Add TP-Link SG2428LP switch hardening as repeatable IaC + static IP/DNS

Secure and configure the SG2428LP managed switch (10.0.0.10) as repeatable
Ansible, driven from the control node over SSH by an expect engine.

- tplink_switch_harden role (verified JetStream CLI syntax): HTTP off/
  HTTPS-only, Telnet off, SSH v2, SNMP off, RSTP + loopback-detection,
  then copy running-config startup-config (persist).
- playbooks/harden_switch.yml + [switches] inventory group; dry by default,
  -e switch_apply=true to apply, -e switch_apply_network=true for addressing.
- Static mgmt IP 10.0.0.10/24 (off DHCP, below the .20-.199 pool), gw 10.0.0.1,
  hostname sg2428lp.
- op -> ansible-vault bridge (etc/set-switch-1password.sh); admin password in
  group_vars/switches/vault.yml.
- DNS: A/PTR sg2428lp.prole.org -> 10.0.0.10 (records added to dns.yml/ptr_records).
- Fix: tag the samba_reverse_dns zone-list set_fact so tag-limited runs define
  samba_reverse_zones.

Engine notes (hard-won): forces password-only SSH auth (the switch drops
publickey-first logins); always saves config (unsaved changes revert on reboot).

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

etc/set-switch-1password.sh

@@ -0,0 +1,62 @@
+#!/usr/bin/env bash
+# set-switch-1password.sh
+# 1Password -> Ansible Vault bridge for the managed switch admin credential.
+#
+# Reads the switch admin password from 1Password (op) and writes it, encrypted
+# with the repo .vault_pass, into the switches group_vars vault file. This keeps
+# 1Password the human source of truth while ansible-vault is what the playbook
+# consumes at run time.
+#
+#   op://knoey/TP-Link SG2428LP/password  --(op read)-->  ENV  --(ansible-vault)-->
+#       infrastructure/inventory/group_vars/switches/vault.yml
+#
+# Usage:
+#   etc/set-switch-1password.sh                # default item/account below
+#   OP_ACCOUNT=... OP_REFERENCE=... etc/set-switch-1password.sh
+#
+# Requires: op (1Password CLI, unlocked), ansible-vault, repo .vault_pass.
+
+set -euo pipefail
+
+SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+REPO_ROOT="$(cd "${SCRIPT_DIR}/.." && pwd)"
+cd "${REPO_ROOT}"
+
+OP_ACCOUNT="${OP_ACCOUNT:-VM7LPYA4XFF4TE2DEI7ZHMZWEA}"
+OP_REFERENCE="${OP_REFERENCE:-op://knoey/TP-Link SG2428LP/password}"
+VAR_NAME="${VAR_NAME:-vault_sg2428lp_admin_password}"
+VAULT_FILE="${VAULT_FILE:-infrastructure/inventory/group_vars/switches/vault.yml}"
+
+command -v op >/dev/null 2>&1 || { echo "ERROR: 1Password CLI (op) not found. brew install 1password-cli" >&2; exit 1; }
+command -v ansible-vault >/dev/null 2>&1 || { echo "ERROR: ansible-vault not found." >&2; exit 1; }
+[[ -f .vault_pass ]] || { echo "ERROR: .vault_pass not found at repo root." >&2; exit 1; }
+
+# op desktop integration sometimes needs an explicit unlock.
+if ! op read --account "$OP_ACCOUNT" "$OP_REFERENCE" >/dev/null 2>&1; then
+  echo "Unlocking 1Password (op signin)..." >&2
+  op signin --account "$OP_ACCOUNT" >/dev/null || true
+fi
+
+SECRET="$(op read --account "$OP_ACCOUNT" "$OP_REFERENCE" 2>/dev/null || true)"
+if [[ -z "$SECRET" ]]; then
+  echo "ERROR: could not read secret from $OP_REFERENCE (is 1Password unlocked?)" >&2
+  exit 1
+fi
+echo "Read secret from 1Password (${#SECRET} chars)." >&2
+
+# Encrypt as an inline !vault var. Rely on ansible.cfg's vault_password_file so
+# we don't create a duplicate 'default' vault-id (which ansible-vault rejects).
+ENC="$(printf '%s' "$SECRET" | ansible-vault encrypt_string --stdin-name "$VAR_NAME")"
+unset SECRET
+
+{
+  echo "---"
+  echo "# Managed-switch admin password — DO NOT EDIT BY HAND."
+  echo "# Source of truth: ${OP_REFERENCE}"
+  echo "# Regenerate with: etc/set-switch-1password.sh"
+  echo "${ENC}"
+} > "${VAULT_FILE}"
+
+echo "Wrote encrypted ${VAR_NAME} -> ${VAULT_FILE}"
+echo "Verify (length only):"
+echo "  ansible localhost -m debug -a \"msg={{ ${VAR_NAME} | length }}\" -e @${VAULT_FILE}"

infrastructure/inventory/group_vars/all/dns.yml

@@ -179,6 +179,8 @@ prole_internal_a_records:
     ipv4s: [10.0.0.4]
   - fqdn: retropie.prole.org
     ipv4s: [10.0.0.207]
+  - fqdn: sg2428lp.prole.org
+    ipv4s: [10.0.0.10]
   - fqdn: synology.prole.org
     ipv4s: [10.0.0.203]
   - fqdn: zinfandel.prole.org

infrastructure/inventory/group_vars/all/vars.yml

@@ -23,6 +23,8 @@ ptr_records:
     fqdn: "raspberry.prole.org"
   - last_octet: "5"
     fqdn: "pi.prole.org"
+  - last_octet: "10"
+    fqdn: "sg2428lp.prole.org"
   - last_octet: "207"
     fqdn: "retropie.prole.org"
 

infrastructure/inventory/group_vars/switches/vars.yml

@@ -0,0 +1,35 @@
+---
+# Managed-switch group — non-secret facts + connection model.
+#
+# The switch is NOT a Linux host: we don't sudo, and we don't use the global
+# ansible_user/key. The hardening role connects FROM the control node TO the
+# switch over SSH via an expect engine (delegated, connection: local).
+#
+# Secret (vault_sg2428lp_admin_password) lives in this group's vault.yml,
+# populated from 1Password by etc/set-switch-1password.sh.
+
+ansible_connection: local
+ansible_become: false
+
+# Management endpoint + credentials (password resolved from vault).
+switch_mgmt_ip: "{{ ansible_host }}"
+switch_admin_user: "admin"
+switch_admin_password: "{{ vault_sg2428lp_admin_password }}"
+
+# 1Password source-of-truth reference (used by the op -> vault bridge script).
+switch_op_account: "VM7LPYA4XFF4TE2DEI7ZHMZWEA"
+switch_op_reference: "op://knoey/TP-Link SG2428LP/password"
+
+# ── Desired identity + management addressing ─────────────────────────────────
+# Static management IP, OFF DHCP. 10.0.0.10 sits below the DHCP pool
+# (10.0.0.20–10.0.0.199), so it never collides with a lease.
+switch_hostname: "sg2428lp"
+switch_fqdn: "{{ switch_hostname }}.{{ prole_domain }}"   # sg2428lp.prole.org
+switch_mgmt_ip_static: "10.0.0.10"
+switch_mgmt_netmask: "255.255.255.0"
+switch_mgmt_gateway: "10.0.0.1"
+switch_mgmt_vlan: 1
+
+# One-time migration source: the DHCP address the switch currently answers on.
+# After cutover, set this host's ansible_host to switch_mgmt_ip_static (.10).
+switch_dhcp_ip: "10.0.0.153"

infrastructure/inventory/group_vars/switches/vault.yml

@@ -0,0 +1,11 @@
+---
+# Managed-switch admin password — DO NOT EDIT BY HAND.
+# Source of truth: op://knoey/TP-Link SG2428LP/password
+# Regenerate with: etc/set-switch-1password.sh
+vault_sg2428lp_admin_password: !vault |
+          $ANSIBLE_VAULT;1.1;AES256
+          61356331346534643937343732303265376438343536653933393630323963613636313261313662
+          3134636433393836363032346161303462653631373963350a336363333436323735326230353133
+          38343464393263393536356435333464393436313635363062633639323362386432653436343063
+          3931656237356665640a626134363939643763386661313666303238653632373738626163386565
+          30383162633432376366326436663064666463353139623465393966323561353038

infrastructure/inventory/hosts.ini

@@ -52,6 +52,14 @@ gandalf.prole.org
 [merlin_bootstrap]
 merlin ansible_host=10.0.0.36 ansible_user=ansible
 
+# ── Managed network switches ─────────────────────────────────────────────────
+# Hardened via playbooks/harden_switch.yml (role: tplink_switch_harden).
+# Driven from the control node over SSH by an expect engine (connection: local,
+# delegated) — NOT the standard linux ansible_user. Admin password lives in
+# group_vars/switches/vault.yml (op -> vault bridge: etc/set-switch-1password.sh).
+[switches]
+sg2428lp ansible_host=10.0.0.10
+
 # ── Engineer workstations — macOS / Linux ────────────────────────────────────
 # workstation_kerberos.yml targets this group to deploy:
 #   - /etc/krb5.conf  (or /Library/Preferences/edu.mit.Kerberos on macOS)

infrastructure/playbooks/harden_switch.yml

@@ -0,0 +1,19 @@
+---
+# Harden managed network switches (TP-Link JetStream).
+#
+#   Secret pipeline:  1Password --(etc/set-switch-1password.sh)--> ansible-vault
+#                     --> this playbook reads vault_sg2428lp_admin_password.
+#
+# Run:
+#   ansible-playbook infrastructure/playbooks/harden_switch.yml
+#   ansible-playbook infrastructure/playbooks/harden_switch.yml --check   # dry run (no apply)
+#   ansible-playbook infrastructure/playbooks/harden_switch.yml -e switch_apply=true
+#
+# By design this does NOT apply config unless switch_apply=true, so a casual run
+# only connects, validates access, and captures `show running-config`.
+- name: Harden managed switches
+  hosts: switches
+  gather_facts: false
+  connection: local
+  roles:
+    - tplink_switch_harden

infrastructure/roles/samba_reverse_dns/tasks/main.yml

@@ -17,6 +17,7 @@
   ansible.builtin.set_fact:
     samba_reverse_zones: >-
       {{ ([lan_reverse_zone] + (k3s_reverse_zones | default([]))) | unique }}
+  tags: [samba, samba_reverse_dns]
 
 - name: Create reverse DNS zones if missing
   ansible.builtin.command:

infrastructure/roles/tplink_switch_harden/README.md

@@ -0,0 +1,77 @@
+# tplink_switch_harden
+
+Repeatable security hardening for TP-Link JetStream managed switches
+(developed for the **SG2428LP** at `10.0.0.153`), driven from the control node
+over SSH by a rendered `expect` engine.
+
+## Secret pipeline
+
+```
+1Password  ──op read──▶  ENV  ──ansible-vault──▶  group_vars/switches/vault.yml  ──▶  playbook
+   op://knoey/TP-Link SG2428LP/password         (vault_sg2428lp_admin_password)
+```
+
+Refresh the vault from 1Password whenever the password changes:
+
+```bash
+etc/set-switch-1password.sh
+```
+
+The admin password is passed to the engine only via the `SWITCH_PW` environment
+variable (`no_log`), so it never lands in inventory, the rendered script, or
+process arguments.
+
+## Usage
+
+```bash
+# Dry run — log in, capture `show system-info`, change nothing:
+ansible-playbook infrastructure/playbooks/harden_switch.yml
+
+# Apply hardening and save to startup-config:
+ansible-playbook infrastructure/playbooks/harden_switch.yml -e switch_apply=true
+```
+
+`switch_apply` defaults to **false**; nothing is changed until you pass
+`-e switch_apply=true`.
+
+## What it hardens
+
+Config-mode commands live in [`defaults/main.yml`](defaults/main.yml)
+(`switch_harden_commands`): disable plaintext HTTP, ensure HTTPS, disable
+Telnet, SSH v2 only, ensure SNMP off, enable RSTP, enable loopback-detection,
+then `copy running-config startup-config`.
+
+The command list was **verified against SG2428LP firmware** via on-device `?`
+help. JetStream tokens vary by model/firmware — reconfirm with `?` if you point
+this at a different model.
+
+Management IP access-control and CLI idle-timeout are intentionally **opt-in**
+(commented in `defaults/main.yml`) because a wrong ACL can lock you out. Apply
+those last, from a host inside the permitted range.
+
+## Static-IP cutover (DHCP → static)
+
+`switch_network_commands` (toggled by `switch_apply_network`) sets the static
+mgmt IP. In steady state — when the host's `ansible_host` already equals the
+static IP — re-asserting it is a no-op. The **one-time** DHCP→static cutover is
+different: the `ip address` line drops the session, so it cannot save afterward
+over the same connection. Do the cutover deliberately (connect on the DHCP
+address, set the static IP, reconnect on the new IP, then save), and afterward
+set the host's `ansible_host` to the static IP. Same-subnet control hosts reach
+the new IP directly, so gateway correctness isn't required to regain access.
+
+## Gotchas (these cost real debugging time)
+
+- **SSH publickey auth breaks login.** The JetStream SSH server drops the
+  connection when the client offers `publickey` first (the default). The engine
+  forces `-o PubkeyAuthentication=no -o PreferredAuthentications=password`.
+  Symptom without it: "connection closed after KEX, before password prompt",
+  intermittently (depends on which keys your agent offers).
+- **Unsaved config reverts on reboot.** Web-UI changes (incl. the first-login
+  password) live in running-config until you explicitly save. A reboot with no
+  save silently reverts to factory. The engine always ends with
+  `copy running-config startup-config` (look for "Saving user config OK!").
+- **One management session at a time.** A logged-in web-UI session occupies the
+  single mgmt slot and blocks SSH login. Log out of the web UI before running.
+- **`no ip http server` doesn't close port 80.** It disables HTTP *management*;
+  port 80 stays open serving only a JS redirect to HTTPS. That's expected/secure.

infrastructure/roles/tplink_switch_harden/defaults/main.yml

@@ -0,0 +1,59 @@
+---
+# tplink_switch_harden — defaults
+#
+# ─────────────────────────────────────────────────────────────────────────────
+#  SAFETY: switch_apply defaults to FALSE.
+#  A normal run only logs in and captures `show system-info` (proves access,
+#  changes nothing). Config is applied ONLY with -e switch_apply=true.
+# ─────────────────────────────────────────────────────────────────────────────
+switch_apply: false
+switch_expect_timeout: 30
+
+# Network re-addressing (static IP + gateway) is DISRUPTIVE: applying it drops
+# the session because the management IP changes. Kept behind its own toggle and
+# OFF by default, because in steady state (already at the static IP) it is a
+# no-op, and the one-time DHCP->static cutover is handled deliberately, not on
+# every run. Enable with -e switch_apply_network=true during the cutover.
+switch_apply_network: false
+
+# ─────────────────────────────────────────────────────────────────────────────
+#  VERIFIED against SG2428LP firmware via on-device `?` help (2026-06-09).
+#  All entries are idempotent — most are already at these values by factory
+#  default, so re-applying is a harmless no-op. Each runs inside `configure`.
+# ─────────────────────────────────────────────────────────────────────────────
+switch_harden_commands:
+  - { cmd: "no ip http server",        desc: "disable plaintext HTTP management (port 80)" }
+  - { cmd: "ip http secure-server",    desc: "ensure HTTPS management is enabled" }
+  - { cmd: "telnet disable",           desc: "disable Telnet (port 23)" }
+  - { cmd: "ip ssh server",            desc: "ensure SSH server is enabled" }
+  - { cmd: "ip ssh version v2",        desc: "enable SSH protocol v2" }
+  - { cmd: "no snmp-server",           desc: "ensure SNMP agent is disabled" }
+  - { cmd: "spanning-tree",            desc: "enable spanning tree globally" }
+  - { cmd: "spanning-tree mode rstp",  desc: "use RSTP" }
+  - { cmd: "loopback-detection",       desc: "enable global loopback detection" }
+
+# Device identity — safe, non-disruptive (applied with the hardening set).
+# VERIFY token: JetStream may use `hostname` or `system name`.
+switch_identity_commands:
+  - { cmd: "hostname {{ switch_hostname }}", desc: "set system hostname" }
+
+# Static management addressing — DISRUPTIVE (drops session on the IP change).
+# Applied only when switch_apply_network=true. VERIFY exact syntax on-device:
+# the mgmt interface, the dhcp-client disable, and gateway form all vary.
+switch_network_commands:
+  - { cmd: "interface vlan {{ switch_mgmt_vlan }}",                         desc: "enter mgmt VLAN interface" }
+  - { cmd: "ip address {{ switch_mgmt_ip_static }} {{ switch_mgmt_netmask }}", desc: "set static mgmt IP (overrides DHCP alloc)" }
+  - { cmd: "exit",                                                          desc: "leave interface context" }
+  - { cmd: "ip route 0.0.0.0 0.0.0.0 {{ switch_mgmt_gateway }}",            desc: "default route via gateway" }
+
+# ─────────────────────────────────────────────────────────────────────────────
+#  OPT-IN, HIGHER-RISK hardening — left OUT of the default set on purpose.
+#  Management access-control can lock you out if the control node is not inside
+#  the permitted range; only enable after confirming syntax AND your admin
+#  subnet. Provided here as a reference to fold into switch_harden_commands once
+#  verified. Do the ACL LAST and from a host inside the permitted range.
+# ─────────────────────────────────────────────────────────────────────────────
+# switch_harden_commands_optional:
+#   - { cmd: "user idle-timeout 10",                      desc: "10-min CLI idle timeout" }
+#   - { cmd: "user access-control ip-based enable",       desc: "enable mgmt IP allowlist" }
+#   - { cmd: "user access-control ip-based 10.0.0.0 255.255.255.0", desc: "permit admin subnet" }

infrastructure/roles/tplink_switch_harden/tasks/main.yml

@@ -0,0 +1,60 @@
+---
+# tplink_switch_harden — drive the JetStream CLI via a rendered expect engine.
+# Runs on the control node (play uses connection: local); the engine SSHes to
+# the switch. The admin password is passed only through the SWITCH_PW env var.
+
+- name: Assert admin password is available from vault
+  ansible.builtin.assert:
+    that:
+      - switch_admin_password is defined
+      - switch_admin_password | length > 0
+    fail_msg: >-
+      vault_sg2428lp_admin_password is empty/undefined. Populate it with
+      etc/set-switch-1password.sh (op -> ansible-vault bridge).
+
+- name: Create temp dir for the rendered engine
+  ansible.builtin.tempfile:
+    state: directory
+    suffix: switch-harden
+  register: _sw_tmp
+  changed_when: false
+
+- name: Render hardening expect engine
+  ansible.builtin.template:
+    src: harden.exp.j2
+    dest: "{{ _sw_tmp.path }}/harden.exp"
+    mode: "0700"
+  changed_when: false
+
+- name: Run switch hardening engine ({{ 'APPLY' if switch_apply | bool else 'DRY/show-only' }})
+  ansible.builtin.command:
+    cmd: "expect -f {{ _sw_tmp.path }}/harden.exp"
+  environment:
+    SWITCH_PW: "{{ switch_admin_password }}"
+  register: _sw_run
+  no_log: true                       # protects SWITCH_PW in the environment
+  changed_when: "'>>>===APPLY-END===<<<' in _sw_run.stdout"
+  failed_when: false                 # evaluate explicitly below so output prints
+
+- name: Show switch engine output
+  ansible.builtin.debug:
+    var: _sw_run.stdout_lines
+
+- name: Remove temp dir
+  ansible.builtin.file:
+    path: "{{ _sw_tmp.path }}"
+    state: absent
+  changed_when: false
+
+- name: Assert hardening engine succeeded
+  ansible.builtin.assert:
+    that:
+      - _sw_run.rc == 0
+      - "'AUTH-FAILED' not in _sw_run.stdout"
+      - "'FORCED-PASSWORD-CHANGE' not in _sw_run.stdout"
+      - "'LOGIN-TIMEOUT' not in _sw_run.stdout"
+    fail_msg: "Switch engine failed — review the output above (rc={{ _sw_run.rc }})."
+    success_msg: >-
+      {{ 'Hardening applied and saved to startup-config.'
+         if switch_apply | bool else
+         'Access verified (dry run). Re-run with -e switch_apply=true to apply.' }}

infrastructure/roles/tplink_switch_harden/templates/harden.exp.j2

@@ -0,0 +1,92 @@
+#!/usr/bin/expect -f
+# RENDERED by Ansible (role: tplink_switch_harden). Do not edit the rendered copy.
+# Drives a TP-Link JetStream switch CLI over SSH. Password comes from $env(SWITCH_PW)
+# so it never appears in this file, the inventory, or the process arguments.
+set timeout {{ switch_expect_timeout | default(30) }}
+set host "{{ switch_mgmt_ip }}"
+set user "{{ switch_admin_user }}"
+set pw $env(SWITCH_PW)
+set apply {{ '1' if switch_apply | bool else '0' }}
+log_user 1
+
+# Device prompt: User EXEC '>' or Privileged '#', optionally prefixed by hostname.
+set PROMPT {[\w./-]*[>#] ?$}
+
+proc wait_prompt {} {
+    global PROMPT
+    expect {
+        -re {(?i)--more--|press any key|\(q to quit\)} { send -- " "; exp_continue }
+        -re $PROMPT {}
+        timeout { puts "\n>>>TIMEOUT-AT-PROMPT<<<"; exit 21 }
+        eof     { puts "\n>>>EOF-AT-PROMPT<<<"; exit 22 }
+    }
+}
+
+proc do {cmd} {
+    send -- "$cmd\r"
+    wait_prompt
+}
+
+# NOTE: PubkeyAuthentication=no + PreferredAuthentications=password is REQUIRED.
+# The JetStream SSH server drops the connection when the client offers publickey
+# auth first (the default), which made logins appear to "close after KEX" at
+# random. Forcing password-only auth makes login deterministic.
+spawn ssh -tt -o StrictHostKeyChecking=no -o UserKnownHostsFile=/dev/null \
+    -o ConnectTimeout=15 -o NumberOfPasswordPrompts=1 \
+    -o PreferredAuthentications=password -o PubkeyAuthentication=no $user@$host
+
+expect {
+    -re {(?i)password:} { log_user 0; send -- "$pw\r"; log_user 1; exp_continue }
+    -re {(?i)(change.*password|new password|must.*change|set.*new password)} {
+        puts "\n>>>FORCED-PASSWORD-CHANGE<<< switch still at first-login; aborting."; exit 9
+    }
+    -re {(?i)(permission denied|authentication fail)} { puts "\n>>>AUTH-FAILED<<<"; exit 8 }
+    -re $PROMPT {}
+    timeout { puts "\n>>>LOGIN-TIMEOUT<<<"; exit 2 }
+    eof     { puts "\n>>>LOGIN-EOF<<<"; exit 3 }
+}
+
+# Enter privileged EXEC (default enable password is blank).
+send -- "enable\r"
+expect {
+    -re {(?i)password:} { send -- "\r"; wait_prompt }
+    -re $PROMPT {}
+    timeout {}
+}
+
+# Always capture a read-only snapshot — proves access, changes nothing.
+puts "\n>>>===SHOW-START===<<<"
+do "show system-info"
+puts "\n>>>===SHOW-END===<<<"
+
+if {$apply == 1} {
+    puts "\n>>>===APPLY-START===<<<"
+    do "configure"
+{% for c in switch_harden_commands %}
+    do {{ "{" ~ c.cmd ~ "}" }}    ;# {{ c.desc }}
+{% endfor %}
+{% for c in switch_identity_commands %}
+    do {{ "{" ~ c.cmd ~ "}" }}    ;# {{ c.desc }}
+{% endfor %}
+{% if switch_apply_network | bool %}
+    # NETWORK re-addressing. In steady state (already at the static IP) these are
+    # no-ops. During the one-time DHCP->static cutover the `ip address` line drops
+    # the session — do that via the dedicated migration step, not this engine.
+{% for c in switch_network_commands %}
+    do {{ "{" ~ c.cmd ~ "}" }}    ;# {{ c.desc }}
+{% endfor %}
+{% endif %}
+    do "end"
+    # Persist running-config to startup (JetStream may ask Y/N).
+    send -- "copy running-config startup-config\r"
+    expect {
+        -re {(?i)(y/n|are you sure|\[y/n\]|continue)} { send -- "Y\r"; wait_prompt }
+        -re $PROMPT {}
+        timeout {}
+    }
+    puts "\n>>>===APPLY-END===<<<"
+}
+
+send -- "exit\r"
+catch {expect eof}
+puts "\n>>>===DONE===<<<"