Background
Every Yale partner uses the Browser based SAML Protocols. The application does not talk to Shibboleth directly, but instead uses your Browser as the transport mechanism. When the application (Archer, Google, Box, ...) wants you to login, it either
- (SAML Browser POST Protocol) writes a Form to the screen containing a pre-loaded text box containing the SAML Request, then uses JavaScript to Submit the data to the Shibboleth URL that was configured to the application
- (SAML Browser Redirect Procol) redirects the Brower to the configured Shibboleth URL appending "?SAMLRequest=" and the encoded Request content (so Shibboleth received a GET instead of a POST).
Either way, Shibboleth gets an HTTP request (GET or POST) containing a Request XML message identifying the application and providing more specific parameters that may override some of the Metadata information. Shibboleth saves the Request info, then Redirects the Browser to the CAS login function. CAS Redirects the Browser back to Shibboleth with a ticket= string. Shibboleth validates the ticket and obtains the Netid. It uses the Netid in various database and LDAP queries to get data, and prepares a Response object containing a Subject and various Attribute fields. It digitally signs the Response with its private key, so the application can use the distributed Yale Shibboleth Public Key to validate the signature and therefore the data in the Response.The Response is sent back to the application using the SAML Browser POST Protocol (it is put in a text box in a form and POST-ed to the application by JavaScript).
SAML has other protocols, and Shibboleth supports them, but they are not currently used at Yale and therefore do not need to be tested.
As this summary points out, the Application and Shibboleth never talk to each other directly. In all cases, they put data in the Browser (in a Form or in a QueryString at the end of the URL) and using POST or Redirect they get the Browser to send the data from the Application to Shibboleth or from Shibboleth to the Application. Since the Browser mediates between them, the Browser is the place where you can configure all your testing.
There are three plausible techniques.
- If you have a testing tool, you could use the tool to instrument HTTP requests and responses to Shibboleth and to the Application. That is a future objective, but developers do not currently have a reasonable tool for doing this type of test automation.
- The simplest option is to install add-ons to the Browser that allow you to trace the SAML traffic going back and forth, and to replace production Shibboleth in any URL ("https://auth.yale.edu/idp/...") with some new version of Shibboleth that has new code or configuration ("http://localhost:8080/idp...").
- A slightly more powerful option, but a bit more complex, is to insert the Charles Web Debugging Proxy between the Browser and the various network addresses to which it communicates (CAS, the test Shibboleth, and the Application).
Although in the long run the first option would be best, here we will describe how to install and configure options 2 and 3.
Preparation
Get Firefox. You can can certainly use other Browsers in production, but Firefox is needed to test.
Firefox Add-Ons
SAML messages are typically BIN64 encoded and appear to be the contents of a Text Box in a form or else part of the query string. You can line up a set of tools to trace, cut, paste, decode, and format the XML, or you can install the Firefox SAML Trace add on, which does all of this work for you.
A lot of debugging can be done by using Shibboleth on your desktop (Sandbox) and setting a URL starting with http://localhost:8080/idp/..." in the Browser address bar. However, final testing may require access to the actual application using a normal login sequence. This may require the use of the PREPROD Shibboleth VM, because it has credentials (the signing key) identical to production Shibboleth and can produce a Response that the application will accept. PREPROD may have a public URL address through the F5, or it may have an internal URL that can be accessed from a desktop, or it may require that you SSH login to the host and tunnel port 8080 to your desktop. However that works, you need the Redirector Firefox plugin which watches for a particular URL pattern "https://auth.yale.edu/idp/*" and then substitutes a replacement for the original URL "http://localhost:8080/idp/$1" where $1 is replaced by the rest of the original URL (after the matched prefix).
If you click the Firefox Menu icon (three horizontal lines in the upper right corner of the toolbar) then Add-On is an option (which looks like a puzzle piece). Click it.
Go to the Add-Ons, Search for a new Add-On with the word "SAML". Install the SAML Tracer. Look for "Redirector" and install the Redirector Add On (it has a logo of a capital R ending in an arrow).
Now SAML Trace appears in the same menu, and Redirector installs an icon on the toolbar itself.
SAML Tracer requires no configuration. When you turn it on it traces Web activity in a new Window and will highlight, decode, and display the XML in a SAML Request or Response on demand. You turn it off by closing the Trace Window.
Redirector requires you to create URL Remappings. What you are really doing is a Match and Replace, just like you would do in a Text editor. You can match text with either a Wildcard or Regular Expression. Wildcard is simpler and is perfectly adequate.
The first remapping is from "https://auth.yale.edu/idp/*" to "http://localhost:8080/idp/$1". Note that the Match string ends in the "*" wildcard character, so it matches all URLs that begin with the string. The Replace string ends in "$1" which is a variable that represents the data that matched the "*" wildcard. In English, this says, "Match all URLs that begin with https://auth.yale.edu/idp/ and replace those characters with http://localhost:8080/idp/, but leave the end of the URL alone.
You need a second remapping because of a problem with how the Unicon CAS-Shibboleth Integration works. It creates a service= string for CAS using (if it exists) the port number that appears in the Host header sent by the Browser. The original URL for "auth.yale.edu" had no port number, and if that URL was sent to production Shibboleth through the F5 then the Integration code sends CAS a service=https://auth.yale.edu/idp string. However, Redirector replaces that URL with a "localhost:8080" and sends a Host header to Shibboleth that contains the :8080 port. The Integration gets the hostname from a property and the port number from the Host Header. The Yale property is cas.target.url, and the PROD configuration for it is:
cas.target.url=https://auth.yale.edu
So if you are testing in the Sandbox and change this property to say:
cas.target.url=http://localhost
then the CAS-Shib integration takes this hostname and adds the port number from the Host Header and gets the right thing "localhost:8080". However, if you are using an SSH tunnel to a VM in the machine room, and the VM has the production value for the property, now CAS-Shib takes the "https://auth.yale.edu" piece from the property and appends the port number from Host header and gets the strange value "https://auth.yale.edu:8080", which is neither the right URL for the VM nor the right URL for the desktop. Because the previous Redirector configuration only matched "auth.yale.edu/idp", once the :8080 got inserted it will not match and will not get rewitten. So you need a second URL Remapping to handle this case. This one maps "https://auth.yale.edu:8080/idp/*" to "http://localhost:8080/idp/$1"
Now when CAS redirects the Browser back using the URL from the service= string, the Browser will also send the ticket string to the test Shibboleth.
Server Access
If you run Shibboleth under Tomcat on your Sandbox desktop, then it is http://localhost:8080/idp. No setup is required.
This is also the URL if you create an SSH tunnel from local port 8080 to vm-shibxxx-01.its.yale.internal port 8080, but to use it you need to use VPN and SSH.
The PREPROD VM is in a machine room behind a firewall that does allow direct access from desktop computers. In normal production, only HTTP traffic gets to VMs through the F5. If the F5 has a mapping to the PREPROD VM, then you can configure Redirector to just use that public URL and you are done. However, in the new DevOps world the common practice is to put new code on new VMs that may not yet have been configured to the F5. In this case, you need to access the machine room indirectly through a VPN.
First, you need to use the standard Cisco AnyConnect client to establish a VPN session, but instead of going to the public "access.yale.edu" target that normal Yale people use from off campus, you need to use one of the special VPN targets reserved for ITS staff on campus to access a part of the network from which they can connect to VMs in the machine room. Someone will tell you the VPN target name, explain how to download the AnyConnect profile for that VPN, and put you in the AD group that has access to that VPN target name.
After you have made the VPN connection, the next step is to use your preferred SSH Client to login to the VM. Operations must have created a login for your Netid on the VM and installed your SSH public key. In addition to the terminal session on the VM, the SSH client can be configured to "tunnel" one or more port number from your desktop computer to the VM. Since Shibboleth runs on Tomcat and the default Tomcat port number is 8080, you configure the SSH client to tunnel 8080 on your computer to 8080 on the VM. Now when you browse to "http://localhost:8080" the SSH Client forwards the traffic through the SSH session (and through the VPN) to Tomcat running on the VM in the machine room.
Only one program can use port 8080 on your computer at a time. When you test Shibboleth on your local Sandbox, it also uses 8080. Using the same local port number for both the Sandbox and the SSH tunnel will generate an error message if you accidentally run both at the same time. SSH will generate error messages that it cannot create the tunnel if you have forgotten to shut down the Sandbox Tomcat, and Tomcat will generate error messages that it cannot bind to the port if you forget forget to shut down SSH before starting the Sandbox. This is a feature, because you really don't want to spend hours trying to figure out what is wrong only to discover that you are debugging the wrong Shibboleth server.
Charles Web Debugging Proxy
Things get more complicated when you want to do final testing with real applications that do not support IdP Initiated logon. You have to go to the application first, and it is configured with the URL of the real Shibboleth production server (https://auth.yale.edu/idp). It will generate a Request and send it to that URL using your Browser to transport the message. So somehow you need to intercept the Request and send it "PREPROD", the test Shibboleth with the new configuration or code but also with the production Shibboleth credentials so it can generate a Response that the application will accept.
All communication between the application and Shibboleth go through the Browser on your computer. It might be possible to intercept the message inside the Browser, or to create a test program that does HTTP and replaces the Browser. Both would be a major coding project.
Instead, this problem can be solved with Charles Web Debugging Proxy, a light weight debugging tool written in Java that runs on Windows, Mac, and Linux. We may have a some licenses available, or an extra license is not expensive and can be provided by the department before your 30 day free trial expires. Go to charlesproxy.com, download and install it. You also want to install the Charles plug-in for Firefox in the same way you installed the SAML Tracer.
A Web Proxy is a program that sits between the Browser and the Web Server. In the old days with a slower Internet, proxies cached files to speed up browsing. Today the F5 acts as what is called a "reverse proxy", where it appears to the network to be the named Web Server ("auth.yale.edu" in this case) but then it forwards the request to other computers in the machine room that do the real work. For our purposes, Charles acts as a private F5 that we can configure to pretend to be the "auth.yale.edu" server (from the point of view of the Firefox browser on the desktop test machine). Like the F5 it forwards the request to a configured VM, but in this case instead of sending the message to the production VM it sends it to the PREPROD test machine.
Charles only intercepts data while you run it. Close Charles and everything behaves normally. Charles only intercepts the traffic you configure, and it runs locally on your desktop and only intercepts traffic from your Browser. Generally you only run it during testing, and when you are running it you only generate test related traffic.
For this test, you need to start specific services and make specific configuration entries. Each performs a specific function, and you have to do all of them or the test won't work. This is sufficiently complicated that I will explain what each step does, how to do it, and maybe what you will see if you do it wrong.
Start Charles. Start Firefox. When Charles is running, Firefox regards it as its Web Proxy and sends all Internet traffic to Charles.
Suppose you tell the Browser to go to production Shibboleth ("https://auth.yale.edu/idp/...").
Because the URL starts with "https", Firefox is going to establish an SSL/TLS session and encrypt all the data it sends. It expects to be talking to a host named "auth.yale.edu" and part of the SSL session setup requires that host to prove its identity with an X.509 Certificate that says it is "auth.yale.edu". If Charles has no configuration at all, it will allow the data to pass through it to the real host with that name, and although it will be in the middle of the data transfer, all the data will be encrypted and any data logged by Charles will be unreadable.
Charles needs to decrypt the data. The only way it can do that is (like the F5) it has to convince Firefox that it is the host named "auth.yale.edu", and to do that it needs to provide its own Certificate. It would be a violation of security practices to give Charles the real X.509 Certificate that the F5 has (because if that file is compromised then any bad guy can pretend to be Shibboleth). In this case, we do not need a Certificate that everyone trusts. We only need a Certificate that the Firefox Browser on our local desktop computer trusts as "auth.yale.edu". Since the Browser is under our control, that is easy to accomplish.
When Charles is installed, it generates a local self-signed Certificate for itself and uses it to create a mini Certificate Authority (CA). In the Charles menu, you select "Proxy" and then "SSL Proxying" from the pulldown list. Click the "Enable SSL Proxying" box and then add an SSL hostname of "auth.yale.edu" to the list box. Charles internally generates a Certificate for"auth.yale.edu" created by its internal Certificate Authority.
You have also just told Charles to intercept any HTTP traffic issued by your Firefox browser for a URL that begins with "https://auth.yale.edu/..." and to send back to Firefox the dummy Certificate issued by the internal Charles CA for hostname "auth.yale.edu". The Charles CA will not be in the list of real commercial Certificate Authorities that Firefox is distributed by Mozilla to automatically trust. So when it gets the dummy Certificate from Charles, Firefox displays a Warning page saying that the Web server certificate is not from a recognized Authority. You can click on the message page and tell Firefox to configure an Exception and trust this Certificate. It is convenient to tell Firefox to trust it from now on, and then you only get the Warning page the first time.
Now there is an SSL session inside your desktop computer between Firefox and Charles (acting as its Web proxy). Firefox encrypts data and Charles decrypts it. If this was all you configured, Charles would establish a second SSL connection between it and the real "auth.yale.edu" endpoint (the F5) and simply forward messages between Firefox and the F5, although now that it can decrypt the data it can log the information flowing in both directions. This is no big deal because SAML contains mostly biographical information, and you already know your own first name and last name, and really sensitive information like passwords do not go to Shibboleth.
However, we want to do something different. We want to take the data that was originally going to production Shibboleth and reroute it so it goes the the PREPROD VM with new code or new configuration. This is a second step where we now tell Charles to send the data for "auth.yale.edu" to a substitute URL address.
Before we configure the rerouting, we have to get access to the PREPROD VM. It is in a machine room behind a firewall that does allow direct access from desktop computers. In normal production, traffic gets to VMs through the F5. If the F5 has a mapping to the PREPROD VM, then you can configure Charles to just use that public URL and you are done. However, in the new DevOps world the common practice is to put new code on new VMs that may not yet have been configured to the F5. In this case, you need to access the machine room indirectly through a VPN.
First, you need to use the standard Cisco AnyConnect client to establish a VPN session, but instead of going to the public "access.yale.edu" target that normal Yale people use from off campus, you need to use one of the special VPN targets reserved for ITS staff on campus to access a part of the network from which they can connect to VMs in the machine room. Someone will tell you the VPN target name, explain how to download the AnyConnect profile for that VPN, and put you in the AD group that has access to that VPN target name.
After you have made the VPN connection, the next step is to use your preferred SSH Client to login to the VM. Operations must have created a login for your Netid on the VM and installed your SSH public key. In addition to the terminal session on the VM, the SSH client can be configured to "tunnel" one or more port number from your desktop computer to the VM. Since Shibboleth runs on Tomcat and the default Tomcat port number is 8080, you configure the SSH client to tunnel 8080 on your computer to 8080 on the VM. Now when you browse to "http://localhost:8080" the SSH Client forwards the traffic through the SSH session (and through the VPN) to Tomcat running on the VM in the machine room.
Only one program can use port 8080 on your computer at a time. When you test Shibboleth on your local Sandbox, it also uses 8080. Using the same local port number for both the Sandbox and the SSH tunnel will generate an error message if you accidentally run both at the same time. SSH will generate error messages that it cannot create the tunnel if you have forgotten to shut down the Sandbox Tomcat, and Tomcat will generate error messages that it cannot bind to the port if you forget forget to shut down SSH before starting the Sandbox. This is a feature, because you really don't want to spend hours trying to figure out what is wrong only to discover that you are debugging the wrong Shibboleth server.
Now we need to configure Charles to map the production Shibboleth URL (https://auth.yale.edu/idp) through that tunnel to the VM.
In the Charles menu, select "Tools", and then "Map Remote" from the pulldown list. Click the "Enable Map Remote" box and then Add a mapping.
The Map From part of the mapping provides data that must match a URL sent from your Firefox browser to Charles. In this case the Protocol is "https" and the Host is "auth.yale.edu". Generally you leave the other fields (port, path, query) blank and they default to matching anything.
The Map To part of the mapping specifies the changes you want to make to the incoming URL. In this case, you want to change the Protocol to "http" and the host to "localhost". You also want to change the Port to "8080". Leaving the other fields blank means that the path (/idp/...) will be copied from the incoming URL to the outgoing connection.
In certain unusual situations the "context" name of Shibboleth on the VM might not be "idp". Then you will want to configure the path field of the Map To to be the odd context name, but this is not the normal case.
If the F5 has a public URL for the VM you are trying to access, then you can forget the VPN and the SSH part. Just configure the Map To with the virtual host name on the F5 for the VM you want to access and Charles will send the data to the F5 instead of localhost:8080.
There is one last step. Click the "Preserve Host Header" box. When Firefox generated its request, it sent a Host header with the "auth.yale.edu" hostname and the original port. Then it went to Charles, who sent it through "localhost:8080" to a VM with some odd vm-shibxxx-01.yale.internal name. Unless the Host header is preserved, the VM has no way to know what the original host and port were. It turns out that this is important for the Unicon CAS Shibboleth integration, because it uses the Host header to get the port number to send in the service= string. The CAS Service has to be a URL that makes sense from the Browser's point of view, but when the CAS-Shibboleth Integration is building it, it is running on the VM in the machine room. The only way to know what URL the Browser used is to examine original Host header the Browser sent, which means that Charles has to send it on to the server unchanged. If you don't do this, then CAS gets a service string of the form "https://auth.yale.edu:8080/idp" which is not going to work.
Now I want to walk through the process end to end. For this example, suppose you start by pointing Firefox to the "yale.box.com" URL. Charles is still the proxy, but since it has no configuration for this URL, it sends the HTTP Request on to the real Box server on the internet. You press the "I am from Yale" button and Box sends back a SAML Request and tells Firefox to forward the request to production Shibboleth, which Box knows to be "https://auth.yale.edu/idp".
The Firefox Browser sends this URL and data to Charles (its Web proxy). Because Charles has "auth.yale.edu" in its list of SSL Proxy hosts, Charles acts as the server end sending its dummy Certificate and decrypts the data. Because "https://auth.yale.edu" is also in its Map Remote list, Charles substitutes "http" for "https" and "localhost:8080" for "auth.yale.edu" and sends the data to local port 8080 on your machine. Because that is the local end of an SSH tunnel, SSH now sends the data to the VM in the machine room where it is sent to that port 8080, which is the Tomcat running PREPROD Shibboleth. The Shibboleth response goes back through the same path in the opposite direction and comes back to Firefox.
If you have to CAS login, then the Unicon CAS Shibboleth integration running on the VM in the machine room gets the "https://auth.yale.edu" hostname part from its properties file, and it gets the port (which is empty/default) from the HTTP Host header, and it Redirects Firefox to "https://secure.its.yale.edu/cas/login?service=https://auth.yale.edu/idp..." CAS returns the ticket to the URL in the service string, and since that URL is also https://auth.yale.edu Charles does the whole SSL Proxy + Map To thing all over again and sends the ticket back to the PREPROD Shib VM, which logs you on to Shibboleth. Shibboleth builds a SAML Response and sends it back to Firefox along with code to forward the SAML to the login URL of box.com.
Because PREPROD Shibboleth has the same credentials/idp.key file as PROD Shibboleth (and the same configured entityID), the SAML message sent to box.com appears to be from production Shibboleth and the digital signature matches one from production Shibboleth. So box accepts the SAML and logs you on. Of course the actual attributes and Subject came from the PREPROD configuration, so if you screw up and don't generate the right data then box will not log you in and you fix the PREPROD configuration and rerun the test.
The techniques used by Charles are similar to exploits used by some malware. The difference is that Charles only functions when you explicitly run it and it only decodes traffic for hosts you configure it to proxy. If you accidentally leave it running and do some banking, then since bankofamerica.com is not in any of its configuration lists the SSL encrypted data remains secure and no sensitive information is exposed, even to other windows on your desktop. If you use it to debug CAS, then close it when you are done and don't save files that contain your Netid password.