  The AUTHENTICATE cell contains a signature of
   various fields, including the contents of the AUTH_CHALLENGE
   which the server sent, using the client's authentication
   key.  These cells allow the client to authenticate to the server.


4.2. Link protocol changes for ECC ID keys

   We add four new CertType values for use in CERTS cells:
        4: Ed25519 signing key
        5: Link key certificate certified by Ed25519 signing key
        6: Ed25519 TLS authentication key certified by Ed25519 signing key
        7: RSA cross-certificate for Ed25519 identity key
   These correspond to types used in the CERT_TYPE field of
   the certificates.

   The content of certificate type [04] (Ed25519 signing key)
   is as in section 2.5 above, containing an identity key and the
   signing key, both signed by the identity key.

   Certificate type [05] (Link certificate signed with Ed25519
   signing key) contains a SHA256 digest of the X.509 link
   certificate used on the TLS connection in its key field; it is
   signed with the signing key.

   Certificate type [06] (Ed25519 TLS authentication signed with
   Ed25519 signing key) has the signing key used to sign the
   AUTHENTICATE cell described later in this section.

   Certificate type [07] (Cross-certification of Ed25519 identity
   with RSA key) contains the following data:
       ED25519_KEY                       [32 bytes]
       EXPIRATION_DATE                   [4 bytes]
       SIGLEN                            [1 byte]
       SIGNATURE                         [SIGLEN bytes]
   Here, the Ed25519 identity key is signed with router's RSA
   identity key, to indicate that authenticating with a key
   certified by the Ed25519 key counts as certifying with RSA
   identity key.  (The signature is computed on the SHA256 hash of
   the non-signature parts of the certificate, prefixed with the
   string "Tor TLS RSA/Ed25519 cross-certificate".)

   (There's no reason to have a corresponding Ed25519-signed-RSA-key
   certificate here, since we do not treat authenticating with an RSA
   key as proving ownership of the Ed25519 identity.)

   Relays with Ed25519 keys should always send these certificate types
   in addition to their other certificate types.

   Non-bridge relays with Ed25519 keys should generate TLS link keys of
   appropriate strength, so that the certificate chain from the Ed25519
   key to the link key is strong enough.


   We add a new authentication type for AUTHENTICATE cells:
   "Ed25519-TLSSecret", with AuthType value 2. Its format is the same as
   "RSA-SHA256-TLSSecret", except that the CID and SID fields support
   more key types; some strings are different, and the signature is
   performed with Ed25519 using the authentication key from a type-6
   cert.  Clients can send this AUTHENTICATE type if the server
   lists it in its AUTH_CHALLENGE cell.

   Modified values and new fields below are marked with asterisks.

       TYPE: The characters "AUTH0002"* [8 octets]
       CID: A SHA256 hash of the initiator's RSA1024 identity key [32 octets]
       SID: A SHA256 hash of the responder's RSA1024 identity key [32 octets]
       *CID_ED: The initiator's Ed25519 identity key [32 octets]
       *SID_ED: The responder's Ed25519 identity key, or all-zero. [32 octets]
       SLOG: A SHA256 hash of all bytes sent from the responder to the
         initiator as part of the negotiation up to and including the
         AUTH_CHALLENGE cell; that is, the VERSIONS cell, the CERTS cell,
         the AUTH_CHALLENGE cell, and any padding cells.  [32 octets]
       CLOG: A SHA256 hash of all bytes sent from the initiator to the
         responder as part of the negotiation so far; that is, the
         VERSIONS cell and the CERTS cell and any padding cells. [32
         octets]
       SCERT: A SHA256 hash of the responder's TLS link certificate. [32
         octets]
       TLSSECRETS: A SHA256 HMAC, using the TLS master secret as the
         secret key, of the following:
           - client_random, as sent in the TLS Client Hello
           - server_random, as sent in the TLS Server Hello
           - the NUL terminated ASCII string:
             "Tor V3 handshake TLS cross-certification with Ed25519"*
          [32 octets]
       RAND: A 24 byte value, randomly chosen by the initiator. [24 octets]
       *SIG: A signature of all previous fields using the initiator's
          Ed25519 authentication flags.
          [variable length]

   If you've got a consensus that lists an ECC key for a node, but the
   node doesn't give you an ECC key, then refuse this connection.

5. The extend protocol

   We add a new NSPEC node specifier for use in EXTEND2 cells, with
   LSTYPE value [03].  Its length must be 32 bytes; its content is the
   Ed25519 identity key of the target node.

   Clients should use this type only when:
     * They know an Ed25519 identity key for the destination node.
     * The source node supports EXTEND2 cells
     * A torrc option is set, _or_ a consensus value is set.

   We'll leave the consensus value off for a while until more clients
   support this, and then turn it on.

   When picking a channel for a circuit, if this NSPEC value is
   provided, then the RSA identity *and* the Ed25519 identity must
   match.

   If we have a channel with a given Ed25519 ID and RSA identity, and we
   have a request for that Ed25519 ID and a different RSA identity, we
   do not attempt to make another connection: we just fail and DESTROY
   the circuit.

   If we receive an EXTEND or EXTEND2 request for a node listed in the
   consensus, but that EXTEND/EXTEND2 request does not include an
   Ed25519 identity key, the node SHOULD treat the connection as failed
   if the Ed25519 identity key it receives does not match the one in the
   consensus.

   For testing, clients may have the ability to configure whether to
   include Ed25519 identities in EXTEND2 cells.  By default, this should
   be governed by the boolean "ExtendByEd25519ID" consensus parameter,
   with default value '0'.

6. Naming nodes in the interface

   Anywhere in the interface that takes an $identity should be able to
   take an ECC identity too.  ECC identities are case-sensitive base64
   encodings of Ed25519 identity keys. You can use $ to indicate them as
   well; we distinguish RSA identity digests by length.

   When we need to indicate an Ed25519 identity key in a hostname
   format (as in a .exit address), we use the lowercased version of the
   name, and perform a case-insensitive match.  (This loses us a little
   less than one bit per byte of name, leaving plenty of bits to make
   sure we choose the right node.)

   Nodes must not list Ed25519 identities in their family lines; clients and
   authorities must not honor them there.  (Doing so would make different
   clients change paths differently in a possibly manipulatable way.)

   Clients shouldn't accept .exit addresses with Ed25519 names on SOCKS
   or DNS ports by default, even when AllowDotExit is set.  We can add
   another option for them later if there's a good reason to have this.

   We need an identity-to-node map for ECC identity and for RSA
   identity.

   The controller interface will need to accept and report Ed25519
   identity keys as well as (or instead of) RSA identity keys.  That's a
   separate proposal, though.

7. Hidden service changes out of scope

   Hidden services need to be able to identify nodes by ECC keys, just as
   they will need to include ntor keys as well as TAP keys.  Not just
   yet though.  This needs to be part of a bigger hidden service
   revamping strategy.

8. Proposed migration steps

   Once a few versions have shipped with Ed25519 key support, turn on
   "Rule 1" on the authorities.  (Don't allow an Ed25519<->RSA pairing
   to change.)

   Once the release with these changes is in beta or rc, turn on the
   consensus option for everyone who receives descriptors with
   Ed25519 identity keys to check them.

   Once the release with these changes is in beta or rc, turn on the
   consensus option for clients to generate EXTEND2 requests with
   Ed25519 identity keys.

   Once the release with these changes has been stable for a month
   or two, turn on "Rule 2" on authorities.  (Don't allow nodes that
   have advertised an Ed25519 key to stop.)

9. Future proposals

   * Ed25519 identity support on the controller interface
   * Supporting nodes without RSA keys
   * Remove support for nodes without Ed25519 keys
   * Ed25519 support for hidden services
   * Bridge identity support.
   * Ed25519-aware family support

A.1. List of certificate types

   The values marked with asterisks are not types corresponding to
   the certificate format of section 2.1.  Instead, they are
   reserved for RSA-signed certificates to avoid conflicts between
   the certificate type enumeration of the CERTS cell and the
   certificate type enumeration of in our Ed25519 certificates.


   **[00],[01],[02],[03] - Reserved to avoid conflict with types used
          in CERTS cells.

   [04] - signing a signing key with an identity key (Section 2.5)

   [05] - TLS link certificate signed with ed25519 signing key
         (Section 4.2)

   [06] - Ed25519 authentication key signed with ed25519 signing key
          (Section 4.2)

   **[07] - reserved for RSA identity cross-certification (Section 4.2)


A.2. List of extension types


   [01] - signed-with-ed25519-key (section 2.2.1)

A.3. List of signature prefixes

   We describe various documents as being signed with a prefix. Here
   are those prefixes:

      "Tor router descriptor signature v1" (section 2.5)
      "Tor node signing key certificate v1" (section 2.1)

A.4. List of certified key types

   [01] ed25519 key
   [02] SHA256 hash of an RSA key
   [03] SHA256 hash of an X.509 certificate

A.5. Reserved numbers

   We need a new consensus algorithm number to encompass checking
   ed25519 keys and putting them in microdescriptors.


   We need new CertType values for use in CERTS cells.  We reserved
   in section 4.2.

        4: Ed25519 signing key
        5: Link key certificate certified by Ed25519 signing key
        6: TLS authentication key certified by Ed25519 signing key
        7: RSA cross-certificate for Ed25519 identity key


A.6. Related changes

   As we merge this, proposal, we should also extend link key size to
   2048 bits, and use SHA256 as the x509 cert algorithm for our link
   keys. This will improve link security, and deliver better
   fingerprinting resistence.  See proposal 179 for an older discussion
   of this issue.