When the network interface provisioned by the CNI has static ARP table entries,
the runtime calls AddARPNeighbor to propagate these to the agent. As of today,
these calls are simply rejected.
In order to allow the calls, we do some sanity checks on the arguments:
We must ensure that we don't unexpectedly route traffic to the host that was
not intended to leave the VM. In a first approximation, this applies to
loopback IPs and devices. However, there may be other sensitive ranges (for
example, VPNs between VMs), so there should be some flexibility for users to
restrict this further. This is why we introduce a setting, similar to
UpdateRoutes, that allows restricting the neighbor IPs further.
The only valid state of an ARP neighbor entry is NUD_PERMANENT, which has a
value of 128 [1]. This is already enforced by the runtime.
According to rtnetlink(7), valid flag values are 8 and 128, respectively [2],
thus we allow any combination of these.
[1]: https://github.com/torvalds/linux/blob/4790580/include/uapi/linux/neighbour.h#L72
[2]: https://github.com/torvalds/linux/blob/4790580/include/uapi/linux/neighbour.h#L49C20-L53Fixes: #11664
Signed-off-by: Markus Rudy <mr@edgeless.systems>
- "confidential_emptyDir" becomes "emptyDir" in the settings file.
- "confidential_configMap" becomes "configMap" in settings.
- "mount_source_cpath" becomes "cpath".
- The new "root_path" gets used instead of the old "cpath" to point to
the container root path..
- "confidential_guest" is no longer used. By default it gets replaced
by "enable_configmap_secret_storages"=false, because CoCo is using
CopyFileRequest instead of the Storage data structures for ConfigMap
and/or Secret volume mounts during CreateContainerRequest.
- The value of "guest_pull" becomes true by default.
- "image_layer_verification" is no longer used - just CoCo's guest pull
is supported.
- The Request input files from unit tests are changing to reflect the
new default settings values described above.
- tests/integration/kubernetes/tests_common.sh adjusts the settings for
platforms that are not set-up for CoCo during CI (i.e., platforms
other than SNP, TDX, and CoCo Dev).
Signed-off-by: Dan Mihai <dmihai@microsoft.com>
AKS Confidential Containers are using the tarfs snapshotter. CoCo
upstream doesn't use this snapshotter, so remove this Policy complexity
from upstream.
Signed-off-by: Dan Mihai <dmihai@microsoft.com>
This patch fixes the rules.rego file to ensure that the
policy is correctly parsed and applied by opa.
Signed-off-by: Archana Choudhary <archana1@microsoft.com>
This patch introduces some basic checks for the
`image_guest_pull` storage type in the genpolicy tool.
Signed-off-by: Archana Choudhary <archana1@microsoft.com>
The additional GIDs are handled by genpolicy as a BTreeSet. This set is
then serialized to an ordered JSON array. On the containerd side, the
GIDs are added to a list in the order they are discovered in /etc/group,
and the main GID of the user is prepended to that list. This means that
we don't have any guarantees that the input GIDs will be sorted. Since
the order does not matter here, comparing the list of GIDs as sets is
close enough.
Signed-off-by: Markus Rudy <mr@edgeless.systems>
Use $(sandbox-namespace) wildcard in case none is specified in yaml. If wildcard is present, compare
input against annotation value.
Fixes regression introduced in https://github.com/microsoft/kata-containers/pull/273
where samples that use metadata.namespace env var were no longer working.
Signed-off-by: Saul Paredes <saulparedes@microsoft.com>
Validate more process fields for commands enabled using the
ExecProcessRequest "commands" and/or "regex" fields from the
settings file.
Add function to get the container from state based on container_id
matching instead of matching it against every policy container data
Signed-off-by: Dan Mihai <dmihai@microsoft.com>
Signed-off-by: Ankita Pareek <ankitapareek@microsoft.com>
Using process data inputs for allow_process() is easier to
read/understand compared with the older OCI data inputs.
Signed-off-by: Dan Mihai <dmihai@microsoft.com>
With added support for parsing these fields in genpolicy, we can now
enable policy verification of AdditionalGids.
Signed-off-by: Cameron Baird <cameronbaird@microsoft.com>
This patch:
- adds a count check on mounts
- adds various test scenarios for mounts with emptyDir volume source
Signed-off-by: Archana Choudhary <archana1@microsoft.com>
With fixes to align policy GID parsing with the CRI behavior, we can now
enable policy verification of GIDs.
Signed-off-by: Cameron Baird <cameronbaird@microsoft.com>
This commit adds changes to add input container_id and related
container data to state after a CreateContainerRequest is allowed. This
helps constrain reference container data for evaluating request
inputs to one instead of matching against every policy container data,
Ex: in ExecProcessRequest inputs.
Fixes#11109
Signed-off-by: Sumedh Sharma <sumsharma@microsoft.com>
This commit introduces missing validations for input fields in ExecProcessRequest to
harden the security policy.
The changes include:
- Update rules.rego to add null/empty field enforcements for String_user, SelinuxLabel and ApparmorProfile
- Add unit test cases for ExecProcessRequest for each of the validations
Signed-off-by: Ankita Pareek <ankitapareek@microsoft.com>
Validate sandbox name using a regex. If the YAML specifies metadata.name, use a regex that exact matches.
If the YAML specifies metadata.generateName, use a regex that matches the prefix of the generated name.
Signed-off-by: Saul Paredes <saulparedes@microsoft.com>
Introduce rule to block routes from source addresses which are the
loopback. Block routes added to the lo device.
Signed-off-by: Cameron Baird <cameronbaird@microsoft.com>
Sysctls may be added to a container by the Kubernetes pod definition or
by containerd configuration. This commit adds support for the
corresponding PodSecurityContext field and an option to specify
environment-dependent sysctls in the settings file.
The sysctls requested in a CreateContainerRequest are checked against
the sysctls in the pod definition, or if not defined there in the
defaults in genpolicy-settings.json. There is no check for the presence
of expected sysctls, though, because Kubernetes might legitimately
omit unsafe syscalls itself and because default sysctls might not apply
to all containers.
Fixes: #10064
Signed-off-by: Markus Rudy <mr@edgeless.systems>
Peer pods have a linux namespace of type network. We want to make sure that all
container in the same pod use the same namespace. Therefore, we add the first
namespace path to the state and check all other requests against that.
This commit also adds the corresponding integration test in the policy crate
showcasing the benefit of having rust integration tests for the policy.
Signed-off-by: Leonard Cohnen <lc@edgeless.systems>
- Remove default_namespace from settings
- Ensure container namespaces in a pod match each other in case no namespace is specified in the YAML
Signed-off-by: Saul Paredes <saulparedes@microsoft.com>
Before this patch there was a mismatch between the JSON path under which
the state of the rule evaluation is set in comparison to under which
it is retrieved.
This resulted in the behavior that each time the policy was evaluated,
it thought it was the _first_ time the policy was evaluated.
This also means that the consistency check for the `sandbox_name`
was ineffective.
Signed-off-by: Leonard Cohnen <lc@edgeless.systems>
Disabling the UID Policy rule was a workaround for #9928. Re-enable
that rule here and add a new test/CI temporary workaround for this
issue. This new test workaround will be removed after fixing #9928.
Signed-off-by: Dan Mihai <dmihai@microsoft.com>
* genpolicy: deny UpdateEphemeralMountsRequest
Deny UpdateEphemeralMountsRequest by default, because paths to
critical Guest components can be redirected using such request.
Signed-off-by: Dan Mihai <Daniel.Mihai@microsoft.com>
Hooks are executed on the host, so we don't expect to run hooks and thus
require that no hook paths are set.
Additional Kernel modules expand the attack surface, so require that
none are set. If a use case arises, modules should be allowlisted via
settings.
Signed-off-by: Markus Rudy <mr@edgeless.systems>
CopyFile is invoked by the host's FileSystemShare.ShareFile function,
which puts all files into directories with a common pattern. Copying
files anywhere else is dangerous and must be prevented. Thus, we check
that the target path prefix matches the expected directory pattern of
ShareFile, and that this directory is not escaped by .. traversal.
Signed-off-by: Markus Rudy <mr@edgeless.systems>
Generate policy that validates each exec command line argument, instead
of joining those args and validating the resulting string. Joining the
args ignored the fact that some of the args might include space
characters.
The older format from genpolicy-settings.json was similar to:
"ExecProcessRequest": {
"commands": [
"sh -c cat /proc/self/status"
],
"regex": []
},
That format will not be supported anymore. genpolicy will detect if its
users are trying to use the older "commands" field and will exit with
a relevant error message in that case.
The new settings format is:
"ExecProcessRequest": {
"allowed_commands": [
[
"sh",
"-c",
"cat /proc/self/status"
]
],
"regex": []
},
Signed-off-by: Dan Mihai <dmihai@microsoft.com>
Keep track of individual exec args instead of joining them in the
policy text. Verifying each arg results in a more precise policy,
because some of the args might include space characters.
This improved validation applies to commands specified in K8s YAML
files using:
- livenessProbe
- readinessProbe
- startupProbe
- lifecycle.postStart
- lifecycle.preStop
Signed-off-by: Dan Mihai <dmihai@microsoft.com>
Add setting to allow specifying the cpath for a mount source.
cpath is the root path for most files used by a container. For example,
the container rootfs and various files copied from the Host to the
Guest when shared_fs=none are hosted under cpath.
mount_source_cpath is the root of the paths used a storage mount
sources. Depending on Kata settings, mount_source_cpath might have the
same value as cpath - but on TDX for example these two paths are
different: TDX uses "/run/kata-containers" as cpath,
but "/run/kata-containers/shared/containers" as mount_source_cpath.
Signed-off-by: Saul Paredes <saulparedes@microsoft.com>
Reject CreateContainerRequest field values that are not tested by
Kata CI and that might impact the confidentiality of CoCo Guests.
This change uses a "better safe than sorry" approach to untested
fields. It is very possible that in the future we'll encounter
reasonable use cases that will either:
- Show that some of these fields are benign and don't have to be
verified by Policy, or
- Show that Policy should verify legitimate values of these fields
These are the new CreateContainerRequest Policy rules:
count(input.shared_mounts) == 0
is_null(input.string_user)
i_oci := input.OCI
is_null(i_oci.Hooks)
is_null(i_oci.Linux.Seccomp)
is_null(i_oci.Solaris)
is_null(i_oci.Windows)
i_linux := i_oci.Linux
count(i_linux.GIDMappings) == 0
count(i_linux.MountLabel) == 0
count(i_linux.Resources.Devices) == 0
count(i_linux.RootfsPropagation) == 0
count(i_linux.UIDMappings) == 0
is_null(i_linux.IntelRdt)
is_null(i_linux.Resources.BlockIO)
is_null(i_linux.Resources.Network)
is_null(i_linux.Resources.Pids)
is_null(i_linux.Seccomp)
i_linux.Sysctl == {}
i_process := i_oci.Process
count(i_process.SelinuxLabel) == 0
count(i_process.User.Username) == 0
Signed-off-by: Dan Mihai <dmihai@microsoft.com>
CreateContainerRequest objects can specify devices to be created inside
the guest VM. This change ensures that requested devices have a
corresponding entry in the PodSpec.
Devices that are added to the pod dynamically, for example via the
Device Plugin architecture, can be allowlisted globally by adding their
definition to the settings file.
Fixes: #9651
Signed-off-by: Markus Rudy <mr@edgeless.systems>
