I think the best thing to do would be to make the methods protected
(they're legitimately used in derived classes), and offer public
non-getter accessors as necessary.
Daryl Olander wrote:
BTW, What do you think is the best solution. I believe we should just
rename these methods. Even if they are public, they are probably not called
often by page flow authors.
On 2/17/06, Daryl Olander <[EMAIL PROTECTED]> wrote:
It seems to be a risk if someone expects only page A to be accessable from
an action and all of the sudden page Bis being navigated to. Futher, I just
tweaked one thing in those structures. You're really the only one that
would understand what other things could be done :-)
It just seems like anytime a user can override navigation out of an action
that there could be some security issues. I agree that they can't access
things they don't have access to because we always forward to the JSP. But,
we pass page inputs, etc along those links. These may actually have
consequences that differ from just directly hitting the .JSPs.
At this point, it might depend on what the App / PageFlow / JSP all do and
how the data flows in the application. I just worry some where out there
there could be a problem. It's not a huge risk, but it is a risk.
On 2/17/06, Rich Feit <[EMAIL PROTECTED]> wrote:
First, I agree that this needs to be fixed.
Second, just so I understand... this is a hole because the user might
have set up the webapp to prevent direct browser URL access to
whatever's being forwarded to?
Daryl Olander wrote:
OK...We've got a hole...
I have the following form that change the forward path to /bar.jsp
<netui:form action="submit">
<netui:hidden dataSource=" pageFlow.currentPageInfo.forward.path"
dataInput="/bar.jsp"/>
<netui:button value="submit" />
</netui:form>
I also have the following action in my page flow.
@Jpf.Action(
forwards={
@Jpf.Forward(name="index", navigateTo =
Jpf.NavigateTo.currentPage)
}
)
protected Forward submit(Form form)
{
return new Forward("index");
}
If the current page is index.jsp, this should navigate back to that,
when
the form is submitted it will navigate to bar.jsp. In my mind this is
actually a security hole. I can dynamically change the navigation
externally in this situation. I haven't played around with the other
exposed properties (currentPageInfo, previousPageInfo,
previousActionInfo)
all expose the same JavaBean that is not immutable.
I'm going to open a Jiri bug on this. I think this is critical and
needs to
be fixed now. My suggestion is that we rename these methods on the
PageFlowController so they aren't picked up as JavaBean properties.
I suggest we do this to:
currentPageInfo
previousPageInfo
previousActionInfo
modeulConfig
actions
We need to spin a new release on this.
On 2/17/06, Rich Feit <[EMAIL PROTECTED]> wrote:
Well, the Struts ModuleConfig and related objects are all immutable
and
always have been. Are you seeing any other objects we expose that
aren't in our control? I agree that it's brittle -- feel free to add
your option #2 if you're worried about continued support for the
deprecated base class.
Daryl Olander wrote:
I agree we need to move to a POJO model....
I think the issue is that we expose objects like the struts config
that
are
developed independently of Beehive which may have setters which
could
open
up security holes. It's also the case that we expose object that
expose
object and underlying modification to the runtime could open up a
security
hole and no one would know. The just simply added a new feature and
exposed
a setter. This is certainly a brittle design even if we verified
all of
the
current paths.
I don't think opion #3 is viable because we still need to support
for at
least some time the current model even if it's deprecated.
On 2/17/06, Rich Feit <[EMAIL PROTECTED]> wrote:
It's not happenstance. When we still extended Struts Action, I had
workarounds in there to prevent access to dangerous base class
objects
(like getServlet()). In general I allowed public getters for
unmodifiable objects. If we're exposing something dangerous, then
it's
my fault -- it isn't just bad luck.
Access to the shared flow Map is luckily illegal when the
expressions
are being updated.
I think I *did* expose a potential security hole by not returning
Collections.unmodifiableMap () from
FlowControllerFactory.getSharedFlowsForPath() -- this needs to be
fixed. Why is access to the Map illegal currently?
I would vote for this option:
3) Verify that what's currently exposed is safe, and move to
the
POJO-pageflow model (deprecate use of the base class).
Rich
Daryl Olander wrote:
I've been looking at a possible security risk in page flows. At
the
moment,
I don't think we have an actual security hole, but I think we have
a
situation where we could create one very easy.
The issue is that there are a number of public properties on the
PageFlowController class. There are public getters that give
access
to
low
level structures. For example, you can get the ModuleConfig from
Struts,
the ActionForm, ActionServlet, the map of shared flows, etc. This
issue
arises because you can submit a form that contains a hidden field
that
would
update these data items.
<netui:form action="submit">
<netui:hidden dataSource="pageFlow.moduleConfig.prefix"
dataInput="value"/>
<netui:button value="submit" />
</netui:form>
In the above code, this could modify the Struts ModuleConfig
structure
and
set the prefix value to "value".
In fact, in looking around at this for a little while, I couldn't
find
anything you can do that is destructive. The Struts config
information
is
frozen, so the code above results in an
IllegalStateException. Access
to
the shared flow Map is luckily illegal when the expressions are
being
updated.
I think that it's purely happenstance that we are not exposing a
security
hole here. In fact, with a bit more playing round, we might find
that
we
really are exposing a hole. We need to prevent page flow updates
for
these
base class properties. There seems to be a number of ways we
could
solve
this,
1) We could prevent all update to PageFlow. This is a pretty
radical
solution because it's a backward incompatible change.
2) We could create a list of properties that can't be
updated. The
list
could be created automatically through reflection.
Right now, I would lean toward 2, but I think we should have more
discussion
of this issue.
Thoughts?
