From 48f1e22f55118914fc0f045c12a2b1cdf103dff5 Mon Sep 17 00:00:00 2001
From: Simon Rettberg
Date: Fri, 20 Dec 2013 00:45:35 +0100
Subject: [smartctl] New module: smartctl - Will currently check reallocated
 sectors and spin retry count - could be extended, but those seem to be the
 best indicators according to $SEARCH_ENGINE

---
 .../system/basic.target.wants/smartctl.service        |  1 +
 .../smartctl/data/etc/systemd/system/smartctl.service |  8 ++++++++
 .../data/opt/openslx/scripts/systemd-smartctl         | 19 +++++++++++++++++++
 3 files changed, 28 insertions(+)
 create mode 120000 remote/modules/smartctl/data/etc/systemd/system/basic.target.wants/smartctl.service
 create mode 100644 remote/modules/smartctl/data/etc/systemd/system/smartctl.service
 create mode 100755 remote/modules/smartctl/data/opt/openslx/scripts/systemd-smartctl

(limited to 'remote/modules/smartctl/data')

diff --git a/remote/modules/smartctl/data/etc/systemd/system/basic.target.wants/smartctl.service b/remote/modules/smartctl/data/etc/systemd/system/basic.target.wants/smartctl.service
new file mode 120000
index 00000000..b0849c0c
--- /dev/null
+++ b/remote/modules/smartctl/data/etc/systemd/system/basic.target.wants/smartctl.service
@@ -0,0 +1 @@
+../smartctl.service
\ No newline at end of file
diff --git a/remote/modules/smartctl/data/etc/systemd/system/smartctl.service b/remote/modules/smartctl/data/etc/systemd/system/smartctl.service
new file mode 100644
index 00000000..ab268cbc
--- /dev/null
+++ b/remote/modules/smartctl/data/etc/systemd/system/smartctl.service
@@ -0,0 +1,8 @@
+[Unit]
+Description=Check HDD health status
+
+[Service]
+Type=oneshot
+RemainAfterExit=yes
+ExecStart=/opt/openslx/scripts/systemd-smartctl
+
diff --git a/remote/modules/smartctl/data/opt/openslx/scripts/systemd-smartctl b/remote/modules/smartctl/data/opt/openslx/scripts/systemd-smartctl
new file mode 100755
index 00000000..6025c962
--- /dev/null
+++ b/remote/modules/smartctl/data/opt/openslx/scripts/systemd-smartctl
@@ -0,0 +1,19 @@
+#!/bin/ash
+
+# Check Reallocated_Sector_Ct and Spin_Retry_Count of local disk(s)
+
+FILE=$(mktemp)
+for dev in /dev/sd?; do
+	smartctl -A "$dev" -f "brief" > "$FILE" || continue # should we report devices where smartctl doesn't work?
+	REALLOC=$(grep "^ *5 " "$FILE" | awk '{print $8}')
+	SPINRETRY_VAL=$(grep "^ *10 " "$FILE" | awk '{print $4}')
+	SPINRETRY_THR=$(grep "^ *10 " "$FILE" | awk '{print $6}')
+	if [ -n "$REALLOC" ] && [ "$REALLOC" != "0" ]; then
+		slxlog "smartctl-realloc" "Failing HDD: $dev has $REALLOC reallocated sectors!" "$FILE"
+	fi
+	if [ -n "$SPINRETRY_VAL" ] && [ "$SPINRETRY_VAL" -le "$SPINRETRY_THR" ]; then
+		slxlog "smartctl-spinretry" "Failing HDD: $dev has bad spin retry count! ($SPINRETRY_VAL/$SPINRETRY_THR)" "$FILE"
+	fi
+done
+rm -f -- "$FILE"
+
-- 
cgit v1.2.3-55-g7522