iankelling.org Git - distro-setup/blob - btrfsmaint

   1 #!/bin/bash
   2
   3
   4 [[ $EUID == 0 ]] || exec sudo -E "${BASH_SOURCE[0]}" "$@"
   5
   6 set -e; . /usr/local/lib/err; set +e
   7
   8 # inspired from
   9 # https://github.com/kdave/btrfsmaintenance
  10
  11 if [[ $INVOCATION_ID ]]; then
  12   err-cleanup() {
  13     exim -odf -i root <<EOF
  14 From: root@$(hostname -f)
  15 To: root@$(hostname -f)
  16 Subject: btrfsmaint automatically exited on command error
  17
  18 journalctl -u btrfsmaint -n 50:
  19 $(journalctl -u btrfsmaint -n 50)
  20 EOF
  21   }
  22 fi
  23
  24 dusage="5 10"
  25 musage="5"
  26
  27 e() {
  28   echo "btrfsmaint: $*"
  29   if ! $dryrun; then
  30     "$@"
  31   fi
  32 }
  33
  34 check-idle() {
  35   type -p xscreensaver-command &>/dev/null || return 0
  36   export XAUTHORITY=/home/iank/.Xauthority
  37   export DISPLAY=:0
  38   locked=false
  39   if lock_info=$(xscreensaver-command -time); then
  40     if [[ $lock_info != *non-blanked* ]]; then
  41       locked=true
  42     fi
  43   else
  44     locked=true
  45   fi
  46 }
  47
  48 usage() {
  49   cat <<EOF
  50 Usage: ${0##*/} [OPTIONS]
  51 Do btrfs maintence or stop if we have X and xprintidle shows a user
  52
  53 Normally, no options are needed.
  54
  55 --check  Only check if an existing maintence should be cancelled due to
  56          nonidle user and run in a loop every 20 seconds for 10
  57          minutes.
  58
  59 --dryrun Just print out what we would do.
  60
  61 --force  Run regardless of user idle status on all disks and do scrub
  62          regardless of when it was last run.
  63 --no-stats  Avoid checking error statistics. Use this to avoid a rare race
  64             condition when running --check concurrently with normal run.
  65
  66
  67 -h|--help   Show help
  68
  69 Note: Uses util-linux getopt option parsing: spaces between args and
  70 options, short options can be combined, options before args.
  71 EOF
  72   exit $1
  73 }
  74
  75 ##### begin command line parsing ########
  76
  77 # ensure we can handle args with spaces or empty.
  78 ret=0; getopt -T || ret=$?
  79 [[ $ret == 4 ]] || { echo "Install util-linux for enhanced getopt" >&2; exit 1; }
  80
  81 check=false
  82 dryrun=false
  83 force=false
  84 stats=true
  85
  86 temp=$(getopt -l help,check,dryrun,force,no-stats h "$@") || usage 1
  87 eval set -- "$temp"
  88 while true; do
  89   case $1 in
  90     --check) check=true ;;
  91     --dryrun) dryrun=true ;;
  92     --force) force=true ;;
  93     --no-stats) stats=false ;;
  94     -h|--help) usage ;;
  95     --) shift; break ;;
  96     *) echo "$0: unexpected args: $*" >&2 ; usage 1 ;;
  97   esac
  98   shift
  99 done
 100 readonly check dryrun force stats
 101 ##### end command line parsing ########
 102
 103
 104 main() {
 105   if ! $force; then
 106     check-idle
 107     if ! $check; then
 108       min=0
 109       max_min=300
 110       # When the cron kicks in, we may not be idle (physically sleeping) yet, so
 111       # wait.
 112       while ! $locked && (( min < max_min )); do
 113         min=$(( min + 1 ))
 114         sleep 60
 115         check-idle
 116       done
 117       # If we've waited a really long time for idle, just give up.
 118       if (( min == max_min )); then
 119         return
 120       fi
 121     fi
 122   fi
 123
 124
 125   fnd="findmnt --types btrfs --noheading"
 126   for x in $($fnd --output "SOURCE" --nofsroot | sort -u); do
 127     mnt=$($fnd --output "TARGET" --first-only --source $x)
 128     [[ $mnt ]] || continue
 129
 130     #### begin look for diff in stats, eg: increasing error count ####
 131     if $stats; then
 132       tmp=$(mktemp)
 133       # ${mnt%/} so that if mnt is / we avoid making a buggy looking path
 134       stats_path=${mnt%/}/btrfs-dev-stats
 135       if [[ ! -e $stats_path ]]; then
 136         btrfs dev stats -c $mnt >$stats_path ||: # populate initial reading
 137       elif ! btrfs dev stats -c $mnt >$tmp; then
 138         if ! diff -q $stats_path $tmp; then
 139           mv $stats_path $stats_path.1
 140           cat $tmp >$stats_path
 141           diff=$(diff -u $stats_path $tmp 2>&1 ||:)
 142           printf "diff of: btrfs dev stats -c %s\n%s\n" "$mnt" "$diff"
 143           exim -odf -i root <<EOF
 144 From: root@$(hostname -f)
 145 To: root@$(hostname -f)
 146 Subject: btrfsmaint: device stats changed for $mnt
 147
 148 diff of: btrfs dev stats -c $mnt
 149 $diff
 150 EOF
 151         fi
 152       fi
 153       rm -f $tmp
 154     fi
 155     #### end look for diff in stats, eg: increasing error count ####
 156
 157     if $check; then
 158       if ! $locked; then
 159         if $dryrun; then
 160           echo "$0: not idle. if this wasnt a dry run, btrfs scrub cancel $mnt"
 161         else
 162           btrfs scrub cancel $mnt &>/dev/null ||:
 163         fi
 164       fi
 165       continue
 166     fi
 167
 168     # for comparing before and after balance.
 169     # the log is already fairly verbose, so commented.
 170     # e btrfs filesystem df $mnt
 171     # e df -H $mnt
 172     if btrfs filesystem df $mnt | grep -q "Data+Metadata"; then
 173       for usage in $dusage; do
 174         e ionice -c 3 btrfs balance start -dusage=$usage -musage=$usage $mnt
 175       done
 176     else
 177       e ionice -c 3 btrfs balance start -dusage=0 $mnt
 178       for usage in $dusage; do
 179         e ionice -c 3 btrfs balance start -dusage=$usage $mnt
 180       done
 181       e ionice -c 3 btrfs balance start -musage=0 $mnt
 182       for usage in $musage; do
 183         e ionice -c 3 btrfs balance start -musage=$usage $mnt
 184       done
 185     fi
 186     date=
 187     scrub_status=$(btrfs scrub status $mnt)
 188     if printf "%s\n" "$scrub_status" | grep -i '^status:[[:space:]]*finished$' &>/dev/null; then
 189       date=$(printf "%s\n" "$scrub_status" | sed -rn 's/^Scrub started:[[:space:]]*(.*)/\1/p')
 190     fi
 191     if [[ ! $date ]]; then
 192       # output from older versions, at least btrfs v4.15.1
 193       date=$(
 194         printf "%s\n" "$scrub_status" | \
 195           sed -rn 's/^\s*scrub started at (.*) and finished.*/\1/p'
 196           )
 197     fi
 198     if ! $force && [[ $date ]]; then
 199       if $dryrun; then
 200         echo "$0: last scrub finish for $mnt: $date"
 201       fi
 202       date=$(date --date="$date" +%s)
 203       # if date is sooner than 60 days ago
 204       # the wiki recommends 30 days or so, but
 205       # I'm going with 60 days.
 206       if (( date > EPOCHSECONDS - 60*60*24*60 )); then
 207         if $dryrun; then
 208           echo "$0: skiping scrub of $mnt, last was $(( (EPOCHSECONDS - date) / 60/60/24 )) days ago, < 30 days"
 209         fi
 210         continue
 211       fi
 212     fi
 213     # btrfsmaintenance does -c 2 -n 4, but I want lowest pri.
 214     e btrfs scrub start -Bd -c 3 $mnt
 215
 216     # We normally only do one disk since this is meant to be run in
 217     # downtime and if we try to do all disks, we invariably end up doing
 218     # a scrub after downtime. So, just do one disk per day.
 219     if ! $force; then
 220       return 0
 221     fi
 222   done
 223 }
 224
 225 loop-main() {
 226   while true; do
 227     main
 228     sleep 60
 229   done
 230 }
 231
 232 if $check; then
 233   loop-main
 234 else
 235   main
 236 fi