Merge remote-tracking branch 'origin/develop' into benvin/mail-edge-dnsmasq

# Conflicts:
#	hieradata/roles/infra/mail/gateway.yaml
#	site/roles/manifests/infra/mail/gateway.pp

hieradata/country/au/region/drw1/infra/storage/consul.yaml

@@ -2,6 +2,6 @@
 profiles::consul::server::bootstrap_count: 3
 profiles::consul::server::raft_multiplier: 10
 profiles::consul::server::primary_datacenter: 'au-syd1'
-profiles::consul::server::join_remote_regions: false
+profiles::consul::server::join_remote_regions: true
 profiles::consul::server::remote_regions:
   - syd1

hieradata/country/au/region/syd1/infra/storage/consul.yaml

@@ -2,6 +2,6 @@
 profiles::consul::server::bootstrap_count: 3
 profiles::consul::server::raft_multiplier: 10
 profiles::consul::server::primary_datacenter: 'au-syd1'
-profiles::consul::server::join_remote_regions: false
+profiles::consul::server::join_remote_regions: true
 profiles::consul::server::remote_regions:
   - drw1

hieradata/nodes/aumel1nxvm2132.main.unkin.net.yaml

@@ -0,0 +1,4 @@
+---
+profiles::postfix::gateway::myhostname: 'mta2.edge.unkin.net'
+profiles::pki::vault::alt_names:
+  - mta2.edge.unkin.net

hieradata/nodes/ausyd2nxvm2130.main.unkin.net.yaml

@@ -0,0 +1,4 @@
+---
+profiles::postfix::gateway::myhostname: 'mta1.edge.unkin.net'
+profiles::pki::vault::alt_names:
+  - mta1.edge.unkin.net

hieradata/os/AlmaLinux/all_releases.yaml

@@ -147,6 +147,18 @@ profiles::yum::global::repos:
     baseurl: https://artifactapi.k8s.syd1.au.unkin.net/api/v1/remote/postgresql/common/redhat/rhel-%{facts.os.release.major}-%{facts.os.architecture}
     gpgkey: https://artifactapi.k8s.syd1.au.unkin.net/api/v1/remote/postgresql/keys/PGDG-RPM-GPG-KEY-RHEL
 
+  # Rspamd repositories
+  rspamd:
+    ensure: absent
+    name: rspamd
+    descr: rspamd repository
+    target: /etc/yum.repos.d/rspamd.repo
+    baseurl: https://artifactapi.k8s.syd1.au.unkin.net/api/v1/remote/rspamd/rpm-stable/centos-%{facts.os.release.major}/x86_64
+    gpgkey: https://artifactapi.k8s.syd1.au.unkin.net/api/v1/remote/rspamd/rpm/rspamd.asc
+    gpgcheck: 1
+    repo_gpgcheck: 1
+    mirrorlist: absent
+
   # Ceph repositories
   ceph:
     ensure: absent

hieradata/roles/infra/mail/gateway.yaml

@@ -1,25 +1,90 @@
 ---
+hiera_include:
+  - redis
 
-# additional altnames
-profiles::pki::vault::alt_names:
-  - in-mta.main.unkin.net
+# no consul datacentre at the edge, so nothing to register or scrape via consul
+profiles::consul::client::enable: false
+exporters::node_exporter::enable: false
+
+# ssh only from the edge vpcs (syd, mel) and the lab over the vpn route
+ssh::server::options:
+  AllowUsers: '*@198.18.30.0/24 *@198.18.31.0/24 *@198.18.0.0/16 *@10.10.12.200/32'
+
+# local caching resolver for postfix and rspamd, forwarding to the lab resolver over the vpn route
+dnsmasq::no_resolv: true
+dnsmasq::cache_size: 10000
+dnsmasq::servers:
+  - 198.18.200.7
+profiles::dns::base::nameservers:
+  - 127.0.0.1
+
+profiles::yum::global::repos:
+  rspamd:
+    ensure: present
+
+# local redis for rspamd bayes and history; bayes tokens carry a ttl so volatile-ttl can evict
+redis::maxmemory: '128mb'
+redis::maxmemory_policy: 'volatile-ttl'
+
+# local rspamd, deterministic checks only; module defaults cover spf/dkim/dmarc/arc/rbl/phishing/mime_types/fuzzy_check
+profiles::rspamd::edge::rules:
+  edge.cf: 'https://git.unkin.net/unkin/rspamd-rules/raw/branch/main/edge.cf'
+profiles::rspamd::edge::local_d:
+  worker-proxy.inc: |
+    bind_socket = "127.0.0.1:11332";
+    milter = yes;
+    timeout = 120s;
+    count = 1;
+    upstream "local" {
+      default = yes;
+      self_scan = yes;
+    }
+  worker-normal.inc: |
+    enabled = false;
+  redis.conf: |
+    servers = "127.0.0.1:6379";
+  actions.conf: |
+    reject = 13;
+    add_header = 6;
+    rewrite_subject = null;
+    greylist = null;
+  greylist.conf: |
+    enabled = false;
+  gpt.conf: |
+    enabled = false;
+  classifier-bayes.conf: |
+    backend = "redis";
+    expire = 8640000;
+    autolearn {
+      spam_threshold = 6.0;
+      ham_threshold = -0.5;
+      check_balance = true;
+    }
+  spamassassin.conf: |
+    ruleset = "/etc/rspamd/rules/*.cf";
+    alpha = 0.1;
+  settings.conf: |
+    smarthost_outbound {
+      priority = high;
+      ip = "103.216.191.185";
+      want_spam = yes;
+    }
 
 # base postfix configuration (passed to postfix class)
 postfix::relayhost: 'direct'
-postfix::myorigin: 'main.unkin.net'
+postfix::myorigin: 'unkin.net'
 postfix::mydestination: 'blank'
-postfix::mynetworks: '127.0.0.0/8 [::1]/128'
+postfix::mynetworks: '127.0.0.0/8 [::1]/128 103.216.191.185/32'
 postfix::mta: true
 postfix::manage_aliases: true
 
-# profile parameters for customization
-profiles::postfix::gateway::myhostname: 'in-mta.main.unkin.net'
-
 # postfix map content (templates)
 profiles::postfix::gateway::relay_recipients_maps:
+  '@unkin.net': 'OK'
   '@main.unkin.net': 'OK'
 
 profiles::postfix::gateway::relay_domains_maps:
+  'unkin.net': 'OK'
   'main.unkin.net': 'OK'
 
 profiles::postfix::gateway::postscreen_access_maps:
@@ -32,21 +97,28 @@ profiles::postfix::gateway::helo_access_maps:
   'unknown': 'REJECT'
   'localhost': 'REJECT You are not localhost'
 
+profiles::postfix::gateway::smtp_tls_policy_maps:
+  '[mail-in.main.unkin.net]:25': 'secure match=hostname'
+
 # postfix transports
 postfix::transports:
+  'unkin.net':
+    ensure: present
+    destination: 'relay'
+    nexthop: '[mail-in.main.unkin.net]:25'
   'main.unkin.net':
     ensure: present
     destination: 'relay'
-    nexthop: 'mail-in.main.unkin.net:25'
+    nexthop: '[mail-in.main.unkin.net]:25'
 
 # postfix virtuals
 postfix::virtuals:
   'root':
     ensure: present
-    destination: 'ben@main.unkin.net'
+    destination: 'ben@unkin.net'
   'postmaster':
     ensure: present
-    destination: 'ben@main.unkin.net'
+    destination: 'ben@unkin.net'
   'abuse':
     ensure: present
-    destination: 'ben@main.unkin.net'
+    destination: 'ben@unkin.net'

hieradata/roles/infra/storage/consul.yaml

@@ -13,19 +13,9 @@ profiles::consul::server::addresses:
   grpc_tls: "%{::networking.ip}"
 profiles::consul::server::ports:
   grpc: 8502
-  grpc_tls: 8503
   dns: 8600
   http: 8500
   https: -1
-profiles::consul::server::tls:
-  defaults:
-    ca_file: /etc/pki/ca-trust/source/anchors/vaultcaroot.pem
-    cert_file: /etc/pki/tls/vault/full_chain.crt
-    key_file: /etc/pki/tls/vault/private.key
-  internal_rpc:
-    verify_incoming: false
-    verify_outgoing: false
-    verify_server_hostname: false
 profiles::consul::server::acl:
   enabled: true
   default_policy: 'deny'

modules/libs/lib/facter/subnet_facts.rb

@@ -22,27 +22,45 @@ class SubnetAttributes
     '198.18.26.0/24' => { environment: 'prod', region: 'syd1', country: 'au', zone: 'common' }, # common node0010
     '198.18.27.0/24' => { environment: 'prod', region: 'syd1', country: 'au', zone: 'common' }, # common node0011
     '198.18.28.0/24' => { environment: 'prod', region: 'syd1', country: 'au', zone: 'common' }, # common node0012
-    '198.18.29.0/24' => { environment: 'prod', region: 'syd1', country: 'au', zone: 'common' }  # common node0013
+    '198.18.29.0/24' => { environment: 'prod', region: 'syd1', country: 'au', zone: 'common' }, # common node0013
+    '198.18.30.0/24' => { environment: 'prod', region: 'syd2', country: 'au', zone: 'dmz' },    # binarylane syd dmz vpc
+    '198.18.31.0/24' => { environment: 'prod', region: 'mel1', country: 'au', zone: 'dmz' }     # binarylane mel dmz vpc
   }.freeze
 
   # Default attributes if no subnet matches, also defined as a constant
   DEFAULT_ATTRIBUTES = { environment: 'unknown', region: 'unknown', country: 'unknown' }.freeze
 
-  # provide ip to return attributes
-  def self.attributes(ip)
-    SUBNET_TO_ATTRIBUTES.each do |subnet, attrs|
-      return attrs if IPAddr.new(subnet).include?(IPAddr.new(ip))
-    end
+  # loopbacks and wg /31s sit in syd1 ranges, so they must not decide where a host lives
+  SKIPPED_INTERFACES = /\A(lo\z|dum|wg)/.freeze
 
-    DEFAULT_ATTRIBUTES
+  # primary ip first, then the remaining interfaces in name order
+  def self.from_networking(networking)
+    match(networking['ip']) || interface_match(networking['interfaces'] || {}) || DEFAULT_ATTRIBUTES
+  end
+
+  def self.interface_match(interfaces)
+    interfaces.sort
+              .reject { |name, _interface| name.match?(SKIPPED_INTERFACES) }
+              .flat_map { |_name, interface| addresses(interface) }
+              .lazy.filter_map { |address| match(address) }.first
+  end
+
+  def self.addresses(interface)
+    (interface['bindings'] || []).map { |binding| binding['address'] }
+  end
+
+  def self.match(ip)
+    return unless ip
+
+    addr = IPAddr.new(ip)
+    SUBNET_TO_ATTRIBUTES.each do |subnet, attrs|
+      return attrs if IPAddr.new(subnet).include?(addr)
+    end
+    nil
   end
 end
 
-# Use the primary IP address from facts
-ip = Facter.value(:networking)['ip']
-
-# Call the class method directly without creating an instance
-subnet_attributes = SubnetAttributes.attributes(ip)
+subnet_attributes = SubnetAttributes.from_networking(Facter.value(:networking))
 
 # Add separate facts for environment, region, and country
 Facter.add('environment') { setcode { subnet_attributes[:environment] } }

modules/libs/test/subnet_facts_check.rb

@@ -0,0 +1,36 @@
+# frozen_string_literal: true
+
+# ruby modules/libs/test/subnet_facts_check.rb
+module Facter
+  def self.value(_name)
+    {}
+  end
+
+  def self.add(*); end
+end
+load File.expand_path('../lib/facter/subnet_facts.rb', __dir__)
+
+iface = ->(*addresses) { { 'bindings' => addresses.map { |address| { 'address' => address } } } }
+
+checks = {
+  'primary ip wins over other interfaces' => [
+    { 'ip' => '198.18.17.5', 'interfaces' => { 'eth1' => iface.call('198.18.13.5') } }, 'drw1'
+  ],
+  'unmatched primary ip falls back to an interface' => [
+    { 'ip' => '203.0.113.5', 'interfaces' => { 'eth0' => iface.call('203.0.113.5'), 'eth1' => iface.call('198.18.24.11') } }, 'syd1'
+  ],
+  'lo, dum and wg interfaces are skipped' => [
+    { 'ip' => '203.0.113.5',
+      'interfaces' => { 'dum0' => iface.call('198.18.2.161'), 'lo' => iface.call('198.18.19.1'),
+                        'wg0' => iface.call('198.18.2.177') } }, 'unknown'
+  ],
+  'interfaces are scanned in name order' => [
+    { 'ip' => '203.0.113.5', 'interfaces' => { 'eth2' => iface.call('198.18.24.1'), 'eth1' => iface.call('198.18.17.1') } }, 'drw1'
+  ],
+  'missing ip and interfaces resolve to unknown' => [{}, 'unknown']
+}
+
+failures = checks.reject { |_name, (networking, expected)| SubnetAttributes.from_networking(networking)[:region] == expected }
+failures.each_key { |name| warn "FAIL: #{name}" }
+abort if failures.any?
+puts "ok (#{checks.size} checks)"

site/profiles/manifests/consul/client.pp

@@ -12,8 +12,13 @@ class profiles::consul::client (
   Array[Hash]          $node_rules         = [],
   Hash                 $ports              = {},
   Stdlib::IP::Address  $host_addr          = $facts['networking']['ip'],
+  Boolean              $enable             = true,
 ) {
 
+  unless $enable {
+    return()
+  }
+
   if $facts['enc_role'] != $members_role {
 
     # set a datacentre/cluster name

site/profiles/manifests/consul/server.pp

@@ -11,7 +11,6 @@ class profiles::consul::server (
   Hash      $acl = {},
   Hash      $ports = {},
   Hash      $addresses = {},
-  Hash      $tls = {},
   Boolean   $members_lookup     = false,
   String    $members_role       = undef,
   Array     $consul_servers     = [],
@@ -113,8 +112,6 @@ class profiles::consul::server (
           'acl'                  => $acl,
           'ports'                => $ports,
           'addresses'            => $addresses,
-          'tls'                  => $tls,
-          'auto_reload_config'   => true,
           'disable_remote_exec'  => $disable_remote_exec,
           'disable_update_check' => $disable_update_check,
           'domain'               => $domain,

site/profiles/manifests/postfix/gateway.pp

@@ -3,8 +3,7 @@ class profiles::postfix::gateway (
   Stdlib::Absolutepath $tls_key_file  = '/etc/pki/tls/vault/certificate.pem',
   Stdlib::Absolutepath $tls_ca_file   = '/etc/pki/ca-trust/extracted/pem/tls-ca-bundle.pem',
   String $myhostname                  = $trusted['certname'],
-  String $message_size_limit          = '133169152',
-  String $mailbox_size_limit          = '133169152',
+  String $message_size_limit          = '52428800',
   String $local_transport             = 'error:No local mail delivery',
   Boolean $enable_postscreen          = true,
   Array[String] $alias_maps           = [
@@ -12,12 +11,12 @@ class profiles::postfix::gateway (
     'hash:/etc/postfix/aliases',
   ],
   Array[String] $postscreen_dnsbl_sites = [
-    'zen.spamhaus.org*3',
-    'b.barracudacentral.org=127.0.0.[2..11]*2',
-    'bl.spameatingmonkey.net*2',
-    'bl.spamcop.net',
-    'dnsbl.sorbs.net',
-    'swl.spamhaus.org*-4',
+    'zen.spamhaus.org=127.0.0.[2..11]*3',
+    'b.barracudacentral.org=127.0.0.2*2',
+    'bl.mailspike.net=127.0.0.2*1',
+    'psbl.surriel.com=127.0.0.2*1',
+    'bl.spameatingmonkey.net=127.0.0.2*2',
+    'bl.spamcop.net=127.0.0.2',
     'list.dnswl.org=127.[0..255].[0..255].0*-2',
     'list.dnswl.org=127.[0..255].[0..255].1*-4',
     'list.dnswl.org=127.[0..255].[0..255].[2..3]*-6',
@@ -25,7 +24,6 @@ class profiles::postfix::gateway (
   Array[String] $smtpd_client_restrictions = [
     'permit_sasl_authenticated',
     'permit_mynetworks',
-    'reject_rbl_client zen.spamhaus.org',
   ],
   Array[String] $smtpd_sender_restrictions = [
     'permit_sasl_authenticated',
@@ -47,6 +45,7 @@ class profiles::postfix::gateway (
     'permit_mynetworks',
     'reject_unauth_destination',
   ],
+  String $smtpd_milters = 'inet:127.0.0.1:11332',
   Hash[String, String] $smtp_tls_policy_maps = {},
   Hash[String, String] $sender_canonical_maps = {},
   Hash[String, String] $sender_access_maps = {},
@@ -73,7 +72,7 @@ class profiles::postfix::gateway (
         'value' => 'permit_mynetworks, cidr:/etc/postfix/postscreen_access'
       },
       'postscreen_blacklist_action' => {
-        'value' => 'enforce'
+        'value' => 'drop'
       },
       'postscreen_cache_map' => {
         'value' => 'btree:$data_directory/postscreen_cache'
@@ -87,6 +86,10 @@ class profiles::postfix::gateway (
       'postscreen_dnsbl_threshold' => {
         'value' => '2'
       },
+      # dnswl-listed clients (net score <= -1) skip the slow greet tests; el9 postfix 3.5 predates the allowlist rename
+      'postscreen_dnsbl_whitelist_threshold' => {
+        'value' => '-1'
+      },
       'postscreen_greet_action' => {
         'value' => 'enforce'
       },
@@ -108,9 +111,6 @@ class profiles::postfix::gateway (
     'alias_database' => {
       'value' => $alias_maps_string
     },
-    'default_destination_recipient_limit' => {
-      'value' => '1'
-    },
     'disable_vrfy_command' => {
       'value' => 'yes'
     },
@@ -129,9 +129,6 @@ class profiles::postfix::gateway (
     'local_transport' => {
       'value' => $local_transport
     },
-    'mailbox_size_limit' => {
-      'value' => $mailbox_size_limit
-    },
     'message_size_limit' => {
       'value' => $message_size_limit
     },
@@ -141,8 +138,20 @@ class profiles::postfix::gateway (
     'non_smtpd_milters' => {
       'ensure' => 'blank'
     },
-    'qmqpd_authorized_clients' => {
-      'value' => '127.0.0.1 [::1]'
+    'milter_connect_timeout' => {
+      'value' => '10s'
+    },
+    'milter_content_timeout' => {
+      'value' => '60s'
+    },
+    'milter_default_action' => {
+      'value' => 'tempfail'
+    },
+    'milter_mail_macros' => {
+      'value' => 'i {mail_addr} {client_addr} {client_name} {auth_authen}'
+    },
+    'milter_protocol' => {
+      'value' => '6'
     },
     'recipient_canonical_maps' => {
       'value' => 'hash:/etc/postfix/recipient_canonical'
@@ -177,7 +186,7 @@ class profiles::postfix::gateway (
     'smtp_tls_session_cache_database' => {
       'value' => 'btree:/var/lib/postfix/smtp_tls_session_cache'
     },
-    'smtp_use_tls' => {
+    'smtp_send_xforward_command' => {
       'value' => 'yes'
     },
     'smtpd_banner' => {
@@ -211,7 +220,7 @@ class profiles::postfix::gateway (
       'value' => 'check_helo_access hash:/etc/postfix/helo_access, reject_invalid_hostname'
     },
     'smtpd_milters' => {
-      'value' => 'inet:127.0.0.1:33333'
+      'value' => $smtpd_milters
     },
     'smtpd_recipient_restrictions' => {
       'value' => join($smtpd_recipient_restrictions, ', ')
@@ -255,13 +264,10 @@ class profiles::postfix::gateway (
     'smtpd_tls_session_cache_timeout' => {
       'value' => '3600s'
     },
-    'smtpd_use_tls' => {
-      'value' => 'yes'
-    },
     'tls_medium_cipherlist' => {
       'value' => join([
         'ECDSA+AESGCM:ECDH+AESGCM:DH+AESGCM:ECDSA+AES:ECDH+AES:DH+AES',
-        'ECDH+3DES:DH+3DES:RSA+AESGCM:RSA+AES:RSA+3DES:!aNULL:!MD5:!DSS'
+        'RSA+AESGCM:RSA+AES:!aNULL:!MD5:!DSS:!3DES'
       ], ':')
     },
     'tls_preempt_cipherlist' => {

site/profiles/manifests/rspamd/edge.pp

@@ -0,0 +1,68 @@
+# local rspamd milter for the mail edge gateways
+# - local_d: /etc/rspamd/local.d file name => content
+# - rules: /etc/rspamd/rules file name => url, re-fetched every puppet run; fetch failures keep the last good copy
+class profiles::rspamd::edge (
+  Hash[String, String]           $local_d = {},
+  Hash[String, Stdlib::HTTPSUrl] $rules   = {},
+) {
+
+  package { 'rspamd':
+    ensure => installed,
+  }
+
+  $local_d.each |$name, $content| {
+    file { "/etc/rspamd/local.d/${name}":
+      ensure  => file,
+      owner   => 'root',
+      group   => 'root',
+      mode    => '0644',
+      content => $content,
+      require => Package['rspamd'],
+      notify  => Service['rspamd'],
+    }
+  }
+
+  file { '/etc/rspamd/rules':
+    ensure  => directory,
+    owner   => 'root',
+    group   => 'root',
+    mode    => '0755',
+    require => Package['rspamd'],
+  }
+
+  $rules.each |$name, $url| {
+    $path = "/etc/rspamd/rules/${name}"
+
+    # empty placeholder so rspamd starts before the first successful fetch
+    file { $path:
+      ensure  => file,
+      owner   => 'root',
+      group   => 'root',
+      mode    => '0644',
+      content => '',
+      replace => false,
+    }
+
+    # a failed or unchanged fetch skips the swap without failing the run
+    exec { "rspamd_rules_fetch_${name}":
+      command => "/usr/bin/mv -f ${path}.new ${path}",
+      onlyif  => "/usr/bin/curl -fsS --max-time 30 -o ${path}.new ${url} && ! /usr/bin/cmp -s ${path}.new ${path}",
+      require => File[$path],
+      notify  => Exec['rspamd_reload'],
+    }
+  }
+
+  # reload keeps the milter answering while config changes apply
+  service { 'rspamd':
+    ensure  => running,
+    enable  => true,
+    restart => '/usr/bin/systemctl reload rspamd',
+    require => Package['rspamd'],
+  }
+
+  exec { 'rspamd_reload':
+    command     => '/usr/bin/systemctl reload rspamd',
+    refreshonly => true,
+    require     => Service['rspamd'],
+  }
+}

site/roles/manifests/infra/mail/gateway.pp

@@ -6,6 +6,11 @@ class roles::infra::mail::gateway {
   }else{
     include profiles::defaults
     include profiles::base
+    include dnsmasq
+    include profiles::rspamd::edge
     include profiles::postfix::gateway
+
+    # resolv.conf points at dnsmasq, so it must be running first
+    Class['dnsmasq'] -> Class['profiles::dns::resolvconf']
   }
 }