This patch rewrites snap-discard-ns to support user mounts. This
involves scanning for the new per-user mount namespaces and mount
profiles.

The program was simplified to not use the "abstraction" of provided by
"ns-support.h". The abstraction was poor and with the upcoming changes
to snap-confine to accommodate per-user mount namespaces it is easier to
audit and understand the code directly. This change makes "ns-support.h"
private to snap-confine so it can evolve separately.

The manual page was updated to reflect the new functionality and refresh
some cruft like project bug report URL.

Signed-off-by: Zygmunt Krynicki <zygmunt.krynicki@canonical.com>

This patch fixes an issue that only manifests itself on Ubuntu 14.04
where /etc/mtab is a real file and not a symbolic link to /proc/mtab.
On such system mount(1) which uses libmount internally, keeps track
of all the operations by editing the mtab file. Sadly, our hackery inside
snap-confine and snap-discard-ns is totally not editing mtab. This causes
confusion when umount(1) is asked to detach a mount namespace file.

The problem is that at that time the file is already unmounted, thanks
to updated snap-discard-ns. Because the test had atificially created the
.mnt file with touch and mount(1) there is now a stale mount entry in mtab.
In the cleanup section of each test (actually the prepare section but
conceptually it is the cleanup after previous test) we, via snapd.postrm
or snap-mgmt.sh unmount and remove all the preserved mount namespaces.
When umount(1) is given a file that is not a mount point it goes to
search mtab, treating it as a mount _source_ (not target) and proceeds
to umount the target of whatever is found.

To avoid this issue, since we are not using libmount in snap-discard-ns
or snap-confine, simply avoid using libmount driven mount(1) in the test
as well.

This patch implements a thin python wrapper around mount and uses it
for the construction of the fake per-user mount namespace.

Signed-off-by: Zygmunt Krynicki <zygmunt.krynicki@canonical.com>

On Amazon Linux 2 we don't have Python 3 but we do have Python 2.
Implement a Python 2.7 compatible version of mount.py and use it
automatically if Python 3 is not available.

Signed-off-by: Zygmunt Krynicki <zygmunt.krynicki@canonical.com>

Thanks to Maciej for spotting this.

Signed-off-by: Zygmunt Krynicki <zygmunt.krynicki@canonical.com>

Thanks to Maciej for spotting this.

Signed-off-by: Zygmunt Krynicki <zygmunt.krynicki@canonical.com>

Signed-off-by: Zygmunt Krynicki <zygmunt.krynicki@canonical.com>

Signed-off-by: Zygmunt Krynicki <zygmunt.krynicki@canonical.com>

Signed-off-by: Zygmunt Krynicki <zygmunt.krynicki@canonical.com>

Signed-off-by: Zygmunt Krynicki <zygmunt.krynicki@canonical.com>

Signed-off-by: Zygmunt Krynicki <zygmunt.krynicki@canonical.com>

Signed-off-by: Zygmunt Krynicki <zygmunt.krynicki@canonical.com>

Since this is a "clean-slate" due to the rewrite we can introduce
alternate formatting mode. After some experimentation I picked
clang-format and used the following style:

    {BasedOnStyle: Google, IndentWidth: 4, ColumnLimit: 120}

This makes the code read better than with defaults, especially due to
longer column limit.

Signed-off-by: Zygmunt Krynicki <zygmunt.krynicki@canonical.com>