Is there a way to ignore the SSH authenticity checking made by Ansible? For example when I've just setup a new server I have to answer yes to this question:
GATHERING FACTS ***************************************************************
The authenticity of host 'xxx.xxx.xxx.xxx (xxx.xxx.xxx.xxx)' can't be established.
RSA key fingerprint is xx:yy:zz:....
Are you sure you want to continue connecting (yes/no)?
I know that this is generally a bad idea but I'm incorporating this in a script that first creates a new virtual server at my cloud provider and then automatically calls my ansible playbook to configure it. I want to avoid any human intervention in the middle of the script execution.
Two options - the first, as you said in your own answer, is setting the environment variable ANSIBLE_HOST_KEY_CHECKING to False.
The second way to set it is to put it in an ansible.cfg file, and that's a really useful option because you can either set that globally (at system or user level, in /etc/ansible/ansible.cfg or ~/.ansible.cfg), or in an config file in the same directory as the playbook you are running.
To do that, make an ansible.cfg file in one of those locations, and include this:
[defaults]
host_key_checking = False
You can also set a lot of other handy defaults there, like whether or not to gather facts at the start of a play, whether to merge hashes declared in multiple places or replace one with another, and so on. There's a whole big list of options here in the Ansible docs.
Edit: a note on security.
SSH host key validation is a meaningful security layer for persistent hosts - if you are connecting to the same machine many times, it's valuable to accept the host key locally.
For longer-lived EC2 instances, it would make sense to accept the host key with a task run only once on initial creation of the instance:
- name: Write the new ec2 instance host key to known hosts
connection: local
shell: "ssh-keyscan -H {{ inventory_hostname }} >> ~/.ssh/known_hosts"
There's no security value for checking host keys on instances that you stand up dynamically and remove right after playbook execution, but there is security value in checking host keys for persistent machines. So you should manage host key checking differently per logical environment.
Leave checking enabled by default (in ~/.ansible.cfg)
Disable host key checking in the working directory for playbooks you run against ephemeral instances (./ansible.cfg alongside the playbook for unit tests against vagrant VMs, automation for short-lived ec2 instances)
I found the answer, you need to set the environment variable ANSIBLE_HOST_KEY_CHECKING to False. For example:
ANSIBLE_HOST_KEY_CHECKING=False ansible-playbook ...
Changing host_key_checking to false for all hosts is a very bad idea.
The only time you want to ignore it, is on "first contact", which this playbook will accomplish:
---
- name: Bootstrap playbook
# Don't gather facts automatically because that will trigger
# a connection, which needs to check the remote host key
gather_facts: false
tasks:
- name: Check known_hosts for {{ inventory_hostname }}
local_action: shell ssh-keygen -F {{ inventory_hostname }}
register: has_entry_in_known_hosts_file
changed_when: false
ignore_errors: true
- name: Ignore host key for {{ inventory_hostname }} on first run
when: has_entry_in_known_hosts_file.rc == 1
set_fact:
ansible_ssh_common_args: "-o StrictHostKeyChecking=no"
# Now that we have resolved the issue with the host key
# we can "gather facts" without issue
- name: Delayed gathering of facts
setup:
So we only turn off host key checking if we don't have the host key in our known_hosts file.
You can pass it as command line argument while running the playbook:
ansible-playbook play.yml --ssh-common-args='-o StrictHostKeyChecking=no'
forward to nikobelia
For those who using jenkins to run the play book, I just added to my jenkins job before running the ansible-playbook the he environment variable ANSIBLE_HOST_KEY_CHECKING = False
For instance this:
export ANSIBLE_HOST_KEY_CHECKING=False
ansible-playbook 'playbook.yml' \
--extra-vars="some vars..." \
--tags="tags_name..." -vv
If you don't want to modify ansible.cfg or the playbook.yml then you can just set an environment variable:
export ANSIBLE_HOST_KEY_CHECKING=False
Ignoring checking is a bad idea as it makes you susceptible to Man-in-the-middle attacks.
I took the freedom to improve nikobelia's answer by only adding each machine's key once and actually setting ok/changed status in Ansible:
- name: Accept EC2 SSH host keys
connection: local
become: false
shell: |
ssh-keygen -F {{ inventory_hostname }} ||
ssh-keyscan -H {{ inventory_hostname }} >> ~/.ssh/known_hosts
register: known_hosts_script
changed_when: "'found' not in known_hosts_script.stdout"
However, Ansible starts gathering facts before the script runs, which requires an SSH connection, so we have to either disable this task or manually move it to later:
- name: Example play
hosts: all
gather_facts: no # gather facts AFTER the host key has been accepted instead
tasks:
# https://stackoverflow.com/questions/32297456/
- name: Accept EC2 SSH host keys
connection: local
become: false
shell: |
ssh-keygen -F {{ inventory_hostname }} ||
ssh-keyscan -H {{ inventory_hostname }} >> ~/.ssh/known_hosts
register: known_hosts_script
changed_when: "'found' not in known_hosts_script.stdout"
- name: Gathering Facts
setup:
One kink I haven't been able to work out is that it marks all as changed even if it only adds a single key. If anyone could contribute a fix that would be great!
You can simply tell SSH to automatically accept fingerprints for new hosts. Just add
StrictHostKeyChecking=accept-new
to your ~/.ssh/config. It does not disable host-key checking entirely, it merely disables this annoying question whether you want to add a new fingerprint to your list of known hosts. In case the fingerprint for a known machine changes, you will still get the error.
This policy also works with ANSIBLE_HOST_KEY_CHECKING and other ways of passing this param to SSH.
I know the question has been answered and it's correct as well, but just wanted to link the ansible doc where it's explained clearly when and why respective check should be added: host-key-checking
The most problems appear when you want to add new host to dynamic inventory (via add_host module) in playbook. I don't want to disable fingerprint host checking permanently so solutions like disabling it in a global config file are not ok for me. Exporting var like ANSIBLE_HOST_KEY_CHECKING before running playbook is another thing to do before running that need to be remembered.
It's better to add local config file in the same dir where playbook is. Create file named ansible.cfg and paste following text:
[defaults]
host_key_checking = False
No need to remember to add something in env vars or add to ansible-playbook options. It's easy to put this file to ansible git repo.
This one is the working one I used in my environment. I use the idea from this ticket https://github.com/mitogen-hq/mitogen/issues/753
- name: Example play
gather_facts: no
hosts: all
tasks:
- name: Check SSH known_hosts for {{ inventory_hostname }}
local_action: shell ssh-keygen -l -F {{ inventory_hostname }}
register: checkForKnownHostsEntry
failed_when: false
changed_when: false
ignore_errors: yes
- name: Add {{ inventory_hostname }} to SSH known hosts automatically
when: checkForKnownHostsEntry.rc == 1
changed_when: checkForKnownHostsEntry.rc == 1
local_action:
module: shell
args: ssh-keyscan -H "{{ inventory_hostname }}" >> $HOME/.ssh/known_hosts
Host key checking is important security measure so I would not just skip it everywhere. Yes, it can be annoying if you keep reinstalling same testing host (without backing up it's SSH certificates) or if you have stable hosts but you run your playbook for Jenkins without simple option to add host key if you are connecting to the host for a first time. So:
This is what we are using for stable hosts (when running the playbook from Jenkins and you simply want to accept the host key when connecting to the host for the first time) in inventory file:
[all:vars]
ansible_ssh_common_args='-o StrictHostKeyChecking=accept-new'
And this is what we have for temporary hosts (in the end this will ignore they host key at all):
[all:vars]
ansible_ssh_common_args='-o StrictHostKeyChecking=no -o UserKnownHostsFile=/dev/null'
There is also environment variable or you can add it into group/host variables file. No need to have it in the inventory - it was just convenient in our case.
Used some other responses here and a co-worker solution, thank you!
Use the parameter named as validate_certs to ignore the ssh validation
- ec2_ami:
instance_id: i-0661fa8b45a7531a7
wait: yes
name: ansible
validate_certs: false
tags:
Name: ansible
Service: TestService
By doing this it ignores the ssh validation process
Related
What I am trying to accomplish overall is to ssh into systems which are untouched by ansible, and have them set up by ansible, including its account, and ssh keys, and adding to the dynamic inventory... and so on and so forth. In this case, it's via a proxy jump. Unfortunately this means having to ssh into them using the ssh command and the shell module, as well as storing a password. Keep in mind I am on ansible 2.9, and this is a build environment, so passwords can be copied to files during build for use and then deleted at the end of the run, so this isn't a problem. If this succeeds, we can set up accounts and ssh keys, then delete the build files and everyone is happy.
I don't need all that much I hope, I would just like to get one sticky piece of that working better. That part is the ssh options that are needed for a proxyjump connection. ansible-controller doesn't have direct access to host p0, but the ecc67 host does. I have it working in the shell command no problem, but for whatever reason, I can't shift it up to the ansible_ssh_common_args variable where it belongs.
Here is the working example of the task as it functions now:
- name: sshpass attempt with the raw module for testing.
shell: sshpass -p "{{ access_var.ansible_ssh_pass_ssn }}" ssh -o 'ProxyCommand=ssh -W %h:%p bob#ecc67 nc %h %p' bob#p0 "w; exit"
register: output_1
The above works just fine and uses an undefined ansible_ssh_common_args. The nc is the netcat binary and is simply being passed options through the proxy command. Then we have the below playbook in which I tried to complete my stated mission, however, it is not functional, and fails at the sshpass task:
- name: Play that is testing for a successful proxyjump connection to p0 through ecc67.
hosts: ansible-controller
remote_user: bob
become: no
become_method: sudo
gather_facts: no
vars:
ansible_connection: ssh
ansible_ssh_common_args: '-o "ProxyCommand=ssh -W %h:%p bob#ecc67 nc %h %p"'
tasks:
- name: Import the password file so that we have the bob account's password.
include_vars:
file: ~/project/copyable-files/dynamic-files/build/active-vars-repository/access.yml
name: access_var
- name: Set password for the bob account from the file value using previous operator input.
set_fact:
ansible_ssh_pass: "{{ access_var.ansible_ssh_pass_b }}"
ansible_become_password: "{{ access_var.ansible_ssh_pass_b }}"
cacheable: yes
- name: sshpass attempt with the raw module for testing.
shell: sshpass -p "{{ ansible_ssh_pass_b }}" ssh "{{ ansible_ssh_common_args }}" bob#p0 "hostname; exit"
register: output_1
- debug:
var: output_1
The error I get when I attempt to use the above playbook with the reworked task and variables is as follows:
TASK [sshpass attempt with the raw module for testing.] ***********************************************
fatal: [ansible-controller]: UNREACHABLE! => {"changed": false, "msg": "Invalid/incorrect password: Killed by signal 1.", "unreachable": true}
The password is not the issue despite the error stating it is, though it's possible it's accessing something I don't expect. Is there any way to do what I want, heck, is there even just a better way to go about it that I didn't think of? Any suggestions would be helpful thanks!
From your description I understand that there is an issue with special characters in variables, quoting, templating and debugging. Therefore I am explicit not addressing the question "Is there ... a better way to go?".
To address the different topics I've created the following minimal example playbook
---
- hosts: localhost
become: false
gather_facts: false
vars:
ansible_ssh_pass: !unsafe "P4$$w0rd!_%&"
ansible_ssh_common_args: !unsafe '-o "ProxyCommand=ssh -W %h:%p user#jump.example.com nc %h %p"'
tasks:
- name: Debug task to show command content
lineinfile:
path: ssh.file
create: true
line: 'sshpass -p {{ ansible_ssh_pass | quote }} ssh {{ ansible_ssh_common_args }} user#test.example.com "hostname; exit"'
resulting into an output of
sshpass -p 'P4$$w0rd!_%&' ssh -o "ProxyCommand=ssh -W %h:%p user#jump.example.com nc %h %p" user#test.example.com "hostname; exit"
... the content of ssh.file and what the shell would "see"
Further Documentation
Advanced playbook syntax - Unsafe or raw strings for usage of !unsafe
The most common use cases include passwords that allow special characters
Using filters to manipulate data
You can use YAML single quote escaping ... Escaping single quotes within single quotes in YAML is done by doubling the single quote.
Using filters to manipulate data - Manipulating strings for usage of quote
To add quotes for shell usage ... | quote
Templating (Jinja2)
Ansible uses Jinja2 templating to enable dynamic expressions and access to variables and facts.
I am attempting to have playbooks that run once to set up a new user and disable root ssh access.
For now, I am doing that by declaring all of my inventory twice. Each host needs an entry that accesses with the root user, used to create a new user, set up ssh settings, and then disable root access.
Then each host needs another entry with the new user that gets created.
My current inventory looks like this. It's only one host for now, but with a larger inventory, the repetition would just take up a ton of unnecessary space:
---
# ./hosts.yaml
---
all:
children:
master_roots:
hosts:
demo_master_root:
ansible_host: a.b.c.d # same ip as below
ansible_user: root
ansible_ssh_private_key_file: ~/.ssh/id_rsa_infra_ops
masters:
hosts:
demo_master:
ansible_host: a.b.c.d # same ip as above
ansible_user: infraops
ansible_ssh_private_key_file: ~/.ssh/id_rsa_infra_ops
Is there a cleaner way to do this?
Is this an anti-pattern in any way? It is not idempotent. It would be nice to have this run in a way that running the same playbook twice always has the same output - either "success", or "no change".
I am using DigitalOcean and they have a functionality to have this done via a bash script before the VM comes up for the first time, but I would prefer a platform-independent solution.
Here is the playbook for setting up the users & ssh settings and disabling root access
---
# ./initial-host-setup.yaml
---
# References
# Digital Ocean recommended droplet setup script:
# - https://docs.digitalocean.com/droplets/tutorials/recommended-setup
# Digital Ocean tutorial on installing kubernetes with Ansible:
# - https://www.digitalocean.com/community/tutorials/how-to-create-a-kubernetes-cluster-using-kubeadm-on-debian-9
# Ansible Galaxy (Community) recipe for securing ssh:
# - https://github.com/vitalk/ansible-secure-ssh
---
- hosts: master_roots
become: 'yes'
tasks:
- name: create the 'infraops' user
user:
state: present
name: infraops
password_lock: 'yes'
groups: sudo
append: 'yes'
createhome: 'yes'
shell: /bin/bash
- name: add authorized keys for the infraops user
authorized_key: 'user=infraops key="{{item}}"'
with_file:
'{{ hostvars[inventory_hostname].ansible_ssh_private_key_file }}.pub'
- name: allow infraops user to have passwordless sudo
lineinfile:
dest: /etc/sudoers
line: 'infraops ALL=(ALL) NOPASSWD: ALL'
validate: visudo -cf %s
- name: disable empty password login for all users
lineinfile:
dest: /etc/ssh/sshd_config
regexp: '^#?PermitEmptyPasswords'
line: PermitEmptyPasswords no
notify: restart sshd
- name: disable password login for all users
lineinfile:
dest: /etc/ssh/sshd_config
regexp: '^(#\s*)?PasswordAuthentication '
line: PasswordAuthentication no
notify: restart sshd
- name: Disable remote root user login
lineinfile:
dest: /etc/ssh/sshd_config
regexp: '^#?PermitRootLogin'
line: 'PermitRootLogin no'
notify: restart sshd
handlers:
- name: restart sshd
service:
name: sshd
state: restarted
Everything after this would use the masters inventory.
EDIT
After some research I have found that "init scripts"/"startup scripts"/"user data" scripts are supported across AWS, GCP, and DigitalOcean, potentially via cloud-init (this is what DigitalOcean uses, didn't research the others), which is cross-provider enough for me to just stick with a bash init script solution.
I would still be interested & curious if someone had a killer Ansible-only solution for this, although I am not sure there is a great way to make this happen without a pre-init script.
Regardless of any ansible limitations, it seems that without using the cloud init script, you can't have this. Either the server starts with a root or similar user to perform these actions, or the server starts without a user with those powers, then you can't perform these actions.
Further, I have seen Ansible playbooks and bash scripts that try to solve the desired "idempotence" (complete with no errors even if root is already disabled) by testing root ssh access, then falling back to another user, but "I can't ssh with root" is a poor test for "is the root user disabled" because there are plenty of ways your ssh access could fail even though the server is still configured to allow root to ssh.
EDIT 2 placing this here, since I can't use newlines in my response to a comment:
β.εηοιτ.βε responded to my assertion:
"but "I can't ssh with root" is a poor test for "is the root user disabled" because there are plenty of ways your ssh access could fail even though the server is still configured to allow root to ssh
with
then, try to ssh with infraops and assert that PermitRootLogin no is in the ssh daemon config file?"
It sounds like the suggestion is:
- attempt ssh with root
- if success, we know user/ssh setup tasks have not completed, so run those tasks
- if failure, attempt ssh with infraops
- if success, go ahead and run everything except the user creation again to ensure ssh config is as desired
- if failure... ? something else is probably wrong, since I can't ssh with either user
I am not sure what this sort of if-then failure recovery actually looks like in an Ansible script
You can overwrite host variables for a given play by using vars.
- hosts: masters
become: 'yes'
vars:
ansible_ssh_user: "root"
ansible_ssh_private_key_file: "~/.ssh/id_rsa_infra_ops"
tasks:
You could only define the demo_master group and alter the ansible_user and ansible_ssh_private_key_file at run time, using command flags --user and --private-key.
So with an host.yaml containing
all:
children:
masters:
hosts:
demo_master:
ansible_host: a.b.c.d # same ip as above
ansible_user: infraops
ansible_ssh_private_key_file: ~/.ssh/id_rsa_infra_ops
And run on - hosts: master, the first run would, for example be with
ansible-playbook initial-host-setup.yaml \
--user root \
--private-key ~/.ssh/id_rsa_root
When the subsequent runs would simply by
ansible-playbook subsequent-host-setup.yaml
Since all the required values are in the inventory already.
SO I would like to provision a proxy-host ( i can do this), add it to the dynamic ansible inventory via add_host (done),
Then in the next play, run tasks on that proxy-host, to find another machine behind it, update something ansible side to know this new host's location, and that It needs to be proxy jumped via this current proxy-host,
Then in the next play target this new machine behind the proxy-host.
I am at a lost here, i was hoping to do it without all of this ssh_config changes... is this possible, has anyone done this, thoughts?
I have an answer to my question. I think that it is a perfectly valid question, and alot the documentation from Ansible semi-answers this question, it is not put into the context of being dynamic, nor is it stated that it can be done completely dynamically.
Pretext: Using terraform, within ansible to generate hosts, with the following configuration:
control_box (running ansible/terraform from)----> dynamically created Bastion/proxy/jump_host ---> some_server(behind the bastion)
Playbook:
#Make the bastion host, and add it to the just_created group
- hosts: 127.0.0.1
roles:
- terraform_logic_add_host_logic
- hosts: just_created #aka bastion
tasks:
- name: Include task list in play
include: "get_the_private_ip_and_add_to_behind_bastion_group.yml"
# Login into behind_bastion group.....
- hosts: behind_bastion_group
vars:
- ansible_connection: ssh
- ansible_ssh_common_args: '-o ProxyCommand="ssh -i {{ some_pem_key }} -o StrictHostKeyChecking=no -W %h:%p -q ec2-user#{{ the_bastion_ip }}"'
tasks:
- name: Include task list in play
include: "do_stuff_finally.yml"
I have done my research as well FYI:
Posts such as this, do not show the complete End2End solution of doing this all dynamically...
https://blog.scottlowe.org/2015/12/24/running-ansible-through-ssh-bastion-host/
Ansible with a bastion host / jump box?
https://selivan.github.io/2018/01/29/ansible-ssh-bastion-host.html
I am using Ansible to provision a Vagrant environment. As part of the provisioning process, I need to connect from the currently-provisioning VM to a private external repository using an ssh key in order to use composer to pull in modules for an application. I've done a lot of reading on this before asking this question, but still can't seem to comprehend what's going on.
What I want to happen is:
As part of the playbook, on the Vagrant VM, I add the ssh key to the private repo to the ssh-agent
Using that private key, I am then able to use composer to require modules from the external source
I've read articles which highlight specifying the key in playbook execution. (E.g. ansible-play -u username --private-key play.yml) As far as I understand, this isn't for me, as I'm calling the playbook via Vagrant file. I've also read articles which mention ssh forwarding. (SSH Agent Forwarding with Ansible). Based on what I have read, this is what I've done:
On the VM being provisioned, I insert a known_hosts file which consists of the host entries of the machines which house the repos I need:
On the VM being provisioned, I have the following in ~/.ssh/config:
Host <VM IP>
ForwardAgent yes
I have the following entries in my ansible.cfg to support ssh forwarding:
[defaults]
transport = ssh
[ssh_connection]
ssh_args=-o ForwardAgent=yes -o ControlMaster=auto -o ControlPersist=60s -o ControlPath=/tmp/ansible-ssh-%h-%p-%r
[privilege_escalation]
pipelining = False
I have also added the following task to the playbook which tries to
use composer:
- name: Add ssh agent line to sudoers
become: true
lineinfile:
dest: /etc/sudoers
state: present
regexp: SSH_AUTH_SOCK
line: Defaults env_keep += "SSH_AUTH_SOCK"
I exit the ansible provisioner and add the private key on the provisioned VM to the agent via a shell provisioner (This is where I suspect I'm going wrong)
Then, I attempt to use composer, or call git via the command module. Like this, for example, to test:
- name: Test connection
command: ssh -T git#github.com
Finally, just in case I wasn't understanding ssh connection forwarding correctly, I assumed that what was supposed to happen was that I needed to first add the key to my local machine's agent, then forward that through to the provisioned VM to use to grab the repositories via composer. So I used ssh-add on my local machine before executing vagrant up and running the provisioner.
No matter what, though, I always get permission denied when I do this. I'd greatly appreciate some understanding as to what I may be missing in my understanding of how ssh forwarding should be working here, as well as any guidance for making this connection happen.
I'm not certain I understand your question correctly, but I often setup machines that connect to a private bitbucket repository in order to clone it. You don't need to (and shouldn't) use agent forwarding for that ("ssh forwarding" is unclear; there's "authentication agent forwarding" and "port forwarding", but you need neither in this case).
Just to be clear with terminology, you are running Ansible in your local machine, you are provisioning the controlled machine, and you want to ssh from the controlled machine to a third-party server.
What I do is I upload the ssh key to the controlled machine, in /root/.ssh (more generally $HOME/.ssh where $HOME is the home directory of the controlled machine user who will connect to the third-party server—in my case that's root). I don't use the names id_rsa and id_rsa.pub, because I don't want to touch the default keys of that user (these might have a different purpose; for example, I use them to backup the controlled machine). So this is the code:
- name: Install bitbucket aptiko_ro ssh key
copy:
dest: /root/.ssh/aptiko_ro_id_rsa
mode: 0600
content: "{{ aptiko_ro_ssh_key }}"
- name: Install bitbucket aptiko_ro ssh public key
copy:
dest: /root/.ssh/aptiko_ro_id_rsa.pub
content: "{{ aptiko_ro_ssh_pub_key }}"
Next, you need to tell the controlled machine ssh this: "When you connect to the third-party server, use key X instead of the default key, and logon as user Y". You tell it in this way:
- name: Install ssh config that uses aptiko_ro keys on bitbucket
copy:
dest: /root/.ssh/config
content: |
Host bitbucket.org
IdentityFile ~/.ssh/aptiko_ro_id_rsa
User aptiko_ro
My playbook needs to change the ssh port and updates the firewall rules. (unfortunately, I cannot "get" a new server directly with the desired custom port).
Managing the change during the execution is easy.
However I do not know how to have an idempotent playbook.
The first run must be initiated on the default port (22).
The next runs must be initiated on the custom port.
A solution could be done but with performance issues.
Is there any other possibility with Ansible 2.0+?
You could approach this a couple of ways really.
The simplest way might be to simply separate the SSH port configuration into a separate playbook/role that specifies the SSH port as 22 but then your inventory would normally define the SSH port as your custom one.
ssh_port.yml
- hosts: all
vars:
ansible_ssh_port: 22
tasks:
- name: change the default ssh port
lineinfile ...
notify: restart ssh
handlers:
- name: restart ssh
service:
name : sshd
state: restarted
You would then only run this playbook on the creation of the machine and then only re-run your main playbook again and again, sidestepping the idempotency of this step.
Alternatively, as Mikko Ohtamaa pointed out in the comments, you could have Ansible modify your inventory file when you change the port. This will mean that you can run the whole thing end to end idempotently as the next run through will connect on the non default SSH port and then simply (pointlessly obviously) check that the SSH port is still set to the desired one. You can get at the inventory file by using the "magic variable" inventory_file. A rough example might look like this:
- name: change the default ssh port
lineinfile ...
notify: restart ssh
- name: change ssh port used by ansible
set_fact:
ansible_ssh_port: {{ custom_ssh_port }}
- name: change ssh port in inventory
lineinfile:
dest: inventory_file
insert_after: '[all:vars]'
line: 'ansible_ssh_port="{{ custom_ssh_port }}"'
Just make sure you have an inline group variables block for all in the inventory file and this will mean all future runs of any playbook against this inventory will connect to all of the hosts contained inside it on your custom SSH port.
If you use source control then you will also need a local_action task to push the change back to your remote.