From: paul.elliott@hrnowl.lonestar.org (Paul Elliott)
Date: Tue, 21 May 1996 01:12:52 +0800
Subject: PGP MIME INTERNET DRAFT considered harmful.
Message-ID: <31a0442b.flight@flight.hrnowl.lonestar.org>
MIME-Version: 1.0
Content-Type: text/plain


-----BEGIN PGP SIGNED MESSAGE-----

Summary: The PGP MIME INTERNET DRAFT, draft-elkins-pem-pgp-03.txt,
contains a design error with respect to signatures on binary data.


This error results from the failure to recognize the distinction between
those features of MIME which are necessary to represent complex data and
those features of MIME that are used to transport the data.

The design error will result in the following negative results if
PGP-MIME is widely used with binary data.

(1) Huge unnecessary inefficiency whenever binary data is sent signed
and encrypted.

(2) The signatures on PGP MIMEd objects can not be extracted from a MIME
context and used where MIME programs are not available.

(3) Many users will rightly refuse to sign the entities the PGP-MIME
draft envisions. This reduces the utility of PGP-MIME.

(4) If users do sign these files, they will be signing data for which
there are no commonly available inspection tools. This will eventually
result in a security breach.

- ----------------------------------------------------------------------


The problem is that when binary data is to be signed the data is to be
PGPed _after_ base64 has been applied and the MIME headers added.

This is required by the draft:
>
>3.  Content-Transfer-Encoding restrictions
>
>     Multipart/signed and multipart/encrypted are to be treated by agents
>     as opaque, meaning that the data is not to be altered in any way
>     [1].  However, many existing mail gateways will detect if the next
>     hop does not support MIME or 8-bit data and perform conversion to
>     either Quoted-Printable or Base64.  This presents serious problems
>     for multipart/signed, in particular, where the signature is
>     invalidated when such an operation occurs.  For this reason it is
>     necessary to REQUIRE that ALL data signed according to this protocol
>     be constrained to 7 bits (8-bit data should be encoded using either
>     Quoted-Printable or Base64).  Note that this also includes the case
>     where a signed object is also encrypted (see section 6).  This
>     restriction will increase the likelihood that the signature will be
>     valid upon receipt.
>
>     Data that is only to be encrypted is allowed to contain 8-bit
>     characters and therefore need not be converted to a 7-bit format.
>
>
In this the draft follows RFC1847.

[Encrypted & Signed binary data.] 
Now when there is a data path for PGP's cyphertext, PGP provides a
binary data path for its plain text. Thus, the inner base64 that PGP
MIME internet draft requires is totally unnecessary. It will cause a 30%
increase in the size of those messages that are encrypted and signed and
large amounts of CPU time will be used applying & removing the base64.

It is worth noting that huge amounts of binary data will be transferred
by MIME, so the above represents a significant problem.

[Signed binary data.]
Now let us consider the question of what PGP-MIME draft requires users
to sign. Suppose we want to send a signed .gif file to a sysop. The
sysop wants to store the .gif in his download section. Suppose the sysop
wants to store the signature as a detached signature so that people who
download it can check the authorship. But the signature proposed by the
PGP-MIME draft is useless for this purpose. It has MIME headers attached
and it has been base64'ed. People who download such a file from a BBS
have no use for it, unless they have MIME.

Or suppose we send as signed .gif file to the maintainer of a WEB page.
He stores the .gif on an insecure UNIX system connected to the internet.
Suppose, a year later the maintainer wants to check if the .gif has been
tampered with. Can the maintainer store the signature on a floppy and
use the signature for later checking? No, the only way the signatures
specified by the draft can be used, would be to add MIME headers and
apply base64. The maintainer will have to store the entire MIME message,
.gif and all if he wants to check the signature later.

Or let us consider an .gif artist. The artist has a policy of only
signing works that he can be proud of. He does not sign his sketches,
because he does not want sketches to tarnish his reputation. Before
signing and releasing a work, he examines it with several different gif
viewers and paint programs. But what does the draft PGP-MIME want the
artist to sign? It wants him to sign a file that has been base64'ed and
with mime headers added. The artist can not examine the file to be
signed with any of his gif viewers or paint programs. Everyone's mother
has told them to "Never sign anything unless you have read the fine
print first." But here we have a file that has been scrambled so that it
can not be inspected with the commonly available tools. The artist
refuses to sign. Not only does he not know what he is signing, but the
base64 offends his artistic standards. Who can be proud of base64?
Necessary perhaps, but lets face it, base64 is an horrible kludge built
to meet the deficiencies of a network.

If users get in the habit of signing binary files which represent
multimedia data, and which can not be examined with commonly available
inspection tools, it is inevitable and predictable that sooner or later
this will cause some kind of negative security event.

Now there is some justification for the way the draft handles text.
Different operating systems and machine architectures represent text in
different ways, so that it is necessary that digital signatures be taken
over some "canonical" format so that signatures will check on different
operating systems. Even after text has been mangled by Quoted-Printable 
it can still be read after a fashion by the person asked to sign it.

Operating systems and machine architectures also differ in the way they
represent binary data. The differences in the ways integers, floating
point numbers and other such thingies are represented are  well known.
However, such differences must be handled at the application level. The
location within a file of integers, floating points, etc must be set by
the application programmer/designer. PGP, MIME, and base64 can not deal
with these differences, because the location of integers and floats can
not be specified in advance for an arbitrary data file. Thus, from the
point of view of PGP and base64, these differences do not exist and
binary data may as well be a stream of bytes. Thus, in the case of
binary data, base64 is not more "canonical" than the original data.
There is no good reason to sign the base64 rather than the original
data. Once a file has been base64ed, the file can not be examined
with the usual inspection tools.

The draft has chosen to treat text and binary data similarly. This
results in negative results mentioned above, but the developer and draft
author do not have to deal with any logic to handle text and binary data
separately. User utility and security have been sacrificed for the
convenience of simplicity for the draft author and the PGP-MIME
developer.


The typical user of MIME software is not necessarily technically
sophisticated. When the deficiencies and disasters associated with
software patterned on this draft become apparent, not everyone will know
exactly which software component is at fault. The problems associated
with the draft (or its successors) may adversely affect the reputation
of PGP.

Now some descendant of the draft could become a standard or the draft
could become a de-facto standard through wide-spread  use. Such a
standard could become a barrier to the acceptance of other software
without the draft's deficiencies. Thus, the draft could permanently
inflict poor software on the world. (Look at the memory architecture of
the IBMPC for one example. Or look at the MSDOS operating system for
another example.)

The draft should be withdrawn. People should rethink and create a better
plan to combine the benefits of PGP and MIME.

It should accommodate the user who wishes to mail a generally usable PGP
signature (that is, one that can be used outside the context of MIME)
along with multimedia binary data.

It should not ask a user to apply a signature to any data that cannot be
examined with commonly available tools.

It should not require anyone to sign an artifact of a data transfer
system such as base64.

It should not require any additional space overhead  (more than that
which may be necessary for transport) when signing and encrypting.

- --
Paul Elliott                                  Telephone: 1-713-781-4543
Paul.Elliott@hrnowl.lonestar.org              Address:   3987 South Gessner #224
                                              Houston Texas 77063

-----BEGIN PGP SIGNATURE-----
Version: 2.6.3
Charset: cp850

iQCVAgUBMaBPRPBUQYbUhJh5AQH2hwP+J1ADSzD3Yx4gvUIvAwN+EDikIN2IaHhM
j+znIlt9QPzl5SSp44H+JnhoivhKR3562ACI1nexNMZ9E2MrPNioiGmrmz0uGwM6
Px/k2HbioQrgqmmP0IO/98cTZGA71pK7iNk7AZbWpEW4XfWkyRDW9hQzrCEZXXw8
jQwM/VHUPl8=
=BvoZ
-----END PGP SIGNATURE-----