[Bug 1641243] Re: Provide full AppArmor confinement for snaps on 14.04
Tyler Hicks
tyhicks at canonical.com
Thu Dec 1 21:26:17 UTC 2016
The old apparmor upload has been rejected and I'll be uploading a new
version shortly.
** Changed in: apparmor (Ubuntu Trusty)
Status: Incomplete => In Progress
--
You received this bug notification because you are a member of Ubuntu
Foundations Bugs, which is subscribed to dbus in Ubuntu.
https://bugs.launchpad.net/bugs/1641243
Title:
Provide full AppArmor confinement for snaps on 14.04
Status in apparmor package in Ubuntu:
Invalid
Status in dbus package in Ubuntu:
Invalid
Status in apparmor source package in Trusty:
In Progress
Status in dbus source package in Trusty:
In Progress
Bug description:
= apparmor SRU =
[Rationale]
For backporting snapd to 14.04 LTS, we need to provide proper AppArmor confinement for snaps when running under the 16.04 hardware enablement kernel. The apparmor userspace package in 14.04 is missing support key mediation features such as UNIX domain socket rules, AppArmor policy namespaces, and AppArmor profile stacking. UNIX domain socket mediation is needed by nearly all snaps. AppArmor policy namespaces and profile stacking are needed by the lxd snap.
Unfortunately, it was not feasible to backport the individual features
to the 14.04 apparmor package as they're quite complex and have a
large number of dependency patches. Additionally, the AppArmor policy
abstractions from Ubuntu 16.04 are needed to provide proper snap
confinement. Because of these two reasons, the decision to bring
16.04's apparmor package to 14.04 was (very carefully) made.
[Test Case]
https://wiki.ubuntu.com/Process/Merges/TestPlans/AppArmor
This update will go through the Test Plan as well as manual testing to
verify that snap confinement on 14.04 does work. Manual tests include
installing snapd in 14.04 and running simple snaps such as pwgen-
tyhicks and hello-world, as well as a much more complex snap such as
lxd.
The following regression tests from lp:qa-regression-testing (these
packages ship an AppArmor profile) can be used to verify that their
respective packages do not regress:
test-apache2-mpm-event.py
test-apache2-mpm-itk.py
test-apache2-mpm-perchild.py
test-apache2-mpm-prefork.py
test-apache2-mpm-worker.py
test-bind9.py
test-clamav.py
test-cups.py
test-dhcp.py
test-mysql.py
test-ntp.py
test-openldap.py
test-rsyslog.py
test-squid.py
test-strongswan.py
test-tcpdump.py
I have a branch of lp:qa-regression-testing (unmerged, currently at
https://code.launchpad.net/~tyhicks/+git/qa-regression-testing/+ref
/apparmor-trusty-sru) that pulls in the parser and regression tests
from the apparmor 2.8.95~2430-0ubuntu5.3 package currently shipping in
Trusty, in addition to the tests in the 2.10.95 based package.
Additionally, manually testing evince, which is confined by an
AppArmor profile, should be done. The manual test should check basic
functionality as well as for proper confinement (`ps auxZ` output).
[Regression Potential]
High. We must be extremely careful to not regress existing, confined applications in Ubuntu 14.04. We are lucky that the upstream AppArmor project has extensive regression tests and that the Ubuntu Security team adds even more testing via the AppArmor Test Plan.
Care was taken to minimally change how the AppArmor policies are
loaded during the boot process. I also verified that the abstractions
shipped in apparmor and the profiles shipped in apparmor-profiles are
the same across this SRU update.
= dbus SRU =
[Rationale]
For backporting snapd to 14.04 LTS, we need to provide proper D-Bus mediation for snaps when running under the 16.04 hardware enablement kernel. The dbus package in 14.04 is missing support for blocking unrequested reply messages. This functionality was added to the D-Bus AppArmor mediation patches after 14.04 was released but before the patches were merged upstream in dbus. The idea is to prevent a malicious snap from attacking another snap, over D-Bus, with unrequested reply messages and also to prevent two connections from subverting the snap confinement by communicating via unrequested reply messages.
[Test Case]
The upstream AppArmor userspace project has thorough tests for D-Bus
mediation, including unrequested replies. Its
tests/regression/apparmor/dbus_*.sh tests should be ran before and
after updating to the dbus SRU. Before updating, the
dbus_unrequested_reply.sh should fail and should pass after updating.
To run the dbus_*.sh tests:
$ sudo apt-get install -y bzr libdbus-1-dev
$ bzr branch lp:apparmor # apt-get source apparmor to test the current apparmor
$ cd apparmor/tests/regression/apparmor/
$ make USE_SYSTEM=1 \
dbus_{eavesdrop,message,service,unrequested_reply} uservars.inc
$ for t in dbus_{eavesdrop,message,service,unrequested_reply}.sh; \
do sudo VERBOSE=1 bash $t || break; done
The exit code should be 0 and all output lines should start with
"ok:".
In addition, the test-dbus.py tests from lp:qa-regression-testing
should be ran to verify basic D-Bus functionality.
This update will go through the Test Plan as well as manual testing to
verify that snap confinement on 14.04 does work. Manual tests include
installing snapd in 14.04 and running simple snaps such as pwgen-
tyhicks and hello-world, as well as a much more complex snap such as
lxd.
[Regression Potential]
Low. There's no use for unrequested D-Bus reply messages and silently dropping them for AppArmor confined applications should have no unintended side effects. The unrequested reply protections have been present in releases after 14.04 and have not caused any issues.
To manage notifications about this bug go to:
https://bugs.launchpad.net/ubuntu/+source/apparmor/+bug/1641243/+subscriptions
More information about the foundations-bugs
mailing list