vCloud Director, VMware

CA-signed vCloud Director certificates no longer trusted – SAN missing

Hello again! Today’s adventures drove me a little wild…

Some background first. In my test environment, I have a full vCloud Director v8.10.1 deployment, load balanced with an F5 LTM. The certificates are loaded on the F5 so that traffic is terminated and re-encrypted on it’s way to the vCloud cells. Since deployment, both the http and console FQDNs functioned as expected. This all changed just a few months ago…

Some users of the test vCloud deployment reported certificate validation errors when accessing the ‘http’ site for vCloud. They also reported that the console had stopped working. I checked the certificates’ validity periods and they seemed OK. I spoke to the networking team and they confirmed the certificates on the F5 were also A-OK.

I dug deeper into the certificate validity message in Chrome and found this:

san-missing

I looked around online and found that starting with Chrome v58 and Firefox v48 (only source I could find), support for SSL certificates without Subject Alternative Names had been deprecated.

This is very interesting! Why would this issue happen to this environment, when I’m almost certain that SAN attributes are included as part of the VMware doco. In fact, I’m definitely certain it’s there…

keytool 
   -keystore certificates.ks
   -alias consoleproxy 
   -storepass passwd
   -keypass passwd
   -storetype JCEKS
   -genkeypair
   -keyalg RSA
   -keysize 2048
   -validity 365 
   -dname "CN=vcd2.example.com, OU=Engineering, O=Example Corp, L=Palo Alto S=California C=US" 
   -ext "san=dns:vcd2.example.com,dns:vcd2,ip:10.100.101.10"

Yup, there it is.

The linked VMware doco will step you through generating a keypair in the form of a self-signed certificate and private key into a keystore you specify (one will be created if it does not already exist). The SAN attribute specified in the example above will go into this self-signed certificate. However, if you attempt to create a CSR from this self-signed certificate using the instructions from VMware, you will be left with a CSR with no SAN attributes.

You can check the CSR yourself by running:

openssl req -in {csr-file} -noout -text

You’ll see that there are no Subject Alternative Names specified. Without knowing this the first time around, I submitted this newly generated CSR to my internal Microsoft CA. While the certificate was issued successfully, none of the SAN attributes had been included.

This is due to VMware leaving a very important switch off the CSR generation command to make sure that SAN attributes are included in the CSR.

To get the SAN attributes included in the CSR, you’ll need to modify VMware’s example from the doco. Instead of running this command to generate the CSR from the self-signed cert:

keytool -keystore certificates.ks -storetype JCEKS -storepass {password} -certreq -alias http -file http.csr

You’ll want to add your SAN attributes to the keytool certreq command so it looks like this:

keytool -keystore certificates.ks -storetype JCEKS -storepass {password} -certreq -alias http -file http.csr -ext SAN=dns:vcd2.example.com,dns:vcd2,ip:10.100.101.10

Huge credit to Eric Lawrence from textslashplain for sending me down the rabbit hole. Even bigger credit to StackOverflow user MrPatol for basically spelling out the fix (original SO thread here).

vCenter, vCloud Director, VMware

SYSTEM_REFRESH_VIMSERVER – Could not register vCloud Director as an extension to vCenter Server

While trying to troubleshoot another problem, we tried Refreshing vCloud to vCenter which includes registering/updating the extension. This is when we hit a beauty we’d never seen before:

vimserver refresh

Alright, calm down. Probably something with the network, right? And  if it’s not the network then it’s probably DNS. Right? Wrong.

I dug around in the vCenter MOB and found the vCloud Director extension. As expected it already had a “vCloud Director-1” named extension. What I found odd was the last heartbeat time was back in 2013. Interestingly enough the last version recorded was also v5.1.2. I say interestingly because we are running v8.10.1 for SP.

Jumping into our test environment, I performed a Refresh of our test vCloud instance to vCenter and lo and behold it happened there too! I couldn’t find anything in the vCloud logs reporting the why behind this failure, but I needed to get this running and quick, too.

Knowing that the vCloud DB stores its own references to the vCenter MOB, and that vCloud would try to register itself as vCloud Director-1 again, I theorised that we could remove the existing extension and perform another Refresh without causing any issues.

So, that’s what I did right in the test environment. It went without a hitch. Rolled the same change out in production and it went beautifully.

If you’re getting this error, I’d suggest taking a backup of your vCenter server/DB and removing the existing vCloud Director extension.

Removing the extension (from KB1025360):

  1. In a web browser, navigate to http://vCenter_Server_name_or_IP/mob.
    Where vCenter_Server_name_or_IP/mob is the name of your vCenter Server or its IP address.
  2. Click Content.
  3. Click ExtensionManager.
  4. Select and copy the name of the plug-in you want to remove from the list of values under Properties. For a list of default plug-ins, see the Additional Information section of this article.
  5. Click UnregisterExtension. A new window appears.
  6. Paste the key of the plug-in and click Invoke Method. This removes the plug-in and results in void.
  7. Close the window.
  8. Refresh the Managed Object Type:ManagedObjectReference:ExtensionManager window to verify that the plug-in is removed successfully

Now go back to vCloud and perform a Refresh against your vCenter server. You should be back in action now!

vCloud Director, VMware

‘java.lang.NullPointerException’ received when modifying objects in vCloud Director

Problem

Roughly 2 weeks ago one of our vCloud Director tenants reported an error when attempting to increase a disk on their VM. They were told to contact their cloud administrator (yay). When we tried to perform the increase, we received an error we’d never seen before: “java.lang.NullPointerException”.

javanull

Here is what we checked:

  1. Confirm the tenant Org vDC has the appropriate resources available (this was an ‘Allocation’ style vDC).
  2. Check the status of vCloud to vCenter connection and perform a vCenter Reconnect followed by a Refresh. This actually exposed another issue written about here.
  3. Check Log Insight for entries similar to this. We found the entries, but even after viewing the log in context we couldn’t find a cause or correlated action.
  4. We tested the same changes against other Org vDCs. We found that newly created test Org vDCs were fine and unaffected by whatever the root issue was. Only some of our existing Org vDCs experienced these issues.
  5. We found that it wasn’t limited to disk changes. Performing any action on infrastructure within these affected Org vDCs resulted in the same error.

We spent longer than we should’ve trawling through the vCloud logs, and ended up logging an SR with VMware.

After going through the intricate details of what we were experiencing and the testing we’d performed, VMware requested a copy of our vCloud databases and a list of affected VMs, vApps and Org vDCs.

It was a nail biting few days, but eventually our assigned tech got back to us and found the cause of this error. There was a stale resgroup-id record in the org_prov_vdc table. Let me explain…

Cause

There are 3 tables in the vCloud Director database that track Org vDC entities and their corresponding Org vDC resource pool and the parent vCenter resource pool. These 3 are supposed to be kept up to date/in sync by vCD:

  • vrp – this tracks Org vDC resource pool names in vCenter that correspond to Org vDCs. It also tracks the resource model (allocated, PAYG) and any compute resource settings that are applied to the resource pool.
  • vrp_rp – stores the vrp_id from the vrp table, along with the sub_rp_moref value for that vrp.
  • org_prov_vdc – stores data related more to the Org vDC entity itself (name, description, network pools, VM folder Moref IDs, resource pool Moref IDs etc)

Notice the bolded resource pool Moref IDs” comment above. This is important, as this value should be the same as what’s stored in the sub_rp_moref column in the vrp_rp table.

Disclaimer: all of the steps below were performed with VMware support.

You can find out if a particular Org vDC has a stale record in the org_prov_vdc table by performing the following queries against your vCloud Director database:

SELECT id FROM vrp WHERE name LIKE '%My Org vDC Name%'

This will return the vrp ID for your Org vDC. Replace the bold ID in the following query with the ID you received in the last step:

SELECT sub_rp_moref FROM vrp_rp WHERE vrp_id = 0x31JSD81AA0923NAFV801234UASD2BF76

Note; this ID has been changed to protect the innocent. Yours will differ.

From the above SELECT query, you’ll get a resource group ID similar to this:

resgroup-6000

Run this query to find out what the current value is in org_prov_vdc table. Make sure to change the Org vDC name inside the percent signs.

SELECT sub_rp_moref FROM org_prov_vdc WHERE name LIKE '%My Org vDC Name%'

If the values from the vrp_rp table and the org_prov_vdc table do not match, then you’ve got a stale moref in the org_prov_vdc table.

Fix

To fix this stale record take the resgroup ID, and the Org vDC name and run the following query:

UPDATE org_prov_vdc SET sub_rp_moref = 'resgroup-6000' where name = 'My Org vDC Name'

All done. You should now be able to make changes to your vCloud objects.

If you’d like to find all Org vDCs with stale moref IDs in the database, I’ve written a small query that can do that for you:

SELECT vrp.name as vrpNAME, org_prov_vdc.name as orgprovNAME, vrp_rp.sub_rp_moref as correctMOREF, org_prov_vdc.sub_rp_moref as staleMOREF
FROM vrp
 JOIN org_prov_vdc
 ON vrp.name LIKE '%' + org_prov_vdc.name + '%'
 JOIN vrp_rp
 ON vrp_rp.vrp_id = vrp.id
 
WHERE vrp_rp.sub_rp_moref != org_prov_vdc.sub_rp_moref

 

The root cause for all of this has not been found. I’m hoping VMware support can provide us with a little more information so I can update this post.

vCloud Director, VMware

vCloud Director and SAML Federation

I had a few issues getting vCloud Director and SAML federation playing nicely. By issues, I mean there wasn’t an explicit how-to in VMware’s pubs. The big issues were group-based authentication and authenticating against a user’s email address instead of their UPN.

Using the following article from pablovirtualization I was able to get vCloud Director federated to an ADFS SAML endpoint.

https://pablovirtualization.wordpress.com/2015/01/13/vcloud-director-and-microsoft-ad-fs-active-director-federation-service-authentication/

This allowed users to login using their UPN. That’s all well and good until you need user’s to log into their account using their email address which may differ from their UPN.

Login via email address

First, if you haven’t already due to some other requirement, allow your ADFS deployment to use the ‘mail’ attribute as an alternate login ID:

Set-ADFSClaimsProviderTrust -TargetIdentifier “AD AUTHORITY” -alternateloginID mail -lookupforest {your forest fqdn here} e.g contoso.corp

Now, brief difference between Pablo’s steps and this. When configuring the NameID transformation rule you’ll need to specify “Email” instead of “Unspecified”

adfs_transform

Group-based authentication

While you’re still adding transform rules, make sure you add this one too:

adfs_transform

Now all you have to do is enter the group name when importing groups in vCloud Director. Any users that are a member of that group will be able to login and receive the role specified when importing the group.