SILC Protocol
=============

Possible SILC protocol and specification document changes.  All of these
are tentative and doesn't mean that any of them would be done at any
point.

 o Full rework of the documents as requested by RFC Editor.  The plan
   is to create only two documents:

   silc-architecture-xx.txt
   silc-specification-xx.txt

 o Make @ reserved character in channel names.  Accept channel@server
   names in all commands and notify types.

 o Add acknowlegments section to specification documents.

 o Group Diffie-Hellman protocol for establishig key with two or more
   users on a channel.

 o Change CTR mode description:

    Truncated HASH from SKE (4 bytes) - This value is the first 4
    bytes from the HASH value that was computed as a result of SKE
    protocol.  This acts as session identifier and each rekey MUST
    produce a new HASH value.

   to

    Truncated HASH from SKE (4 bytes) - This value is the first 4
    bytes from the HASH value that was computed in SKE.  In each rekey
    the value MUST be recomputed as follows:

      HASH = hash(new Sending/Receiving IV from SKE)

    The hash function is the one used in SKE.  The 'new Sending/Receiving
    IV from SKE' is the first 8 bytes of the new value computed during
    rekey.  The first 4 bytes are used from the recomputed HASH.

 o Consider for future authenticated encryption modes, especially GCM.

 o Extend the Channel ID port to be actually a counter, allowing the
   2^32 channels per cell, instead of 2^16 like now.  The port with
   compliant implementation would always be 706, and it could be used
   as a counter, starting from 706... For interop, with old code.

 o In SKE with UDP/IP responder doesn't have to do retransmissions.
   Initiator will retransmit its packet.  Initiator can be considered
   the one that actually WANTs to establish the keys.  So no need for
   responder to retransmit.  Define this clearly in the specs.

 o Define clearly that the DSS signature format is the the Dss-Sig-Value
   ASN.1 encoding defined for PKIX.

 o Define clearly the SSH2 signature format is the one specified for SSH2
   protocol.

 o Dynamic server and router connections, ala Jabber.  SILC has allowed
   this from the beginning.  It should be written out clearly in the
   specs.  Connection would be created with nick strings (which are of
   format nick@server).

 o NAT detection protocool during SKE so that party behind NAT can
   detect if it is behind NAT and receive the public IP address and port
   that it may need (servers need it to create valid Server ID).  (***DONE)

 o Counter block send/receive IV 64 bits instead of 32 bits, and the
   value itself is used as 64-bit MSB ordered counter, which must
   be reset before the packet sequence counter wraps.  It's basically
   a counter which is initially set to a random value. (***DONE)

 o Nickname to NEW_CLIENT packet. (***DONE)

 o Add Source and Destination ID in message MAC computation to fully
   associate the Message Payload with the true sender and the true
   recipient of the message.  This will fix some security issues that
   currently exists.  It is currently possible in some specific set of
   conditions to mount a replay attack using Message Payload.  This change
   will remove the possibility of these attacks.

   After including Source and Destination ID in message MAC, ONLY replay
   attack possible is the following and with ONLY following conditions:

   1. the attacker is able to record encrypted Message Payloads and has
      the ability to replay them.
   2. the message payload is encrypted with static private message key
   3. the original sender of the message is not anymore in the network,
      has changed nickname, has detached and resumed, or has reconnected
      to other server.
   4. the original receiver of the message is still in the network, has
      not changed nickname, has not detached and resumed, and has not
      reconnected to any other server, or, some other user has the same
      client ID.
   5. the attacker is able to get the same client ID as the original
      sender.
   6. the original receiver still has the static key set for the same
      remote client ID (for original sender's client ID).

   All this is possible to happen though likelyhood is quite small.  It
   does illustrate how inappropriate the use of static keys is. (***DONE)

 o The SILC public key identifier separator is ', ' not ','.  The
   whitespace is mandatory. (***DONE)

 o Definition of EAP as new authentication method for connection auth
   protocol (RFC 3748).

 o Count limit to LIST command?

 o Strict announces if Channel ID is different than on router?  To not
   allow any modes, topic, etc changes from server if the ID was wrong
   initially?  Meaning: riding with netsplits not possible since the
   channel created during split will not enforce is modes to the
   router.  Or more liberal solution, like now?  Read emails on
   silc-users.  (This is very old issue)

 o The time values in STATS is 32-bits.  After 2038 it's over 32-bits.