mirror of
https://git.proxmox.com/git/mirror_zfs.git
synced 2024-12-27 11:29:36 +03:00
ca0b376604
This adds the SMART attributes required to probe Samsung SSD and NVMe (and possibly others) disks when using the "zpool status -c" command. Reviewed-by: loli10K <ezomori.nozomu@gmail.com> Reviewed-by: Giuseppe Di Natale <dinatale2@llnl.gov> Reviewed-by: Tony Hutter <hutter2@llnl.gov> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: bunder2015 <omfgbunder@gmail.com> Closes #7183 Closes #7193
143 lines
4.2 KiB
Bash
Executable File
143 lines
4.2 KiB
Bash
Executable File
#!/bin/sh
|
|
#
|
|
# Show SMART stats
|
|
#
|
|
|
|
helpstr="
|
|
smart: Show SMART temperature and error stats (specific to drive type)
|
|
smartx: Show SMART extended drive stats (specific to drive type).
|
|
temp: Show SMART drive temperature in celsius (all drives).
|
|
health: Show reported SMART status (all drives).
|
|
r_proc: Show SMART read GBytes processed over drive lifetime (SAS).
|
|
w_proc: Show SMART write GBytes processed over drive lifetime (SAS).
|
|
r_ucor: Show SMART read uncorrectable errors (SAS).
|
|
w_ucor: Show SMART write uncorrectable errors (SAS).
|
|
nonmed: Show SMART non-medium errors (SAS).
|
|
defect: Show SMART grown defect list (SAS).
|
|
hours_on: Show number of hours drive powered on (all drives).
|
|
realloc: Show SMART reallocated sectors count (ATA).
|
|
rep_ucor: Show SMART reported uncorrectable count (ATA).
|
|
cmd_to: Show SMART command timeout count (ATA).
|
|
pend_sec: Show SMART current pending sector count (ATA).
|
|
off_ucor: Show SMART offline uncorrectable errors (ATA).
|
|
ata_err: Show SMART ATA errors (ATA).
|
|
pwr_cyc: Show SMART power cycle count (ATA).
|
|
serial: Show disk serial number.
|
|
nvme_err: Show SMART NVMe errors (NVMe).
|
|
"
|
|
|
|
script=$(basename "$0")
|
|
|
|
if [ "$1" = "-h" ] ; then
|
|
echo "$helpstr" | grep "$script:" | tr -s '\t' | cut -f 2-
|
|
exit
|
|
fi
|
|
|
|
smartctl_path=$(which smartctl)
|
|
|
|
if [ -b "$VDEV_UPATH" ] && [ -x "$smartctl_path" ]; then
|
|
raw_out=$(eval "sudo $smartctl_path -a $VDEV_UPATH")
|
|
|
|
# What kind of drive are we? Look for the right line in smartctl:
|
|
#
|
|
# SAS:
|
|
# Transport protocol: SAS
|
|
#
|
|
# SATA:
|
|
# ATA Version is: 8
|
|
#
|
|
# NVMe:
|
|
# SMART/Health Information (NVMe Log 0xnn, NSID 0xnn)
|
|
#
|
|
type=$(echo "$raw_out" | grep -m 1 -Eo '^ATA|NVMe|SAS$')
|
|
out=$(echo "$raw_out" | awk '
|
|
# SAS specific
|
|
/read:/{print "rrd="$4"\nr_cor="$5"\nr_proc="$7"\nr_ucor="$8}
|
|
/write:/{print "rwr="$4"\nw_cor="$5"\nw_proc="$7"\nw_ucor="$8}
|
|
/Non-medium error count/{print "nonmed="$4}
|
|
/Elements in grown defect list/{print "defect="$6}
|
|
|
|
# SAS common
|
|
/Drive Temperature:/{print "temp="$4}
|
|
# Status can be a long string, substitute spaces for '_'
|
|
/SMART Health Status:/{printf "health="; for(i=4;i<=NF-1;i++){printf "%s_", $i}; printf "%s\n", $i}
|
|
/number of hours powered up/{print "hours_on="$7}
|
|
/Serial number:/{print "serial="$3}
|
|
|
|
# SATA specific
|
|
/Reallocated_Sector_Ct/{print "realloc="$10}
|
|
/Reported_Uncorrect/{print "rep_ucor="$10}
|
|
/Command_Timeout/{print "cmd_to="$10}
|
|
/Current_Pending_Sector/{print "pend_sec="$10}
|
|
/Offline_Uncorrectable/{print "off_ucor="$10}
|
|
/ATA Error Count:/{print "ata_err="$4}
|
|
/Power_Cycle_Count/{print "pwr_cyc="$10}
|
|
|
|
# SATA common
|
|
/Temperature_Celsius/{print "temp="$10}
|
|
/Airflow_Temperature_Cel/{print "temp="$10}
|
|
/SMART overall-health self-assessment test result:/{print "health="$6}
|
|
/Power_On_Hours/{print "hours_on="$10}
|
|
/Serial Number:/{print "serial="$3}
|
|
|
|
# NVMe common
|
|
/Temperature:/{print "temp="$2}
|
|
/SMART overall-health self-assessment test result:/{print "health="$6}
|
|
/Power On Hours:/{gsub("[^0-9]","",$4); print "hours_on="$4}
|
|
/Serial Number:/{print "serial="$3}
|
|
/Power Cycles:/{print "pwr_cyc="$3}
|
|
|
|
# NVMe specific
|
|
/Media and Data Integrity Errors:/{print "nvme_err="$6}
|
|
|
|
END {ORS="\n"; print ""}
|
|
');
|
|
fi
|
|
|
|
# if type is not set by now, either we don't have a block device
|
|
# or smartctl failed. Either way, default to ATA and set out to
|
|
# nothing
|
|
if [ -z "$type" ]; then
|
|
type="ATA"
|
|
out=
|
|
fi
|
|
|
|
case $script in
|
|
smart)
|
|
# Print temperature plus common predictors of drive failure
|
|
if [ "$type" = "SAS" ] ; then
|
|
scripts="temp|health|r_ucor|w_ucor"
|
|
elif [ "$type" = "ATA" ] ; then
|
|
scripts="temp|health|ata_err|realloc|rep_ucor|cmd_to|pend_sec|off_ucor"
|
|
elif [ "$type" = "NVMe" ] ; then
|
|
scripts="temp|health|nvme_err"
|
|
fi
|
|
;;
|
|
smartx)
|
|
# Print some other interesting stats
|
|
if [ "$type" = "SAS" ] ; then
|
|
scripts="hours_on|defect|nonmed|r_proc|w_proc"
|
|
elif [ "$type" = "ATA" ] ; then
|
|
scripts="hours_on|pwr_cyc"
|
|
elif [ "$type" = "NVMe" ] ; then
|
|
scripts="hours_on|pwr_cyc"
|
|
fi
|
|
;;
|
|
*)
|
|
scripts="$script"
|
|
esac
|
|
|
|
with_vals=$(echo "$out" | grep -E "$scripts")
|
|
if [ ! -z "$with_vals" ]; then
|
|
echo "$with_vals"
|
|
without_vals=$(echo "$scripts" | tr "|" "\n" |
|
|
grep -v -E "$(echo "$with_vals" |
|
|
awk -F "=" '{print $1}')" | awk '{print $0"="}')
|
|
else
|
|
without_vals=$(echo "$scripts" | tr "|" "\n" | awk '{print $0"="}')
|
|
fi
|
|
|
|
if [ ! -z "$without_vals" ]; then
|
|
echo "$without_vals"
|
|
fi
|