Add drive SMART attributes and health status

bin/malpd

@@ -24,6 +24,45 @@ def drive_type(name)
   rotational == "1" ? "hdd" : "ssd"
 end
 
+def read_smart(name)
+  out = `smartctl -j -A -H /dev/#{name} 2>/dev/null`
+  return nil if out.empty?
+  data = JSON.parse(out) rescue nil
+  return nil unless data
+
+  result = {}
+  passed = data.dig("smart_status", "passed")
+  result["passed"] = passed unless passed.nil?
+
+  if (t = data.dig("temperature", "current"))
+    result["temperature"] = t
+  end
+  if (poh = data.dig("power_on_time", "hours"))
+    result["power_on_hours"] = poh
+  end
+
+  if name.start_with?("nvme")
+    log = data["nvme_smart_health_information_log"] || {}
+    result["percentage_used"] = log["percentage_used"] if log.key?("percentage_used")
+    result["media_errors"]    = log["media_errors"]    if log.key?("media_errors")
+    result["critical_warning"] = log["critical_warning"] if log.key?("critical_warning")
+  else
+    attrs = data.dig("ata_smart_attributes", "table") || []
+    attrs.each do |a|
+      raw = a.dig("raw", "value")
+      next if raw.nil?
+      case a["id"]
+      when 5   then result["reallocated_sectors"] = raw
+      when 10  then result["spin_retry"] = raw
+      when 197 then result["pending_sectors"] = raw
+      when 199 then result["crc_errors"] = raw
+      end
+    end
+  end
+
+  result
+end
+
 def handle_info(conn)
   info = []
 
@@ -37,12 +76,17 @@ def handle_info(conn)
     model = (File.read("#{block_path}/device/model").strip rescue "Unknown")
     bytes = File.read("#{block_path}/size").strip.to_i * 512
 
-    info << {
+    entry = {
       "type" => "drive",
       "drivetype" => drive_type(name),
+      "name" => name,
       "model" => model,
       "capacity" => human_capacity(bytes)
     }
+    if (smart = read_smart(name))
+      entry["smart"] = smart
+    end
+    info << entry
   end
 
   if conn