On Thu, Sep 15, 2022 at 2:45 PM Petr Lautrbach <plautrba@xxxxxxxxxx> wrote: > `fixfiles -M relabel` temporary bind mounts filestems before relabeling > but it leaves a directory mounted in /tmp/tmp.XXXX when a user hits > CTRL-C. It means that if the user run `fixfiles -M relabel` again and > answered Y to clean out /tmp directory, it would remove all data from > mounted fs. > > Fixes: https://bugzilla.redhat.com/show_bug.cgi?id=2125355 > > Signed-off-by: Petr Lautrbach <plautrba@xxxxxxxxxx> > --- > policycoreutils/scripts/fixfiles | 11 +++++++++++ > 1 file changed, 11 insertions(+) > > diff --git a/policycoreutils/scripts/fixfiles b/policycoreutils/scripts/fixfiles > index c72ca0eb9d61..6811921970f2 100755 > --- a/policycoreutils/scripts/fixfiles > +++ b/policycoreutils/scripts/fixfiles > @@ -207,6 +207,15 @@ rpm -q --qf '[%{FILESTATES} %{FILENAMES}\n]' "$1" | grep '^0 ' | cut -f2- -d ' ' > [ ${PIPESTATUS[0]} != 0 ] && echo "$1 not found" >/dev/stderr > } > > +# unmount tmp bind mount before exit > +umount_TMP_MOUNT() { > + if [ -n "$TMP_MOUNT" ]; then > + umount "${TMP_MOUNT}${m}" || exit 130 > + rm -rf "${TMP_MOUNT}" || echo "Error cleaning up." > + fi > + exit 130 > +} > + > # > # restore > # if called with -n will only check file context > @@ -251,6 +260,7 @@ case "$RESTORE_MODE" in > else > # we bind mount so we can fix the labels of files that have already been > # mounted over > + trap umount_TMP_MOUNT SIGINT > for m in `echo $FILESYSTEMSRW`; do > TMP_MOUNT="$(mktemp -d)" > test -z ${TMP_MOUNT+x} && echo "Unable to find temporary directory!" && exit 1 > @@ -261,6 +271,7 @@ case "$RESTORE_MODE" in > umount "${TMP_MOUNT}${m}" || exit 1 > rm -rf "${TMP_MOUNT}" || echo "Error cleaning up." > done; > + trap SIGINT > fi > else > echo >&2 "fixfiles: No suitable file systems found" > -- > 2.37.3 > And what if the fixfiles process is terminated via SIGTERM or even SIGKILL? Or a power failure occurs at the wrong time? What if some other process leaves behind a bind mount / other mount in /tmp? My suggestion would be to: 1) Change the trap from SIGINT to EXIT. That will cover both SIGTERM and SIGINT. 2) Additionally modify fullrelabel() to not traverse across mounts when doing the removing (+ possibly exit with an error whenever a mountpoint is encountered - OR - try to unmount the mounts instead of removing their contents). -- Ondrej Mosnacek Senior Software Engineer, Linux Security - SELinux kernel Red Hat, Inc.
