My plan was unclear! Let's go with another example:
Step 1 - the Gateway receives the following HTTP request:
----
POST /apiman-gateway/MyOrg/soap-api/2.7 HTTP/1.1
Host:
Content-Type: application/soap+xml; charset=utf-8
X-API-Key: API-KEY-FOR-ACTIVE-CLIENT
SOAPAction: ExampleAction
<?xml version="1.0"?>
<soap:Envelope
<soap:Header>
<ns1:Header1 xmlns:ns1="uri:ns1">foo</ns1:Header1>
<ns2:Header2 xmlns:ns2="uri:ns2">bar</ns2:Header2>
</soap:Header>
<soap:Body>
<m:GetStockPrice
<m:StockName>Red Hat</m:StockName>
</m:GetStockPrice>
</soap:Body>
</soap:Envelope>
----
We'll parse the first part of the envelope so that we can read the
headers and make them available to any policies. After that's done,
we'll invoke the policy chain as per normal. However, because it's a
SOAP api, there will exist a SOAPRequestInfo object in the policy
context. So policies can read and/or modify the soap information. This
class might look something like this:
public class SOAPRequestInfo {
private String action;
private String encoding;
private Map<QName, SOAPHeader> headers;
}
This allows interested policies (like your ws-security) policy to have
easy access to all the soap related stuff. It also allows you to alter
these things. Including adding/removing/modifying the SOAP headers.
So let's assume that we have a policy which *adds* a SOAP header
(ns3:AddedHeader) and another policy which *removes* one (ns2:Header2).
The policy code might look like this:
SOAPRequestInfo soapInfo = context.getAttribute(
Constants.SOAP_INFO, (SOAPRequestInfo) null);
soapInfo.getHeaders().remove(new QName("uri:ns2", "Header2"));
soapInfo.getHeaders().put(
new QName("uri:ns3", "AddedHeader"),
createSoapHeader()
);
In that case, this is the HTTP request that will be sent/proxied to the
back-end API:
----
POST / HTTP/1.1
Host:
Content-Type: application/soap+xml; charset=utf-8
SOAPAction: ExampleAction
<?xml version="1.0"?>
<soap:Envelope
<soap:Header>
<ns1:Header1 xmlns:ns1="uri:ns1">foo</ns1:Header1>
<ns3:AddedHeader xmlns:ns3="uri:ns3">bar</ns3:AddedHeader>
</soap:Header>
<soap:Body>
<m:GetStockPrice
<m:StockName>Red Hat</m:StockName>
</m:GetStockPrice>
</soap:Body>
</soap:Envelope>
----
Sound reasonable?
-Eric
On 3/29/2016 1:42 PM, Benjamin Kastelic wrote:
If I understand you correctly Eric, only the soap:Body part will be
forwarded to the registered API?
What if the API still needs to receive the soap:Header part?
2016-03-29 15:34 GMT+02:00 Keith Babo <kbabo(a)redhat.com
<mailto:kbabo@redhat.com>>:
Sounds cool to me. Header policy will need to be QName and
DOM-aware, so that namespace-qualified headers can be added and
appropriate namespace definitions can be added to the DOM if
required. Of course you already know all this, but pointing it out
makes me feel useful.
> On Mar 29, 2016, at 8:38 AM, Eric Wittmann
<eric.wittmann(a)redhat.com <mailto:eric.wittmann@redhat.com>> wrote:
>
> That's an interesting idea. It'd be harder to inject the headers
into the request than it is to inject HTTP request headers,
obviously. But not impossible. We'd need to be aware of the change
to any Content-Length that may be set.
>
> This makes the implementation slightly more complicated, because
I think the HTTP connector will need to be made smarter (it will
need to send the <soap:Envelope> and <soap:Header> sections first,
then just stream the remaining request body as-is.
>
> So perhaps what we have is this:
>
> 1. <?xml version="1.0"?>
> 2. <soap:Envelope
xmlns:soap="http://www.w3.org/2003/05/soap-envelope">
> 3. <soap:Header>
> 4. <ns1:Foo actor="..."
xmlns:ns1="uri:ns1">BAR</ns1:Foo>
> 5. </soap:Header>
> 6. <soap:Body>
> 7. <m:GetStockPrice
xmlns:m="http://www.example.org/stock/Surya">
> 8. <m:StockName>IBM</m:StockName>
> 9. </m:GetStockPrice>
> 10. </soap:Body>
> 11. </soap:Envelope>
>
> Lines 1-5 will be read and consumed by apiman *before* any
policies are executed. We'll actually keep reading until we find
<soap:Body> in the request body. We'll throw out everything before
line #6 from our in-memory buffer, resulting in a buffer with just
line #6 (and any extra bytes after that based on our I/O chunk size).
>
> The policies will be executed, which may result in soap headers
being added, modified, or removed. If all policies pass, then we
proxy the request to the back-end. Normally the HTTP connection
would simply send all bytes from the HTTP request as-is. Instead,
we'll need to *generate* new content for lines 1-5, with the newly
modified soap headers. Once the generated content is sent, then we
send the contents of the in-memory buffer (which contains line #6+
any additional bytes). After that, we proxy the remaining bytes
from the HTTP request as-is.
>
> This may be starting to take shape. :)
>
> Additional thoughts?
>
> -Eric
>
> PS: @Keith - we'll likely have a separate Policy for
manipulating SOAP headers, rather than re-use the existing HTTP
headers policy.
>
> On 3/29/2016 8:13 AM, Keith Babo wrote:
>> Sounds like a reasonable first step to me. Just to make life
slightly
>> more complicated, will the headers policy be updated to allow
add/remove
>> of SOAP:Headers? :-)
>>
>> ~ keith
>>
>>> On Mar 29, 2016, at 7:52 AM, Eric Wittmann
<eric.wittmann(a)redhat.com <mailto:eric.wittmann@redhat.com>
>>> <mailto:eric.wittmann@redhat.com
<mailto:eric.wittmann@redhat.com>>> wrote:
>>>
>>> OK so here's what I propose (feedback welcome):
>>>
>>> *If* an API's 'type' is set to SOAP, then we will *always*
look
for a
>>> soap envelope in the body. If no body is found or no soap
envelope is
>>> found in the body, then a standard apiman error will be thrown.
>>>
>>> If an envelope *is* found, then we will read the body of the HTTP
>>> request until we find "<soap:Body>". We'll extract
the
<soap:Header>
>>> and parse its children. While parsing, we'll obviously keep
the data
>>> we read in a memory buffer. Once parsing is done, we'll
include the
>>> soap headers, soap action, and the global encoding type in some
sort
>>> of soapinfo object and include that in the policy context.
>>>
>>> Finally, after all that is done, we'll process the request as
normal,
>>> executing the policy chain, then processing the request body,
etc. The
>>> entire request payload will still be processed (remember that
we saved
>>> the bytes we read in a memory buffer).
>>>
>>> So from the perspective of a policy, everything will look identical
>>> except that a SOAPInfo object will be available in the policy
context.
>>>
>>> Thoughts?
>>>
>>> -Eric
>>>
>>> On 3/28/2016 1:48 PM, Benjamin Kastelic wrote:
>>>> Yup, I agree. That would probably be best, since several
validators
>>>> (wss4j for example) require DOM Elements
(javax.xml.soap.SOAPHeader) to
>>>> function.
>>>>
>>>> Best regards,
>>>> Benjamin
>>>>
>>>> 2016-03-28 19:14 GMT+02:00 Eric Wittmann
<eric.wittmann(a)redhat.com <mailto:eric.wittmann@redhat.com>
>>>> <mailto:eric.wittmann@redhat.com
<mailto:eric.wittmann@redhat.com>>
>>>> <mailto:eric.wittmann@redhat.com
<mailto:eric.wittmann@redhat.com>>>:
>>>>
>>>> Thanks! In that case, making the headers available as DOM
Element
>>>> objects (perhaps with a simple QName based lookup) would be
best.
>>>>
>>>> -Eric
>>>>
>>>> On 3/28/2016 12:39 PM, Keith Babo wrote:
>>>>
>>>> SOAP:Headers can be complex types. WS-Security is a good
>>>> example of
>>>> this in practice.
>>>>
>>>> ~ keith
>>>>
>>>> On Mar 28, 2016, at 11:37 AM, Eric Wittmann
>>>> <eric.wittmann(a)redhat.com
<mailto:eric.wittmann@redhat.com>
>>>> <mailto:eric.wittmann@redhat.com
<mailto:eric.wittmann@redhat.com>><mailto:eric.wittmann@redhat.com
<mailto:eric.wittmann@redhat.com>>
>>>> <mailto:eric.wittmann@redhat.com
<mailto:eric.wittmann@redhat.com>
>>>> <mailto:eric.wittmann@redhat.com
<mailto:eric.wittmann@redhat.com>>>> wrote:
>>>>
>>>> That's a bit hacky, but also sort of a genius
approach as
>>>> well. I'm
>>>> actually a little bummed I didn't think of it. :)
>>>>
>>>> As for extending SOAP support - I was thinking that
I could
>>>> make the
>>>> relevant changes to apiman if you would be willing
to provide
>>>> feedback/guidance/testing. My SOAP expertise is
quite stale
>>>> at this
>>>> point, so having some eyeballs on these changes would be
>>>> very useful.
>>>>
>>>> To start off with, what pieces of the SOAP envelope
should
>>>> be extracted
>>>> prior to calling the policy chain? Some candidates are:
>>>>
>>>> * The encoding style
>>>> * All SOAP headers
>>>> * SOAPAction (already available as an HTTP header)
>>>> * ???
>>>>
>>>> For the soap headers, all of the examples I've seen
take the
>>>> following
>>>> form:
>>>>
>>>> <HeaderName
xmlns="elementNS">Header-Value</HeaderName>
>>>>
>>>> It can also have the optional "actor" or
"mustUnderstand"
>>>> attributes.
>>>> The SOAP envelope schema is pretty lax though, so
I'm not
>>>> sure if the
>>>> above is a convention or a rule. Can headers be more
>>>> complex than the
>>>> above?
>>>>
>>>> -Eric
>>>>
>>>> On 3/26/2016 7:06 AM, Benjamin Kastelic wrote:
>>>>
>>>> Hi,
>>>>
>>>> I temporarily solved the problem by storing the
request
>>>> body, which is
>>>> contained in ApiRequest.rawRequest object, in a
>>>> temporary buffer. I then
>>>> process the data (authentication) and based on the
>>>> results proceed with
>>>> the policy chain or report a failure. Then in
the end()
>>>> method of the
>>>> requestDataHandler method I write the contents of my
>>>> temporary buffer
>>>> using super.write(IApimanBuffer). That way I can
forward
>>>> the request to
>>>> then ext policy in the chain. But this is still
a hacky
>>>> way of doing
>>>> this.
>>>>
>>>> I would be glad to help with extending SOAP
support. But
>>>> I would need a
>>>> few pointers where to start. The way of storing SOAP
>>>> headers in the
>>>> ApiRequest object seems like a good idea.
>>>>
>>>> 2016-03-24 18:40 GMT+01:00 Eric Wittmann
>>>> <eric.wittmann(a)redhat.com
<mailto:eric.wittmann@redhat.com>
>>>> <mailto:eric.wittmann@redhat.com
<mailto:eric.wittmann@redhat.com>><mailto:eric.wittmann@redhat.com
<mailto:eric.wittmann@redhat.com>>
>>>> <mailto:eric.wittmann@redhat.com
<mailto:eric.wittmann@redhat.com>
>>>> <mailto:eric.wittmann@redhat.com
<mailto:eric.wittmann@redhat.com>>>
>>>> <mailto:eric.wittmann@redhat.com
<mailto:eric.wittmann@redhat.com>
>>>> <mailto:eric.wittmann@redhat.com
<mailto:eric.wittmann@redhat.com>>>>:
>>>>
>>>> Hi Benjamin - thanks for the excellent
question. I
>>>> will do my best
>>>> to answer and note that I am CC'ing the
apiman-dev
>>>> mailing list so
>>>> others can chime in.
>>>>
>>>> First let me say that a WS-Security policy
sounds
>>>> great - we haven't
>>>> focused much on the WS-* protocols because
we get
>>>> much more demand
>>>> for managing REST APIs than SOAP APIs. That
said,
>>>> better SOAP
>>>> support is certainly on the radar. When that
>>>> happens, my hope is
>>>> that processing the envelope might be a core
part of
>>>> the gateway and
>>>> so implementing policies that use information in
>>>> there will be
>>>> easier. Perhaps your implementation can be the
>>>> genesis of some of
>>>> that work!
>>>>
>>>> To your question - without core changes to
apiman,
>>>> the approach you
>>>> *need* to take is to have your policy implement
>>>> IDataPolicy. I
>>>> believe you may have already tried that, and
>>>> observed that you
>>>> cannot send proper policy failures from that
>>>> method. You are right
>>>> - that's something we will need to fix! I
think you
>>>> should be able
>>>> to throw a runtime exception from the
>>>> write(IApimanBuffer chunk)
>>>> method if you detect an error. However,
this is a
>>>> little bit hacky!
>>>>
>>>> Instead, I suggest (if you're up for it) that
we
>>>> perhaps work
>>>> together to bake SOAP support directly into
the core
>>>> of apiman, such
>>>> that the SOAP envelope is read/parsed
*before* the
>>>> policy chain is
>>>> executed. We could expose, for example, the
SOAP
>>>> headers as a
>>>> proper Map<> stored either in the context
or
on the
>>>> ApiRequest.
>>>> This would allow you to properly implement most
>>>> (all?) WS-*
>>>> protocols as proper apiman policies in the
>>>> apply(ApiRequest request)
>>>> method.
>>>>
>>>> Thoughts?
>>>>
>>>> -Eric
>>>>
>>>>
>>>> On 3/24/2016 7:58 AM, Benjamin Kastelic wrote:
>>>>
>>>> Greetings,
>>>>
>>>> I first thought to write this question as an
>>>> issue on Github,
>>>> but it
>>>> seemed better to write you a direct email.
>>>>
>>>> I am making a custom WS Security policy,
that
>>>> reads the body and
>>>> check
>>>> the UsernameToken security header. This
works
>>>> OK, but now I've
>>>> hit a wall.
>>>>
>>>> In the doApply method I get the rawRequest
>>>> object and read the
>>>> body from
>>>> the ServletInputStream of the request. The
>>>> problem I'm facing
>>>> now is
>>>> that the input stream was read and it
can't be
>>>> reset back to it's
>>>> initial state.
>>>>
>>>> I was also trying to implement the same
logic
>>>> in the
>>>> requestDataHandler
>>>> method, but I don't know if it is even
possible
>>>> to send a failure
>>>> message to the request chain from there.
>>>>
>>>> Any suggesstions ?
>>>>
>>>> Best regards,
>>>> Benjamin
>>>>
>>>>
>>>>
>>>>
>>>> --
>>>> Lp, Benjamin
>>>>
>>>> _______________________________________________
>>>> Apiman-dev mailing list
>>>> Apiman-dev(a)lists.jboss.org
<mailto:Apiman-dev@lists.jboss.org>
<mailto:Apiman-dev@lists.jboss.org <mailto:Apiman-dev@lists.jboss.org>>
>>>> <mailto:Apiman-dev@lists.jboss.org
<mailto:Apiman-dev@lists.jboss.org>>
>>>> <mailto:Apiman-dev@lists.jboss.org
<mailto:Apiman-dev@lists.jboss.org>
>>>> <mailto:Apiman-dev@lists.jboss.org
<mailto:Apiman-dev@lists.jboss.org>>>
>>>>
https://lists.jboss.org/mailman/listinfo/apiman-dev
>>>>
>>>>
>>>>
>>>>
>>>>
>>>> --
>>>> Lp, Benjamin
>>
--
Lp, Benjamin