1994-01-30 - PGP Toolkit

Header Data

From: nobody@shell.portal.com
To: cypherpunks@toad.com
Message Hash: bdeacd93c605a7016529855a05a0ebb7b7f434fd1e43818c518dd30c829bf767
Message ID: <199401300552.VAA20150@jobe.shell.portal.com>
Reply To: N/A
UTC Datetime: 1994-01-30 05:52:45 UTC
Raw Date: Sat, 29 Jan 94 21:52:45 PST

Raw message

From: nobody@shell.portal.com
Date: Sat, 29 Jan 94 21:52:45 PST
To: cypherpunks@toad.com
Subject: PGP Toolkit
Message-ID: <199401300552.VAA20150@jobe.shell.portal.com>
MIME-Version: 1.0
Content-Type: text/plain


PGP Tools Version 1.0 - The Crypto Construction Set
Brought To You By Pr0duct Cypher
Available by ftp from csn.org /mpj

What is PGP Tools?

PGP Tools is a set of functions which allows you to easily write PGP-
compatible encryption programs. It will make it easy to write graphical-
interface versions of PGP for Windows, Macintosh, X, and other GUIs, as
well as to integrate encryption into applications. A few possible
applications include point-and-click PGP, all-in-one secure mail programs,
newsreaders with digital signature capability, keyservers, digital cash
servers, and login authentication.

The toolkit allows you to do almost everything PGP does. The user interface
and program design are up to you. PGP Tools uses a system of dynamically-
allocated FIFO buffers for all operations. These FIFOs can exist either in
memory or on disk. Small amounts of data can be processed entirely in
memory, with no disk access or risky temporary files. It is easy to
manipulate PGP packets directly, if you need to construct anything not
already implemented.

The toolkit has functions to RSA encrypt and decrypt, sign and check
signatures, process plaintext and IDEA ciphertext, take MD5 hashes, perform
ZIP compression and decompression, create and extract ASCII armor, and
perform all types of key management. Keys can be generated, signatures
checked, and trust parameters updated, using simple function calls.

There is no console I/O built into the library, and all file I/O is
concentrated in fifo.c. In those cases where I/O is necessary, such as in
the keyring trust update, the user provides a function pointer. There was
a problem with stack space under DOS, so most of the functions were changed
to avoid storing large objects on the stack.

Most of the documentation is in the .h files. If you aren't sure how to use
a function, look at PTD or at the function code itself.

The mini-application PTDEMO is included as a demonstration and debugging
aid. It will encrypt, sign, decrypt, check signatures, display keyrings,
add keys to a keyring, and perform keyring maintenance. It also has the -x
option which displays a list of packets in any PGP-format file. You will
find this useful in debugging your programs. It is not a full PGP, but it
does attempt to test all the functions in the library. The coding in PTDEMO
is admittedly horrible. PTDEMO was not systematically written; it just grew
as needed to test the rest of the code.

FIFO contains the fifo system. PGPMEM has safemalloc(), and a hash table
for storing keys in memory. The crypto functions are in PGPTOOLS. Key
management, with the exception of the trust update, is in PGPKMGT. PGPKTRU
contains code to do a keyring trust update. PGPARMOR does ASCII armor,
PGPKGEN generates keys, and FIFOZIP performs compression.

How Does It Work?

Everything is done with FIFOs, which take the place of temporary files. You
access a fifo using a pointer (struct fifo *). You can create a fifo in
memory or on disk. Memory and disk fifos are similar with only a few
exceptions, and all the PGP functions work with either type transparently.

You create a memory fifo by calling fifo_mem_create, which returns a
pointer to a new, empty fifo. Then you can use fifo_put to put bytes into
the fifo, and fifo_get to read bytes out. Naturally, the first byte put in
is the first one read out. Fifo_aput and fifo_aget allow you to put or get
an array in one operation. Fifo_length will return the length of a fifo.

A fifo is represented in memory as a linked list of fixed size blocks. Each
fifo structure points to a first and last fifo_block structure. All fifos,
even empty ones, have at least one fifo_block and one fifo_data. Each
fifo_block structure points to a fifo_data structure, and to the next
fifo_block in the chain. Memory is automatically allocated as characters
are put into a fifo, and freed as characters are read out. This means that
in most cases there is only one copy of an object in memory. For example,
if you use pgp_create_idea to encrypt some data, the memory used by the
plaintext is being freed as the ciphertext is being created.

Fifo_copy is used to duplicate a fifo. It does not actually make a copy of
the data, except for the last fifo_data block. Instead, it creates new
fifo_blocks which point to the existing fifo_datas. The fifo_data structure
has a links field which keeps track of how many fifo_blocks point to it.
This is incremented when a fifo is copied and decremented when a fifo_block
is freed. When it goes to zero, the fifo_data is freed. Two fifos can
remain logically independent while sharing a common portion of memory.

Fifo_destroy is used to deallocate a fifo. Some functions (generally those
which are guaranteed to use up their input) do this automatically. Others
(those which may leave some data in the input) do not. Abandoning a fifo
without destroying it will cause a memory leak. Destroying or otherwise
accessing a fifo which does not exist will crash the system. A fifo is not
automatically destroyed when its length reaches zero. An empty fifo can
have more data put into it, if it has not been destroyed.

There are several more functions which work with fifos. Fifo_append adds
one fifo to the end of another, destroying the second one. Fifo_rget allows
you to "look ahead" non-destructively. Fifo_rput is used to update keyring
trust parameters on disk. If they are copies of a fifo, they will change
when fifo_rput is used. Do not use with memory fifos if there are copies.

Disk fifos are used to access files. You can create one with
fifo_file_create, by passing it a FILE pointer. Disk fifos behave much like
memory fifos, except that they access files. Fifo_get will read bytes
starting at the beginning of the file. Fifo_put will put bytes onto the
end. Fifo_length will return the length, equal to the size of the file
minus the number of bytes read. Fifo_copy makes a copy of the fifo. If you
need to use two copies of a disk fifo simultaneously, you should call
fifo_clearlac to reset the file pointers for each one, before using it and
after accessing the other one. Files should be opened with appropriate
modes for the use intended. Fifo_destroy gets rid of the fifo, but does not
close the file.

The fifo system provides a convenient and memory-efficient way to work with
PGP packets. Generally, to create (encrypt/sign) a PGP message, you take
the plaintext and build up the message from it. To extract, you check the
input fifo to see which type a packet is, and call the appropriate extract
function to recover the data.

Some of the key-management functions are high-level, such as checking all
the signatures on a key, merging a set of new keys with the keyring, and
updating the trust parameters for a keyring. Others allow the low-level
manipulation of keyrings. See PTDEMO for examples of their use. A good
description of the PGP file formats can be found in PGFORMAT.DOC in the PGP
source distribution.

Most of the extract functions do not perform error checking on the incoming
packets. Instead, there is a separate pgp_check_packets function which
checks a series of packets for proper format, version, length, etc. This
should be used on any input whose integrity is untrusted.

The random-number generator, pgp_randombyte, is initialized by an input
fifo which contains random data. This can be the plaintext file you are
encrypting, characters and timing from random keypresses, or anything else
which is random. This data is put into a buffer to create a seed. Random
numbers are generated by MD5ing the seed and a counter. The counter is
incremented after each MD5, and is initialized by time(). Pgp_randombyte
can be replaced by any good random number generator. The random number
source will depend on your application. The random generator must be
initialized before calling any function which uses it.

Memory allocation (except within the ZIP compression) goes through
safemalloc(), which calls out_of_memory() if it can't allocate. You should
put in your own out of memory handler, because the default one just prints
an error and exits. There is also an error bailout in the ZIP routines,
although I've never seen it happen. The ZIP functions need more cleaning
up to prevent these bailouts.

There are several more features I'd like to add. Some systems, such as
digital cash banks, will involve servers, possibly accessing thousands of
keys. The sequential PGP key is too slow for this. Servers could use a
separate hash file, taking the least significant bits of the key id as the
hash value. This file would provide pointers into the keyring for fast
access by keyid. Fast access by userid is hard, because userids are looked
up by substrings. If anyone knows of a file format which allows fast
substring searches, without a huge index, please let me know.

PTDEMO has been tested under DOS and Unix. Endian() is used for endian
swapping. On a big-endian machine, defining HIGHFIRST will define out
endian. This should work, but I don't have a big-endian machine to test it.

Some of this code, including the MPI library, the key generation, the ZIP
routines, and some of the ASCII-armoring functions, was written by others
for PGP, and they own that code. The parts I wrote (everything which isn't
in PGP) are not copyrighted and may be used for any purpose, commercial or
otherwise. I don't care how it is used, as long as it helps to make public-
key cryptography more readily available.

Unlike some people, I'm not out to get PKP. If they would sell, for a
reasonable price, a personal RSA license, I would buy one. This license
would allow you to use RSA yourself and to share source code for the use
of other licensed people, but not to sell commercial software without
royalties. If they would agree to allow free non-commercial use, I would
continue to write code, which would simplify commercial implementations.
This would make money for PKP in the long run.

The problem is that every available RSA implementation has strings
attached: you are forced to use DES or ask their specific permission to do
otherwise (RSAREF/RIPEM), to get keys or signatures from RSADSI, to buy a
new key after a mandatory expiration date (DSS), are not allowed to see the
source code to verify the security (ViaCrypt), etc. My interest is in
promoting the use of cryptography. If PKP makes a reasonable profit along
the way, that's okay with me.

I would like to ask PKP's permission up front to release this library, in
the hope of legitimizing PGP-compatible encryption. Unfortunately, the
present grey-area status of PGP, and the oppressive environment created by
the Phil Zimmermann investigations, prevent me from doing so.

I will read and possibly participate in any discussions of this library on
alt.security.pgp, so if you have any ideas, post them there. Any messages
from me will be signed with my key:

Type bits/keyID   Date       User ID
pub  1024/558A1D 1994/01/10  Pr0duct Cypher <alt.security.pgp>
Key fingerprint =  9D AF 6D 4D 8E 64 43 FC  D5 CB 9C 7A 36 C7 6D B9

Version: 2.3a


Version: 2.3a