Users can specify extra arguments for virtiofsd in a pod spec using the
io.katacontainers.config.hypervisor.virtio_fs_extra_args annontation.
However, this annotation was ignored so far by the runtime. This commit
fixes the issue by processing the annotation value (if present) and
translating it to the corresponding hypervisor configuration item.
Fixes#1523
Signed-off-by: Pavel Mores <pmores@redhat.com>
Regenerate the client code with the added `go-fmt` step. No functional
changes.
Fixes: #1606
Signed-off-by: Bo Chen <chen.bo@intel.com>
(cherry picked from commit 0c38d9ecc4)
This patch extends the current process of generating client code for
cloud-hypervisor API with an additional step, `go-fmt`, which will remove
the generated `client/go.mod` file and format all auto-generated code.
Fixes: #1606
Signed-off-by: Bo Chen <chen.bo@intel.com>
(cherry picked from commit 52cacf8838)
Highlights for cloud-hypervisor version 0.14.0 include: 1) Structured
event monitoring; 2) MSHV improvements; 3) Improved aarch64 platform; 4)
Updated hotplug documentation; 6) PTY control for serial and
virtio-console; 7) Block device rate limiting; 8) Plan to deprecate the
support of "LinuxBoot" protocol and support PVH protocol only.
Highlights for cloud-hypervisor version 0.13.0 include: 1) Wider VFIO
device support; 2) Improve huge page support; 3) MACvTAP support; 4) VHD
disk image support; 5) Improved Virtio device threading; 6) Clean
shutdown support via synthetic power button.
Details can be found:
https://github.com/cloud-hypervisor/cloud-hypervisor/releases
Note: The client code of cloud-hypervisor's OpenAPI is automatically
generated by `openapi-generator` [1-2]. As the API changes do not
impact usages in Kata, no additional changes in kata's runtime are
needed to work with the latest version of cloud-hypervisor.
[1] https://github.com/OpenAPITools/openapi-generator
[2] https://github.com/kata-containers/kata-containers/blob/main/src/runtime/virtcontainers/pkg/cloud-hypervisor/README.mdFixes: #1591
Signed-off-by: Bo Chen <chen.bo@intel.com>
(cherry picked from commit 84b62dc3b1)
A wrong path was being used for container directory when
virtiofs is utilized. This resulted in a warning message in
logs when a container is killed, or completes:
level=warning msg="Could not remove container share dir"
Without proper removal, they'd later be cleaned up when the shared
path is removed as part of stopping the sandbox.
Fixes: #1559
Signed-off-by: Eric Ernst <eric.g.ernst@gmail.com>
Right now we rely heavily on mount propagation to share host
files/directories to the guest. However, because virtiofsd
pivots and moves itself to a separate mount namespace, the remount
mount is not present in virtiofsd's mount. And it causes guest to be
able to write to the host RO volume.
To fix it, create a private RO mount and then move it to the host mounts
dir so that it will be present readonly in the host-guest shared dir.
Signed-off-by: Peng Tao <bergwolf@hyper.sh>
bindmount remount events are not propagated through mount subtrees,
so we have to remount the shared dir mountpoint directly.
E.g.,
```
mkdir -p source dest foo source/foo
mount -o bind --make-shared source dest
mount -o bind foo source/foo
echo bind mount rw
mount | grep foo
echo remount ro
mount -o remount,bind,ro source/foo
mount | grep foo
```
would result in:
```
bind mount rw
/dev/xvda1 on /home/ubuntu/source/foo type ext4 (rw,relatime,discard,data=ordered)
/dev/xvda1 on /home/ubuntu/dest/foo type ext4 (rw,relatime,discard,data=ordered)
remount ro
/dev/xvda1 on /home/ubuntu/source/foo type ext4 (ro,relatime,discard,data=ordered)
/dev/xvda1 on /home/ubuntu/dest/foo type ext4 (rw,relatime,discard,data=ordered)
```
The reason is that bind mount creats new mount structs and attaches them to different mount subtrees.
However, MS_REMOUNT only looks for existing mount structs to modify and does not try to propagate the
change to mount structs in other subtrees.
Fixes: #1061
Signed-off-by: Peng Tao <bergwolf@hyper.sh>
On some setups, starting multiple kata pods (qemu) simultaneously on the same node
might cause kata VMs booting time to increase and the pods to fail with:
Failed to check if grpc server is working: rpc error: code = DeadlineExceeded desc = timed
out connecting to vsock 1358662990:1024: unknown
Increasing default dialing timeout to 30s should cover most cases.
Signed-off-by: Snir Sheriber <ssheribe@redhat.com>
Fixes: #1543
(backport https://github.com/kata-containers/kata-containers/pull/1544)
Since the kata's hypervisor process is in the network namespace,
which is close to container's process, and some host metrics
such as cadvisor can use this pid to access the network namespace
to get some network metrics. Thus this commit replace the shim's
pid with the hypervisor's pid.
Fixes: #1451
Signed-off-by: fupan.lfp <fupan.lfp@antfin.com>
(backport https://github.com/kata-containers/kata-containers/pull/1452)
Signed-off-by: Francesco Giudici <fgiudici@redhat.com>
Highlights for cloud-hypervisor version v0.12.0 include: removal of
`vhost-user-net` and `vhost-user-block` self spawning, migration of
`vhost-user-fs` backend, ARM64 enhancements with full support of
`--watchdog` for rebooting, and enhanced `info` HTTP API to include the
details of devices used by the VM including VFIO devices.
Fixes: #1315
Signed-off-by: Bo Chen <chen.bo@intel.com>
Today we only clear out the cpuset details when doing an update call on
existing container/pods. This works in the case of Kubernetes, but not
in the case where we are explicitly setting the cpuset details at boot
time. For example, if you are running a single container via docker ala:
docker run --cpuset-cpus 0-3 -it alpine sh
What would happen is the cpuset info would be passed in with the
container spec for create container request to the agent. At that point
in time, there'd only be the defualt number of CPUs available in the
guest (1), so you'd be left with cpusets set to 0. Next, we'd hotplug
the vCPUs, providing 0-4 CPUs in the guest, but the cpuset would never
be updated, leaving the application tied to CPU 0.
Ouch.
Until the day we support cpusets in the guest, let's make sure that we
start off clearing the cpuset fields.
Fixes: #1405
Signed-off-by: Eric Ernst <eric.g.ernst@gmail.com>
Update cloud-hypervisor to commit 2706319.
Fixes a limitation in OpenAPITools/openapi-generator tool,
it's impossible to send go zero types, like false and 0 to
cloud-hypervisor because `omitempty` is added if a field is not
required.
See cloud-hypervisor/cloud-hypervisor#1961 for more information
Signed-off-by: Julio Montes <julio.montes@intel.com>
Guest consumes 120Mb more of memory when DAX is enabled and the default
FS cache size (8G) is used. Disable dax when it is not required
reducing guest's memory footprint.
Without this patch:
```
7fdea4000000-7fdee4000000 rw-s 18850589 /memfd:ch_ram (deleted)
Size: 1048576 kB
KernelPageSize: 4 kB
MMUPageSize: 4 kB
Rss: 187876 kB
```
With this patch:
```
7fa970000000-7fa9b0000000 rw-s 612001 /memfd:ch_ram (deleted)
Size: 1048576 kB
KernelPageSize: 4 kB
MMUPageSize: 4 kB
Rss: 57308 kB
Pss: 56722 kB
```
fixes#1100
Signed-off-by: Julio Montes <julio.montes@intel.com>
The release v0.11.0 of cloud-hypervisor features the following changes:
1) Improved Linux Boot Time, 2) `SIGTERM/SIGINT` Interrupt Signal,
Handling 3) Default Log Level Changed, 4) `io_uring` support by default
for `virtio-block` (on host kernel version 5.8+), 5) Windows Guest
Support, 6) New `--balloon` Parameter Added, 7) Experimental
`virtio-watchdog` Support, 8) Bug fixes.
Fixes: #1089
Signed-off-by: Bo Chen <chen.bo@intel.com>
On pod delete, we were looking to read files that we had just deleted. In particular,
stopSandbox for QEMU was called (we cleanup up vmpath), and then QEMU's
save function was called, which immediately checks for the PID file.
Let's only update the persist store for QEMU if QEMU is actually
running. This'll avoid Error messages being displayed when we are
stopping and deleting a sandbox:
```
level=error msg="Could not read qemu pid file"
```
I reviewed CLH, and it looks like it is already taking appropriate
action, so no changes needed.
Ideally we won't spend much time saving state to persist.json unless
there's an actual error during stop/delete/shutdown path, as the persist will
also be removed after the pod is removed. We may want to optimize this,
as currently we are doing a persist store when deleting each container
(after the sandbox is stopped, VM is killed), and when we stop the sandbox.
This'll require more rework... tracked in:
https://github.com/kata-containers/kata-containers/issues/1181Fixes: #1179
Signed-off-by: Eric Ernst <eric.g.ernst@gmail.com>
If the upcast from resultingRoutes to *grpc.IRoutes fails, we return
(nil, err), but previous code ensures that err is nil at that point, so we
return no error.
fixes#1206
Forward port of
0ffaeeb5d8
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
If the upcast from resultingInterfaces to *grpc.Interfaces fails, we
return (nil, err), but previous code ensures that err is nil at that
point, so we return no error.
Forward port of
b86e904c2dfixes#1206
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Add the verification of some basic protections, namely that:
- EnableAnnotations is honored
- Dangerous paths cannot be modified if no match
- Errors are returned when expected
Fixes: #901
Signed-off-by: Christophe de Dinechin <dinechin@redhat.com>
Warning from gocyclo during make check:
virtcontainers/pkg/oci/utils.go:404:1: cyclomatic complexity 37 of func `addHypervisorConfigOverrides` is high (> 30) (gocyclo)
func addHypervisorConfigOverrides(ocispec specs.Spec, config *vc.SandboxConfig, runtime RuntimeConfig) error {
^
Fixes: #901
Signed-off-by: Christophe de Dinechin <dinechin@redhat.com>
There are a few interesting corner cases to consider for this
function.
Fixes: #901
Suggested-by: James O.D. Hunt <james.o.hunt@intel.com>
Signed-off-by: Christophe de Dinechin <dinechin@redhat.com>
James O.D Hunt: "But also, regexpContains() and
checkPathIsInGlobList() seem like good candidates for some unit
tests. The "look" obvious, but a few boundary condition tests would be
useful I think (filenames with spaces, backslashes, special
characters, and relative & absolute paths are also an interesting
thought here)."
There aren't that many boundary conditions on a list with regexps,
if you assume the regexp match function itself works. However, the
tests is useful in documenting expectations.
Fixes: #901
Suggested-by: James O.D. Hunt <james.o.hunt@intel.com>
Signed-off-by: Christophe de Dinechin <dinechin@redhat.com>
Add a field "enable_annotations" to the runtime configuration that can
be used to whitelist annotations using a list of regular expressions,
which are used to match any part of the base annotation name, i.e. the
part after "io.katacontainers.config.hypervisor."
For example, the following configuraiton will match "virtio_fs_daemon",
"initrd" and "jailer_path", but not "path" nor "firmware":
enable_annotations = [ "virtio.*", "initrd", "_path" ]
The default is an empty list of enabled annotations, which disables
annotations entirely.
If an anontation is rejected, the message is something like:
annotation io.katacontainers.config.hypervisor.virtio_fs_daemon is not enabled
Fixes: #901
Suggested-by: Peng Tao <tao.peng@linux.alibaba.com>
Signed-off-by: Christophe de Dinechin <dinechin@redhat.com>
When filtering annotations that correspond to paths,
e.g. hypervisor.path, it is better to use a glob syntax than a regexp
syntax, as it is more usual for paths, and prevents classes of matches
that are undesirable in our case, such as matching .. against .*
Fixes: #901
Signed-off-by: Christophe de Dinechin <dinechin@redhat.com>
A comment talking about runtime related annotations describes them as
being related to the agent. A similar comment for the agent
annotations is missing.
Fixes: #901
Signed-off-by: Christophe de Dinechin <dinechin@redhat.com>
This one could theoretically be used to overwrite data on the host.
It seems somewhat less risky than the earlier ones for a number
of reasons, but worth protecting a little anyway.
Fixes: #901
Signed-off-by: Christophe de Dinechin <dinechin@redhat.com>
This also adds annotation for ctlpath which were not present
before. It's better to implement the code consistenly right now to make
sure that we don't end up with a leaky implementation tacked on later.
Fixes: #901
Signed-off-by: Christophe de Dinechin <dinechin@redhat.com>
The jailer_path annotation can be used to execute arbitrary code on
the host. Add a jailer_path_list configuration entry providing a list
of regular expressions that can be used to filter annotations that
represent valid file names.
Fixes: #901
Signed-off-by: Christophe de Dinechin <dinechin@redhat.com>
The annotation is provided, so it should be respected.
Furthermore, it is important to implement it with the appropriate
protetions similar to what was done for virtiofsd.
Fixes: #901
Signed-off-by: Christophe de Dinechin <dinechin@redhat.com>
Sending the virtio_fs_daemon annotation can be used to execute
arbitrary code on the host. In order to prevent this, restrict the
values of the annotation to a list provided by the configuration
file.
Fixes: #901
Signed-off-by: Christophe de Dinechin <dinechin@redhat.com>
Pulled from 1.18.4 Kubernetes, adding the cpuset pkg for managing
CPUSet calculations on the host. Go mod'ing the original code from
k8s.io/kubernetes was very painful, and this is very static, so let's
just pull in what we need.
Signed-off-by: Eric Ernst <eric.g.ernst@gmail.com>
Kata doesn't map any numa topologies in the guest. Let's make sure we
clear the Cpuset fields before passing container updates to the
guest.
Note, in the future we may want to have a vCPU to guest CPU mapping and
still include the cpuset.Cpus. Until we have this support, clear this as
well.
Fixes: #932
Signed-off-by: Eric Ernst <eric.g.ernst@gmail.com>
CPUSet cgroup allows for pinning the memory associated with a cpuset to
a given numa node. Similar to cpuset.cpus, we should take cpuset.mems
into account for the sandbox-cgroup that Kata creates.
Signed-off-by: Eric Ernst <eric.g.ernst@gmail.com>
