add ATA disk mapping, zpool check, netdev check, and a config file

scripts/ugreen-diskiomon

@@ -6,6 +6,8 @@ exit-ugreen-diskiomon() {
     rm "/var/run/ugreen-diskiomon.lock"
   fi
   kill $smart_check_pid 2>/dev/null
+  kill $zpool_check_pid 2>/dev/null
+  kill $disk_online_check_pid 2>/dev/null
 }
 
 # trap exit and remove lockfile
@@ -23,42 +25,50 @@ if [[ -f /boot/config/plugins/ugreenleds-driver/settings.cfg ]]; then
   source /boot/config/plugins/ugreenleds-driver/settings.cfg
 fi
 
+# load environment variables
+if [[ -f /etc/ugreen-leds.conf ]]; then
+  source /etc/ugreen-leds.conf
+fi
+
 # led-disk mapping (see https://github.com/miskcoo/ugreen_dx4600_leds_controller/pull/4)
-MAPPING_METHOD=${MAPPING_METHOD:=hctl} # hctl, serial
+MAPPING_METHOD=${MAPPING_METHOD:=ata} # ata, hctl, serial
 led_map=(disk1 disk2 disk3 disk4 disk5 disk6 disk7 disk8)
 
 # hctl, $> lsblk -S -x hctl -o hctl,serial,name 
 # NOTE: It is reported that the order below should be adjusted for each model. 
 #       Please check the disk mapping section in https://github.com/miskcoo/ugreen_dx4600_leds_controller/blob/master/README.md.
 hctl_map=("0:0:0:0" "1:0:0:0" "2:0:0:0" "3:0:0:0" "4:0:0:0" "5:0:0:0" "6:0:0:0" "7:0:0:0")
+# serial number, $> lsblk -S -x hctl -o hctl,serial,name 
+serial_map=(${DISK_SERIAL})
+# ata number, $> ls /sys/block | egrep ata\d
+ata_map=("ata1" "ata2" "ata3" "ata4" "ata5" "ata6" "ata7" "ata8")
+
 if which dmidecode; then
     product_name=$(dmidecode --string system-product-name)
     case "${product_name}" in 
-        "DXP6800 Pro")
-          echo "Found UGREEN DXP6800 Pro" 
-          hctl_map=("2:0:0:0" "3:0:0:0" "4:0:0:0" "5:0:0:0" "0:0:0:0" "1:0:0:0")
-          ;;
-        "DX4600 Pro")
-          echo "Found UGREEN DX4600 Pro" 
-          # using the default mapping
-          ;;
-        "DXP8800 Plus")
-          echo "Found UGREEN DXP8800 Plus" 
-          # using the default mapping
-          ;;
+        DXP6800*)   # tested on DXP6800 Pro
+            echo "Found UGREEN DXP6800 series" 
+            hctl_map=("2:0:0:0" "3:0:0:0" "4:0:0:0" "5:0:0:0" "0:0:0:0" "1:0:0:0")
+            ata_map=("ata3" "ata4" "ata5" "ata6" "ata1" "ata2")
+            ;;
+        DX4600*)   # tested on DX4600 Pro
+            echo "Found UGREEN DX4600 series" 
+            ;;
+        DXP8800*)  # tested on DXP8800 Plus
+            echo "Found UGREEN DXP8800 series" 
+            # using the default mapping
+            ;;
         *)
-          if [[ "${MAPPING_METHOD}" = "hctl" ]]; then
-              echo "Using the default HCTL order. Please check it maps to your disk slots correctly."
-              echo "If you confirm that the HCTL order is correct, or find it is different, you can submit an issue to let us know, so we can update the script."
-              echo "(Read the disk mapping section in https://github.com/miskcoo/ugreen_dx4600_leds_controller/blob/master/README.md for more details)"
-          fi
-          ;;
+            if [[ "${MAPPING_METHOD}" = "hctl" ]]; then
+                echo "Using the default HCTL order. Please check it maps to your disk slots correctly."
+                echo "If you confirm that the HCTL order is correct, or find it is different, you can submit an issue to let us know, so we can update the script."
+                echo "(Read the disk mapping section in https://github.com/miskcoo/ugreen_dx4600_leds_controller/blob/master/README.md for more details)"
+            fi
+            ;;
     esac
 elif [[ "${MAPPING_METHOD}" = "hctl" ]]; then
     echo "installing the tool `dmidecode` is suggested; otherwise the script cannot detect your device and adjust the hctl_map"
 fi 
-# serial number, $> lsblk -S -x hctl -o hctl,serial,name 
-serial_map=("placeholder0" "placeholder1" "placeholder2" "placeholder3" "placeholder4" "placeholder5" "placeholder6" "placeholder7")
 declare -A devices
 
 # set monitor SMART information to true by default if not running unRAID
@@ -72,46 +82,73 @@ CHECK_SMART_INTERVAL=${CHECK_SMART_INTERVAL:=360}
 # refresh interval from disk leds
 LED_REFRESH_INTERVAL=${LED_REFRESH_INTERVAL:=0.1}
 
+# whether to check zpool health
+CHECK_ZPOOL=${CHECK_ZPOOL:=false}
+# polling rate for checking zpool health. 5 seconds by default
+CHECK_ZPOOL_INTERVAL=${CHECK_ZPOOL_INTERVAL:=5}
+
+# polling rate for checking disk online. 5 seconds by default
+CHECK_DISK_ONLINE_INTERVAL=${CHECK_DISK_ONLINE_INTERVAL:=5}
 
+COLOR_DISK_HEALTH=${COLOR_DISK_HEALTH:="255 255 255"}
+COLOR_DISK_UNAVAIL=${COLOR_DISK_UNAVAIL:="255 0 0"}
+COLOR_ZPOOL_FAIL=${COLOR_ZPOOL_FAIL:="255 0 0"}
+COLOR_SMART_FAIL=${COLOR_SMART_FAIL:="255 0 0"}
 
-{ lsmod | grep ledtrig_oneshot ; } || modprobe -v ledtrig_oneshot
+
+{ lsmod | grep ledtrig_oneshot > /dev/null; } || modprobe -v ledtrig_oneshot
 
 sleep 2
 
+function disk_enumerating_string() {
+    if [[ $MAPPING_METHOD = ata ]]; then
+        ls -ahl /sys/block | sed 's/\/$//' | awk '{
+            if (match($0, /ata[0-9]+/)) {
+                ata = substr($0, RSTART, RLENGTH);
+                if (match($0, /[^\/]+$/)) {
+                    basename = substr($0, RSTART, RLENGTH);
+                    print basename, ata;
+                }
+            }
+        }'
+    elif [[ $MAPPING_METHOD = hctl ]] || [[ $MAPPING_METHOD = serial ]]; then
+        lsblk -S -o name,${MAPPING_METHOD},tran | grep sata
+    else
+        echo Unsupported mapping method: ${MAPPING_METHOD}
+        exit 1
+    fi
+}
+
 echo Enumerating disks based on $MAPPING_METHOD...
 declare -A dev_map
 while read line
 do
     blk_line=($line)
-    if [[ $MAPPING_METHOD = hctl ]]; then
-        key=${blk_line[1]}
-        val=${blk_line[0]}
-    elif [[ $MAPPING_METHOD = serial ]]; then
-        key=${blk_line[1]}
-        val=${blk_line[0]}
-    else
-        echo Unsupported mapping method: ${MAPPING_METHOD}
-        exit 1
-    fi
+    key=${blk_line[1]}
+    val=${blk_line[0]}
     dev_map[${key}]=${val}
     echo $MAPPING_METHOD ${key} ">>" ${dev_map[${key}]}
-done <<< "$(lsblk -S -o name,${MAPPING_METHOD} | tail -n +2)"
+done <<< "$(disk_enumerating_string)"
 
+# initialize LEDs
+declare -A dev_to_led_map
 for i in "${!led_map[@]}"; do
     led=${led_map[i]} 
     if [[ -d /sys/class/leds/$led ]]; then
         echo oneshot > /sys/class/leds/$led/trigger
         echo 1 > /sys/class/leds/$led/invert
         echo 100 > /sys/class/leds/$led/delay_on
         echo 100 > /sys/class/leds/$led/delay_off
-        echo "255 255 255" > /sys/class/leds/$led/color
+        echo "$COLOR_DISK_HEALTH" > /sys/class/leds/$led/color
 
         # find corresponding device
         _tmp_str=${MAPPING_METHOD}_map[@]
         _tmp_arr=(${!_tmp_str})
         dev=${dev_map[${_tmp_arr[i]}]}
+
         if [[ -f /sys/class/block/${dev}/stat ]]; then
             devices[$led]=${dev}
+            dev_to_led_map[$dev]=$led
         else
             # turn off the led if no disk installed on this slot
             echo 0 > /sys/class/leds/$led/brightness
@@ -120,19 +157,94 @@ for i in "${!led_map[@]}"; do
     fi
 done
 
+# construct zpool device mapping
+declare -A zpool_ledmap
+if [ "$CHECK_ZPOOL" = true ]; then
+    echo Enumerating zpool devices...
+    while read line
+    do
+        zpool_dev_line=($line)
+        zpool_dev_name=${zpool_dev_line[0]} 
+        zpool_scsi_dev_name="unknown"
+        # zpool_dev_state=${zpool_dev_line[1]}
+        case "$zpool_dev_name" in
+            sd*)
+                # remove the trailing partition number
+                zpool_scsi_dev_name=$(echo $zpool_dev_name | sed 's/[0-9]*$//')
+                ;;
+            dm*)
+                # find the underlying block device of the encrypted device
+                dm_slaves=($(ls /sys/block/${zpool_dev_name}/slaves))
+                zpool_scsi_dev_name=${dm_slaves[0]}
+                ;;
+            *)
+                echo Unsupported zpool device type ${zpool_dev_name}.
+                ;;
+        esac
+
+        # if the detected scsi device can be found in the mapping array 
+        #echo zpool $zpool_dev_name ">>" $zpool_scsi_dev_name ">>" ${dev_to_led_map[${zpool_scsi_dev_name}]}
+        if [[ -v "dev_to_led_map[${zpool_scsi_dev_name}]" ]]; then
+            zpool_ledmap[$zpool_dev_name]=${dev_to_led_map[${zpool_scsi_dev_name}]}
+            echo "zpool device" $zpool_dev_name ">>" $zpool_scsi_dev_name ">> LED:"${zpool_ledmap[$zpool_dev_name]}
+        fi
+    done <<< "$(zpool status -L | egrep ^\\s*\(sd\|dm\))"
+
+    function zpool_check_loop() {
+        while true; do
+            while read line
+            do
+                zpool_dev_line=($line)
+                zpool_dev_name=${zpool_dev_line[0]} 
+                zpool_dev_state=${zpool_dev_line[1]}
+
+                # TODO: do something if the pool is unhealthy?
+
+                if [[ -v "zpool_ledmap[${zpool_dev_name}]" ]]; then
+                    led=${zpool_ledmap[$zpool_dev_name]}
+
+                    if [[ "$(cat /sys/class/leds/$led/color)" != "$COLOR_DISK_HEALTH" ]]; then
+                        continue;
+                    fi
+
+                    if [[ "${zpool_dev_state}" != "ONLINE" ]]; then
+                        echo "$COLOR_ZPOOL_FAIL" > /sys/class/leds/$led/color
+                        echo Disk failure detected on /dev/$dev at $(date +%Y-%m-%d' '%H:%M:%S)
+                    fi
+
+                    # ==== To recover from an error, you should restart the script ====
+                    ## case "${zpool_dev_state}" in
+                    ##     ONLINE)
+                    ##         # echo "$COLOR_DISK_HEALTH" > /sys/class/leds/$led/color
+                    ##         ;;
+                    ##     *)
+                    ##         echo "255 0 0" > /sys/class/leds/$led/color
+                    ##         ;;
+                    ## esac
+                fi
+            done <<< "$(zpool status -L | egrep ^\\s*\(sd\|dm\))"
+
+            sleep ${CHECK_ZPOOL_INTERVAL}s
+        done
+    }
+
+    zpool_check_loop &
+    zpool_check_pid=$!
+fi
+
 # check disk health if enabled
 if [ "$CHECK_SMART" = true ]; then
 (
     while true; do
         for led in "${!devices[@]}"; do
-            if [[ "$(cat /sys/class/leds/$led/color)" = "255 0 0" ]]; then
+            if [[ "$(cat /sys/class/leds/$led/color)" != "$COLOR_DISK_HEALTH" ]]; then
                 continue;
             fi
 
             dev=${devices[$led]}
 
             if [[ -z $(smartctl -H /dev/${dev} | grep PASSED) ]]; then
-                echo "255 0 0" > /sys/class/leds/$led/color
+                echo "$COLOR_SMART_FAIL" > /sys/class/leds/$led/color
                 echo Disk failure detected on /dev/$dev at $(date +%Y-%m-%d' '%H:%M:%S)
                 continue
             fi
@@ -143,40 +255,42 @@ if [ "$CHECK_SMART" = true ]; then
 smart_check_pid=$!
 fi
 
-# check for zpool-leds.sh and set variable
-if [[ -f /usr/bin/zpool-leds.sh ]]; then
-    ZPOOL_LEDS="bash /usr/bin/zpool-leds.sh"
-else
-    ZPOOL_LEDS=""
-fi
+# check disk online status
+(
+    while true; do
+        for led in "${!devices[@]}"; do
+            dev=${devices[$led]}
+
+            if [[ "$(cat /sys/class/leds/$led/color)" != "$COLOR_DISK_HEALTH" ]]; then
+                continue;
+            fi
 
+            if [[ ! -f /sys/class/block/${dev}/stat ]]; then
+                echo "$COLOR_DISK_UNAVAIL" > /sys/class/leds/$led/color 2>/dev/null
+                echo Disk /dev/$dev went offline at $(date +%Y-%m-%d' '%H:%M:%S)
+                continue
+            fi
+        done
+        sleep ${CHECK_DISK_ONLINE_INTERVAL}s
+    done
+) &
+disk_online_check_pid=$!
+
+# monitor disk activities
 declare -A diskio_data_rw
 while true; do
     for led in "${!devices[@]}"; do
-        dev=${devices[$led]}
-        diskio_old_rw="${diskio_data_rw[$led]}"
-
-        if [[ "$(cat /sys/class/leds/$led/color)" = "255 0 0" ]]; then
-            continue;
-        fi
 
-        if [[ ! -f /sys/class/block/${dev}/stat ]]; then
-            echo "255 0 0" > /sys/class/leds/$led/color 2>/dev/null
-            echo Disk /dev/$dev went offline at $(date +%Y-%m-%d' '%H:%M:%S)
-            continue
-        fi
-
-        diskio_new_rw="$(cat /sys/block/${dev}/stat)"
+        # if $dev does not exist, diskio_new_rw="", which will be safe
+        diskio_new_rw="$(cat /sys/block/${devices[$led]}/stat 2>/dev/null)"
 
-        if [ "${diskio_old_rw}" != "${diskio_new_rw}" ]; then
+        if [ "${diskio_data_rw[$led]}" != "${diskio_new_rw}" ]; then
             echo 1 > /sys/class/leds/$led/shot
         fi
 
         diskio_data_rw[$led]=$diskio_new_rw
     done
 
-    ${ZPOOL_LEDS}
-
     sleep ${LED_REFRESH_INTERVAL}s
 
 done