fix lxc ZFS filesystem cannot destroy error.

README.md

Fix LXD error when cannot destroy filesystem dataset busy

Sometimes lxd with zfs gets hosed in a way that:

$ lxc delete --force golden-piranha
Error: Failed to destroy ZFS filesystem: cannot destroy 'default/containers/golden-piranha': dataset is busy

The script lxcl-fix-filesystem-busy below helps to fix that.

lxc-fix-filesystem-busy

#!/bin/sh
#
# lxc delete --force b3
# Error: Failed to destroy ZFS filesystem: cannot destroy 
#   'default/containers/b3': dataset is busy

fail() { [ $# -eq 0 ] || error "$@"; exit 1; }
error() { echo "$@" 1>&2; }

[ "$1" = "--help" -o "$1" = "-h" ] && { echo "give path for dataset"; exit 1; }
[ -n "$1" ] || fail "give path for 'dataset busy'"
path="$1"

out=$(gawk '
    BEGINFILE { if (ERRNO != "") nextfile; }
    $9 == path {
    t=FILENAME;
    sub(/\/[^/]*\//, "", t);
    sub(/\/.*/, "", t);
    printf("%s %s\n", t, $5); nextfile; }' "path=$path" \
       /proc/*/mountinfo)

[ -n "$out" ] || { error "no mounts found of $path"; exit 1; }

if [ "$(id -u)" = "0" ]; then
    dry=""
else
    echo "# dry run, be root"
    dry="error"
fi

echo "$out" |
    while read pid path; do
        grep -q "$path" "/proc/$pid/mountinfo" || {
            #error "$pid: already unmounted";
            continue;
        }
        $dry nsenter -t "$pid" -m -- umount "$path" ||
           fail "failed nsenter -t $pid -m -- umount $path"
        if [ "$dry" = "error" ]; then
            # normally just one umount needs doing. so in dry mode just exit.
            exit 0
        fi
        error "umounted $path in $pid"
    done