zfs-patches/0036-Add-SMART-attributes-for-SSD-and-NVMe.patch

   1 From 0000000000000000000000000000000000000000 Mon Sep 17 00:00:00 2001
   2 From: bunder2015 <omfgbunder@gmail.com>
   3 Date: Wed, 21 Feb 2018 16:52:47 -0500
   4 Subject: [PATCH] Add SMART attributes for SSD and NVMe
   5 MIME-Version: 1.0
   6 Content-Type: text/plain; charset=UTF-8
   7 Content-Transfer-Encoding: 8bit
   8
   9 This adds the SMART attributes required to probe Samsung SSD and NVMe
  10 (and possibly others) disks when using the "zpool status -c" command.
  11
  12 Reviewed-by: loli10K <ezomori.nozomu@gmail.com>
  13 Reviewed-by: Giuseppe Di Natale <dinatale2@llnl.gov>
  14 Reviewed-by: Tony Hutter <hutter2@llnl.gov>
  15 Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
  16 Signed-off-by: bunder2015 <omfgbunder@gmail.com>
  17 Closes #7183
  18 Closes #7193
  19 (cherry picked from commit c705d8386b21b08aefdc62b6b1a556aab6717316)
  20 Signed-off-by: Fabian Grünbichler <f.gruenbichler@proxmox.com>
  21 ---
  22  cmd/zpool/Makefile.am      |  2 ++
  23  cmd/zpool/zpool.d/nvme_err |  1 +
  24  cmd/zpool/zpool.d/smart    | 23 +++++++++++++++++++++--
  25  3 files changed, 24 insertions(+), 2 deletions(-)
  26  create mode 120000 cmd/zpool/zpool.d/nvme_err
  27
  28 diff --git a/cmd/zpool/Makefile.am b/cmd/zpool/Makefile.am
  29 index 6eff1d143..c7b8b76e3 100644
  30 --- a/cmd/zpool/Makefile.am
  31 +++ b/cmd/zpool/Makefile.am
  32 @@ -60,6 +60,7 @@ dist_zpoolexec_SCRIPTS = \
  33         zpool.d/pend_sec \
  34         zpool.d/off_ucor \
  35         zpool.d/ata_err \
  36 +       zpool.d/nvme_err \
  37         zpool.d/pwr_cyc \
  38         zpool.d/upath \
  39         zpool.d/vendor
  40 @@ -98,6 +99,7 @@ zpoolconfdefaults = \
  41         pend_sec \
  42         off_ucor \
  43         ata_err \
  44 +       nvme_err \
  45         pwr_cyc \
  46         upath \
  47         vendor
  48 diff --git a/cmd/zpool/zpool.d/nvme_err b/cmd/zpool/zpool.d/nvme_err
  49 new file mode 120000
  50 index 000000000..94f22861f
  51 --- /dev/null
  52 +++ b/cmd/zpool/zpool.d/nvme_err
  53 @@ -0,0 +1 @@
  54 +smart
  55 \ No newline at end of file
  56 diff --git a/cmd/zpool/zpool.d/smart b/cmd/zpool/zpool.d/smart
  57 index 3721f30ed..4bc3af39d 100755
  58 --- a/cmd/zpool/zpool.d/smart
  59 +++ b/cmd/zpool/zpool.d/smart
  60 @@ -23,6 +23,7 @@ off_ucor:     Show SMART offline uncorrectable errors (ATA).
  61  ata_err:       Show SMART ATA errors (ATA).
  62  pwr_cyc:       Show SMART power cycle count (ATA).
  63  serial:                Show disk serial number.
  64 +nvme_err:      Show SMART NVMe errors (NVMe).
  65  "
  66
  67  script=$(basename "$0")
  68 @@ -37,7 +38,7 @@ smartctl_path=$(which smartctl)
  69  if [ -b "$VDEV_UPATH" ] && [ -x "$smartctl_path" ]; then
  70         raw_out=$(eval "sudo $smartctl_path -a $VDEV_UPATH")
  71
  72 -       # Are we a SAS or ATA drive?  Look for the right line in smartctl:
  73 +       # What kind of drive are we?  Look for the right line in smartctl:
  74         #
  75         # SAS:
  76         #       Transport protocol:   SAS
  77 @@ -45,7 +46,10 @@ if [ -b "$VDEV_UPATH" ] && [ -x "$smartctl_path" ]; then
  78         # SATA:
  79         #       ATA Version is:   8
  80         #
  81 -       type=$(echo "$raw_out" | grep -m 1 -Eo '^ATA|SAS$')
  82 +       # NVMe:
  83 +       #       SMART/Health Information (NVMe Log 0xnn, NSID 0xnn)
  84 +       #
  85 +       type=$(echo "$raw_out" | grep -m 1 -Eo '^ATA|NVMe|SAS$')
  86         out=$(echo "$raw_out" | awk '
  87  # SAS specific
  88  /read:/{print "rrd="$4"\nr_cor="$5"\nr_proc="$7"\nr_ucor="$8}
  89 @@ -71,10 +75,21 @@ if [ -b "$VDEV_UPATH" ] && [ -x "$smartctl_path" ]; then
  90
  91  # SATA common
  92  /Temperature_Celsius/{print "temp="$10}
  93 +/Airflow_Temperature_Cel/{print "temp="$10}
  94  /SMART overall-health self-assessment test result:/{print "health="$6}
  95  /Power_On_Hours/{print "hours_on="$10}
  96  /Serial Number:/{print "serial="$3}
  97
  98 +# NVMe common
  99 +/Temperature:/{print "temp="$2}
 100 +/SMART overall-health self-assessment test result:/{print "health="$6}
 101 +/Power On Hours:/{gsub("[^0-9]","",$4); print "hours_on="$4}
 102 +/Serial Number:/{print "serial="$3}
 103 +/Power Cycles:/{print "pwr_cyc="$3}
 104 +
 105 +# NVMe specific
 106 +/Media and Data Integrity Errors:/{print "nvme_err="$6}
 107 +
 108  END {ORS="\n"; print ""}
 109  ');
 110  fi
 111 @@ -94,6 +109,8 @@ smart)
 112                 scripts="temp|health|r_ucor|w_ucor"
 113         elif [ "$type" = "ATA" ] ; then
 114                 scripts="temp|health|ata_err|realloc|rep_ucor|cmd_to|pend_sec|off_ucor"
 115 +       elif [ "$type" = "NVMe" ] ; then
 116 +               scripts="temp|health|nvme_err"
 117         fi
 118         ;;
 119  smartx)
 120 @@ -102,6 +119,8 @@ smartx)
 121                 scripts="hours_on|defect|nonmed|r_proc|w_proc"
 122         elif [ "$type" = "ATA" ] ; then
 123                 scripts="hours_on|pwr_cyc"
 124 +       elif [ "$type" = "NVMe" ] ; then
 125 +               scripts="hours_on|pwr_cyc"
 126         fi
 127         ;;
 128  *)
 129 --
 130 2.14.2
 131