Add SMART self-test results to zpool status -c

Add in SMART self-test results to zpool status|iostat -c.  This
works for both SAS and SATA drives.

Also, add plumbing to allow the 'smart' script to take smartctl
output from a directory of output text files instead of running
it against the vdevs.

Reviewed-by: Giuseppe Di Natale <[email protected]>
Reviewed-by: Brian Behlendorf <[email protected]>
Signed-off-by: Tony Hutter <[email protected]>
Closes #7178

7 files changed, 133 insertions, 18 deletions

diff --git a/cmd/zpool/Makefile.am b/cmd/zpool/Makefile.am
index c7b8b76e3..d07f8d616 100644
--- a/cmd/zpool/Makefile.am
+++ b/cmd/zpool/Makefile.am
@@ -63,7 +63,12 @@ dist_zpoolexec_SCRIPTS = \
 	zpool.d/nvme_err \
 	zpool.d/pwr_cyc \
 	zpool.d/upath \
-	zpool.d/vendor
+	zpool.d/vendor \
+	zpool.d/smart_test \
+	zpool.d/test_type \
+	zpool.d/test_status \
+	zpool.d/test_progress \
+	zpool.d/test_ended
 
 zpoolconfdefaults = \
 	enc \
@@ -102,7 +107,12 @@ zpoolconfdefaults = \
 	nvme_err \
 	pwr_cyc \
 	upath \
-	vendor
+	vendor \
+	smart_test \
+	test_type \
+	test_status \
+	test_progress \
+	test_ended
 
 install-data-hook:
 	$(MKDIR_P) "$(DESTDIR)$(zpoolconfdir)"
diff --git a/cmd/zpool/zpool.d/smart b/cmd/zpool/zpool.d/smart
index 4bc3af39d..64b5f6e4e 100755
--- a/cmd/zpool/zpool.d/smart
+++ b/cmd/zpool/zpool.d/smart
@@ -24,8 +24,44 @@ 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).
+smart_test:	Show SMART self-test results summary.
+test_type:	Show SMART self-test type (short, long... ).
+test_status:	Show SMART self-test status.
+test_progress:	Show SMART self-test percentage done.
+test_ended:	Show when the last SMART self-test ended (if supported).
 "
 
+# Hack for developer testing
+#
+# If you set $samples to a directory containing smartctl output text files,
+# we will use them instead of running smartctl on the vdevs.  This can be
+# useful if you want to test a bunch of different smartctl outputs.  Also, if
+# $samples is set, and additional 'file' column is added to the zpool output
+# showing the filename.
+samples=
+
+# get_filename_from_dir DIR
+#
+# Look in directory DIR and return a filename from it.  The filename returned
+# is chosen quasi-sequentially (based off our PID).  This allows us to return
+# a different filename every time this script is invoked (which we do for each
+# vdev), without having to maintain state.
+get_filename_from_dir()
+{
+	dir=$1
+	pid="$$"
+	num_files=$(find "$dir" -maxdepth 1 -type f | wc -l)
+	mod=$((pid % num_files))
+	i=0
+	find "$dir" -type f -printf "%f\n" | while read -r file ; do
+		if [ "$mod" = "$i" ] ; then
+			echo "$file"
+			break
+		fi
+		i=$((i+1))
+	done
+}
+
 script=$(basename "$0")
 
 if [ "$1" = "-h" ] ; then
@@ -35,8 +71,16 @@ fi
 
 smartctl_path=$(which smartctl)
 
-if [ -b "$VDEV_UPATH" ] && [ -x "$smartctl_path" ]; then
-	raw_out=$(eval "sudo $smartctl_path -a $VDEV_UPATH")
+if [ -b "$VDEV_UPATH" ] && [ -x "$smartctl_path" ] || [ -n "$samples" ] ; then
+	if [ -n "$samples" ] ; then
+		# cat a smartctl output text file instead of running smartctl
+		# on a vdev (only used for developer testing).
+		file=$(get_filename_from_dir $samples)
+		echo "file=$file"
+		raw_out=$(cat "$samples/$file")
+	else
+		raw_out=$(eval "sudo $smartctl_path -a $VDEV_UPATH")
+	fi
 
 	# What kind of drive are we?  Look for the right line in smartctl:
 	#
@@ -49,7 +93,6 @@ if [ -b "$VDEV_UPATH" ] && [ -x "$smartctl_path" ]; then
 	# 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="$8}
@@ -58,10 +101,11 @@ if [ -b "$VDEV_UPATH" ] && [ -x "$smartctl_path" ]; then
 /Elements in grown defect list/{print "defect="$6}
 
 # SAS common
+/SAS/{type="sas"}
 /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}
+/number of hours powered up/{print "hours_on="$7; hours_on=int($7)}
 /Serial number:/{print "serial="$3}
 
 # SATA specific
@@ -74,13 +118,16 @@ if [ -b "$VDEV_UPATH" ] && [ -x "$smartctl_path" ]; then
 /Power_Cycle_Count/{print "pwr_cyc="$10}
 
 # SATA common
+/SATA/{type="sata"}
 /Temperature_Celsius/{print "temp="$10}
 /Airflow_Temperature_Cel/{print "temp="$10}
+/Current Temperature:/{print "temp="$3}
 /SMART overall-health self-assessment test result:/{print "health="$6}
-/Power_On_Hours/{print "hours_on="$10}
+/Power_On_Hours/{print "hours_on="$10; hours_on=int($10)}
 /Serial Number:/{print "serial="$3}
 
 # NVMe common
+/NVMe/{type="nvme"}
 /Temperature:/{print "temp="$2}
 /SMART overall-health self-assessment test result:/{print "health="$6}
 /Power On Hours:/{gsub("[^0-9]","",$4); print "hours_on="$4}
@@ -90,39 +137,92 @@ if [ -b "$VDEV_UPATH" ] && [ -x "$smartctl_path" ]; then
 # NVMe specific
 /Media and Data Integrity Errors:/{print "nvme_err="$6}
 
-END {ORS="\n"; print ""}
+# SMART self-test info
+/Self-test execution status:/{progress=tolower($4)} # SAS
+/SMART Self-test log/{test_seen=1} # SAS
+/SMART Extended Self-test Log/{test_seen=1} # SATA
+/# 1/{
+	test_type=tolower($3"_"$4);
+	# Status could be one word ("Completed") or multiple ("Completed: read
+	# failure").  Look for the ":" to see if we need to grab more words.
+
+	if ($5 ~ ":")
+		status=tolower($5""$6"_"$7)
+	else
+		status=tolower($5)
+	if (status=="self")
+		status="running";
+
+	if (type == "sas") {
+		hours=int($(NF-4))
+	} else {
+		hours=int($(NF-1))
+		# SATA reports percent remaining, rather than percent done
+		# Convert it to percent done.
+		progress=(100-int($(NF-2)))"%"
+	}
+	# When we int()-ify "hours", it converts stuff like "NOW" and "-" into
+	# 0.  In those cases, set it to hours_on, so they will cancel out in
+	# the "hours_ago" calculation later on.
+	if (hours == 0)
+		hours=hours_on
+
+	if (test_seen) {
+		print "test="hours_on
+		print "test_type="test_type
+		print "test_status="status
+		print "test_progress="progress
+	}
+	# Not all drives report hours_on
+	if (hours_on && hours) {
+		total_hours_ago=(hours_on-hours)
+		days_ago=int(total_hours_ago/24)
+		hours_ago=(total_hours_ago % 24)
+		if (days_ago != 0)
+			ago_str=days_ago"d"
+		if (hours_ago !=0)
+			ago_str=ago_str""hours_ago"h"
+		print "test_ended="ago_str
+	}
+}
+
+END {print "type="type; ORS="\n"; print ""}
 ');
 fi
+type=$(echo "$out" | grep '^type=' | cut -d '=' -f 2)
 
-# 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 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"
+	type="sata"
 	out=
 fi
 
 case $script in
 smart)
 	# Print temperature plus common predictors of drive failure
-	if [ "$type" = "SAS" ] ; then
+	if [ "$type" = "sas" ] ; then
 		scripts="temp|health|r_ucor|w_ucor"
-	elif [ "$type" = "ATA" ] ; then
+	elif [ "$type" = "sata" ] ; then
 		scripts="temp|health|ata_err|realloc|rep_ucor|cmd_to|pend_sec|off_ucor"
-	elif [ "$type" = "NVMe" ] ; then
+	elif [ "$type" = "nvme" ] ; then
 		scripts="temp|health|nvme_err"
 	fi
 	;;
 smartx)
 	# Print some other interesting stats
-	if [ "$type" = "SAS" ] ; then
+	if [ "$type" = "sas" ] ; then
 		scripts="hours_on|defect|nonmed|r_proc|w_proc"
-	elif [ "$type" = "ATA" ] ; then
+	elif [ "$type" = "sata" ] ; then
 		scripts="hours_on|pwr_cyc"
-	elif [ "$type" = "NVMe" ] ; then
+	elif [ "$type" = "nvme" ] ; then
 		scripts="hours_on|pwr_cyc"
 	fi
 	;;
+smart_test)
+	scripts="test_type|test_status|test_progress|test_ended"
+	;;
 *)
 	scripts="$script"
 esac
diff --git a/cmd/zpool/zpool.d/smart_test b/cmd/zpool/zpool.d/smart_test
new file mode 120000
index 000000000..94f22861f
--- /dev/null
+++ b/cmd/zpool/zpool.d/smart_test
@@ -0,0 +1 @@
+smart
\ No newline at end of file
diff --git a/cmd/zpool/zpool.d/test_ended b/cmd/zpool/zpool.d/test_ended
new file mode 120000
index 000000000..94f22861f
--- /dev/null
+++ b/cmd/zpool/zpool.d/test_ended
@@ -0,0 +1 @@
+smart
\ No newline at end of file
diff --git a/cmd/zpool/zpool.d/test_progress b/cmd/zpool/zpool.d/test_progress
new file mode 120000
index 000000000..94f22861f
--- /dev/null
+++ b/cmd/zpool/zpool.d/test_progress
@@ -0,0 +1 @@
+smart
\ No newline at end of file
diff --git a/cmd/zpool/zpool.d/test_status b/cmd/zpool/zpool.d/test_status
new file mode 120000
index 000000000..94f22861f
--- /dev/null
+++ b/cmd/zpool/zpool.d/test_status
@@ -0,0 +1 @@
+smart
\ No newline at end of file
diff --git a/cmd/zpool/zpool.d/test_type b/cmd/zpool/zpool.d/test_type
new file mode 120000
index 000000000..94f22861f
--- /dev/null
+++ b/cmd/zpool/zpool.d/test_type
@@ -0,0 +1 @@
+smart
\ No newline at end of file