zpool iostat/status -c improvements

Users can now provide their own scripts to be run
with 'zpool iostat/status -c'. User scripts should be
placed in ~/.zpool.d to be included in zpool's
default search path.

Provide a script which can be used with
'zpool iostat|status -c' that will return the type of
device (hdd, sdd, file).

Provide a script to get various values from smartctl
when using 'zpool iostat/status -c'.

Allow users to define the ZPOOL_SCRIPTS_PATH
environment variable which can be used to override
the default 'zpool iostat/status -c' search path.

Allow the ZPOOL_SCRIPTS_ENABLED environment
variable to enable or disable 'zpool status/iostat -c'
functionality.

Use the new smart script to provide the serial command.

Install /etc/sudoers.d/zfs file which contains the sudoer
rule for smartctl as a sample.

Allow 'zpool iostat/status -c' tests to run in tree.

Reviewed-by: Tony Hutter <hutter2@llnl.gov>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Giuseppe Di Natale <dinatale2@llnl.gov>
Closes #6121 
Closes #6153

cmd/zpool/zpool.d/ata_err

@@ -0,0 +1 @@
+smart

cmd/zpool/zpool.d/cmd_to

@@ -0,0 +1 @@
+smart

cmd/zpool/zpool.d/defect

@@ -0,0 +1 @@
+smart

cmd/zpool/zpool.d/health

@@ -0,0 +1 @@
+smart

cmd/zpool/zpool.d/hours_on

@@ -0,0 +1 @@
+smart

cmd/zpool/zpool.d/lsblk

@@ -38,16 +38,15 @@
 #   DISC-ZERO  discard zeroes data
 #
 # If the script is run as just 'lsblk' then print out disk size, vendor,
-# model number and serial number.
+# and model number.
 
 
 helpstr="
 label:	Show filesystem label.
 model:	Show disk model number.
-serial:	Show disk serial number.
 size:	Show the disk capacity.
 vendor:	Show the disk vendor.
-lsblk:	Show the disk size, vendor, model number, and serial number."
+lsblk:	Show the disk size, vendor, and model number."
 
 script=$(basename "$0")
 
@@ -57,7 +56,7 @@ if [ "$1" = "-h" ] ; then
 fi
 
 if [ "$script" = "lsblk" ] ; then
-	list="size vendor model serial"
+	list="size vendor model"
 else
 	list=$(echo "$script" | tr '[:upper:]' '[:lower:]')
 fi

cmd/zpool/zpool.d/media

@@ -0,0 +1,27 @@
+#!/bin/sh
+#
+# Print out the type of device
+#
+
+if [ "$1" = "-h" ] ; then
+	echo "Show whether a vdev is a file, hdd, or ssd."
+	exit
+fi
+
+if [ -b "$VDEV_UPATH" ]; then
+	device=$(basename "$VDEV_UPATH")
+	val=$(cat "/sys/block/$device/queue/rotational" 2>/dev/null)
+	if [ "$val" = "0" ]; then
+		MEDIA="ssd"
+	fi
+
+	if [ "$val" = "1" ]; then
+		MEDIA="hdd"
+	fi
+else
+	if [ -f "$VDEV_UPATH" ]; then
+		MEDIA="file"
+	fi
+fi
+
+echo "media=$MEDIA"

cmd/zpool/zpool.d/nonmed

@@ -0,0 +1 @@
+smart

cmd/zpool/zpool.d/off_ucor

@@ -0,0 +1 @@
+smart

cmd/zpool/zpool.d/pend_sec

@@ -0,0 +1 @@
+smart

cmd/zpool/zpool.d/pwr_cyc

@@ -0,0 +1 @@
+smart

cmd/zpool/zpool.d/r_proc

@@ -0,0 +1 @@
+smart

cmd/zpool/zpool.d/r_ucor

@@ -0,0 +1 @@
+smart

cmd/zpool/zpool.d/realloc

@@ -0,0 +1 @@
+smart

cmd/zpool/zpool.d/rep_ucor

@@ -0,0 +1 @@
+smart

cmd/zpool/zpool.d/serial

@@ -1 +1 @@
-lsblk
+smart

cmd/zpool/zpool.d/ses

@@ -6,10 +6,10 @@
 helpstr="
 enc:		Show disk enclosure w:x:y:z value.
 slot:		Show disk slot number as reported by the enclosure.
-encdev:		Show the /dev/sg* device for the enclosure associated with the disk slot.
-fault_led:	Show the value of the disk enclosure slot fault LED.
-locate_led:	Show the value of the disk enclosure slot locate LED.
-ses:		Show disk's enclosure, enclosure dev, slot number, and fault/locate LED values."
+encdev:		Show /dev/sg* device associated with the enclosure disk slot.
+fault_led:	Show value of the disk enclosure slot fault LED.
+locate_led:	Show value of the disk enclosure slot locate LED.
+ses:		Show disk's enc, enc device, slot, and fault/locate LED values."
 
 script=$(basename "$0")
 if [ "$1" = "-h" ] ; then

cmd/zpool/zpool.d/smart

@@ -0,0 +1,123 @@
+#!/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.
+"
+
+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 $VDEV_UPATH")
+
+	# Are we a SAS or ATA drive?  Look for the right line in smartctl:
+	#
+	# SAS:
+	#	Transport protocol:   SAS
+	#
+	# SATA:
+	#	ATA Version is:   8
+	#
+	type=$(echo "$raw_out" | grep -m 1 -Eo '^ATA|SAS$')
+	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}
+/SMART overall-health self-assessment test result:/{print "health="$6}
+/Power_On_Hours/{print "hours_on="$10}
+/Serial Number:/{print "serial="$3}
+
+END {ORS="\n"; print ""}
+');
+fi
+
+# if type is not set by now, either we don't have a block device
+# or smartctl failed. Either way, default to ATA and set out to
+# nothing
+if [ -z "$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 [ "$type" = "ATA" ] ; then
+		scripts="temp|health|ata_err|realloc|rep_ucor|cmd_to|pend_sec|off_ucor"
+	fi
+	;;
+smartx)
+	# Print some other interesting stats
+	if [ "$type" = "SAS" ] ; then
+		scripts="hours_on|defect|nonmed|r_proc|w_proc"
+	elif [ "$type" = "ATA" ] ; then
+		scripts="hours_on|pwr_cyc"
+	fi
+	;;
+*)
+	scripts="$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

cmd/zpool/zpool.d/smartx

@@ -0,0 +1 @@
+smart

cmd/zpool/zpool.d/temp

@@ -0,0 +1 @@
+smart

cmd/zpool/zpool.d/w_proc

@@ -0,0 +1 @@
+smart

cmd/zpool/zpool.d/w_ucor

@@ -0,0 +1 @@
+smart