Hi All,
I am choosing this point (in the email threads) to continue discussion
of the issues with the Hierarchical profile as I think it contains the
best context to start from.
As indicated in the last couple of TC meetings, the update I proposed:
http://lists.oasis-open.org/archives/xacml/200901/msg00079.html
only goes part-way to addressing the problems I have found in the
Hierarchical profile specification. The update only includes
effectively a glossary with definitions of the terms that are currently
used in the spec that make the rest of the spec comprehensible (at
least from my perspective), and based on that glossary, the spec
includes a few clarifying edits, which are primarily intended to make
the statements in the spec consistent in terms of the definitions in
the glossary.
Primarily, the glossary simply defines "hierarchy" and "DAG" (Directed
Acyclic Graph), where DAG is a collection of hierarchies that are
mapped onto a flat collection of resources, where any resource can be a
member of multiple hierarchies within the DAG.
Based on this starting point and Erik's suggestion as to how the
concepts might be applied in the real world, I did further
investigation and have come to some additional preliminary conclusions
on which I think the next step to address this spec should be based.
The most significant conclusion I have reached is that the
primary "problem" with the non-xml resources (section 3.2) is that the
details here are solving a much more complicated problem than is
necessary for most common use cases, which is that the resources are
identified with URIs as described in section 2.2 and 4.3.
In particular, this "more complicated problem" is one where the
resource-id, itself, is not sufficient to describe the hierarchy of
which it is a member. An example might be resources identified by
serial number (a flat namespace), as opposed to resources identified by
URI, which is an inherently hierarchical namespace
(http://www.ietf.org/rfc/rfc2396.txt).
With a hierarchical namespace there is no need to collect parent and
ancestor nodes since they are already built in to the URI itself. In
fact, there is an example document that I "found", that was produced as
an artifact of the XACML 2.0 development effort, that appears to have
gotten lost from the active XACML collection:
http://www.oasis-open.org/committees/download.php/7315/xacml-profile-hierarchical-resources-nonXML-1.0-draft01.pdf
that demonstrates this quite clearly: section 4.2 shows policies based
on resource-ancestor and resource-parent attributes, and section 4.3
shows the same policies based on matching anyURI. i.e. if the resources
are identified by URI there is no need to use resource-parent and
resource-ancestor attributes either in the RequestContext or in the
Policies.
Based on this situation, my proposal is to enhance the existing
Hierarchical spec using the URI representation in section 3.2 and
indicating that this approach can be used instead of resource-parent
and ancestor when the hierarchies are represented by nodes with URI
resource-ids.
A 2nd significant problem I believe I have found is that the
algorithms defined in section 3.2 are incorrect in some significant
ways, which I believe should be addressed (I will give an example
below).
My proposal here is to simply fix the algorithms.
A 3rd significant problem I believe I have found is that there
are important concepts that are either left out or not given sufficient
emphasis in the spec, which, imo, cause conceptual ambiguities that
make it effectively impossible to say what problem the specification is
actually addressing.
In particular, there is no explicit definition of "normative identity",
which is a key term used in the algorithms defined in section 3.2.
My proposal here is to assert that the normative identity may use
either the recommended representation of URI described in section 2.2,
or a functionally equivalent representation, such as can be constructed
using resource-parent and resource-ancestor node collections.
In particular, there is insufficient development of the concept of
"consistent representations for identities" as mentioned in section 1,
lines 68-72, section 2, lines 175-178, section 3, lines 227-229. While
the point is made that such consistent representations are necessary,
the essential fact that in a DAG that a single node can have multiple
normative identities AND that those identities each belong to a
separate distinct hierarchy within the DAG is not established. The
result is that it is conceptually ambiguous that when a node is said to
have multiple parents, the essential fact is that the node must then
have multiple normative identities, each of which is traceable to one
and only one parent (or self, if the current node happens to be the top
of one of the hierarchies within the DAG).
My proposal here is to adjust the text to emphasize these notions
accordingly.
One final note: I appreciate the significant effort that went into the
development of the Hierarchical specification. A review of the emails
from April 2004 -> September 2004 will show anyone interested that a
lot of work went into the development of this specification. My own
interpretation of the problems I have had with this spec is that it was
a relatively small, but considered to be important, part of XACML 2.0
and that while the ideas collected in this spec are generally properly
targeted, it appears there was simply not sufficient time available for
the TC to consolidate the findings of their studies of the hierarchy
problem and to structurally sort the results into separable pieces that
would have made the results easier to understand and apply.
This effort is intended to build on the existing spec and emphasize
aspects of it that currently are not, and to correct any problems that
may be found in the process.
As indicated above, I will give an example of what I consider to be an
"error" in the algorithms of section 3.2:
In section 3.2 the descriptions appear to be lterally
"over-specified": for example, consider lines 324-327:
" For each immediate parent of the node specified in the
“resource-id” attribute or attributes, and for each normative
representation of that parent node, an <Attribute> element with
AttributeId “urn:oasis::names:tc:xacml:2.0:resource:resource-parent”. "
This statement is indicating that the node "specified in the
'resource-id' attribute or attributes" can have multiple parents and
that each of those parents themselves can have multiple normative
representations.
I claim that the quoted statement is incorrect because while the given
node can have multiple parents, because the current node may belong to
multiple hierarchies, there should be no concern for hierarchies that
the parent belongs to that the current node does not belong to.
The following example will show why this is doubly and actually triply
specified and also why it is essential that the changes described in
1-3 above are needed.
Consider 3 hierarchies:
/a1/b1/c1
/a2/b2/c2
/a3/b3/c3
Let's say that a1, a2, and a3 are distinct roots and represent 3
separate nodes (or resources) in the overall collection.
However, let us also say that b1 is a node that belongs to all 3
hierarchies and that, in fact, /a1/b1, /a2/b2, and /a3/b3 all represent
normative identities of the same node.
Now let us further assume that c1 and c2 represent the same physical
node, but that c3 is a different node.
Let's also assume that these physical nodes have serial numbers r*,
which serve as alternate unique identifiers of the nodes. We can then
see the relation as follows:
Let r1, r2, r3 be serial numbers of the 3 separate root nodes.
Let r4 be the serial number of the b1,b2,b3 node.
Let r5 be the serial number of the c1,c2 node.
Let r6 be the serial number of the c3 node.
The logical and physical resources and hierarchies can now be
represented together as follows:
/a1(r1)/b1(r4)/c1(r5)
/a2(r2)/b2(r4)/c2(r5)
/a3(r3)/b3(r4)/c3(r6)
From the above, it is clear that if we are requesting access to, say
node /a1/b1/c1, then to apply the rules quoted above for parents, we
would need to first include:
/a1(r1)/b1(r4)/c1(r5)
/a2(r2)/b2(r4)/c2(r5)
where the 2 parents of c1(r5) are
/a1/b1
/a2/b2
i.e. c1, which is r5, which is then also c2, effectively has these 2
parents.
However, the rule then requires us to collect all the normative
identities of each of the parents.
First of all in this case the parent rule requires us to collect the
normative identities of r4 twice, once in its a1/b1 incarnation at
which point it also picks up a2/b2, and once in its a2/b2 incarnation,
where it picks up a1/b1 again. Therefore it is already "over-specified"
because it requires collecting these identities twice resulting in 4
instead of 2 "normative parent identities".
Furthermore, the algorithm also requires us to pick up /a3/b3, which is
another "normative identity" of this parent node, which means we will
have 3 copies of each normative identity of the "parents" of c1 for a
total of 9 "normative parent identities", which are, of course triply
redundant.
In fact, we really only need /a1/b1 and /a2/b2 since r5 does not belong
to the /a3/b3 hierarchy.
Therefore this algorithm needs at least 2 corrections:
Correct it so that it only includes one copy of each normative
identity of each parent.
Correct it so that it does not include normative identities of
the parent which are part of hierarchies to which it does not belong,
since policies on those hierarchies can have no bearing on this node
because the normative representations within those hierarchies cannot
refer to this node because it does not have a normative representation
in those hierarchies to which it does not belong.
Comments and suggestions are welcome. I will wait for feedback on
emails and discussion at the next TC meeting before proceeding with the
recommended changes.
Thanks,
Rich
Rich.Levinson wrote:
--------
Original Message --------
Subject:
Re: [xacml] Issue: Hierarchical profile appears ambiguous
and inconsistent
Date:
Thu, 15 Jan 2009 11:37:10 -0500
From:
Rich.Levinson <>
To:
Erik Rissanen <>
CC:
Daniel Engovatov <>, xacml
<>
References:
<>
<>
<>
<>
Hi Erik,
Thanks for this feedback. Unfortunately I did not have time to process
this email before today's meeting, but now that I have, it addresses
one of my major concerns which was the motivational context. i.e. by
seeing the actual example you provided, I can see that a Policy can now
base decisions knowing that some node happens to be an ancestor of the
requested node.
In addition, for your example, I think it would be instructive to show
when a node belongs to two or more hierarchies, that the collection of
attributes should probably have a mechanism to indicate which hierarchy
a node belongs to. For example, if C had an alias C', and parent B' and
ancestors A'->D' where, while (C = C'), that in general (B != B')
and (A != A') and obviously D' has no relation to the unprimed nodes at
all. We would then have a request:
<Resource>
resource-id = C
parent-id = B
self-or-ancestor = C
self-or-ancestor = B
self-or-ancestor = A
resource-id = C'
parent-id = B'
self-or-ancestor = C'
self-or-ancestor = B'
self-or-ancestor = A'
self-or-ancestor = D'
</Resource>
It would seem to me that there needs to be a mechanism whereby one
would be able to tell the primed from unprimed attributes. Possibly
using Issuer
In any event, it is useful information to have this additional context
for understanding the current spec.
As agreed at the meeting, I will try to find some cycles to say what I
think needs to be done to make the spec easier to understand, which is
possibly just including the above information (i.e. your email extended
to multiple hierarchies with some example policy concepts, also such as
you provided).
Also, I think, as I mentioned at the end of the meeting that "scope"
may also have a meaningful role to include in this profile as well.
i.e. one can easily see that if policies are defined whereby certain
conditions apply when a resource-id node is within scope (as defined by
multi-resource spec) of some other node, that some if that ""other
node" happens to be a parent or ancestor of the resource-id node, then
those "certain conditions" would apply to the current resource-id being
requested.
Thanks,
Rich
Erik Rissanen wrote:
Rich.Levinson wrote:
I am trying to understand what policies are
supposed to do with the definitions in the spec. i.e. it is the spec
that says in section 3.2 that all the parent and ancestor nodes need to
be assembled in the request context. What "policy evaluation" are you
referring to? Are you saying what I indicated in original email that a
policy does not need to know anything about hierarchies that the
resource-id node does not belong to?
Hi Rich,
I don't understand all the questions you have, but here's the basic
approach of the profile in a simple example.
Assume the following simple hierarchy:
A <- B <- C
If someone requests access to C, the request will contain these
attributes. this is from the top of my head, so it might be slightly
inaccurate and I might have forgotten some of the attributes, but
hopefully you get the idea.
<Resource>
resource-id = C
parent-id = B
self-or-ancestor = C
self-or-ancestor = B
self-or-ancestor = A
</Resource>
All these attributes are there so it is possible to write policies
which apply to parts of the hierarchy, not just individual nodes.
For example:
<Target>
resource-id = C
</Target>
Matches only the resource C, nothing else.
<Target>
parent-id = B
</Target>
matches the immediate children of B. In the example this is C, but if C
had a sibling, it would also match.
<Target>
ancestor-or-self = B
</Target>
Matches B or any node below B. In this case also C.
Best regards,
Erik
--------
Original Message --------
Subject:
[xacml] Issue: Hierarchical profile appears ambiguous and
inconsistent
Date:
Thu, 15 Jan 2009 01:54:24 -0500
From:
Rich.Levinson <>
To:
xacml <>
I am finding the Hierarchical profile ambiguous and inconsistent and
therefore incomprehensible, which may, similarly to the Multi-Resource
profile be because there are essential contextual paradigms missing
from the specification.
For the Multi-Resource profile, the missing contextual paradigms
included:
* It is necessary to pre-process the multi-resource request and
submit individual requests to the PDP, and then to package up the
responses into a single response.
* The PDP, itself, can only process Request elements that in 2.0
contain a single Resource element, single Action element, single
Environment element, and multiple Subject elements, but each
Subject element has to have a distinct SubjectCategory attribute,
and in 3.0 there can be multiple Attributes elements, but each has
to have a distinct Category attribute.
Once the above paradigms are established, the Multi-Resource profiles
become comprehensible.
I suspect something similar is going on w the Hierarchical profile
also.
Here is the background on the Hierarchical profile that I am trying to
work thru (each bullet represents a potential issue to resolve or
suggestion for improvement):
* There needs to be a definition of "hierarchy". In particular, a
"hierarchy" defn should state that the fundamental properties are
that there must be a single root node with no parent, and that
every other node in the hierarchy must have one and only one
parent, and can have zero, one, or more children.
* Section 3.2 is the biggest problem. To begin, the following needs
confirmation: it appears that the key structural sentence for 3.2
that explains what the four bullets that follow it are is lines
314-315 (which is sandwiched between a negative statement on
ResourceContent and subsequent 3 sentence "note") that state:
o "The request context <Resource> element SHALL contain
the
following elements and XML attributes."
* Each of the 4 bullets in section 3.2 contains the phrase "(for
each) normative representation (of some node ...)". I think this
is the core of the problem I am having understanding this spec.
What I think it means is that for a given collection of physical
resources, such as files in a file system, that in additional to
the conventional file identification by directory path, which
includes all the files in the file system, that other logical
hierarchies that can span part or all of this set of resources
also needs to be taken into account. And that the "normative
identity" of a specific "file" is within the context of a specific
hierarchy. i.e. each hierarchy that covers all or part of this set
of resources has its own "namespace" and within that namespace has
a scheme for hierarchically naming its members. As such, a given
physical resource can belong to any number of these hierarchies
simultaneously, and has one "normative representation" for each
hierarchy it belongs to. And presumably, the point is that each
hierarchy the resource belongs to may have its own restrictions to
apply to resource, so we have to identify all of them to get the
full picture.
* Given the above characterization of the multiple "hierarchies" a
given resource can belong to, it then appears that the four
bullets in section 3.2 state that in order to submit a request,
one has to somehow identify all the hierarchies the given node
belongs to, all the hierarchies the node's parent(s) and ancestors
to, and include an Attribute element for each.
* There appears to be some over-specification above and beyond what
is presumably intended (assuming the above characterizes what is
"presumably intended"). For example, since a node in a hierarchy
can have only one parent, and if a node belongs to multiple
hierarchies, then that node will have one and only one parent for
each hierarchy it belongs to, the following sentence (lines
324-325) appears to doubly state this relationship:
o "For each immediate parent of the node specified in the
“resource-id” attribute or attributes, and for each
normative representation of that parent node ..."
i.e. the node specified in the resource-id can only have one
parent wrt the the node's normative representation with a
specific hierarchy. There seems to be no point to looking at
a parent node in any other context than the normative
context of the resource-id node. i.e. if a node only belongs
to one hierarchy, but its parent belongs to many, then the
other hierarchies the parent belongs to should have no
impact on the child since the child is not a member of those
hierarchies.
Assuming the above characterization is even close to correct, I am
still wondering "why" all this information is being assembled and what
the Policy is expected to do with it. i.e. what is the policy
evaluation paradigm that is being represented here? I suspect that at
most one would need to collect all the normative representations of
only the resource-id node (i.e. identify all the hierarchies it belongs
to), then for each hierarchy, one would evaluate the policies that
apply to that hierarchy.
Thanks,
Rich
---------------------------------------------------------------------
To unsubscribe from this mail list, you must leave the OASIS TC that
generates this mail. Follow this link to all your TCs in OASIS at:
https://www.oasis-open.org/apps/org/workgroup/portal/my_workgroups.php
