From 065e5a524cc57f9b23f3e071b4fbdb8120bdda75 Mon Sep 17 00:00:00 2001
From: "andrea.noni" <andrea.noni@hangar.org>
Date: Wed, 30 Oct 2024 16:56:34 +0100
Subject: [PATCH] added dinamic btrfs partition device errors check

---
 disks_checks.sh | 35 +++++++++++++++++++++++++++++++----
 1 file changed, 31 insertions(+), 4 deletions(-)

diff --git a/disks_checks.sh b/disks_checks.sh
index ec27aff..f5fbe10 100755
--- a/disks_checks.sh
+++ b/disks_checks.sh
@@ -1,9 +1,10 @@
 #!/bin/bash
 
-### I lost the link where I copied this script from, sorry, i'll update as soon as I find it back! The btrfs raid1 check part is written by tech_at_hangar.org
+### I lost the link where I copied the first part of this script from, sorry, i'll update as soon as I find it back! The btrfs raid1 check part (second) and the btrfs device errors part (third) are written by tech_at_hangar.org
 
-## The first part of this script checks for given minimum space on a mounted volume, and contacts ntfy in case the limit is surpassed.
-## The second part of this script checks for missing disks on btrfs raid1 devices and contacs ntfy in case of missing devices
+## The first part of this script checks for given minimum space on a mounted volume, and alerts ntfy in case the limit is surpassed.
+## The second part of this script checks for missing disks on btrfs raid1 devices and alerts ntfy in case of missing devices
+## The third part of this script checks for read/write errors on disks in mounted partitions and alerts ntfy in case of errors
 
 # root volume 
 mingigs=10
@@ -33,7 +34,7 @@ if [ -n "$avail" ]; then
     $topicurl
 fi
 
-# check btrfs raid 1 status on /mnt/OTHER_VOLUME NAME
+### check btrfs raid status on btrfs devices
 
 if (btrfs fi show | grep -wic "Some devices missing"); then
     echo Bad: - Btrfs is missing devices, some disk is broken
@@ -45,3 +46,29 @@ if (btrfs fi show | grep -wic "Some devices missing"); then
     -H "Tags: warning,boar" \
     $topicurl
 fi
+
+### check errors on btrfs devices
+
+topicurl=https://NTFY_SERVER_IP/disk_alerts
+# get list of btrfs mounted partition, put all results in a single line separated by a comma and remove the last comma
+mounted_partitions="$(mount | grep btrfs | awk '{print $3}' |  tr '\n' ',' | awk '{sub(/,$/,""); print}')"
+
+for i in ${mounted_partitions//,/ }
+do
+	# check btrfs errors on mounted device, if all output is 0 then return a single "1"
+	check_disk_errors=$(btrfs device stats $i | awk '{print $2}' | grep -c 0 -v)
+
+        if [ ${check_disk_errors} -eq 1 ] ; then
+                    : # do nothing if output is a single "1"
+            else
+                    echo Bad: - Btrfs is detecting errors on partition $i, some disk is about to be broken
+                    echo Sending alert...
+                    curl -k --retry 3 \
+                    -d "Bad - Errors on btrfs mounted device $i: you will have to replace a disk. Check the status of mounted disks with: btrfs device stats $1" \
+                    -H "Title: btrfs errors on partition $i of $(hostname)" \
+                    -H "Priority: high" \
+                    -H "Tags: warning,boar" \
+                    $topicurl
+        fi
+done
+