Add all agent configuration options to README so that users can more easily understand
what these options do and how to configure them at runtime.
Fixes: #9109
Signed-off-by: ChengyuZhu6 <chengyu.zhu@intel.com>
When cgroup v2 is in use, a container should only see its part of the
unified hierarchy in `/sys/fs/cgroup`, not the full hierarchy created
at the OS level. Similarly, `/proc/self/cgroup` inside the container
should display `0::/`, rather than a full path such as :
0::/kubepods.slice/kubepods-besteffort.slice/kubepods-besteffort-podde291f58_8f20_4d44_aa89_c9e538613d85.slice/crio-9e1823d09627f3c2d42f30d76f0d2933abdbc033a630aab732339c90334fbc5f.scope
What is needed here is isolation from the OS. Do that by running the
container in its own cgroup namespace. This matches what runc and
other non VM based runtimes do.
Fixes#9124
Signed-off-by: Greg Kurz <groug@kaod.org>
There is a race condition in agent HVSOCK_STREAMS hashmap, where a
stream may be taken before it is inserted into the hashmap. This patch
add simple retry logic to the stream consumer to alleviate this issue.
Fixes: #6714
Signed-off-by: Zixuan Tan <tanzixuan.me@gmail.com>
Linux forbids opening an existing socket through /proc/<pid>/fd/<fd>,
making some images relying on the special file /dev/stdout(stderr),
/proc/self/fd/1(2) fail to boot in passfd io mode, where the
stdout/stderr of a container process is a vsock socket.
For back compatibility, a pipe is introduced between the process
and the socket, and its read end is set as stdout/stderr of the
container process instead of the socket. The agent will do the
forwarding between the pipe and the socket.
Fixes: #6714
Signed-off-by: Zixuan Tan <tanzixuan.me@gmail.com>
In passfd io mode, when not using a terminal, the stdout/stderr vsock
streams are directly used as the stdout/stderr of the child process.
These streams are non-blocking by default.
The stdout/stderr of the process should be blocking, otherwise
the process may encounter EAGAIN error when writing to stdout/stderr.
Fixes: #6714
Signed-off-by: Zixuan Tan <tanzixuan.me@gmail.com>
When container exits, the agent should clean up the term master fd,
otherwise the fd will be leaked.
Fixes: kata-containers#6714
Signed-off-by: Zixuan Tan <tanzixuan.me@gmail.com>
Currently in the kata container, every io read/write operation requires
an RPC request from the runtime to the agent. This process involves
data copying into/from an RPC request/response, which are high overhead.
To solve this issue, this commit utilize the vsock fd passthrough, a
newly introduced feature in the Dragonball hypervisor. This feature
allows other host programs to pass a file descriptor to the Dragonball
process, directly as the backend of an ordinary hybrid vsock connection.
The runtime-rs now utilizes this feature for container process io. It
open the stdin/stdout/stderr fifo from containerd, and pass them to
Dragonball, then don't bother with process io any more, eliminating
the need for an RPC for each io read/write operation.
In passfd io mode, the agent uses the vsock connections as the child
process's stdin/stdout/stderr, eliminating the need for a pipe
to bump data (in non-tty mode).
Fixes: #6714
Signed-off-by: Zixuan Tan <tanzixuan.me@gmail.com>
Two toml options, `use_passfd_io` and `passfd_listener_port` are introduced
to enable and configure dragonball's vsock fd passthrough io feature.
This commit is a preparation for vsock fd passthrough io feature.
Fixes: #6714
Signed-off-by: Zixuan Tan <tanzixuan.me@gmail.com>
- test_volume_capacity_stats: verify the file block size against the fetched size via statfs()
- test_reseed_rng: Correct the request codes for RNDADDTOENTCNT and RNDRESEEDCRNG when platform is ppc64le
- test list_routes: Add the route only if destination is not empty
- test_new_fs_manager: skip the test if cgroups v2 is used by default
- skip test cases rpc::tests::test_do_write_stream, sandbox::tests::test_find_process, sandbox::t
ests::test_find_container_process and sandbox::tests::add_and_get_container on ppc64le as they are fl
aky
Signed-off-by: Amulyam24 <amulmek1@in.ibm.com>
Some linting errors were identified during the enablement of `make check`.
These have not been found by the Jenkins CI job because `make test` was
only triggered.
The errors for the `agent` occurs under the s390x specific tests while
the other ones for the `kata-ctl` are the architecture-specific code.
This commit is to fix those errors.
Signed-off-by: Hyounggyu Choi <Hyounggyu.Choi@ibm.com>
The agent registers an event fd in `memory.oom_control`. An OOM event is
forwarded to containerd when the event is emitted, regardless of the
content in that file.
I observed content indicating that events should not be forwarded, as shown
below. When `oom_kill` is set to 0, it means no OOM has occurred. Therefore,
it is important to check the content to avoid mistakenly forwarding OOM
events.
```
oom_kill_disable 0
under_oom 0
oom_kill 0
```
Fixes: #8715
Signed-off-by: Xuewei Niu <niuxuewei.nxw@antgroup.com>
Don't release the lock between is_allowed and set_policy calls,
because the policy might change in between these calls.
Also, move more policy code into policy.rs.
Fixes: #8734
Signed-off-by: Dan Mihai <dmihai@microsoft.com>
vfio commits introduce quite a lot change in runtime-rs, this commit is
for all the changes related to ci, including compilation errors and so on.
Signed-off-by: Chao Wu <chaowu@linux.alibaba.com>
This PR fixes k8's configmap/secrets etc update propagation when filesystem sharing is disabled.
The commit introduces below changes with some limitations:
- creates new timestamped directory in guest
- updates the '..data' symlink
- creates user visible symlinks to newly created secrets.
- Limitation: The older timestamped directory and stale user visible symlinks exist in guest
due to missing DELETE api in agent.
Fixes: #7398
Signed-off-by: Sumedh Alok Sharma <sumsharma@microsoft.com>
- Disable device cgroup restriction while pod cgroup is not available.
- Remove balcklist-related names and change whitelist-related names to
allowed_all.
Signed-off-by: Xuewei Niu <niuxuewei.nxw@antgroup.com>
FSManager of systemd cgroup manager is responsible for setting up cgroup
path. The container launching will be failed if the FSManager is in
read-only mode.
Signed-off-by: Xuewei Niu <niuxuewei.nxw@antgroup.com>
The changes include:
- Change to debug logging level for resources after processed.
- Remove a todo for pod cgroup cleanup.
- Add an anyhow context to `get_paths_and_mounts()`.
- Remove code which denys access to VMROOTFS since it won't take effect. If
blackmode is in use, the VMROOTFS will be denyed as default. Otherwise,
device cgroups won't be updated in whitelist mode.
- Add a unit test for `default_allowed_devices()`.
Signed-off-by: Xuewei Niu <niuxuewei.nxw@antgroup.com>
The runk is a standard OCI runtime that isnt' aware of concept of sandbox.
Therefore, the `devcg_info` argument of `LinuxContainer::new()` is
unneccessary to be provided.
Signed-off-by: Xuewei Niu <niuxuewei.nxw@antgroup.com>
The target is to guarantee that containers couldn't escape to access extra
devices, like vm rootfs, etc.
Assume that there is a cgroup, such as `/A/B`. The `B` is container cgroup,
and the `A` is what we called pod cgroup. No matter what permissions are
set for the container (`B`), the `A`'s permission is always `a *:* rwm`. It
leads that containers could acquire permission to access to other devices
in VM that not belongs to themselves.
In order to set devices cgroup properly, the order of setting cgroups is
that the pod cgroup comes first and the container cgroup comes after.
The `Sandbox` has a new field, `devcg_info`, to save cgroup states. To
avoid setting container cgroup too early, an initialization should be done
carefully. `inited`, one of the states, is a boolean to indicate if the pod
cgroup is initialized. If no, the pod cgroup should be created firstly, and
set default permissions. After that, the pause container cgroup is created
and inherits the permissions from the pod cgroup.
If whitelist mode which allows containers to access all devices in VM is
enabled, then device resources from OCI spec are ignored.
This feature not supports systemd cgroup and cgroup v2, since:
- Systemd cgroup implemented on Agent hasn't supported devices subsystem so
far, see: https://github.com/kata-containers/kata-containers/issues/7506.
- Cgroup v2's device controller depends on eBPF programs, which is out of
scope of cgroup.
Fixes: #7507
Signed-off-by: Xuewei Niu <niuxuewei.nxw@antgroup.com>
This is to skip a flaky test `create_tmpfs()` on s390x until a root cause is identified and fixed.
Fixes: #4248
Signed-off-by: Hyounggyu Choi <Hyounggyu.Choi@ibm.com>
Remove the ability to block access to kata agent endpoints by using
agent-config.toml. That functionality is now implemented using the
Agent Policy feature (#7573).
The CCv0 branch relied on blocking endpoints using agent-config.toml
but will set-up an equivalent default policy file instead (#8219).
Fixes: #8228
Signed-off-by: Dan Mihai <dmihai@microsoft.com>
This will help to build the agent binary as part of the kata-deploy
localbuild, as we need to pass the DESTDIR to where the agent will be
installed, and also whether we're building the agent with policy support
enabled or not.
Signed-off-by: Fabiano Fidêncio <fabiano.fidencio@intel.com>
Otherwise `make test` will fail with:
```
error[E0583]: file not found for module `version`
```
Fixes: #7974 -- part 0
Signed-off-by: Fabiano Fidêncio <fabiano.fidencio@intel.com>
- Allow `clippy::redundant-closure-call` in `from_cmdline`
which has issues with the guard function passed into
the `parse_cmdline_param` macro
Fixes: #7902
Signed-off-by: stevenhorsman <steven@uk.ibm.com>
1. Directly support CgroupManager::freeze through systemd API.
2. Avoid always passing unit_name by storing it into DBusClient.
3. Realize CgroupManager::destroy more accurately by killing systemd unit rather than stop it.
4. Ignore no such unit error when destroying systemd unit.
5. Update zbus version and corresponding interface file.
Acknowledgement: error handling for no such systemd unit error refers to
Fixes: #7080, #7142, #7143, #7166
Signed-off-by: Yuan-Zhuo <yuanzhuo0118@outlook.com>
Signed-off-by: Yohei Ueda <yohei@jp.ibm.com>
When a storage device is used by more than one container, the second
and forth instances will cause storage device reference count leakage,
thus cause storage device leakage. The reason is:
add_storages() will increase reference count of existing storage device,
but forget to add the device to the `mount_list` array, thus leak the
reference count.
Fixes: #7820
Signed-off-by: Jiang Liu <gerry@linux.alibaba.com>
