Improve filesystem sandboxing for tile services

[chef.git] / cookbooks / tile / recipes / default.rb

diff --git a/cookbooks/tile/recipes/default.rb b/cookbooks/tile/recipes/default.rb
index 241b73a8e9283d6be225a88cbaf035bd067d35e9..576c0f6adf080718f0e35dd637bac1b208a8964c 100644 (file)
--- a/cookbooks/tile/recipes/default.rb
+++ b/cookbooks/tile/recipes/default.rb
@@ -25,6 +25,7 @@ include_recipe "nodejs"
 include_recipe "postgresql"
 include_recipe "prometheus"
 include_recipe "python"
+include_recipe "ruby"
 include_recipe "tools"
 
 blocks = data_bag_item("tile", "blocks")
@@ -41,6 +42,10 @@ apache_module "tile" do
   conf "tile.conf.erb"
 end
 
+apache_conf "renderd" do
+  action :disable
+end
+
 ssl_certificate node[:fqdn] do
   domains [node[:fqdn], "tile.openstreetmap.org", "render.openstreetmap.org"]
   notifies :reload, "service[apache2]"
@@ -49,18 +54,22 @@ end
 remote_file "#{Chef::Config[:file_cache_path]}/fastly-ip-list.json" do
   source "https://api.fastly.com/public-ip-list"
   compile_time true
+  ignore_failure true
 end
 
-tilecaches = search(:node, "roles:tilecache").sort_by { |n| n[:hostname] }
 fastlyips = JSON.parse(IO.read("#{Chef::Config[:file_cache_path]}/fastly-ip-list.json"))
 
 apache_site "default" do
-  action [:disable]
+  action :disable
+end
+
+apache_site "tileserver_site" do
+  action :disable
 end
 
 apache_site "tile.openstreetmap.org" do
   template "apache.erb"
-  variables :caches => tilecaches, :fastly => fastlyips["addresses"]
+  variables :fastly => fastlyips["addresses"]
 end
 
 template "/etc/logrotate.d/apache2" do
@@ -76,16 +85,25 @@ directory "/srv/tile.openstreetmap.org" do
   mode "755"
 end
 
+directory "/srv/tile.openstreetmap.org/conf" do
+  owner "tile"
+  group "tile"
+  mode "755"
+end
+
+file "/srv/tile.openstreetmap.org/conf/ip.map" do
+  owner "tile"
+  group "adm"
+  mode "644"
+end
+
 package "renderd"
 
 systemd_service "renderd" do
-  description "Mapnik rendering daemon"
+  dropin "chef"
   after "postgresql.service"
   wants "postgresql.service"
-  user "www-data"
-  exec_start "/usr/bin/renderd -f"
-  runtime_directory "renderd"
-  standard_error "null"
+  limit_nofile 4096
   private_tmp true
   private_devices true
   private_network true
@@ -95,6 +113,10 @@ systemd_service "renderd" do
   restart "on-failure"
 end
 
+systemd_service "renderd" do
+  action :delete
+end
+
 service "renderd" do
   action [:enable, :start]
   subscribes :restart, "systemd_service[renderd]"
@@ -135,6 +157,7 @@ end
 package %w[
   python3-cairo
   python3-mapnik
+  python3-pyproj
   python3-setuptools
 ]
 
@@ -142,24 +165,6 @@ python_package "pyotp" do
   python_version "3"
 end
 
-package %w[
-  fonts-noto-cjk
-  fonts-noto-hinted
-  fonts-noto-unhinted
-  fonts-hanazono
-  ttf-unifont
-]
-
-["NotoSansArabicUI-Regular.ttf", "NotoSansArabicUI-Bold.ttf"].each do |font|
-  remote_file "/usr/share/fonts/truetype/noto/#{font}" do
-    action :create_if_missing
-    source "https://github.com/googlei18n/noto-fonts/raw/master/hinted/#{font}"
-    owner "root"
-    group "root"
-    mode "644"
-  end
-end
-
 directory "/srv/tile.openstreetmap.org/cgi-bin" do
   owner "tile"
   group "tile"
@@ -272,13 +277,19 @@ nodejs_package "carto"
 
 systemd_service "update-lowzoom@" do
   description "Low zoom tile update service for %i layer"
+  conflicts "render-lowzoom.service"
   user "tile"
   exec_start "/bin/bash /usr/local/bin/update-lowzoom-%i"
+  runtime_directory "update-lowzoom-%i"
   private_tmp true
   private_devices true
   private_network true
-  protect_system "full"
+  protect_system "strict"
   protect_home true
+  read_write_paths [
+    "/srv/tile.openstreetmap.org/tiles/%i",
+    "/var/log/tile"
+  ]
   no_new_privileges true
   restart "on-failure"
 end
@@ -314,15 +325,15 @@ node[:tile][:styles].each do |name, details|
 
   details[:tile_directories].each do |directory|
     directory directory[:name] do
-      owner "www-data"
-      group "www-data"
+      owner "_renderd"
+      group "_renderd"
       mode "755"
     end
 
     directory[:min_zoom].upto(directory[:max_zoom]) do |zoom|
       directory "#{directory[:name]}/#{zoom}" do
-        owner "www-data"
-        group "www-data"
+        owner "_renderd"
+        group "_renderd"
         mode "755"
       end
 
@@ -355,6 +366,17 @@ node[:tile][:styles].each do |name, details|
     group "tile"
   end
 
+  if details[:fonts_script]
+    execute details[:fonts_script] do
+      action :nothing
+      command details[:fonts_script]
+      cwd style_directory
+      user "tile"
+      group "tile"
+      subscribes :run, "git[#{style_directory}]"
+    end
+  end
+
   execute "#{style_directory}/project.mml" do
     action :nothing
     command "carto -a 3.0.0 project.mml > project.xml"
@@ -382,6 +404,11 @@ postgresql_user "tomh" do
   superuser true
 end
 
+postgresql_user "pnorman" do
+  cluster node[:tile][:database][:cluster]
+  superuser true
+end
+
 postgresql_user "tile" do
   cluster node[:tile][:database][:cluster]
 end
@@ -390,6 +417,10 @@ postgresql_user "www-data" do
   cluster node[:tile][:database][:cluster]
 end
 
+postgresql_user "_renderd" do
+  cluster node[:tile][:database][:cluster]
+end
+
 postgresql_database "gis" do
   cluster node[:tile][:database][:cluster]
   owner "tile"
@@ -420,7 +451,32 @@ end
     cluster node[:tile][:database][:cluster]
     database "gis"
     owner "tile"
-    permissions "tile" => :all, "www-data" => :select
+    permissions "tile" => :all, "www-data" => :select, "_renderd" => :select
+  end
+end
+
+package %w[
+  gdal-bin
+  python3-yaml
+  python3-psycopg2
+]
+
+if node[:tile][:database][:external_data_script]
+  execute node[:tile][:database][:external_data_script] do
+    command "#{node[:tile][:database][:external_data_script]} -R _renderd"
+    cwd "/srv/tile.openstreetmap.org"
+    user "tile"
+    group "tile"
+    ignore_failure true
+  end
+
+  Array(node[:tile][:database][:external_data_tables]).each do |table|
+    postgresql_table table do
+      cluster node[:tile][:database][:cluster]
+      database "gis"
+      owner "tile"
+      permissions "tile" => :all, "www-data" => :select, "_renderd" => :select
+    end
   end
 end
 
@@ -438,7 +494,7 @@ end
 
 file node[:tile][:database][:node_file] do
   owner "tile"
-  group "www-data"
+  group "_renderd"
   mode "660"
 end
 
@@ -450,12 +506,23 @@ end
 
 package %w[
   osm2pgsql
-  ruby
   osmium-tool
   pyosmium
   python3-pyproj
 ]
 
+gem_package "apachelogregex" do
+  gem_binary node[:ruby][:gem]
+end
+
+gem_package "file-tail" do
+  gem_binary node[:ruby][:gem]
+end
+
+gem_package "lru_redux" do
+  gem_binary node[:ruby][:gem]
+end
+
 remote_directory "/usr/local/bin" do
   source "bin"
   owner "root"
@@ -466,6 +533,36 @@ remote_directory "/usr/local/bin" do
   files_mode "755"
 end
 
+template "/usr/local/bin/tile-ratelimit" do
+  source "tile-ratelimit.erb"
+  owner "root"
+  group "root"
+  mode "755"
+end
+
+systemd_service "tile-ratelimit" do
+  description "Monitor tile requests and enforce rate limits"
+  after "apache2.service"
+  user "tile"
+  group "adm"
+  exec_start "/usr/local/bin/tile-ratelimit"
+  nice 10
+  private_tmp true
+  private_devices true
+  private_network true
+  protect_system "strict"
+  protect_home true
+  read_write_paths "/srv/tile.openstreetmap.org/conf"
+  no_new_privileges true
+  restart "on-failure"
+end
+
+service "tile-ratelimit" do
+  action [:enable, :start]
+  subscribes :restart, "file[/usr/local/bin/tile-ratelimit]"
+  subscribes :restart, "systemd_service[tile-ratelimit]"
+end
+
 template "/usr/local/bin/expire-tiles" do
   source "expire-tiles.erb"
   owner "root"
@@ -481,7 +578,7 @@ end
 
 directory "/var/lib/replicate/expire-queue" do
   owner "tile"
-  group "www-data"
+  group "_renderd"
   mode "775"
 end
 
@@ -496,13 +593,19 @@ end
 systemd_service "expire-tiles" do
   description "Tile dirtying service"
   type "simple"
-  user "www-data"
+  user "_renderd"
   exec_start "/usr/local/bin/expire-tiles"
+  nice 10
   standard_output "null"
   private_tmp true
   private_devices true
-  protect_system "full"
+  protect_system "strict"
   protect_home true
+  read_write_paths [
+    "/srv/tile.openstreetmap.org/tiles/%i",
+    "/var/lib/replicate/expire-queue",
+    "/var/log/tile"
+  ]
   no_new_privileges true
 end
 
@@ -524,8 +627,13 @@ systemd_service "replicate" do
   exec_start "/usr/local/bin/replicate"
   private_tmp true
   private_devices true
-  protect_system "full"
+  protect_system "strict"
   protect_home true
+  read_write_paths [
+    "/store/database/nodes",
+    "/var/lib/replicate",
+    "/var/log/tile"
+  ]
   no_new_privileges true
   restart "on-failure"
 end
@@ -552,19 +660,21 @@ end
 
 systemd_service "render-lowzoom" do
   description "Render low zoom tiles"
+  condition_path_exists_glob "!/run/update-lowzoom-*"
   user "tile"
   exec_start "/usr/local/bin/render-lowzoom"
   private_tmp true
   private_devices true
   private_network true
-  protect_system "full"
+  protect_system "strict"
   protect_home true
+  read_write_paths "/var/log/tile"
   no_new_privileges true
 end
 
 systemd_timer "render-lowzoom" do
   description "Render low zoom tiles"
-  on_calendar "Sun *-*~07/1 01:00:00"
+  on_calendar "Fri *-*-* 23:00:00 UTC"
 end
 
 service "render-lowzoom.timer" do
@@ -590,7 +700,7 @@ tile_directories.each do |directory|
 
   cron_d "cleanup-tiles#{label}" do
     minute "0"
-    user "www-data"
+    user "_renderd"
     command "ionice -c 3 /usr/local/bin/cleanup-tiles #{directory}"
     mailto "admins@openstreetmap.org"
   end
@@ -609,10 +719,12 @@ munin_plugin "renderd_zoom_time"
 
 munin_plugin "replication_delay"
 
-prometheus_collector "modtile" do
-  interval "1m"
+package "ruby-webrick"
+
+prometheus_exporter "modtile" do
+  port 9494
 end
 
-prometheus_collector "renderd" do
-  interval "1m"
+prometheus_exporter "renderd" do
+  port 9393
 end