gnupg/THOUGHTS

How often have we to do a key lookup by mailaddress?.
can this be accomplished by an external program?



What about using S-Exp to describe the interface to the ciphers instead
of simply iterating over them. This way we can easy register a name which
can be used as the "hintstr" for --load-extension.

EGD
====
Oh, and on embedding egd into the gpg package: I think if you just unpack it
into, say, util/egd/* then you can put something like this into configure.in:

AC_CHECK_PROG(perl_present, perl, true, false)
if $perl_present; then
 AC_PATH_PROG(PERL, perl)
 (cd util/egd; $PERL Makefile.PL FULLPERL=$PERL INSTALLBIN=$sbindir)
fi
AM_CONDITIONAL(WITH_EGD, $perl_present)

and add util/egd to the top-level Makefile directory list inside a WITH_EGD
conditional.

    * What shall we do if we have a valid subkey revocation certificate
      but no subkey binding?  Is this a valid but revoked key?

    * use a mmaped file for secure memory if mlock does not work and
      make sure that this file is always wiped out.  Is this really
      more secure than swapping out to the swap disk?  I don't
      believe so because if an attacker has access to the physical
      box (and he needs this to look at the swap area) he can also
      leave a Trojan horse which is far more easier than to analyze
      memory dumps.  Question:	Is it possible that a Unix pages
      an old (left over by some other process) swap page in for
      another process - this should be considered a serious design
      flow/bug.

Date: Mon, 4 Jan 1999 19:34:29 -0800 (PST)
From: Matthew Skala <mskala@ansuz.sooke.bc.ca>

- Signing with an expired key doesn't work by default, does work with a
  special option.
- Verifying a signature that appears to have been made by an expired key
  after its expiry date but is otherwise good reports the signature as BAD,
  preferably with a message indicating that it's a key-expiry problem rather
  than a cryptographically bad signature.
- Verifying a signature from a key that is now expired, where the
  signature was made before the expiry date, reports the signature as
  GOOD, possibly with a warning that the key has since expired.
- Encrypting to an expired key doesn't work by default, does work with a
  special option.
- Decrypting always works, if you have the appropriate secret key and
  passphrase.



==============================
[ "-->" indicates a comment by me (wk) ]

Hi Werner..

I was looking at some of the PROJECTS items in the recent gpg CVS and wanted
to comment on one of them:

  * Add a way to override the current cipher/md implementations
    by others (using extensions)

As you know I've been thinking about how to use a PalmPilot or an iButton in
some useful way in GPG. The two things that seem reasonable are:
 1) keep the secret key in the device, only transferring it to the host
    computer for the duration of the secret-key operation (sign or decrypt).
    The key is never kept on disk, only in RAM. This removes the chance that
    casual snooping on your office workstation will reveal your key (it
    doesn't help against an active attack, but the attacker must leave a
    tampered version of GPG around or otherwise get their code to run while
    the key-storage device is attached to attack the key)
 2) perform the secret-key operation on the device, so the secret key never
    leaves the confines of that device. There are still attacks possible,
    based upon talking to the device while it is connected and trying to
    convince the device (and possibly the user) that it is the real GPG,
    but in general this protects the key pretty strongly. Any individual
    message is still vulnerable, but that's a tradeoff of the convenience of
    composing that message on a full-sized screen+keyboard (plus the added
    speed of encryption) vs. the security of writing the message on a
    secure device.

I think there are a variety of ways of implementing these things, but a few
extension mechanisms in GPG should be enough to try various ways later on.

1) pass an argument string to loadable extension modules (maybe
    gpg --load-extension foofish=arg1,arg2,arg3 ?)
    --> could also be achived by S-Exps

2) allow multiple instances of the same extension module (presumably with
   different arguments)
   --> set an alias name when loading them
3) allow extension modules to use stdin/stdout/stderr as normal (probably
   already in there), for giving feedback to the user, or possibly asking them
   for a password of some sort
   --> there should really be some kind of callback mechanism.

4) have an extension to provide secret keys:

   It looks like most of the hooks for this are already in place, it just
   needs an extension module which can register itself as a keyblock resource.

   I'm thinking of a module for this that is given an external program name as
   an argument. When the keyblock resource is asked to enumerate its keys, it
   runs the external program (first with a "0" argument, then a "1", and so on
   until the program reports that no more keys are available). The external
--> better use a cookie:  This way we are also stateless but have a more
   general interface.

   program returns one (possibly armored) secret key block each time. The
   program might have some kind of special protocol to talk to the storage
   device.  One thing that comes to mind is to simply include a random number
   in the message sent over the serial port: the program would display this
   number, the Pilot at the other end would display the number it receives, if
   the user sees that both are the same they instruct the Pilot to release the
   key, as basic protection against someone else asking for the key while it
   is attached. More sophisticated schemes are possible depending upon how
   much processing power and IO is available on the device. But the same
   extension module should be able to handle as complex a scheme as one could
   wish.
   --> authenticate the session on startup, using DH and the mentioned
       cookie/screen/keyboard authentication.

   The current keyblock-resource interface would work fine, although it
   might be more convenient if a resource could be asked for a key by id
   instead of enumerating all of them and then searching through the resulting
   list for a match. A module that provided public keys would have to work this
   way (imagine a module that could automatically do an http fetch for a
   particular key.. easily-added automatic key fetching). Without that ability
   to fetch by id (which would require it to fall back to the other keyblock
   resources if it failed), the user's device might be asked to release the
   key even though some other secret key was the one needed.
   --> Right.


5) have an extension to perform a secret-key operation without the actual
   secret key material
   --> Define a clear interface to do this and in the first step write
       a daemon which does exactly this.

 basically something to indicate that any decrypt or sign operations that
 occur for a specific keyid should call the extension module instead. The
 secret key would not be extracted (it wouldn't be available anyway). The
 module is given the keyid and the MPI of the block it is supposed to sign
 or decrypt.

 The module could then run an external program to do the operation. I'm
 imagining a Pilot program which receives the data, asks the user if it can go
 along with the operation (after displaying a hash of the request, which is
 also displayed by the extension module's program to make sure the Pilot is
 being asked to do the right operation), performs the signature or decryption,
 then returns the data. This protocol could be made arbitrarily complex, with
 a D-H key to encrypt the link, and both sides signing requests to
 authenticate one to the other (although this transforms the the problem of
 getting your secret key off your office workstation into the problem of
 your workstation holding a key tells your Pilot that it is allowed to perform
 the secret key operation, and if someone gets a hold of that key they may
 be able to trick your pilot [plugged in somewhere else] to do the same thing
 for them).

 This is basically red/black separation, with the Pilot or iButton having the
 perimeter beyond which the red data doesn't pass. Better than the secret-key
 storage device but requires a lot more power on the device (the new iButtons
 with the exponentiator could do it, but it would take way too much code space
 on the old ones, although they would be fine for just carrying the keys).

The signature code might need to be extended to verify the signature you just
made, since an active intruder pretending to the the Pilot wouldn't be able to
make a valid signature (but they might sign your message with a different key
just to be annoying).

Anyway, just wanted to share my thoughts on some possibilities. I've been
carrying this little Java iButton on my keyring for months now, looking for
something cool to do with it, and I think that secure storage for my GPG key
would be just the right application.

cheers,
 -Brian

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v0.4.5 (GNU/Linux)
Comment: For info finger gcrypt@ftp.guug.de

iD8DBQE2c5oZkDmgv9E5zEwRArAwAKDWV5fpTtbGPiMPgl2Bpp0gvhbfQgCgzJuY
AmIQTk4s62/y2zMAHDdOzK0=
=jr7m
-----END PGP SIGNATURE-----



About a new Keyserver  (discussion with Allan Clark <allanc@sco.com>):
=====================

Some ideas:

o  the KS should verify signatures and only accept those
   which are good.

o  Keep a blacklist of known bad signatures to minimize
   the time needed to check them

o  Should be fast - I'm currently designing a new storage
   system called keybox which takes advantage of the fact
   that the keyID is highly random and can directly be
   used as a hash value and this keyID is (for v4 keys)
   part of the fingerprint: So it is possible to use the
   fingerprint as key but do an lookup by the keyID.

o  To be used as the "public keyring" in a LAN so that there
   is no need to keep one on every machine.

o  Allow more that one file for key storage.

o  Use the HKS protocol and enhance it in a way that binary
   keyrings can be transmitted.  (I already wrote some
   http server and client code which can be used for this)
   And extend it to allow reuse of a connection.

o  Keep a checksum (hash) of the entire keyblock so that a
   client can easy check whether this keyblock has changed.
   (keyblock = the entire key with all certificates etc.)
   Transmitted in the HEAD info.

o  Allow efficient propagation of new keys and revocation
   certificates.


Probably more things but this keyserver is not a goal for the
1.0 release. Someone should be able to fix some of the limitations
of the existing key servers (I think they bail out on some rfc2440
packet formats).


DJGPP
=====
Don't use symlinks but try to do the preprocessing in the config-links script.
DJPGG has problems to distinguish betwen .s and .S becaus the FAT filesystem
is not case sensitive (Mark Elbrecht).
Well, it runs only on one architecture and therefor it might be possible
to add a special case for it, completely bypassing the symlink autselection
trick.

Special procmail addresses
==========================
  * foo+bar@example.net:  Try to match the address without the "+bar".
    Should be done by the MUA, but maybe we can do this.
    --> Yep.  Another reason to utilize a directory service or something
	else for keylookup.



Suggested things which I will not do:
=====================================
  * Let take --help an option to select some topics.
    Using grep is much easier
  * Check if an object (a message, detached sign, public key, or whatever)
    is signed by definite user, i.e. define user
    (userid, or any other unique identification) on command line.
    --> Use a script and --status-fd