Re: [Netconf] draft-ietf-netconf-rfc6536bis: one week review of a specific change

[Andy Bierman <[e-mail]>]

On Mon, Nov 27, 2017 at 7:44 AM, Robert Wilton <[e-mail]> wrote:

>
>
> On 27/11/2017 15:12, Andy Bierman wrote:
>
>
>
> On Mon, Nov 27, 2017 at 2:39 AM, Robert Wilton <[e-mail]> wrote:
>
>> I do get the point that Tom is making here, and have quite a lot of
>> sympathy for it.  E.g. by automatically allowing the side effects of
>> when/choice statements then we are introducing a potential security hole
>> here that operators may find hard to be aware of and mitigate.
>>
>> It also seems odd to me that a client is allowed to implicitly remove
>> some configuration through the side effect of a when/choice statement that
>> they are not allowed to delete explicitly.
>>
>> The alternative of requiring appropriate access for the side-effects,
>> seems like an intuitively safer design, but I also get Andy's concern that
>> security isn't useful if it become so unwieldy that nobody uses it.
>>
>> So, I guess the question here is, do we have some concrete examples using
>> real data models where the extra NACM delete rules would be required?  If
>> not then perhaps requiring the explicit access for side effects would be
>> better.
>>
>
>
> Yes -- please provide some examples where it is better to force the
> operator
> to provide extra "delete" rules to allow false-when data nodes to be
> deleted.
>
> It seems to me that these rules would allow individual data structures to
> be
> removed by the client, that would not otherwise be possible.
>
> Yes, this is true.
>
> But it is still odd that they a client is allowed to delete the
> configuration, just not explicitly.  I.e. it is strange that they are not
> allowed to put in a semantically equivalent request that explicitly deletes
> the configuration, they are only allowed to do it if the delete is
> implicit.  This seems inconsistent.
>
>
It is just as odd that a request  to create /foo/case1 will work if nothing
from
the choice-stmt exists, but will fail if /foo/case2 has to be deleted in
order
to create /foo/case1.



> If 3 data structures need to be in place "when /foo exists" then
> deleting 1 or 2 of them directly may not be desirable.  Deleting
> them all together (i.e., via when-false deletion) may be the intended
> usage.
>
> If there is a dependency that all three exist then that may be better
> expressed with a must statement instead.
>
>
> The vendor and operator need to be aware of the data model dependencies.
> That said, how come CLI-based configuration does not have this problem?
>
> My experience of CLI based configurations do have these sorts of problems,
> and many more :-)
>


>
> (Or rather, why has this problem been ignored for so long? Maybe it not
> that real?)
>
> This may be the real crux of the argument:
>
> It probably doesn't make sense to use when statements on disjoint parts of
> the YANG schema (e.g. it does not make sense for the OSPF YANG model to
> depend on whether an interface it runs over has an IP address configured).
> Instead, a more normal usage would be a flag (e.g. like interface type) to
> allow/prevent certain child nodes from being available.  In these
> scenarios, if a user is allowed to change the type of an interface then
> they are most likely allowed to change the other settings on the interface
> as well.
>
> So for 'normal' when statements, it probably doesn't make much difference
> whether or not explicit NACM access is required to the nodes being
> implicitly deleted because the client probably already has the necessary
> access rights.
>
> But for the 'corner case' when statement scenario, that are unlikely to be
> used, it still seems safer to require explicit access permissions than
> implicitly accepting the side effect with no NACM validation.
>


This is fine, except there is no way to tell a normal when from a
corner-case when.

In all cases, the false-when evaluation means the node is not relevant to
the model anymore.
Same for a case that is being replaced by a different case.
In order for this NACM threat to be real. the YANG modeler needs to be
wrong about
whether the false-when states are actually irrelevant to the model in its
new state.



> Thanks,
> Rob
>


Andy


>
>
>
> Thanks,
>> Rob
>>
>
> Andy
>
>
>>
>> On 25/11/2017 15:53, Andy Bierman wrote:
>>
>>
>>
>> On Sat, Nov 25, 2017 at 4:42 AM, t.petch <[e-mail]> wrote:
>>
>>> ----- Original Message -----
>>> From: "Andy Bierman" <[e-mail]>
>>> To: "t.petch" <[e-mail]>
>>>
>>>
>>> > On Fri, Nov 24, 2017 at 3:02 AM, t.petch <[e-mail]> wrote:
>>> >
>>> > > <inline>
>>> > >
>>> > > Tom Petch
>>> > >
>>> > > ----- Original Message -----
>>> > > From: "Robert Wilton" <[e-mail]>
>>> > > Sent: Thursday, November 23, 2017 10:44 AM
>>> > >
>>> > > > Hi Andy,
>>> > > >
>>> > > > On 22/11/2017 18:09, Andy Bierman wrote:
>>> > > > > On Wed, Nov 22, 2017 at 5:41 AM, Benoit Claise wrote:
>>> > > > >
>>> > > > >     Hi Rob,
>>> > > > >
>>> > > > >     At that points in time, if it clarifies the spec. and the
>>> > > authors
>>> > > > >     agree with this, let's do the right thing.
>>> > > > >
>>> > > > > I will add the new text if that is what the WG wants.
>>> > >
>>> > > > Having read this draft, I was undecided what the intended behavior
>>> > > > actually should be.
>>> > > >
>>> > > > The way that Yumapro and Tail-f have implemented this is
>>> reasonable,
>>> > > and
>>> > > > is probably also the way that I would implemented it.
>>> > > >
>>> > > > But I also think that it would be a reasonable implementation for
>>> a
>>> > > > server to calculate the full change set (taking account of side
>>> > > effects
>>> > > > from when and choice statements) to the datastore before checking
>>> the
>>> > > > "write data node access control" against the resultant change set.
>>> > > >
>>> > > > My interpretation is that the RFC text is ambiguous on what the
>>> > > correct
>>> > > > behavior is, and one could make a reasonable argument that the
>>> current
>>> > > > RFC actually specifies that the alternative behavior is correct,
>>> i.e.
>>> > > > requiring delete access even if a node is implicitly changes as a
>>> side
>>> > > > effect. E.g. section 3.2.3 of RFC 6536 states:
>>> > > >
>>> > > >     If the protocol operation would result in the deletion of a
>>> > > datastore
>>> > > >     node and the user does not have "delete" access permission for
>>> > > that
>>> > > >     node, the protocol operation is rejected with an
>>> "access-denied"
>>> > > >     error.
>>> > >
>>> > > Rob
>>> > >
>>> > > I have been reading that paragraph since this thread started and
>>> > > wondering how it could be thought unclear, what am I missing:-)
>>> > >
>>> > > To me it clearly states that the user must have the appropriate
>>> access
>>> > > for the consequences of whatever they do, be that 'when', 'choice'
>>> or
>>> > > anything else! Indeed, I see exactly that behaviour in operational
>>> > > (non-network) databases of which I am a user with limited
>>> privileges.
>>> > >
>>> > >
>>> >
>>> > This is not correct.
>>> > The server is the entity that is cleaning up false when-stmt or
>>> unselected
>>> > cases
>>> > for a choice-stmt.
>>> >
>>> > NACM does not prevent the sever from making any changes to the system.
>>> > It only affects the operations requested by a client.
>>>
>>> Andy
>>>
>>> My model is slightly different, that the server is a black box, with an
>>> interface through which I can make authorised changes.  If something I
>>> do causes the deletion of something I am not allowed to delete, may be
>>> not even to read, well, we part company on the acceptability of that!
>>>
>>> I do accept, as Randy points out, that the situation is complicated.  I
>>> am reminded of the issues that came up relating to the 'when' statement
>>> when preparing YANG 1.1.  Yes, 'when' makes things possible and is very
>>> useful but at the same time, its side effects can be troublesome.  I
>>> would like to wrap it up in a conceptual bubble so you could not have
>>> the sort of fine-grained access control that allows the case that Rob
>>> postulates to occur but have no idea what that might look like.  And I
>>> don't know how much 'when' is used in that way in existing YANG
>>> modules - uh huh, more reading required.
>>>
>>
>>
>> When the access control is too complicated, it does not get used.
>> Instead, the most likely access control is "everybody is the root user".
>> The server pruning of false when/choice data nodes is considered cleanup.
>> The pruned data is no longer relevant to the data model. This is the
>> indended
>> use, but when-stmt can easily be abused.
>>
>> The complex tangled web of dependencies is no worse
>> than it has been with CLI for 30 years, where every dependency is ad-hoc
>> and probably undocumented. The granularity of CLI-based access control
>> is less granular than NACM.
>>
>> It is possible that vendors and operators are willing to spend hours and
>> days
>> figuring out how to tune the NACM rules to allow a specific edit to work.
>> Of course the operator will not even be told by the server which nodes
>> are blocking the edit. That would be a security risk.
>>
>> IMO NACM will be completely unusable if rules are needed to allow
>> server cleanup of dead nodes.  The additional NACM rules required
>> might provide more access than intended (unless lots of delete-only
>> rules are added). The huge jump in NACM rule management complexity is
>> a new security vulnerability, which might even be worse than the
>> vulnerability it is intended to fix.
>>
>>
>>
>>> Tom Petch
>>>
>>> >
>>> > Andy
>>> >
>>>
>>>
>>>
>>>
>> Andy
>>
>>
>>
>>>
>>>
>>>
>>>
>>>
>>>
>>> > > My other thought was that given all the complexities of 'when' that
>>> > > emerged during the specification of YANG 1.1, perhaps, were I an
>>> > > implementor of this, I would take the easy option and ignore the
>>> side
>>> > > effects of 'when'; but I might feel guilty about doing so.
>>> > >
>>> > > If the new paragraphs go in as proposed, then I think that the two
>>> > > paragraphs you quote need changing else that section looks to me
>>> like an
>>> > > oxymoron.
>>> > >
>>> > > Tom Petch
>>> > >
>>> > > > An <edit-data> request that causes a when constraint to evaluate
>>> > > > differently does result in a deletion of those data nodes from the
>>> > > > datastore, and hence according to the text above, requires
>>> explicit
>>> > > > "delete" access permission to accomplish that.
>>> > > >
>>> > > > Clearly it would be an interop issue if different servers
>>> implemented
>>> > > > the NACM path based filtering differently.
>>> > > >
>>> > > > Hence why I think that it would be prudent for the draft to be
>>> more
>>> > > > explicit on when and choice statement handling.
>>> > > >
>>> > > > Rob
>>> > > >
>>> > > > > There have not been any comments on this issue.
>>> > > > >
>>> > > > > Andy
>>> > > > >
>>> > > > >     Regards, Benoit.
>>> > > > >>
>>> > > > >>     Hi Benoit,
>>> > > > >>
>>> > > > >>     There is also one further, unrelated change that I am
>>> proposing
>>> > > > >>     is made the draft before it is published, to help better
>>> > > clarify
>>> > > > >>     the expected behavior. It isn't the end of the world if
>>> this
>>> > > > >>     doesn't go in, but I think that it prevents sometime taking
>>> a
>>> > > > >>     different, but IMO reasonable, interpretation of how "when"
>>> > > > >>     statements are considered, and then having a future
>>> argument
>>> > > > >>     about what behavior it specified in the standard.
>>> > > > >>
>>> > > > >>     If we clarify it now, then it closes that door :-)
>>> > > > >>
>>> > > > >>     I've proposed text to Andy and Martin on Monday, but I've
>>> not
>>> > > > >>     heard back yet.
>>> > > > >>
>>> > > > >>     Netconf email with proposed text attached. The text doesn't
>>> > > > >>     necessarily have to match this, but personally I think that
>>> it
>>> > > is
>>> > > > >>     useful if the draft says something on this.
>>> > > > >>
>>> > > > >>     Thanks,
>>> > > > >>     Rob
>>> > > > >>
>>> > > > >>
>>> > > > >>     On 22/11/2017 13:25, Benoit Claise wrote:
>>> > > > >>>     On 11/10/2017 7:23 PM, Andy Bierman wrote:
>>> > > > >>>>     Hi,
>>> > > > >>>>
>>> > > > >>>>     Here are some proposed edits to make the data rule
>>> consistent
>>> > > > >>>>     with the examples.
>>> > > > >>>>     Note that this issue is not related to the edit in the
>>> > > original
>>> > > > >>>>     1-week change.
>>> > > > >>>     That's right, but we found a source of misinterpretation
>>> in
>>> > > the
>>> > > > >>>     draft and you have rightly corrected it in the github v9.
>>> > > > >>>
>>> > > > >>>     Regards, Benoit
>>> > > > >>>>
>>> > > > >>>>
>>> > > > >>>>     sec. 3.3.5:
>>> > > > >>>>
>>> > > > >>>>     OLD:
>>> > > > >>>>
>>> > > > >>>>
>>> > > > >>>>     data node rule: controls access for a specific data
>>> > > > >>>>     node, identified
>>> > > > >>>>     by its path location within the conceptual XML document
>>> > > > >>>>     for the
>>> > > > >>>>     data node.
>>> > > > >>>>
>>> > > > >>>>
>>> > > > >>>>     NEW:
>>> > > > >>>>
>>> > > > >>>>     data node rule: controls access for a specific data node
>>> > > > >>>>     and its descendants,
>>> > > > >>>>     identified by its path location within the conceptual XML
>>> > > > >>>>     document for the
>>> > > > >>>>     data node.
>>> > > > >>>>
>>> > > > >>>>
>>> > > > >>>>     sec 3.4.5, step 6, bullet 2:
>>> > > > >>>>
>>> > > > >>>>
>>> > > > >>>>     OLD:
>>> > > > >>>>
>>> > > > >>>>     * The rule does not have a "rule-type" defined or the
>>> > > > >>>>     "rule-
>>> > > > >>>>     type" is "data-node" and the "path" matches the
>>> > > > >>>>     requested
>>> > > > >>>>     data node, action node, or notification node.
>>> > > > >>>>
>>> > > > >>>>     NEW:
>>> > > > >>>>     * The rule does not have a "rule-type" defined or the
>>> > > > >>>>     "rule-
>>> > > > >>>>     type" is "data-node" and the "path" matches the
>>> > > > >>>>     requested
>>> > > > >>>>     data node, action node, or notification node. A path is
>>> > > > >>>>     considered to match if the current data node is the
>>> > > > >>>>     data node
>>> > > > >>>>     specified by the path, or is a descendant data node
>>> > > > >>>>     of this data node.
>>> > > > >>>>     appendix B.4: (2 bugs in explanation)
>>> > > > >>>>     OLD:
>>> > > > >>>>     deny-nacm: This rule denies the "guest" group any access
>>> to
>>> > > the
>>> > > > >>>>     <nacm> subtree. Note that the default namespace is only
>>> > > > >>>>     applicable because this subtree is defined in the same
>>> > > > >>>>     namespace as the <data-rule> element.
>>> > > > >>>>     NEW:
>>> > > > >>>>     deny-nacm: This rule denies the "guest" group any access
>>> to
>>> > > the
>>> > > > >>>>     <nacm> subtree.
>>> > > > >>>>     Andy
>>> > > > >>>>
>>> > > > >>>>     On Fri, Nov 10, 2017 at 9:24 AM, Robert Wilton
>>> > > > >>>>     <[e-mail] <mailto:[e-mail]>> wrote:
>>> > > > >>>>
>>> > > > >>>>
>>> > > > >>>>
>>> > > > >>>>         On 10/11/2017 16:33, Andy Bierman wrote:
>>> > > > >>>>>
>>> > > > >>>>>
>>> > > > >>>>>         On Fri, Nov 10, 2017 at 8:16 AM, Robert Wilton
>>> > > > >>>>>         <[e-mail] <mailto:[e-mail]>>
>>> wrote:
>>> > > > >>>>>
>>> > > > >>>>>
>>> > > > >>>>>
>>> > > > >>>>>             On 10/11/2017 15:49, Andy Bierman wrote:
>>> > > > >>>>>>
>>> > > > >>>>>>
>>> > > <snip>
>>> > >
>>> > >
>>> >
>>>
>>>
>>
>>
>
>