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