xfs-defrag-helper2

#!/bin/bash
# XFS defrag helper script
# Copyright 2010-2018 Robin H. Johnson <robbat2@gentoo.org>
# Originally written for IsoHunt, later improved for BC Libraries Cooperative & Dreamhost
#
# For each filesystem
# 1. get complete output of inode/fragments
# 2. select all inodes w/ more than 10 excess fragments (actual-ideal)
# 3. just the inode lines, not directories or other stats
# 4. sort inodes by number of fragments, descending
# 5. extract the inode number
# 6. convert inode number to file name
# 7. dump file names to temp file
# 8. defrag files in order from tempfile
#
#
# You must NOT pipe xfs_db & xfs_fsr together
# That WILL cause a deadlock in XFS!

XFS_VOL="$1"
# set the number of fragments past which we fire
: ${FRAGLIMIT:=10}
export TMPDIR=$(mktemp -d /tmp/xfs-defrag.XXXXXX)
export TMPFILE=$(mktemp -p $TMPDIR files-to-defrag.XXXXXX)
trap "rm -rf $TMPDIR" EXIT
echo $TMPDIR
cd $TMPDIR
AWKSCRIPT="${TMPDIR}/xfs_db_parse.awk"
cat >"${AWKSCRIPT}" <<'EOF'
BEGIN { 
	# Truncate file
    printf "" >"inode.frags"; 
	# set limit over excess fragments
    if (limit < 1) {
        limit = 1;
    }   
}
# find lines
/^inode [0-9]+ actual [0-9]+ ideal [0-9]+/ {
  inode=$2;
  actual=$4;
  ideal=$6;
  if(actual - ideal > limit) {
    print inode, actual-ideal >>"inode.frags";
  }
  next;
}
# stats
/actual.*ideal.*fragmentation.factor/ { print > "stats"; next }
# useless output
/Note, this number is largely meaningless./ { next }
/Files on this filesystem average/ { next }
# other lines
// {
  printf "xfs-defrag: unknown xfs_db output: '%s'\n", $0 > "/dev/stderr"
}
EOF
found=0
grep -w xfs /proc/mounts | while read dev mnt fstype rest ; do
  if [ -n "$XFS_VOL" ]; then
    [ "$XFS_VOL" != "$dev" ] && [ "$XFS_VOL" != "$mnt" ] && continue
	found=1
  fi
  echo Scanning $mnt

  # This command will generate a LOT of output, we must pipe
  # this will write (inode,fragments) to the file 'inode.frags'.
  xfs_db -r -c 'frag -v -f' $dev |
	  awk -v limit=$FRAGLIMIT -f "${AWKSCRIPT}"

  # sort by inode number
  sort -k 1b,1 <inode.frags >inode.frags.tmp && mv -f inode.frags.tmp inode.frags

  # If there are no fragmented files, skip to next FS
  [ -s inode.frags ] || continue

  # find the filenames for each inode, print 'inode filename'
  # WARNING: filename may contain spaces!
  find $mnt \
	\( \
	-false \
	$(awk '{printf "-or -inum %d ",$1}' inode.frags) \
	\) \
    -printf "%i %h/%f\n" \
	>inode.filename
  # sort by inode number
  # WARNING: filename may contain spaces!
  sort -k 1b,1 <inode.filename >inode.filename.tmp && mv inode.filename.tmp -f inode.filename

  # Now we need to join the stats, so we can visit files by name, in order of
  # how many fragments they have.
  join -j 1 inode.frags inode.filename >joined

  # 'joined' now contains
  # inode numfrag filename
  # WARNING: filename may contain spaces!
  sort -k +2nr joined |cut -d' ' -f3- >>$TMPFILE

  # TMPFILE now contains
  # filename

  # Do not scan more
  [ -n "$XFS_VOL" -a "$found" = "1" ] && break
done

# You must NOT pipe xfs_db & xfs_fsr together
# That WILL cause a deadlock in XFS!

# Run FSR on files listed
if [ -z "$DRYRUN" ] && [ -s "${TMPFILE}" ]; then
	# xfs_fsr does not run properly under xargs normally, because it wants an
	# interactive stdin.
	# if stdin is not interactive [tested with isatty(3)], then xfs_fsr output
	# goes to syslog!
	# to work around this, we use 'script -c "$CMD" /dev/null' to force the
	# command to be in a new interactive shell.
	cat $TMPFILE \
		| tr '\n' '\0' \
		| ionice -c 3 \
		xargs -0 --replace \
		/usr/bin/script -c 'xfs_fsr -v {}' /dev/null \
		| grep --line-buffered -e '^extents ' -e '(skipping)'
fi