Thanks for your attention.
[DeMartini, Thomas] I think your
analysis of the crux of the difference
is correct. I do not conclude that just because Signatures are
not
security tokens they may not carry Claims.
Yes, this is the crux. What then is the criteria for whether
something is a "security token". The only definition in
the spec is
A security token
represents a collection (one or more) of claims.
If something can represent a collection of claims, but yet not be a
security token, then we need to say that explicitly.
I have added a few more comments in blue, in line
below.
At 04:41 PM 3/26/2003, DeMartini, Thomas wrote:
See my comments inline after
[DeMartini, Thomas].
Thanks,
Thomas.
-----Original Message-----
From: Jerry Schwarz
[mailto:]
Sent: Wednesday, March 26, 2003 3:58 PM
To:
Subject: [wss] What constitutes a claim
My idea of what constitutes a "claim" seems so straightforward
to me
that I
have a hard time understanding what other definitions might be
possible.
Yet discussions with several members of the committee have convinced me
that they have a different idea and in the interests of clearer
communication I hope this email will lead to a better
understanding.
Since we don't have a formal language in which to express claims, I use
English and propose below a number of English statements that I
belive
are
the kinds of claims that would be related to SOAP messages. I
discuss
how
I believe those claims might be carried within a <Security>
element
below.
A. The message was sent from a computer owned by Widget Inc.
B. The request in the message was authorized by Joe
C. Joe is an employee of Widget Inc.
D. Joe authorized this request at 8AM on April 1, 2003.
E. Sam vouches that Joe was a purchasing agent of Widget Inc on April
1,
2003.
F. This request was sent on May 1, 2003.
If you think any of the above isn't a "claim" then you're using
the word
"claim" with a technical meaning rather than its plain English
sense.
If
so, I'm curious to see what the definition is and there is a
consensus
on
that definition then it should be added to clarify the meaning of
"security
token".
[DeMartini, Thomas] I agree that all of the above are claims.
Conceivably, we could have security tokens claiming any of them.
However, we don't currently have security tokens that make all
these
claims (and we don't need security tokens that make all these
claims).
This is, I think, where the confusion lies: they are valid claims,
but
they are not the claims made by the security tokens we are dealing
with.
I will elaborate more on this below.
But they are exactly the kinds of claims that need
to be made about messages. If we aren't providing a mechanism to make
such claims then we aren't doing our job.
A, B, D and F specifically
refer to the message. In the context of WSS
that's what makes them interesting. C and E which don't refer to
the
message are interesting because the are claims about entities refered
to
in E.
Observe that the above claims mention four entities: The computer
that
sent
the message, Joe, Sam and Widget Inc. All these must somehow
be
referenced
in the XML. Let's assume that these entities are all identified by X509
certificates within <BinarySecurityToken> element. As people have
repeatedly pointed out to me, X509 certificates carry claims.
However,
they
can't carry A, B, D or F because those claims refer specifically to the
message.
Consider the claims in turn
A. That the computer is owned by Widget Inc can reasonably be part
of
the
X509 certificate.
[DeMartini, Thomas] This is correct.
The client and server might agree that any certificate
(i.e. the <BinarySecurityToken> that contains the certificate) that
is
an
immediate sub-element of the <Security> header is the certificate
of the
machine that is sending the message.
[DeMartini, Thomas] I think it would be bad design if this were the
case.
While this is possible, it seems an
awkward agreement.
[DeMartini, Thomas] Agreed.
The most natural thing is to specify this with a usage
attribute of an STR that references this certificate.
[DeMartini, Thomas] I think the more natural thing, actually, would
be
to have the computer sign the message. It is the signature of
the
message that makes the additional claim that the sender of the
message
is the computer. (NB: I am not implying the signature is a
security
token, only that security tokens are not the only things that may
make
claims.)
Actually what I had in mind was that the computer
would encrypt parts of the message. But the principal is the
same.
B. That Joe authorized the
request might be indicated by the fact that
Joe's certificate was used to sign the body. To be more explicit
about
it
the STR that references Joe's certificate could have a usage
attribute.
B
seems to be the crux of the difference between myself and other
members
of
the committee. I've been told repeatedly that signature's
aren't
security
tokens, from which I conclude that they don't make claims. So
either
people have in mind some other way of carrying (B) or that don't
think
(B)
is a claim. I would like clarification.
[DeMartini, Thomas] I think your analysis of the crux of the
difference
is correct. I do not conclude that just because Signatures are
not
security tokens they may not carry Claims.
C. That Joe is an employee of Widget might reasonably be carried as
a
claim
in Joe's certificate.
[DeMartini, Thomas] Yes.
D. The time at which the request was made is carried in a
<Timestamp>
element. There is no usage attribute in a <Timestamp>, and we
need
something to distinguish that the <Timestamp> is the authorization
time
and
not the time at which the message was sent. On possibility is to
treat
the
<Timestamp> as an embedded "security token" in an
STR and add a usage
attribute. Note that in this context it is the STR itself that
asserts
the
claim and the embedded "security token".
[DeMartini, Thomas] I think it would be bad design if we had to
embed
<Timestamp> in STR in order to indicate what its meaning is.
If the
specification of the <Timestamp> element does not adequately
indicate
its semantics (either authorization or sent time) them we must fix
the
specification of the <Timestamp> element. Perhaps it is
simply the case
that we do not have a security token to express one of the claims
(either D or F).
E. This might be carried as a SAML
security token.
[DeMartini, Thomas] Agreed.
F. This is essentially the same as D except the usage attribute would
presumably be different.
[DeMartini, Thomas] Same comment as D.
All this is hypothetical and there are many alternative ways to
carry
these
claims. The discussion is intended to explore what kinds of
elements
might
be used. In particular note that signatures and STR's carry many of
the
claims in the above. If we want to exclude them from being allowed as
embedded "security tokens" in STR's then we need to do so
based
explicity
on the tag's of the elements and not based on whether or not they
"assert a
claim".
[DeMartini, Thomas] I agree with the last sentence. I think it
would be
highly undesirable for STRs to carry claims beyond those of the ST
they
reference (anything they provide should be a hint only).
It seems unnatural to me too, but it was done when
we put the usage attribute into the STR. Possibly the people who voted
for that alternative weren't thinking of this as a place to make a claim,
but that's what it looks like to me. In my example there are two
times associated with the message. The time at which Joe authorized it,
and the time at which it was sent. I propose to use the usage attribute
of an STR to distinguish the two. How do you propose to do it?
I am fine with
Signatures and Messages carrying claims. In fact, if we were to
decide
that Signatures do not carry claims (which hopefully we don't), then
we
would have to redefine Claim in some more technical sense.
I think, at this point, we can agree that we can require that
security
tokens assert claims, without requiring that all claims be asserted
by
security tokens?
Yes we can. But I'm still unhappy because I
have no way of looking at a <Security> header and determining if
some elements assert claims.
For example, suppose I have a web service that is completely open. That
is it does not require any encryption, or signatures, or
authorization.
<wsse:Security>
<wsse:UsernameToken>
<wsse:Username> Jerry
</Username>
</UsernameToken> </Security>
The application can just ignore this, but does the <UsernameToken>
actually make a claim? Does it matter? I think the answer to both
questions is no.
