For those of you who haven't already seen this -- my apologies for the post if you already have.... Forwarded message:
From: pgut1@cs.aukuni.ac.nz (Peter Gutmann) Newsgroups: alt.security.pgp,sci.crypt,talk.politics.crypto Subject: A new PGP Followup-To: alt.security.pgp Date: 13 Sep 1994 16:01:08 GMT Organization: University of Auckland Lines: 263 Sender: pgut1@cs.aukuni.ac.nz (Peter Gutmann) Message-ID: <354ic4$t54@ccu2.auckland.ac.nz> NNTP-Posting-Host: cs13.cs.aukuni.ac.nz X-Newsreader: NN version 6.5.0 #7 (NOV) Xref: news.sprintlink.net alt.security.pgp:18448 sci.crypt:31730 talk.politics.crypto:7169
[NB: Followups redirected to alt.security.pgp]
This posting is a call for participation in a complete rehash of PGP to parallel the existing PGP effort in the US. All contributions are welcome, although any cryptographic material from the US won't be usable by anyone else, so it'd be best if all crypto work was done outside the US. I believe Phil is aware of this project but can't comment on it due to legal advice - please don't pester him (or me) about this.
The following sections are a preliminary sketch of what's to be done, see the end of this message for more information on how to participate.
Design goals for the new PGP:
- Proper key database for fast lookup of keys. - Something like DER-encoded ASN.1 keys and data, PKCS/X.509 compatible with extensions for PGP's trust model and also i18n features not found in X.509 (exact details still under debate). - Ability to plug in alternative routines, eg RSAREF instead of independant RSA code. - Backwards compatibility with 2.x. - MIME security extension compliant encoding of messages. - Flexible, planned message/key format which allows easy expandability.
The universal PGP rewrite emphasizes flexibility of design, modularity and Chinese-wall seperation for ease of development, and abstraction of messy details like low-level I/O. Some rules to go by:
- All functional units are kept in seperate modules. There is one header file giving the interface to that module. Only the interface given in the header file is available to the outside world. All non-public information should be in header files which are private to that module. This should vastly simplfy development by multiple people since all the content-coupling of the existing code is eliminated.
The whole of PGP is too big for one person to get an overview of, by enforcing modularity with a preset interface we can allow independant teams to work on the parts they're best at, as well as allow independant developers to user the libraries they want in PGP support software without having to worry about all the API's breaking every time there's a new release.
- None of the library routines perform any user I/O. This is all handled by higher-level code, using callbacks if really necessary. The libraries should be usable in any CLI or GUI environment without needing changes made for different operating environments.
- All routines return an integer status with well-defined codes. Status values are defined in the module interface header and are of the form <3-char module name>ERR_<error type>, eg DBXERR_READ would indicate a read error in the database access code. No mysterious constants buried seven nesting levesl down in routines somewhere!
The basic modules are:
Streams -------
A generalized stream I/O library. A stream can be a block of memory, a FILE *, a file-descriptor-based file, a TCP socket, an X.25 virtual circuit, or whatever. The use of memory streams does away with PGP2's overuse of temporary files.
int newStream( STREAM *stream ); int attachStream( STREAM *stream, ??? ); int ioctlStream( STREAM *stream, ??? ); int readStream( STREAM *stream, void *buffer, size_t length ); int writeStream( STREAM *stream, void *buffer, size_t length ); int lengthStream( STREAM *stream ); int deleteStream( STREAM *stream );
The attachStream() and ioctlStream() attach a stream to an interface, eg a block of memory, a file, or a network connection, and change its characteristics. The end user is never aware of any difference, or of the fact that, for example, for a small message they might be R/W memory, but for a larger one they might be R/W a temporary file. Perhaps an existing PD streams library can be adapted to the task.
Error codes: STMERR_xxx
PKC ---
A PKC library.
int rsaEncrypt( PUBKEY *pubKey, STREAM *stream ); int rsaDecrypt( PRIVKEY *privKey, STREAM *stream ); int rsaKeyGen( ??? );
Error codes: PKCERR_xxx
Crypto ------
A conventional-key crypto library.
int encrypt( CRYPTINFO *cryptInfo, STREAM *stream ); int decrypt( CRYPTINFO *cryptInfo, STREAM *stream );
The CRYPTINFO struct contains all the information you need to specify encryption algorithms and modes, so you'd have something like:
cryptInfo->algorithm = CRYPT_ALGO_IDEA; cryptInfo->mode = CRYPT_MODE_CFB; encrypt( cryptInfo, stream );
This fixes another complaint with PGP, that for every job you do there are 15 different functions to do it, all with slightly different parameters and options. With the unified interface, the call to encrypt data is *always* called encrypt() (rather than encryptIdeaCfb(), encryptFooBar(), encryptThisThatAndTheOther(), etc), and the CRYPT_INFO struct contains all the parameters you need. A library could then implement a number of different algorithms and you choose which one you want without having to know that the function name for that option is encryptQwertyFoo(). You could even have a getCryptInfo() call which queries a library as to which algorithms and modes it implements, or the call could return CRYERR_ALGO_UNAVAIL (algorithm unavailable) or CRYERR_MODE_UNAVAIL (encryption mode unavailable).
Error codes: CRYERR_xxx
Configuration -------------
Get configuration information for PGP routines. Use a proper grammar definition, perhaps lex+yacc (tcl has also been suggested), read into (private) vars, make available to outside world via getXXX() calls. NO GLOBAL CONFIG VARS!
int readConfigFiles( void ); int getFooInfo( char *fooPtr ); int getBarInfo( int *barPtr );
Error codes: CFGERR_xxx
Random Number Handling ----------------------
A random number management library.
int openRandomStream( STREAM *stream ); int readRandomStream( STREAM *stream, void *buffer, size_t count ); int closeRandomStream( STREAM *stream );
This could use the existing keystroke latency method, or connect the stream to custom hardware, or whatever.
Error codes: RNDERR_xxx
Key Database Management -----------------------
All keys can (and should be) stored in a database, which encapsulate the full DER-encoded keys inside a database allowing fast lookup. The general format will be:
{ keyID, userID, encapsulated key }
The keyID is a hash of the key (making it independant of the key itself and not prone to denial-of-service attacks), the userID is simply the user ID/address/whatever, the encapsulated key contains the full key. Lookup is done on the keyID and userID which locate the required key. Details of the exact mechanism to be decided by the database experts - need to define message format, kludges like storage of encapsulated keys in XMS for peecee's, etc etc.
En/decoding -----------
ASN.1 DER-encoding of keys/data, MIME en/decoding. Needs to be discussed.
Compatibility -------------
Backwards-compatibility code to handle old PGP keys and messages. Presumably keys will only need to be converted once, then we need to read old messages and (possibly) write them.
Low-Level ---------
Low-level OS interface code. All OS-specific code is hidden in this module.
I18n ----
A proper i18n system which reads the appropriate messages from a database once and then keeps them in memory.
int readInternationalizationDatabase( void );
All user I/O then uses strings from the i18n database. The database is generated by a precompiler from some user-friendly input format, so that a group of seperate language scripts for eg English, French, German, Russian, etc are fed into one and, and the output is the full i18n database. Adding a new language to PGP then involves simply creating a new script in that language and running the database compiler on it.
Installation ------------
Not part of PGP, but we need someone to do a decent install script which removes from end users the need to fiddle makefiles and options.
Others ------
Other routines as required. For example, we each library should have an init call which registers an exit handler with a function called by atexit() to allow a clean shutdown no matter how we exit the program. This does away with the current need for PGP to explicitly perform exit handling all over the place, and hides the messy details from the end user (who may not even think about things like this).
What we need:
- A well-connected site outside the US to run mailing lists, one per developers group:
pgp-streams - streams interface for I/O pgp-crypto - public and private-key encryption, random number management pgp-dbx - key database management pgp-compat - PGP 2.x backwards-compatibility management pgp-protocol - PGP data/key format and protocol redesign pgp-lowlevel - low-level OS-specific glue code pgp-misc - everything else, eg config code, i18n, installation
There may also be a need for a seperate group to handle key-related routines such as trust-level checks and whatnot, which is a fairly complex task. The idea is to mimimze the amount of unnecessary traffic people will need to read in order to get their job done.
The only communication which should be necessary between these groups is the header file which defines the data structures and interface and/or a written specification of the interface to a module. This means that the PKC group can work on the PKC code without having to worry about what the streams, database, crypto, and os-specific groups are doing.
An attached FTP site for distribution of code would also be nice.
- Groups of developers with expertise in PKC crypto, conventional crypto, database management, low-level OS-specific stuff, code optimization, protocol design, software i18n, and everything else imaginable.
Until the mailing lists are set up (someone volunteer a site, quickly!) I'll be handling things in email, which means that it could take quite a while for me to extract myself from the flood of messages I'm expecting. Please try and keep non-essential messages to a minimum, and be patient when waiting for replies.
Peter (Hmm, maybe I'd better mount /usr/spool/mail on a seperate drive).
- paul