Check a baseline without changing anything
Before anyone lets a tool change production, they ask what’s wrong with it today. Compliance teams ask first, and they want the answer without a single setting moved.
Conn3x answers from a plan. The conn3x_command data source runs a read-only command on each host when you plan. An OpenTofu check block compares what came back with the rule, and reports every host that breaks it. Nothing on the host changes, and a failed check is a warning, not an error, so the plan still finishes.
The three checks below follow items found in the CIS benchmarks for Linux: root can’t log in over SSH, passwords expire within a year, and there are no world-writable files. They were run against two Ubuntu 24.04 hosts, one set up to pass and one to fail.
The hosts
Section titled “The hosts”The checks read a map of hosts, so they cover the same fleet as your circuits:
variable "hosts" { type = map(string) default = { web1 = "10.0.1.10" web2 = "10.0.1.11" }}Each data source below runs once per host with for_each. Its ssh block takes the same settings as a circuit’s, and become = true runs the command through sudo.
Root login over SSH
Section titled “Root login over SSH”sshd -T prints the settings sshd really uses, after every Include and Match, so a drop-in file in sshd_config.d can’t hide a PermitRootLogin yes:
data "conn3x_command" "root_login" { for_each = var.hosts command = "sshd -T | grep -i '^permitrootlogin '"
ssh { host = each.value user = "ops" become = true }}
locals { root_login = [for h, d in data.conn3x_command.root_login : h if trimspace(nonsensitive(d.stdout)) != "permitrootlogin no"]}
check "ssh_root_login" { assert { condition = length(local.root_login) == 0 error_message = "Root can log in over SSH on ${join(", ", local.root_login)}." }}A command’s output is marked sensitive, since it could hold a secret. nonsensitive is fine here because a setting’s name and value aren’t secret. Don’t wrap anything that could be.
Password ageing
Section titled “Password ageing”data "conn3x_command" "pass_max_days" { for_each = var.hosts command = "awk '$1 == \"PASS_MAX_DAYS\" { print $2 }' /etc/login.defs"
ssh { host = each.value user = "ops" }}
locals { pass_age = [for h, d in data.conn3x_command.pass_max_days : h if tonumber(trimspace(nonsensitive(d.stdout))) > 365]}
check "password_max_age" { assert { condition = length(local.pass_age) == 0 error_message = "PASS_MAX_DAYS is over 365 on ${join(", ", local.pass_age)}." }}This one reads a file anyone can read, so it doesn’t need become. PASS_MAX_DAYS applies to passwords set from now on. Accounts that exist keep the maximum chage gave them.
World-writable files
Section titled “World-writable files”data "conn3x_command" "world_writable" { for_each = var.hosts command = "find / -xdev -type f -perm -0002 2>/dev/null | head -n 20"
ssh { host = each.value user = "ops" become = true }}
locals { writable = { for h, d in data.conn3x_command.world_writable : h => split("\n", trimspace(nonsensitive(d.stdout))) if trimspace(nonsensitive(d.stdout)) != "" }}
check "world_writable_files" { assert { condition = length(local.writable) == 0 error_message = join("; ", [for h, files in local.writable : "${h}: ${join(", ", files)}"]) }}-xdev keeps find on the root filesystem, out of /proc and network mounts. It still reads every directory, which takes a while on a big disk, and it runs at every plan. Keep a check like this in its own configuration that you plan on a schedule, rather than next to your circuits.
What the plan says
Section titled “What the plan says”On the two test hosts, web1 passed everything and web2 failed all three:
No changes. Your infrastructure matches the configuration.
Warning: Check block assertion failed
on cis.tf line 54, in check "ssh_root_login": 54: condition = length(local.root_login) == 0
Root can log in over SSH on web2.
Warning: Check block assertion failed
on cis.tf line 61, in check "password_max_age":
PASS_MAX_DAYS is over 365 on web2.
Warning: Check block assertion failed
on cis.tf line 68, in check "world_writable_files":
web2: /etc/oopsFailing a CI job
Section titled “Failing a CI job”The plan exits 0 even when checks fail. To fail a scheduled job instead, read the warnings from -json:
conn3x plan -input=false -json > checks.jsonjq -r 'select(.type == "diagnostic" and .diagnostic.summary == "Check block assertion failed") | .diagnostic.detail' checks.jsontest -z "$(jq -r 'select(.type == "diagnostic" and .diagnostic.summary == "Check block assertion failed")' checks.json)"The last line exits 1 when any check failed.
Then fix them
Section titled “Then fix them”Each finding has a step that fixes it. A line step with regexp = "^PermitRootLogin" sets the SSH rule. Another on /etc/login.defs sets PASS_MAX_DAYS. Put those in a circuit when you’re ready to change things, and keep the checks to prove they stayed fixed.