From 97abacccc3105677a587c782bf57e128dbd706b7 Mon Sep 17 00:00:00 2001
From: Jan Schulz-Hofen <jan@plan.io>
Date: Mon, 25 Dec 2017 19:34:33 +0100
Subject: [PATCH] Check hard drive health using S.M.A.R.T.

---
 management/status_checks.py | 22 ++++++++++++++++++++++
 setup/system.sh             | 17 +++++++++++++++++
 2 files changed, 39 insertions(+)

diff --git a/management/status_checks.py b/management/status_checks.py
index a9d0595c..57599ed2 100755
--- a/management/status_checks.py
+++ b/management/status_checks.py
@@ -165,6 +165,7 @@ def run_system_checks(rounded_values, env, output):
 	check_miab_version(env, output)
 	check_system_aliases(env, output)
 	check_free_disk_space(rounded_values, env, output)
+	check_smart_status(env, output)
 	check_free_memory(rounded_values, env, output)
 
 def check_ufw(env, output):
@@ -250,6 +251,27 @@ def check_free_disk_space(rounded_values, env, output):
 		if rounded_values: disk_msg = "The disk has less than 15% free space."
 		output.print_error(disk_msg)
 
+def check_smart_status(env, output):
+	devices = shell('check_output', ['smartctl', '--scan-open'])
+	for device in devices.splitlines():
+		device = device.split()[0]
+		info = shell('check_output', ['smartctl', '-i', device])
+		if 'SMART support is: Available' in info:
+			if 'SMART support is: Enabled' in info:
+				code, health = shell('check_output', ['smartctl', '-H', device, '--quietmode=errorsonly'], trap=True)
+				if code == 0:
+					output.print_ok('Disk %s passed all S.M.A.R.T. checks and seems healthy.' % device)
+				else:
+					output.print_error('Disk %s failed the S.M.A.R.T health check. Consider replacing the hard drive. Detailed information:' % device)
+					output.print_line("")
+					for line in health.split('=== START OF READ SMART DATA SECTION ===')[1].splitlines():
+						output.print_line(line)
+			else:
+				output.print_warning('Disk %s supports S.M.A.R.T, but it is disabled. You should activate it using \'sudo smartctl -s on %s\'.' % (device, device))
+
+		else:
+			output.print_ok('Disk %s does not support S.M.A.R.T. Health checks are skipped.' % device)
+
 def check_free_memory(rounded_values, env, output):
 	# Check free memory.
 	percent_free = 100 - psutil.virtual_memory().percent
diff --git a/setup/system.sh b/setup/system.sh
index 28043b16..6c522750 100755
--- a/setup/system.sh
+++ b/setup/system.sh
@@ -357,3 +357,20 @@ cp -f conf/fail2ban/filter.d/* /etc/fail2ban/filter.d/
 # scripts will ensure the files exist and then fail2ban is given another
 # restart at the very end of setup.
 restart_service fail2ban
+
+# ### S.M.A.R.T. Monitoring Tools
+
+# Install and configure smartmontools so we can check for failing hard drives.
+# We'll perform a first single smartd startup to check if any devices are found
+# which can be monitored. If none are found (virtual machines!), we will configure
+# smartd to not start automatically. Otherwise, smartd will run.
+apt_install smartmontools
+smartd --quit=onecheck > /dev/null 2>&1 || { smart=$? ; }
+if [ "${smart:-0}" -ne 17 ]; then # smartd manpage: 17 means smartd didn't find any devices to monitor.
+	echo "S.M.A.R.T. capable hard drives found, setting up smartd..."
+	tools/editconf.py /etc/default/smartmontools start_smartd=yes
+	restart_service smartmontools
+else
+	echo "No S.M.A.R.T. capable hard drives found, disabling smartd..."
+	tools/editconf.py /etc/default/smartmontools start_smartd=no
+fi