X-Git-Url: https://iankelling.org/git/?a=blobdiff_plain;f=btrfsmaint;h=0204ed5aa78a4e9043e0aae3a624066252a77955;hb=72c18f3a6a7f1ed0ca16af654a1f804ab96e1ff9;hp=33cae4a36280351c170155f11fc83ba8ca3b2102;hpb=ea108a03dfa2d7f73447c0b14210d766e5ee5d9b;p=distro-setup

diff --git a/btrfsmaint b/btrfsmaint
index 33cae4a..0204ed5 100755
--- a/btrfsmaint
+++ b/btrfsmaint
@@ -3,20 +3,29 @@
 
 [[ $EUID == 0 ]] || exec sudo -E "${BASH_SOURCE[0]}" "$@"
 
-source /a/bin/errhandle/err
+set -e; . /usr/local/lib/err; set +e
 
 # inspired from
 # https://github.com/kdave/btrfsmaintenance
 
+if [[ $INVOCATION_ID ]]; then
+  err-cleanup() {
+    exim -odf -i root <<EOF
+From: root@$(hostname -f)
+To: root@$(hostname -f)
+Subject: btrfsmaint automatically exited on command error
+
+journalctl -u btrfsmaint -n 50:
+$(journalctl -u btrfsmaint -n 50)
+EOF
+  }
+fi
 
-# Man page says we could also use a range, i suppose it would be
-# logical to use a pattern like 5..10 10..20,
-# but I don't know if this would help us at all.
 dusage="5 10"
 musage="5"
 
 e() {
-  echo "cron: $*"
+  echo "btrfsmaint: $*"
   if ! $dryrun; then
     "$@"
   fi
@@ -43,41 +52,58 @@ check-idle() {
 
 usage() {
   cat <<EOF
-Usage: ${0##*/} [ARGS]
-Do btrfs maintence or stop if xprintidle shows a user
+Usage: ${0##*/} [OPTIONS]
+Do btrfs maintence or stop if we have X and xprintidle shows a user
+
+Normally, no options are needed.
 
-force  Run regardless of user idle status on all disks.
-check  Only check if an existing maintence should be cancelled due to
-         nonidle user. Also, runs in a loop every 20 seconds for 10
+--check  Only check if an existing maintence should be cancelled due to
+         nonidle user and run in a loop every 20 seconds for 10
          minutes.
 
+--dryrun Just print out what we would do.
+
+--force  Run regardless of user idle status on all disks and do scrub
+         regardless of when it was last run.
+--no-stats  Avoid checking error statistics. Use this to avoid a rare race
+            condition when running --check concurrently with normal run.
+
+
+-h|--help   Show help
+
 Note: Uses util-linux getopt option parsing: spaces between args and
 options, short options can be combined, options before args.
 EOF
   exit $1
 }
 
+##### begin command line parsing ########
+
+# ensure we can handle args with spaces or empty.
+ret=0; getopt -T || ret=$?
+[[ $ret == 4 ]] || { echo "Install util-linux for enhanced getopt" >&2; exit 1; }
 
-force=false
 check=false
 dryrun=false
-if [[ $1 ]]; then
+force=false
+stats=true
+
+temp=$(getopt -l help,check,dryrun,force,no-stats h "$@") || usage 1
+eval set -- "$temp"
+while true; do
   case $1 in
-    check)
-      check=true
-      ;;
-    force)
-      force=true
-      ;;
-    dryrun)
-      dryrun=true
-      ;;
-    *)
-      echo "$0: error: unexpected arg" >&2
-      usage 1
-      ;;
+    --check) check=true ;;
+    --dryrun) dryrun=true ;;
+    --force) force=true ;;
+    --no-stats) stats=false ;;
+    -h|--help) usage ;;
+    --) shift; break ;;
+    *) echo "$0: unexpected args: $*" >&2 ; usage 1 ;;
   esac
-fi
+  shift
+done
+readonly check dryrun force stats
+##### end command line parsing ########
 
 
 main() {
@@ -102,21 +128,32 @@ main() {
   fi
 
 
-  tmp=$(mktemp)
   fnd="findmnt --types btrfs --noheading"
   for x in $($fnd --output "SOURCE" --nofsroot | sort -u); do
     mnt=$($fnd --output "TARGET" --first-only --source $x)
     [[ $mnt ]] || continue
 
     #### begin look for diff in stats, eg: increasing error count ####
-
-    # Only run for $check, since it runs in parallel to non-check, avoid
-    # race condition.
-    if $check; then
-      if ! btrfs dev stats -c $mnt >$tmp; then
-        if diff -q $mnt/btrfs-dev-stats $tmp; then
-          diff -u $mnt/btrfs-dev-stats $tmp | mail -s "$HOSTNAME: error: btrfs dev stats -c $mnt" root@localhost
-          cat $tmp >$mnt/btrfs-dev-stats
+    if $stats; then
+      tmp=$(mktemp)
+      # ${mnt%/} so that if mnt is / we avoid making a buggy looking path
+      stats_path=${mnt%/}/btrfs-dev-stats
+      if [[ ! -e $stats_path ]]; then
+        btrfs dev stats -c $mnt >$stats_path ||: # populate initial reading
+      elif ! btrfs dev stats -c $mnt >$tmp; then
+        if ! diff -q $stats_path $tmp; then
+          mv $stats_path $stats_path.1
+          cat $tmp >$stats_path
+          diff=$(diff -u $stats_path $tmp 2>&1 ||:)
+          printf "diff of: btrfs dev stats -c %s\n%s\n" "$mnt" "$diff"
+          exim -odf -i root <<EOF
+From: root@$(hostname -f)
+To: root@$(hostname -f)
+Subject: btrfsmaint: device stats changed for $mnt
+
+diff of: btrfs dev stats -c $mnt
+$diff
+EOF
         fi
       fi
       rm -f $tmp
@@ -164,7 +201,7 @@ main() {
           sed -rn 's/^\s*scrub started at (.*) and finished.*/\1/p'
           )
     fi
-    if [[ $date ]]; then
+    if ! $force && [[ $date ]]; then
       if $dryrun; then
         echo "$0: last scrub finish for $mnt: $date"
       fi
@@ -179,12 +216,12 @@ main() {
         continue
       fi
     fi
-    # -c 2 -n 4 is from btrfsmaintenance, does ionice
-    e btrfs scrub start -Bd -c 2 -n 4 $mnt
+    # btrfsmaintenance does -c 2 -n 4, but I want lowest pri.
+    e btrfs scrub start -Bd -c 3 $mnt
 
-    # We normally only do one disk since this is meant to be run while I sleep
-    # and if we try to do all disks, we invariably end up doing a scrub still
-    # after I've woken up. So, just do one per day.
+    # We normally only do one disk since this is meant to be run in
+    # downtime and if we try to do all disks, we invariably end up doing
+    # a scrub after downtime. So, just do one disk per day.
     if ! $force; then
       return 0
     fi