no subject (file transmission)

12 Dec 1992

      Cypherpunks!

Like several people (I guess), I am working on an implementation of
digital cash.  Because of the possible legal repurcussions, I'd prefer
to remain anonymous at this time.  Thanks to the efforts of the people
on this list, this is now possible.

My implementation is pretty far along.  It uses PGP modules for the
arithmetic, so the speed is good.  It works on Unix and I should be
able to get it working on MSDOS in a day or two.  Sorry, I don't have
the ability to work on a Mac version.

Here are some of the features.  The basic cash algorithm is the Chaum
system which was posted here.  Multiple denominations are supported,
using different exponents for each denomination.  The program is
presented in the context of a package to be used for an email based
game like Monopoly where the program would be used to allow cash
transfers.  One player is the "banker", and he uses a different
execution module than the other players.  The banker program keeps a
database of used note numbers which is used to detect money reuse.
The database is maintained using a freeware version of the "dbm"
package, so it should be fast even for large note number databases.

The mapping between exponents and values is as follows:

    exponent	    value
	17		1
	19		2
	23		5
	29		10
	31		20
	37		50
	41		100
	43		200
	47		500
	53		1000
	59		2000
	61		5000
	67		10000

and so on, up to a value of 2e9.  The exponents are ascending primes
starting with 17.  This was chosen because you want them to be
smallish for fast note checking, but it was too slow to find primes p
and q for the bank key such that (p-1)(q-1) was not divisible by any
small primes.  The program chooses random p and then tests p-1 to make
sure it passes the divisibility test, and rejects it if not.  Too many
were being rejected when I started with an exponent of 3.  Starting
with 17 rejects about 1/2 the exponents, compared to something like
80% when I started with 3.  The "value" fields are presumed to be
cents, but could be whatever you like.

The code I've written does not do anything other than the basic
electronic cash algorithms.  It does not do bank account maintenance.
It doesn't do PGP encryption.  It doesn't send mail.  (It does have
some functions to scan and check the files created by the program.)

Cash generation is a 3 step process.  First, the user creates what I
call a "withdrawal request" packet.  This is a set of triplets of the
form (e, s, refx), where e is the exponent from the table above, s is
a 16-byte "unique identifier" used solely to link these withdrawal
requests with the returned messages from the bank, and refx is r^e * f(x).
f(x) is MD5 of x, padded to the size of the bank's modulus n using the
PGP routine which pads MD5 signatures.  This padding helps make sure the
arithmetic is more "mixing".  x is the random input to MD5, which I've
chosen as 64 bytes since that is the block size MD5 works on.  (The
output of MD5 is 16 bytes.)  r is the blinding factor.  This is now
128 bits long; longer r's take too long to calculate r^-1 in the third
step, below.  (It takes longer for PGP to calculate r^-1 than to do an
RSA decryption, for r = n = 1024 bits!)

Second, the bank program converts the withdrawal request packet into
what I call a "withdrawal" packet, by just RSA-decrypting the third
entry using the inverse exponent "d" for the value exponent "e".
(These "d" values are calculated at keygen time and stored with the
bank's key information in a private file.)  I call the return triplet
(e, s, rfxd), where e is the value exponent again, s is the same
unique identifier, and rfxd is r * f(x)^d.

(As I said above, the code I've written does not try to maintain
account balances or do any other banking functions.  It just does the
cash algorithms.  There is a routine to scan a withdrawal request and
return the total value being requested for withdrawal.  The idea was
that this could be used as input to a banking program to decide
whether to allow the withdrawal.)

Third, the "player" (e.g. user) program transforms the withdrawal
packet into a "money" packet, by un-blinding it.  To do this, it has
to recover the x and r which correspond with each triplet.  This is
done by the use of a dbm database of "pending withdrawal requests"
which is written during step 1, above.  The database entries are keyed
by (e, s) and return the corresponding (x, r) which were generated
during step 1.  Using x and r, the user transforms (e, s, rfxd) into
(e, x, fxd), the digital cash.  e is the value exponent, x is the
random 64-byte number, and fxd is f(x)^d, the signed version of the
MD5'd and padded x.

There are also player functions to scan and check a money file
(comparing a calculated f(x) to fxd^e), merge money files, and extract
some items from a money file into another money file (this last is
what is to be used for payment).  There are banker functions to check
incoming money and compare against the used-note database, and to add
incoming money to that database.  (The database consists of the
16-byte f(x) values for each note.)

I am pretty happy with the basic routines, but the user interface
needs work.  There are three kinds of files floating around
(withdrawal requests, withdrawals, and money files) and I'm worried
that this will be confusing to the user.  If he accidentally deletes
one at the wrong time he could lose money permanently.  Or if he
accidentally reuses one he could be accused of fraud.  I'm not sure
what the best model is for the user.

The specific issues of creating withdrawal messages and extracting
"bills" from a money file are areas where the user interface should be
made nice.  We want to make it easy for the user to specify exactly
what denominations he wants to work with.  One possibility is to
simply have him input the amount (e.g. $10.55) and the program
calculates that that's a 1000, a 50, and a 5, but this isn't really
flexible enough.  A nice system would be to give him a list of options
and let him fill out a form on-screen, but that's hard to do portably.

Another idea I've had is that there should be a special money file
called the user's "wallet" which is the default place where incoming
money from the bank should go.  This might help organize things.  He
still needs to be able to create other money files for paying other
people, and remembering to delete them after he sends them.

Any suggestions or thoughts on these interface issues, or comments
about how this program could or should be integrated into a larger
system, would be appreciated.

no subject (file transmission)

nobody＠soda.berkeley.edu