We've noticed this caused regressions with the k8s-oom tests, and then
decided to take a step back and do this in the same way it was done
before 67972ec48a.
Moreover, this step back is also more reasonable in terms of the
controlling logic.
And by doing this we can re-enable the k8s-oom.bats tests, which is done
as part of this PR.
Fixes: #7271
Depends-on: github.com/kata-containers/tests#5705
Signed-off-by: Yushuo <y-shuo@linux.alibaba.com>
Let's take the same approach of the go runtime, instead, and allocate
the maximum allowed number of vcpus instead.
Fixes: #7270
Signed-off-by: Fabiano Fidêncio <fabiano.fidencio@intel.com>
This reverts commit 25d2fb0fde.
The reason we're reverting the commit is because it to check whether
it's the cause for the regression on devmapper tests.
Fixes: #7253
Depends-on: github.com/kata-containers/tests#5705
Signed-off-by: Fabiano Fidêncio <fabiano.fidencio@intel.com>
Currently, even when using devmapper, if the VMM supports virtio-fs /
virtio-9p, that's used to share a few files between the host and the
guest.
This *needed*, as we need to share with the guest contents like secrets,
certificates, and configurations, via Kubernetes objects like configMaps
or secrets, and those are rotated and must be updated into the guest
whenever the rotation happens.
However, there are still use-cases users can live with just copying
those files into the guest at the pod creation time, and for those
there's absolutely no need to have a shared filesystem process running
with no extra obvious benefit, consuming memory and even increasing the
attack surface used by Kata Containers.
For the case mentioned above, we should allow users, making it very
clear which limitations it'll bring, to run Kata Containers with
devmapper without actually having to use a shared file system, which is
already the approach taken when using Firecracker as the VMM.
Fixes: #7207
Signed-off-by: Fabiano Fidêncio <fabiano.fidencio@intel.com>
There is nothing in them that requires them to be macros. Converting
them to functions allows for better error messages.
Fixes: #7201
Signed-off-by: Wedson Almeida Filho <walmeida@microsoft.com>
There is nothing in it that requires it to be a macro. Converting it to
a function allows for better error messages.
Fixes: #7201
Signed-off-by: Wedson Almeida Filho <walmeida@microsoft.com>
Having a function allows for better error messages from the type checker
and it makes it clearer to callers what can happen. For example:
is_allowed!(req);
Gives no indication that it may result in an early return, and no simple
way for callers to modify the behaviour. It also makes it look like
ownership of `req` is being transferred.
On the other hand,
is_allowed(&req)?;
Indicates that `req` is being borrowed (immutably) and may fail. The
question mark indicates that the caller wants an early return on
failure.
Fixes: #7201
Signed-off-by: Wedson Almeida Filho <walmeida@microsoft.com>
Since it is never modified, it doesn't really need a lock of any kind.
Removing the `RwLock` wrapper allows us to remove all `.read().await`
calls when accessing it.
Additionally, `AGENT_CONFIG` already has a static lifetime, so there is
no need to wrap it in a ref-counted heap allocation.
Fixes: #5409
Signed-off-by: Wedson Almeida Filho <walmeida@microsoft.com>
Vfio support introduce build error on AArch64. Remove arch related
annotation can avoid this error.
Fixes: #7187
Signed-off-by: Jianyong Wu <jianyong.wu@arm.com>
The failure mainly caused by the encoded volume path and
the mount/src. As the src will be validated with stat,but
it's not a full path and encoded, which causes the stat
mount source failed.
Fixes: #7186
Signed-off-by: alex.lyn <alex.lyn@antgroup.com>
Unlike the previous usage which requires creating
/dev/xxx by mknod on the host, the new approach will
fully utilize the DirectVolume-related usage method,
and pass the spdk controller to vmm.
And a user guide about using the spdk volume when run
a kata-containers. it can be found in docs/how-to.
Fixes: #6526
Signed-off-by: alex.lyn <alex.lyn@antgroup.com>
When running on a VM, the kernel parameter "unrestricted_guest" for
kernel module "kvm_intel" is not required. So, return success when running
on a VM without checking value of this kernel parameter.
Signed-off-by: Archana Shinde <archana.m.shinde@intel.com>
Implement functionality to add to the env output if the host is capable
of running a VM.
Fixes: #6727
Signed-off-by: Archana Shinde <archana.m.shinde@intel.com>
A new choice of using vfio devic based volume for kata-containers.
With the help of kata-ctl direct-volume, users are able to add a
specified device which is BDF or IOMMU group ID.
To help users to use it smoothly, A doc about howto added in
docs/how-to/how-to-run-kata-containers-with-kinds-of-Block-Volumes.
Fixes: #6525
Signed-off-by: alex.lyn <alex.lyn@antgroup.com>
Limitations:
As no ready rust vmm's vfio manager is ready, it only supports
part of vfio in runtime-rs. And the left part is to call vmm
interfaces related to vfio add/remove.
So when vmm/vfio manager ready, a new PR will be pushed to
narrow the gap.
Fixes: #6525
Signed-off-by: alex.lyn <alex.lyn@antgroup.com>
The `-o` option is the legacy way to configure virtiofsd, inherited
from the C implementation. The rust implementation honours it for
compatibility but it logs deprecation warnings.
Let's use the replacement options in the go shim code. Also drop
references to `-o` from the configuration TOML file.
Fixes#7111
Signed-off-by: Greg Kurz <groug@kaod.org>
The C implementation of virtiofsd had some kind of limited support
for remote POSIX locks that was causing some workflows to fail with
kata. Commit 432f9bea6e hard coded `-o no_posix_lock` in order
to enforce guest local POSIX locks and avoid the issues.
We've switched to the rust implementation of virtiofsd since then,
but it emits a warning about `-o` being deprecated.
According to https://gitlab.com/virtio-fs/virtiofsd/-/issues/53 :
The C implementation of the daemon has limited support for
remote POSIX locks, restricted exclusively to non-blocking
operations. We tried to implement the same level of
functionality in #2, but we finally decided against it because,
in practice most applications will fail if non-blocking
operations aren't supported.
Implementing support for non-blocking isn't trivial and will
probably require extending the kernel interface before we can
even start working on the daemon side.
There is thus no justification to pass `-o no_posix_lock` anymore.
Signed-off-by: Greg Kurz <groug@kaod.org>
The rust implementation of virtiofsd always runs foreground and
spits a deprecation warning when `-f` is passed.
Signed-off-by: Greg Kurz <groug@kaod.org>
If we override the cold, hot plug with an annotation
we need to reset the other plugging mechanism to NoPort
otherwise both will be enabled.
Signed-off-by: Zvonko Kaiser <zkaiser@nvidia.com>
In Virt the vhost-user-block is an PCIe device so
we need to make sure to consider it as well. We're keeping
track of vhost-user-block devices and deduce the correct
amount of PCIe root ports.
Signed-off-by: Zvonko Kaiser <zkaiser@nvidia.com>
Now it is possible to configure the PCIe topology via annotations
and addded a simple test, checking for Invalid and RootPort
Signed-off-by: Zvonko Kaiser <zkaiser@nvidia.com>
Removed the configuration of PCIeRootPort and PCIeSwitchPort, those
values can be deduced in createPCIeTopology
Signed-off-by: Zvonko Kaiser <zkaiser@nvidia.com>
Refactor the bus assignment so that the call to GetAllVFIODevicesFromIOMMUGroup
can be used by any module without affecting the topology.
Signed-off-by: Zvonko Kaiser <zkaiser@nvidia.com>
The hypervisor_state file was the wrong location for the PCIe Port
settings, moved everything under device umbrella, where it can be
consumed more easily and we do not get into circular deps.
Signed-off-by: Zvonko Kaiser <zkaiser@nvidia.com>
Kubernetes and Containerd will help calculate the Sandbox Size and pass it to
Kata Containers through annotations.
In order to accommodate this favorable change and be compatible with the past,
we have implemented the handling of the number of vCPUs in runtime-rs. This is
This is slightly different from the original runtime-go design.
This doc introduce how we handle vCPU size in runtime-rs.
Fixes: #5030
Signed-off-by: Yushuo <y-shuo@linux.alibaba.com>
Signed-off-by: Ji-Xinyou <jerryji0414@outlook.com>
In this commit, we refactored the logic of static resource management.
We defined the sandbox size calculated from PodSandbox's annotation and
SingleContainer's spec as initial size, which will always be the sandbox
size when booting the VM.
The configuration static_sandbox_resource_mgmt controls whether we will
modify the sandbox size in the following container operation.
Signed-off-by: Yushuo <y-shuo@linux.alibaba.com>
Signed-off-by: Ji-Xinyou <jerryji0414@outlook.com>
Some vmms, such as dragonball, will actively help us
perform online cpu operations when doing cpu hotplug.
Under the old onlineCpuMem interface, it is difficult
to adapt to this situation.
So we modify the semantics of nb_cpus in onlineCpuMemRequest.
In the original semantics, nb_cpus represents the number of
newly added CPUs that need to be online. The modified
semantics become that the number of online CPUs in the guest
needs to be guaranteed.
Fixes: #5030
Signed-off-by: Yushuo <y-shuo@linux.alibaba.com>
Signed-off-by: Ji-Xinyou <jerryji0414@outlook.com>
The declaration of the cpu number in the cpuset is greater
than the actual number of vcpus, which will cause an error when
updating the cgroup in the guest.
This problem is difficult to solve, so we temporarily clean up
the cpuset in the container spec before passing in the agent.
Fixes: #5030
Signed-off-by: Yushuo <y-shuo@linux.alibaba.com>
Signed-off-by: Ji-Xinyou <jerryji0414@outlook.com>
Updating vCPU resources and memory resources of the sandbox and
updating cgroups on the host will always happening together, and
they are all updated based on the linux resources declarations of
all the containers.
So we merge update_cgroups into the update_linux_resources, so we
can better manage the resources allocated to one pod in the host.
Fixes: #5030
Signed-off-by: Yushuo <y-shuo@linux.alibaba.com>
Signed-off-by: Ji-Xinyou <jerryji0414@outlook.com>
Support vcpu resizing on runtime side:
1. Calculate vcpu numbers in resource_manager using all the containers'
linux_resources in the spec.
2. Call the hypervisor(vmm) to do the vcpu resize.
3. Call the agent to online vcpus.
Fixes: #5030
Signed-off-by: Ji-Xinyou <jerryji0414@outlook.com>
Signed-off-by: Yushuo <y-shuo@linux.alibaba.com>
Nobody has volunteered to maintain the (currently broken) snap build, so
remove it.
Fixes: #6769.
Signed-off-by: James O. D. Hunt <james.o.hunt@intel.com>
