CDR design document

[Adam Back <aba@dcs.ex.ac.uk>]

Here is a proposal for consideration for inclusion in the OpenPGP
standard:

	http://www.dcs.ex.ac.uk/~aba/cdr/

(Also attached text version below [1])

It describes a mechanism for commercial data recovery which attempts
to minimise the security risks inherent in allowing disaster recovery
of stored encrypted documents.  The CDR design also avoids the
security risks in building systems which send recovery information
over open communications networks, as the CMR proposal currently does.

The CDR design offers the option of more ergonomic recovery from
forgotten passphrase.  CDR allows for more secure use to be made of
the existing pgp5.0 key expiry functionality when applied to
communications-only keys because they can be deleted with out loss of
data-availability, providing a form of PFS which is backwards
compatible with pgp5.0 and pgp5.5 clients.

Comments?

The only change addition to the standard required is to create a
distinction between a storage only key and a communications only key.
This distinction could be made via the use of an extra flag, ideally
so that keyrings can remain standardised and interchangeable between
different implementations.

The document can also be viewed as a recommendation to implementors of
how to avoid security risks (and related government abuse risks) in
implementing data recovery systems.

I would like to see the OpenPGP standard accomodate both the CMR and
CDR design so that each can fly on it's technical merits.  (And
political merits, cryptography being a highly political field, a
politically unpopular design may suffer for other than technical
reasons).

Adam


[1]	http://www.dcs.ex.ac.uk/~aba/cdr/
==============================8<==============================
Commercial Data Recovery

This document describes a mechanism, Commercial Data Recovery (CDR), to
allow companies to recover stored encrypted information as part of a
disaster recovery plan.

Introduction

This document is centered around the OpenPGP standard, and therefore
explores first what is possible within the existing standard. Several of the
CDR variants presented can be achieved entirely within the pgp5.0 standard
without adding any new functionality.

Commercial disaster recovery requirement

PGP is commonly used for two purposes: file encryption, and encryption of
email communications. The situation arises in a commercial setting that
individuals are protecting information which the company would like to be
able to recover in the event of disaster: in the event of the user
forgetting his passphrase, or in the event of the unexpected death of the
employee. Stored files which the company may require the ability to recover
include encrypted files stored in the files system, on backup tapes, or on
floppy disks; and encrypted sent and recieved emails stored in mail client
folders.

Security vs Availability

There is a security risk in adding data recovery mechanisms: the additional
risk that recovery procedure is used by other than the owners of the data to
recover plaintext.

A central goal of the CDR design described in this document is to minimise
this security risk. Keys used to encrypt email which is transmitted over the
Internet are more valuable to an attacker than keys used to encrypt stored
files because of the relative ease with which an attacker can obtain copies
of emailed ciphertext. Stored encrypted files in contrast are protected by
all the physical security systems the company is relying on to protect it's
paper files, plaintext data stored on disks, and backup tapes.

For this reason the CDR approach treats communications keys as more
sensitive than keys used to encrypt locally stored information.

Separate Signature and Encryption Keys added in pgp5.0

This section describes a feature added to PGP with the pgp5.0 release:
separate signature and encryption keys. Readers familiar with pgp5.0
functionality may like to skip this section.

pgp2.x RSA keys

pgp2.x (and pgp1.x) only had direct support for one type of public key: the
RSA public key. RSA public keys were used both as signature keys, and as
encryption keys to transfer session keys. The ability to generate RSA keys,
and to interoperate with pgp2.x is retained in pgp5.x.

pgp5.x El Gamal (DH) encryption and DSA signature keys

One reason for the introduction of separate key functionalities is the
addition, starting with pgp5.0, of the DSA digital signature algorithm which
can only function as a signature system. In addition the El Gamal (EG)
public key system which was also added with pgp5.0 is used only as a public
key encryption system. (There are some El Gamal variants which can be used
to produce digital signatures, but these are not used in pgp5.0). (El Gamal
is a Diffie-Hellman (DH) variant which is sometimes also referred to
confusingly as DH.) The DSA is itself a DH signature variant.

Patent issues

Another significant reason for the adoption of EG and DSA in particular is
the freedom from patent encumbrances these algorithms now enjoy. RSA by
contrast is patented, and unpatented algorithms are generally preferred for
internet standards.

Key expiry feature added in pgp5.0

One general security principle is that keys used to protect communications
should be replaced periodically to reduce the value to an attacker of an
individual key.

Migration path

The pgp5.0 standard provides support for this functionality in terms of the
ability to attach expiry dates to signature and encryption keys. (In the
pgp5.0 and pgp5.5 clients the signature and encryption keys are treated as a
pair, and are given the same expiry information, but both the pgp5.0 and
pgp5.5 clients can already cope with signature-only keys and encryption-only
keys with different expiry information.)

Web of Trust

In pgp5.0 with the EG public keys, the Web of Trust (WoT) is based on the
signature keys only. The users signature key is authenticated in the web of
trust. The user signs his own encryption key thereby transferring trust to
it. This ensures that encryption keys can be expired and replaced without
affecting the web of trust.

Limited Perfect Forward Secrecy

Perfect Forward Secrecy (PFS) refers the technique of destroying old
communications keys to ensure that past traffic can not be compromised after
the point of re-keying. One way to use the pgp5.0 key expiry facilities is
to have signature and encryption keys with very different expiry periods.
For example a signature key could be given no expiry period (the pgp5.0
standard supports expiry periods of `forever'), and encryption keys could be
given an expiry period of 6 months.

To obtain PFS, a more secure way of dealing with expired encryption keys is
to actually irretrievably delete them. This then simulates PFS: an attacker
is unable to decrypt traffic encrypted with old communications keys even if
he is able to coerce co-operation from the recipient. No one can be coerced
because the key has been deleted.

It seems reasonable that some users may want to use key expiry for this
purpose, and this does not require any modifications to the existing pgp5.0
standard.

Key expiry problems for stored encrypted data

One problem introduced by key expiry is that any stored data encrypted to
the expired key becomes less available. For security reasons the expired key
should ideally be deleted, and in this case access to the data is lost.
Alternatives are to attempt to store the old keys more securely, this then
presents the problem that some stored data is harder to access than other
data, and presents additional key management problems for old keys which
will be confusing for users.

Re-encrypting after key expiry

One method to solve this problem is to re-encrypt all of the encrypted files
at the point of re-keying. However this method has disadvantages in that not
all of the encrypted data may be easily available. Some of the encrypted
data may be on backup tapes, on floppy disks, or stored inside `zip' or
`tar' archives on the disk and hence may be missed.

Separate storage keys

The simplest solution to this problem seems to be to encrypt to a separate
storage key. Within the pgp5.0 standard this storage key would be a separate
El Gamal key with a long expiry date. With this system we then have three
types of keys: signature-only keys, communications-only keys and
storage-only keys. This could require minimal or even no modifications to
the pgp5.0 standard, storage and encryption keys could be differentiated
between by:

   * location storage-only key is stored in (for example a separate keyring)
   * application being configured to use a chosen encryption key as the
     storage-only key by giving a keyID
   * convention on expiry dates (expiry date `forever' could be storage key,
     other values could be communications keys)
   * extra flag to indicate a storage-only key

Once the separation of key types is made, the security benefits of re-keying
can be more fully obtained.

Sent and recieved email folders

To maintain availability after re-keying of encrypted sent and received
emails archived in mail folders, the email plaintext would be encrypted with
the storage-only key. This then means that data availability for encrypted
files on disk, floppy disk, and backup tape is maintained; and that
availability for archived sent and received emails is maintained.

Storage Recovery

If the system is being used in a commercial setting, disaster recovery can
be provided by storing recovery information to allow recovery of the
storage-only key. There is still some additional danger over the case where
there is no recovery procedure: the risk that an attacker may be able to
compromise the recovery information. However because the encrypted data is
all protected by the companies existing physical security mechanisms the
security risk is minimised.

Another form of storage recovery would be to include recovery information
with each file, and this would be a valid combination achieved by placing
CMR-style requests for additional crypto-recipients on the CDR storage-only
key. However this practice makes recovery from forgotten passphrase more
problematic: to restore user access to the data after forgotten passphrase,
the recovery agent would have to decrypt all stored files (be they on hard
disk, backup tape, floppy disk, or stored inside `zip' or other archives),
then the user would have to generate a new storage-only key, and re-encrypt
all of the files. Given the frequency with which users are prone to
forgetting passphrases implementors may decide to avoid this option for
ergonomics reasons.

CDR vs CMR

PGP Inc has an alternative data recovery proposal called Commercial Message
Recovery (CMR). The CMR proposal involves sending recovery information with
the message in the form of a second crypto recipient. In addition mechanisms
are provided in the form of a proposed CMR public key extension to the IETF
OpenPGP standard which indicates a request to the sender to include this
recovery information. In addition facilities are provided in the form of PGP
Inc's SMTP policy enforcer to partially enforce the inclusion of this
recovery information by bouncing mail which does not include it.

Security comparison

PGP Inc's CMR proposal makes re-keying problematic because data is archived
in email folders still encrypted to the encryption key. This tends to
encourage the practice of having very long term communication encryption
keys. Indeed the beta pgp5.0 implementation the author tried makes the
recommendation that most users would give keys an expiry period of
`forever'.

In addition a security risk with the CMR proposal is that an attacker is
able to obtain via coercion, bribery or burglary the CMR key used to encrypt
the traffic for this key. With a purloined CMR recovery key, and given the
long life time encouraged for such keys the previously passive attacker may
be able to recover years worth of old traffic.

Another risk is that companies may for simplicity have only one CMR key,
thereby putting at risk the entire companies secured communications over
periods of years. This is naturally a bad practice, and one which is
discouraged by PGP Inc also, but the author remains cynically confident that
some corporate users will ignore such advice.

Privacy Comparison

PGP Inc proposese the CMR communications recovery mechanism as a privacy
respecting method of achieving data recovery. The privacy features of the
CMR proposal are that there are proposed flags which are attached to a
public key to indicate statement of intent about plaintext handling to the
sender who is about to use the public key.

Three statements of intent are possible with the current CMR proposal:

   * personal key (no flags set)
   * company use key, company may read message, please encrypt to listed
     company CMR recovery key(s) (recovery flag set)
   * company may read message, encrypt to listed CMR recovery key(s) or the
     message will be bounced (strict recovery flag set)

PGP Inc's statement of intent in plaintext handling is a useful concept, in
that it explains explicitly to the sender who will be able to read the
message, and how the plaintext will be handled. Statement of intents are
always advisory, in that they are impossible to enforce; however they are
useful in encouraging ethical behaviour in this regard.

Statement of intent messages are independent of recovery mechanism, and
apply equally to the CDR mechanism described in this document. In fact the
author would like to encourage a fuller set of statement of intent flags for
both sender and recipient, allowing keys to be marked with intents:

   * personal key, messages not archived
   * personal key, messages archived
   * key used for company business, messages not archived
   * key used for company business, messages archived
   * key used for company business, messages archived and recoverable

And, in addition a similar privacy argument could be made to allow the
sender to express his preferences with regard to plaintext handling on a per
message basis. This could generalise the -m function of pgp2.x which
instructs the recipient not to save to a file. (The -m flag is the sort of
electronic version of `please burn this document after reading', this flag
is mildly enforced by the pgp2.x command line application). These set of
flags would allow a sender to send a `normal communication message', or a
`official statement, please archive with recovery if available', or
`sensitive do not archive' and so on. A privacy respecting implementation
could provide support for these functionalities. Whilst obviously easy to
over-ride, this is felt to be a useful exercise in encouraging respect for
privacy.

In addition the recipient should have ability where company policy allows to
choose which messages to archive, and which to archive with company recovery
information.

>From a privacy perspective the CDR and CMR proposals are approximately
equivalent; the statement of intent flags, and the social value of
encouraging companies to respect privacy are useful in both.

Government access to communication key politics

Another aspect of the whole data recovery area is that the US government,
French government, and UK government (as well as other governments) have
expressed an unhealthy interest in obtaining private individuals, and
companies communications keys. Many privacy advocates view this development
with fear, as it seems to many a particularly Orwellian development. Usually
the Government Access to Communication Key (GACK) attempts try to deflect
criticism by focussing on unrelated issues as a vehicle to introducing GACK:

   * Terrorist attacks and requirement for emergency government access to
     communications traffic
   * Government assistance for companies to recover stored plaintext via key
     escrow infrastructure.
   * Assisting electronic commerce by helping put in place Certification
     Authorities and digital signature laws

However readers who have followed GACK politics will recognise the above for
what they are: attempts to put in place GACK via indirect routes
(respectively: the four horsemen of the infocalypse argument (the fallacy
that terrorists will obey laws and use encryption systems which have
government backdoors), the Clipper I, II and III attemps (various attempts
by the US government to bribe and bully companies into building a GACK
architecture), and the UK TTP (Trusted Third Party) proposal where `TTPs'
are a euphamism for a Certification Authority (CA) which has been coerced by
law to keep individuals private keys).

Scenarios governments might shortly try this to request master keys for
communications might be France, where the government is changing from a
position of no encryption software without license, to a position of
unlicensed encryption software being allowed provided that the governments
has a back door into the communications. Other examples being perhaps the UK
(with it's highly controversial TTP licensing proposals), or the US with the
administration and law enforcement community pushing hard for access to
communications keys. It is likely, particularly in the UK, and US in the
authors opinion, that government backdoors would be demanded in carefully
orchestrated stages, designed to minimise public opposition. There might be
trials where the traditionally more regulated financial organisations were
required to comply, next perhaps government funded research labs, and
business organisations accustomed to large amounts of government regulation
Only towards the end would people using government funded networks
(academics), Internet services providers (ISPs) and individuals be
regulated. The final phases may use a terrorist activity, or national
emergency, or other scandal as justification for increasing the
requirements. It is likely also that the justification will be in terms of
wholesome sounding public safety safeguards.

Political Comparison

PGP Inc's CMR proposal loses heavily in the political argument in the
author's opinion because the protocol design supports the practice of
allowing third and fourth party access to communications traffic.
Particularly with the multiple CMR key requests per key which are rumored to
be coming in the next version of PGP, it will be easy technically for a
government to request that one of the recovery keys belong to the
government. Governments already have the political will to have access to
communications; the barriers are technical, and individual's resistance to
the notion. The danger with the CMR proposal is that it could become the
enabling technology for push button suveillance, the enabling technology for
wide scale secret service signals intelligence keyword scanning. This
probably not what it's proposers intend, however the danger is there in
building the technology.

The CDR proposal on the other hand focusses on recovery information of
stored information only, and as a design principle avoids sending recovery
information over open networks. Whislt governments may be interested to
obtain storage keys, they have much less value, because the ciphertext is
much harder to obtain. Also non-compliance is much harder to detect: a
government law enforcement agency would not know if an individual was
actually using storage software with government backdoor key access without
physically obtaining the storage media. For most individuals the point at
which governments executes a dawn raid is a point at which the individual is
already in trouble. The CDR proposal is much closer to the status quo in
political terms.