Puppet Certificate's

[Sean McGrath]

Firstly my apologies for posting this if it has been answered
elsewhere and I missed it while looking.

I'm starting to look at using Puppet to manage our fleet of Mac's
running OS X in our lab environment and I'm quite impressed with it
from my testing so far.

I have tested the functionality of the autosign.conf file with the
hostnames of the trusted clients in it.

However, if I re-image one of the Mac's as we occasionally do that
destroys the client certificate that it uses for the puppetca request.
Thus the puppet master see's a request with a different certificate
from a node with a hostname that has had its trust relationship
established with a different certificate.

This is probably a noob question but I haven't been able to figure it
out. How do I get around this in an automated manner. I don't want to
have to revoke certificates each time I re-image a Mac so they can be
re-trusted by the puppet master. Is there something like a root
certificate I could build into the image to establish the trust
relationship easily and securely each time a Mac is re-imaged?

many thanks

Sean

[Gary Larizza]

Hey Sean,

First - congrats on wrangling your Macs with Puppet! Next, I understand and have shared your pain regarding timely imaging of workstations and Puppet cert-wrangling.  Generally, I've seen folks do one of a couple of things:

1. Autosign
2. Utilize a CGI script to sign/revoke certs on the master (which can largely be replaced through the use of the `puppet cert` face)
3. Use the same private key everywhere and change the individual node_name

Numbers 1 and 2 are largely process around signing individual certs for every node.  You COULD even backup the $ssldir on your clients, image the machine, install puppet, restore the $ssldir, and then run Puppet again and Puppet will work fine for your clients.

Number 3 is a bit different.  With #3, you would have the SAME private cert for EVERY node in your infrastructure.  Because of this, the certname must be THE SAME for every node.  When you do this, however, Puppet treats every node as if it were the SAME node - so you need a way to de-couple the name of the node as Puppet knows it with the name of the node as the Certificate knows it.  The solution is the 'node_name_fact' and 'node_name_value' configuration item in puppet.conf --> http://docs.puppetlabs.com/references/stable/configuration.html#nodenamefact  You would essentially ship the private cert around to EVERY node, set the node_name_{fact,value} in puppet.conf, and then Puppet would treat each machine as a separate node (even though the certificate is the same everywhere).  Obviously there are security implications for this, but some people prefer it to Autosigning.

Hopefully, this should help you on your way. 

Sean

--
You received this message because you are subscribed to the Google Groups "Puppet Users" group.
To post to this group, send email to puppet...@googlegroups.com.
To unsubscribe from this group, send email to puppet-users...@googlegroups.com.
For more options, visit this group at http://groups.google.com/group/puppet-users?hl=en.

--

Gary Larizza

Professional Services Engineer

Puppet Labs

[Allister Banks]

Hey there,

I've got fire in ma belly to get this solved, as I'm impatient to use
Puppet's CA to bootstrap Simian, but due to this same sticking point
the bootstrapping process is more ungainly then it should be. My point
is we customize things about each machine while naming and binding,
why can't cert distribution (preceded by generation if necessary) be
part of that step in the process? Then it's a (debatably) more secure
and simple process, in my opinion.

Pardon I haven't looked in to puppet cert yet(it allows me to generate
certs for clients in advance based on supplied info/UUID's/serial
numbers, etc?), and I also need to figure out how generic I can make
the puppet install/config in the deployed image so only this cert
needs to be a custom addition.

My proposal is scripting DeployStudio to add a filedrop (or similar)
workflow to seed the right cert on the imaged system. I'll report back
as soon as I have a proof-of-concept.

Allister

[Sean McGrath]

Gary,

Thanks very much for getting back to me on this. While I have been
researching Puppet for our Mac fleet I have used a lot of the very
useful information you have published about this so thank you very
much for that.

In regards your response. Firstly, apologies for my lack of knowledge
in this matter and any help that can be provided will be very much
appreciated.

I am interested in #3 and am having looking at it now but the
following are some of the things that have crossed my mind

The private cert to distribute to the client nodes, which one from the
master is it? There are a few there and I cannot tell which one it is
and where does it go on the client?

Is there any other configuration that is needed on the master?

In the clients /etc/puppet/puppet.conf file would the following
configuration be correct for this approach.

[puppetd]
node_name_fact = hostname
node_name_value = certname # i.e. the one from the server

Is this correct?

Thats it for now, I'm going to stay playing around with this and see
where I can get but any help and guidance that anyone can provide will
be very much appreciated.

Regards

Sean

On Apr 11, 5:32 pm, Gary Larizza <g...@puppetlabs.com> wrote:
> Hey Sean,
>
> First - congrats on wrangling your Macs with Puppet! Next, I understand and
> have shared your pain regarding timely imaging of workstations and Puppet
> cert-wrangling.  Generally, I've seen folks do one of a couple of things:
>

>    1. Autosign
>    2. Utilize a CGI script to sign/revoke certs on the master (which can

>    largely be replaced through the use of the `puppet cert` face)

>    3. Use the same private key everywhere and change the individual

>    node_name
>
> Numbers 1 and 2 are largely process around signing individual certs for
> every node.  You COULD even backup the $ssldir on your clients, image the
> machine, install puppet, restore the $ssldir, and then run Puppet again and
> Puppet will work fine for your clients.
>
> Number 3 is a bit different.  With #3, you would have the SAME private cert
> for EVERY node in your infrastructure.  Because of this, the certname must
> be THE SAME for every node.  When you do this, however, Puppet treats every
> node as if it were the SAME node - so you need a way to de-couple the name
> of the node as Puppet knows it with the name of the node as the Certificate
> knows it.  The solution is the 'node_name_fact' and 'node_name_value'

> configuration item in puppet.conf -->http://docs.puppetlabs.com/references/stable/configuration.html#noden...

> You would essentially ship the private cert around to EVERY node, set
> the
> node_name_{fact,value} in puppet.conf, and then Puppet would treat each
> machine as a separate node (even though the certificate is the same
> everywhere).  Obviously there are security implications for this, but some
> people prefer it to Autosigning.
>
> Hopefully, this should help you on your way.
>

[Gary Larizza]

Hey Guys,

Let me try to step through this one step at a time.  I recommend downloading one of our VMs that we host for testing puppet (I'm using the VM we use for training located here --> http://downloads.puppetlabs.com/training/puppet-vmware.zip but there's also one for VirtualBox too --> http://downloads.puppetlabs.com/training/puppet-vbox.zip) to test this setup.  If you do that, make sure DNS and networking are working (So, setup host entries on your laptop for each VM, setup the hostname and FQDN ON EACH VM, and make sure each VM can ping the VM Master).  I'm using three VMs (a master and two agents): nodemaster.puppetlabs.vm, node1.puppetlabs.vm, and node2.puppetlabs.vm.  Once DNS resolves and is working, setup Puppet on the master FIRST, and then setup the remaining nodes as agents.  Again, I recommend using the VMs and using Puppet Enterprise because it makes it MUCH EASIER to setup Puppet - just use the puppet-enterprise-installer script (once you get the process of using a single cert and multiple node_names DOWN, THEN you can start on some actual machines).

Now we can start customizing Puppet.  There are a couple of key steps that need to be taken since we're shipping around a single cert for all machines:

1. When you stand up a PE Agent, it's going to contact the master and setup a Certificate Signing Request.  Because of this, go to the master node's $ssldir (which, on Puppet Enterprise, is /etc/puppetlabs/puppet/ssl.  If you want to find the $ssldir on YOUR master, just run `puppet config print ssldir` and it will print it out for you) and remove the CSRs that are in the ca/requests/ directory (they should be named after your nodes).

2. Next, lets generate the single certificate that will be shopped around.  You will need to use the same 'certname' on every node.  This is the name of the node as SSL is aware (in my case I'm using 'macs.puppetlabs.vm').  Do this by doing `puppet cert generate macs.puppetlabs.vm` on your MASTER machine.  The master will generate the certs and your $ssldir will look something like this:

ssl

|-- ca

|   |-- ca_crl.pem

|   |-- ca_crt.pem

|   |-- ca_key.pem

|   |-- ca_pub.pem

|   |-- inventory.txt

|   |-- private

|   |   `-- ca.pass

|   |-- requests

|   |-- serial

|   `-- signed

|       |-- macs.puppetlabs.vm.pem

|       |-- nodemaster.puppetlabs.vm.pem

|       |-- pe-internal-broker.pem

|       |-- pe-internal-dashboard.pem

|       |-- pe-internal-mcollective-servers.pem

|       |-- pe-internal-peadmin-mcollective-client.pem

|       `-- pe-internal-puppet-console-mcollective-client.pem

|-- certificate_requests

|-- certs

|   |-- ca.pem

|   |-- macs.puppetlabs.vm.pem

|   |-- nodemaster.puppetlabs.vm.pem

|   |-- pe-internal-broker.pem

|   |-- pe-internal-mcollective-servers.pem

|   |-- pe-internal-peadmin-mcollective-client.pem

|   `-- pe-internal-puppet-console-mcollective-client.pem

|-- crl.pem

|-- private

|-- private_keys

|   |-- macs.puppetlabs.vm.pem

|   |-- nodemaster.puppetlabs.vm.pem

|   |-- pe-internal-broker.pem

|   |-- pe-internal-mcollective-servers.pem

|   |-- pe-internal-peadmin-mcollective-client.pem

|   `-- pe-internal-puppet-console-mcollective-client.pem

`-- public_keys

    |-- foo.puppetlabs.vm.pem

    |-- macs.puppetlabs.vm.pem

    |-- nodemaster.puppetlabs.vm.pem

    |-- pe-internal-broker.pem

    |-- pe-internal-mcollective-servers.pem

    |-- pe-internal-peadmin-mcollective-client.pem

    `-- pe-internal-puppet-console-mcollective-client.pem

3.  There are three files you need to collect on your master and ship around to all of your nodes.  They are ALL in the $ssldir, so these paths are relative to THAT directory.  The three files are:

• $ssldir/private_keys/macs.puppetlabs.vm.pem on the master  ->  gets copied to the agent's $ssldir/private_keys directory
• $ssldir/public_keys/macs.puppetlabs.vm.pem on the master   ->  gets copied to the agent's $ssldir/public_keys directory
• $ssldir/ca/signed/macs.puppetlabs.vm.pem on the master       ->  gets copied to the agent's $ssldir/certs directory

4.  You need to modify /etc/puppetlabs/puppet/auth.conf on the master so every node can access the find action on the catalog REST endpoint for the master (i.e. The Mac nodes can get their catalog).  Look for this stanza in auth.conf:

# allow nodes to retrieve their own catalog (ie their configuration)

path ~ ^/catalog/([^/]+)$

method find

allow $1

Add the following line under 'allow $1': (Remember that I'm using the certname of macs.puppetlabs.vm - you would substitute the certname you will use in your infrastructure)

• allow macs.puppetlabs.vm

5.  Now, we need to modify /etc/puppetlabs/puppet/puppet.conf on the NODES themselves.  There are two changes that need to be made: the certname and the nodename.  Remember that we need to decouple the name that SSL uses to identify the node (macs.puppetlabs.vm) with the name that PUPPET uses to CLASSIFY the node (node1.puppetlabs.vm and node2.puppetlabs.vm in our case).  The first line you need to change in puppet.conf is the certname configuration item, set that to the following on ALL of your nodes:  'certname = macs.puppetlabs.vm'.  If you're using Puppet Enterprise, that item should already be in the [agent] stanza, so you'll need to change it.  The next configuration item could either go in the [agent] or [main] stanza, and that's the node_name_fact OR the node_name_value item.  NOTE:  THESE ITEMS ARE MUTUALLY EXCLUSIVE - YOU CAN ONLY USE ONE OR THE OTHER.  For more information, see --> http://docs.puppetlabs.com/references/stable/configuration.html#nodenamefact  The node_name_fact allows you to set the nodename based on a Facter fact.  In my case I'm going to set 'node_name_fact = fqdn' on all of my nodes because I want the nodename to match the FQDN on the machine.  If you're managing Macs, however, you might want to use something relatively static like a serial number (if you use something like the hostname, it will change any time someone renames their machine in the sharing pane).  You could also do 'node_name_value = thisnode' and Puppet would classify the node based on the name of 'thisnode', but you would need to have that item be unique for every one of your nodes.  This is a change you will need to evaluate in your own environment.

6.  Once you've generated the cert, put it in the appropriate directories on the client, changed auth.conf, set puppet.conf on the agents, and everything is installed, you should be able to run `puppet agent -t` and watch things work!  Note that if you're using the console, you will see the node records listed based on the nodename.

Let me know if you have any other questions on this process :)

[Sean McGrath]

Gary, thanks very much for that. I got that to work as I wanted after
a bit of fighting with other issues of my own creation. So I am very
grateful to you for your help.

If you ever in Dublin, Ireland. Look me up, I owe you a favor so I'm
willing to pay back, within reason of course.

Best

Sean

On Apr 12, 5:21 pm, Gary Larizza <g...@puppetlabs.com> wrote:
> Hey Guys,
>
> Let me try to step through this one step at a time.  I recommend
> downloading one of our VMs that we host for testing puppet (I'm using the

> VM we use for training located here -->http://downloads.puppetlabs.com/training/puppet-vmware.zipbut there's also

>    - $ssldir/private_keys/macs.puppetlabs.vm.pem on the master  ->  gets

>    copied to the agent's $ssldir/private_keys directory

>    - $ssldir/public_keys/macs.puppetlabs.vm.pem on the master   ->  gets

>    copied to the agent's $ssldir/public_keys directory

>    - $ssldir/ca/signed/macs.puppetlabs.vm.pem on the master       ->  gets

>    copied to the agent's $ssldir/certs directory
>
> 4.  You need to modify /etc/puppetlabs/puppet/auth.conf on the master so
> every node can access the find action on the catalog REST endpoint for the
> master (i.e. The Mac nodes can get their catalog).  Look for this stanza in
> auth.conf:
>
> # allow nodes to retrieve their own catalog (ie their configuration)
> path ~ ^/catalog/([^/]+)$
> method find
> allow $1
>
> Add the following line under 'allow $1': (Remember that I'm using the
> certname of macs.puppetlabs.vm - you would substitute the certname you will
> use in your infrastructure)
>

>    - allow macs.puppetlabs.vm

>
> 5.  Now, we need to modify /etc/puppetlabs/puppet/puppet.conf on the NODES
> themselves.  There are two changes that need to be made: the certname and
> the nodename.  Remember that we need to decouple the name that SSL uses to
> identify the node (macs.puppetlabs.vm) with the name that PUPPET uses to
> CLASSIFY the node (node1.puppetlabs.vm and node2.puppetlabs.vm in our
> case).  The first line you need to change in puppet.conf is the certname
> configuration item, set that to the following on ALL of your nodes:
>  'certname = macs.puppetlabs.vm'.  If you're using Puppet Enterprise, that
> item should already be in the [agent] stanza, so you'll need to change it.
>  The next configuration item could either go in the [agent] or [main]
> stanza, and that's the node_name_fact OR the node_name_value item.  NOTE:
>  THESE ITEMS ARE MUTUALLY EXCLUSIVE - YOU CAN ONLY USE ONE OR THE OTHER.

>  For more information, see -->http://docs.puppetlabs.com/references/stable/configuration.html#noden...

> The node_name_fact allows you to set the nodename based on a Facter
> fact.
>  In my case I'm going to set 'node_name_fact = fqdn' on all of my nodes
> because I want the nodename to match the FQDN on the machine.  If you're
> managing Macs, however, you might want to use something relatively static
> like a serial number (if you use something like the hostname, it will
> change any time someone renames their machine in the sharing pane).  You
> could also do 'node_name_value = thisnode' and Puppet would classify the
> node based on the name of 'thisnode', but you would need to have that item
> be unique for every one of your nodes.  This is a change you will need to
> evaluate in your own environment.
>
> 6.  Once you've generated the cert, put it in the appropriate directories
> on the client, changed auth.conf, set puppet.conf on the agents, and
> everything is installed, you should be able to run `puppet agent -t` and
> watch things work!  Note that if you're using the console, you will see the
> node records listed based on the nodename.
>
> Let me know if you have any other questions on this process :)
>

> ...
>
> read more »

[Allister Banks]

Sorry to resurrect an old(er) thread, but:
http://projects.puppetlabs.com/issues/3360#note-33
leads me to believe none of those workarounds are necessary, just
allow_duplicate_cert

However,
https://gist.github.com/0c76fb5b28abfcb2f9d6
That's a proof of concept that I started testing on the DeployStudio
side, and will probably fire up some python (once conference
extravaganza passes) to iterate over a csv of serial numbers and
therefore generate a bunch of certs at once.

Allister

On Apr 11, 12:32 pm, Gary Larizza <g...@puppetlabs.com> wrote:
> Hey Sean,
>
> First - congrats on wrangling your Macs with Puppet! Next, I understand and
> have shared your pain regarding timely imaging of workstations and Puppet
> cert-wrangling.  Generally, I've seen folks do one of a couple of things:
>

>    1. Autosign
>    2. Utilize a CGI script to sign/revoke certs on the master (which can

>    largely be replaced through the use of the `puppet cert` face)

>    3. Use the same private key everywhere and change the individual

>    node_name
>
> Numbers 1 and 2 are largely process around signing individual certs for
> every node.  You COULD even backup the $ssldir on your clients, image the
> machine, install puppet, restore the $ssldir, and then run Puppet again and
> Puppet will work fine for your clients.
>
> Number 3 is a bit different.  With #3, you would have the SAME private cert
> for EVERY node in your infrastructure.  Because of this, the certname must
> be THE SAME for every node.  When you do this, however, Puppet treats every
> node as if it were the SAME node - so you need a way to de-couple the name
> of the node as Puppet knows it with the name of the node as the Certificate
> knows it.  The solution is the 'node_name_fact' and 'node_name_value'

> configuration item in puppet.conf -->http://docs.puppetlabs.com/references/stable/configuration.html#noden...

> You would essentially ship the private cert around to EVERY node, set
> the
> node_name_{fact,value} in puppet.conf, and then Puppet would treat each
> machine as a separate node (even though the certificate is the same
> everywhere).  Obviously there are security implications for this, but some
> people prefer it to Autosigning.
>
> Hopefully, this should help you on your way.
>

[Gary Larizza]

On Fri, Apr 27, 2012 at 2:22 PM, Allister Banks <a...@aru-b.com> wrote:

Sorry to resurrect an old(er) thread, but:
http://projects.puppetlabs.com/issues/3360#note-33
leads me to believe none of those workarounds are necessary, just
allow_duplicate_cert

Have you tried this out?  I've personally never used the allow_duplicate_cert feature, but if it's not working properly then reporting that in the bug is totally the best way to go?

 

However,
https://gist.github.com/0c76fb5b28abfcb2f9d6
That's a proof of concept that I started testing on the DeployStudio
side, and will probably fire up some python (once conference
extravaganza passes) to iterate over a csv of serial numbers and
therefore generate a bunch of certs at once.

Commented on the gist - that should work as long as you generate the certs ON the puppet master and use the certname of the node you're wanting to provision.

[Gary Larizza]

Gah, should not end with a question mark -->  "I've personally never used the allow_duplicate_cert feature, but if it's not working properly then reporting that in the bug report is totally the best way to go?"