PGP 2.6 Application Note: Integrating PGP with mailers
Derek Atkins <warlord@MIT.EDU>
22-May-94
This document will try to describe how to create an
application around PGP version 2.6, in order to incorporate PGP into
mailers and news readers. There are a number of changes in 2.6 which
make it easier to write wrappers for PGP. In writing this, I may have
inadvertantly left out some functionality, some mailers, or some
machine types. I should warn readers now that I am a UNIX person, and
that most of this document will probably be in UNIX terms.
The first thing to remember is that PGP usually needs to have
a controlling TTY in order to gather information, like the password,
or for the user to answer questions, like to whether to add or sign
keys. This is useful for text-based mail agents. In fact, it is most
useful for agents which use files for messages, like MH.
One way to use PGP is to just decrypt the message into a file
and display that file. This can be a security risk, since wiping
files off disk is not always successful, and by having the plain-text
go to a file means more time for a possible attacker to get ahold of
the plain-text of the message.
A better way to accomplish this is to use filter_mode, which
is the -f option. This tells PGP to read the message from stdin, and
to put the output onto stdout. Unfortunately, in this manner, the
signature information is output onto stderr, so you will either lose
it, or it and all other PGP output will be put in the same stream with
the message, but this depends on your piping ability.
PGP tries to send all "interesting" data to standard out, and
error messages to standard error. This lets you pick out the
interesting information and discard the rest. This also means that
you can use PGP in filter-mode as a back-end to some user interface,
and obtain the data in the manner. But remember that the current
implementation of PGP uses temporary files to store intermediate data,
so you are still at a risk, although it is much less of a risk than
just decrypting into a file.
This works well when dealing with a command-line mailer, or a
mailer that is run in a terminal. There are problems with this
approach, however, if you do not have a TTY in which to get a password
to decrypt or sign messages. It seems that there would not be a good
way around this, but then again, PEM is going to have this same
problem. (An example that I can think of is integrating with xmh).
However, there is a way around this in some cases. PGP has
numerous ways to accept the passphrase other than just promping for
it. One way, which is not recommended, is to use the "-z" option to
set the passphrase. Again, this is *NOT* recommended, since some
operating systems will not allow the program to erase the process
table, and someone can retreive the pass phrase from there via a "ps"
listing of active processes. A similar way to get the pass phrase in
is to use the PGPPASS environment variable. Again, this has the same
problems as "-z" with regards to an attacker finding the passphrase in
the OS kernel memory of the process table.. An example of this usage
is:
pgp -sat -z "This is my pass phrase" inputfile
There is a better way of doing this in PGP 2.6, which is an
environment variable called "PGPPASSFD". If this is set, it means
that the FIRST thing PGP will do is read the pass phrase from this
file descriptor. So, for example, one can set PGPPASSFD to "0"
(zero), and then PGP will read the pass phrase from stdin as the first
thing. This allows you to send the passphrase to PGP in a manner
invisible to someone armed with the process listing.
For example, an emacs utility could grab the block to be
encrypted (or decrypted), ask the user for the pass phrase in the
mini-buffer, and then do the equivalent of this shell script, using
something like:
(send-string PROCESS "PassPhrase")
(send-region PROCESS (point-min) (point-max))
---begin--#!/bin/sh
PGPPASSFD=0; export PGPPASSFD
(echo "PassPhraseHere"; cat filename ) | pgp -feast recipient1
recipient2...
---end--I must admit, this is a crude script, since it doesn't strip
out stderr, which included the bannerlines and error messages, but
that is not difficult to do out of band.
This is an example perl script that demonstrates the use of PGPPASSFD:
---begin--#!/usr/local/bin/perl
#
# perl example for PGPPASSFD,
# encrypts stream with password 'test'
#
pipe(READER,WRITER);
if (!fork) {
close(WRITER);
$ENV{'PGPPASSFD'}=fileno(READER);
# the $^F (Ctrl-F) variable controls close-on-exec of files
$ =fileno(READER);
exec "pgp -acf";
die "can't exec pgp\n";
}
close(READER);
syswrite(WRITER, "test\n", 5);
close(WRITER);
wait
---end--Another feature of 2.6 which can be utilized in mailer scripts
is the batchmode feature. This is used in the key-server software
(see keyserv.doc), to allow a process to call PGP and have it perform
without prompting the user for anything. It will take the default
answer to most questions, which may not be what the user wants. This
is switched by adding "+batchmode" to the command line.
One more mailer I should mention, and this is probably the
most important of all of them, is MIME compatibility. In order to use
MIME, a user needs to create a proper entry for PGP. Unfortunately
there is not yet a standard MIME content-type for PGP-MIME. However
there is a recommended set of mailcap entries which would be useful
for using metamail:
application/pgp; pgp -f < %s | metamail; needsterminal; \
test=test %{encapsulation}=entity
application/pgp; pgp %s; needsterminal
I hope that this document has helped people understand some of
the work being done to integrate PGP with mailers. There is some work
going on already to integrate it even more. If you have a mailer for
which there is no PGP handler, and you want to write one, please let
me know, so that we don't duplicate work. In addition, if you have
written a mailer application, and its not included here in the
release, again let me know.
A second contact for this is Colin Plumb <colin@nyx.cs.du.edu>.
Have fun!
-derek
<warlord@MIT.EDU>