It should do nothing instead of return an error when
hot-unplug the memory to the size smaller than static
plugged memory size.
Signed-off-by: Fupan Li <fupan.lfp@antgroup.com>
Since the network device hotplug is an asynchronous operation,
it's possible that the hotplug operation had returned, but
the network device hasn't ready in guest, thus it's better to
retry on this operation to wait until the device ready in guest.
Signed-off-by: Fupan Li <fupan.lfp@antgroup.com>
With the change made to the matrix when the CC GPU runner was added,
there was a change in the job name (@sprt saw that coming, but I
didn't).
Signed-off-by: Fabiano Fidêncio <ffidencio@nvidia.com>
Same deal as the previous commut, just enabling the tests here, with the
same list of improvements that we will need to go through in order to
get is working in a perfect way.
Signed-off-by: Manuel Huber <manuelh@nvidia.com>
Signed-off-by: Fabiano Fidêncio <ffidencio@nvidia.com>
While the primary goal of this change is to detect regressions to the
NVIDIA SNP GPU scenario, various improvements to reflect a more
realistic CC setting are planned in subsequent changes, such as:
* moving away from the overlayfs snapshotter
* disabling filesystem sharing
* applying a pod security policy
* activating the GPUs only after attestation
* using a refined approach for GPU cold-plugging without requiring
annotations
* revisiting pod timeout and overhead parameters (the podOverhead value
was increased due to CUDA vectorAdd requiring about 6Gi of
podOverhead, as well as the inference and embedqa requiring at least
12Gi, respectively, 14Gi of podOverhead to run without invoking the
host's oom-killer. We will revisit this aspect after addressing
points 1. and 2.)
Signed-off-by: Manuel Huber <manuelh@nvidia.com>
Signed-off-by: Fabiano Fidêncio <ffidencio@nvidia.com>
For the nvidia-gpu-snp and nvidia-gpu-tdx we must set containerd to
allow the CDI annotation to be passed to down.
This solution may become obsolete soon enough, but the cleanest way to
have it properly working is by adding it here (even if we remove it
before the next release).
Signed-off-by: Manuel Huber <manuelh@nvidia.com>
Signed-off-by: Fabiano Fidêncio <ffidencio@nvidia.com>
It's been noticed that as more RAM is needed to run the CC tests, we
also need to update the podOverhead of the NVIDIA CC runtime classes to
avoid getting OOM Killed.
Signed-off-by: Manuel Huber <manuelh@nvidia.com>
Signed-off-by: Fabiano Fidêncio <ffidencio@nvidia.com>
Since the nerdctl's network hook would call pselect6 syscall
by xtables-nft-multi, thus we'd better add it to the seccomp's
whitelist.
Signed-off-by: Fupan Li <fupan.lfp@antgroup.com>
Let's add a new NVIDIA machine, which later on will be used for CC
related tests.
For now the current tests are skipped in the CC capable machine.
Signed-off-by: Manuel Huber <manuelh@nvidia.com>
Signed-off-by: Fabiano Fidêncio <ffidencio@nvidia.com>
Let's now make sure that we don't add duplicated values to any of our
entries, making the script as sane as possible for sequential runs.
Vibed with Cursor's help!
Signed-off-by: Fabiano Fidêncio <ffidencio@nvidia.com>
Let's add some helper functions, not yet used, to avoid adding
duplicated items.
This idea is an expansion of Choi's idea to avoid setting duplicated
items, and it'll help on making the whole script idempotent on
sequential runs.
Vibed with Cursor's help!
Signed-off-by: Fabiano Fidêncio <ffidencio@nvidia.com>
I know, this is not simplifying much things for now, but it has a good
intent in the background and will serve as base for making the
kata-deploy helm chart more user friendly.
With that said, let's add ALLOWED_HYPERVISOR_ANNOTATIONS per arch, while
adding support to set something like "qemu:foo,bar clh:bar foobar
barfoo". Why? Because in the future we'll have a better way to set this
per shim (and the shim is per arch ...).
More details of what we'll do in the future are being discussed here:
https://github.com/kata-containers/kata-containers/issues/12024
Anyways, the variables are **DELIBERATELY** not exposed to the chart for
now, as those will be later on when addressing the issue mentioned
above.
Signed-off-by: Fabiano Fidêncio <ffidencio@nvidia.com>
When the runtimeClasses were added, as part of 7cfa826804, the
firecracker runtimeClass ended up missing from the dictionary.
Signed-off-by: Fabiano Fidêncio <ffidencio@nvidia.com>
The Firecracker installation docs had an outaded containerd configuration for the devmapper plugin.
This commit updates the instructions so that they are compatible with more recent versions of containerd.
Signed-off-by: Anton Ippolitov <anton.ippolitov@datadoghq.com>
When added, I've mistakenly used the wrong test-type name, which is now
fixed and should be enough to trigger the tests correctly.
Signed-off-by: Fabiano Fidêncio <ffidencio@nvidia.com>
On IBM actionspz P/Z runners, the following error was observed during
runtime tests:
```
host system doesn't support vsock: stat /dev/vhost-vsock: no such file or directory
```
Since loading the vsock module on the fly is not permitted, this commit
moves the runtime tests back to self-hosted runners for P/Z.
Signed-off-by: Hyounggyu Choi <Hyounggyu.Choi@ibm.com>
On IBM actionspz Z runners, the following error occurs when running
`modprobe`:
```
modprobe: FATAL: Module bridge not found in directory /lib/modules/6.8.0-85-generic
```
Additionally, there are no files under `/lib/modules`, for example:
```
total 0
drwxr-xr-x 1 root root 0 Aug 5 13:09 .
drwxr-xr-x 1 root root 2.0K Oct 1 22:59 ..
```
This commit skips the `test_load_kernel_module` test if the module is
not found or if running `modprobe` is not permitted.
Signed-off-by: Hyounggyu Choi <Hyounggyu.Choi@ibm.com>
On IBM actionspz Z runners, write operations on network interfaces
are not allowed, even for the root user.
This commit skips the `add_update_addresses` test if the operation
fails with EACCES (-13, permission denied).
Signed-off-by: Hyounggyu Choi <Hyounggyu.Choi@ibm.com>
On IBM actionspz Z runners, the ioctl system call is not allowed even
for the root user. There is likely an additional security mechanism
(such as AppArmor or seccomp) in place on Ubuntu runners.
This commit introduces a new helper, `is_permission_error()`,
which skips the test if ioctl operations in `reseed_rng()` are not
permitted.
Signed-off-by: Hyounggyu Choi <Hyounggyu.Choi@ibm.com>
The IBM actionspz Z runners mount /dev as tmpfs, while other systems
use devtmpfs. This difference causes an assertion failure for
test_already_baremounted.
This commit sets the detected filesystem for bare-mounted points
as the expected value.
Signed-off-by: Hyounggyu Choi <Hyounggyu.Choi@ibm.com>
The root filesystem for IBM actionspz Z runners is `btrfs` instead of `ext4`.
The error message differs when an unprivileged user tries to perform a bind mount.
This commit adjusts the handling of error messages based on the detected root
filesystem type.
Signed-off-by: Hyounggyu Choi <Hyounggyu.Choi@ibm.com>
Since the qemu & cloud-hypervisor support the cpu & memory
hotplug now, thus disable the static resource management
for qemu and cloud-hypervisor by default.
Signed-off-by: Fupan Li <fupan.lfp@antgroup.com>
Since qemu-coco-dev-runtime-rs and qemu-coco-dev had disabled the
cpu&memory hotplug by enable static_sandbox_resource_mgmt, thus
we should disable the cpu hotplug test for those two runtime.
Signed-off-by: Fupan Li <fupan.lfp@antgroup.com>
Since the qemu, cloud-hypervisor and dragonball had supported the
cpu hotplug on runtime-rs, thus enable the cpu hotplug test in CI.
Signed-off-by: Fupan Li <fupan.lfp@antgroup.com>
Pod annotations from the outer runtime are being used for cold-plugging
CDI devices. We need to ensure that these annotations don't leak into
the inner runtime for which specific container (sibling) annotations
are being created. Without this change, the inner runtime receives both
annotations, leading to failing CDI injection as an outer runtime
annotation observed in the guest translates to an unresolvable CDI
device, for example, cdi.k8s.io/gpu: "nvidia.com/pgpu=0".
Signed-off-by: Manuel Huber <manuelh@nvidia.com>
Stratovirt has been failing for a considerable amount of time, with no
sign of someone watching it and being actively working on a fix.
With this we also stop building and shipping stratovirt as part of our
release as we cannot test it.
Signed-off-by: Fabiano Fidêncio <ffidencio@nvidia.com>
A few weeks ago we've tested nydus-snapshotter with this approach, and
we DID find issues with it.
Now, let's also test this with `experimental_force_guest_pull`.
Signed-off-by: Fabiano Fidêncio <ffidencio@nvidia.com>
It's just a follow-up on the previous commit where we move away from the
runtimeClass creation inside the script, and instead we do it using the
chart itself.
Signed-off-by: Fabiano Fidêncio <ffidencio@nvidia.com>
This reverts commit be05e1370c, which is
not a problem as we never released such option.
Conflicts:
tools/packaging/kata-deploy/helm-chart/README.md
Signed-off-by: Fabiano Fidêncio <ffidencio@nvidia.com>
We had this logic inside the script when we didn't use the helm chart.
However, this only makes the shim script more convoluted for no reason.
Signed-off-by: Fabiano Fidêncio <ffidencio@nvidia.com>
In order to fix:
```
=== Running govulncheck on containerd-shim-kata-v2 ===
Vulnerabilities found in containerd-shim-kata-v2:
=== Symbol Results ===
Vulnerability #1: GO-2025-4015
Excessive CPU consumption in Reader.ReadResponse in net/textproto
More info: https://pkg.go.dev/vuln/GO-2025-4015
Standard library
Found in: net/textproto@go1.24.6
Fixed in: net/textproto@go1.24.8
Vulnerable symbols found:
#1: textproto.Reader.ReadResponse
Vulnerability #2: GO-2025-4014
Unbounded allocation when parsing GNU sparse map in archive/tar
More info: https://pkg.go.dev/vuln/GO-2025-4014
Standard library
Found in: archive/tar@go1.24.6
Fixed in: archive/tar@go1.24.8
Vulnerable symbols found:
#1: tar.Reader.Next
Vulnerability #3: GO-2025-4013
Panic when validating certificates with DSA public keys in crypto/x509
More info: https://pkg.go.dev/vuln/GO-2025-4013
Standard library
Found in: crypto/x509@go1.24.6
Fixed in: crypto/x509@go1.24.8
Vulnerable symbols found:
#1: x509.Certificate.Verify
#2: x509.Certificate.Verify
Vulnerability #4: GO-2025-4012
Lack of limit when parsing cookies can cause memory exhaustion in net/http
More info: https://pkg.go.dev/vuln/GO-2025-4012
Standard library
Found in: net/http@go1.24.6
Fixed in: net/http@go1.24.8
Vulnerable symbols found:
#1: http.Client.Do
#2: http.Client.Get
#3: http.Client.Head
#4: http.Client.Post
#5: http.Client.PostForm
Use '-show traces' to see the other 9 found symbols
Vulnerability #5: GO-2025-4011
Parsing DER payload can cause memory exhaustion in encoding/asn1
More info: https://pkg.go.dev/vuln/GO-2025-4011
Standard library
Found in: encoding/asn1@go1.24.6
Fixed in: encoding/asn1@go1.24.8
Vulnerable symbols found:
#1: asn1.Unmarshal
#2: asn1.UnmarshalWithParams
Vulnerability #6: GO-2025-4010
Insufficient validation of bracketed IPv6 hostnames in net/url
More info: https://pkg.go.dev/vuln/GO-2025-4010
Standard library
Found in: net/url@go1.24.6
Fixed in: net/url@go1.24.8
Vulnerable symbols found:
#1: url.JoinPath
#2: url.Parse
#3: url.ParseRequestURI
#4: url.URL.Parse
#5: url.URL.UnmarshalBinary
Vulnerability #7: GO-2025-4009
Quadratic complexity when parsing some invalid inputs in encoding/pem
More info: https://pkg.go.dev/vuln/GO-2025-4009
Standard library
Found in: encoding/pem@go1.24.6
Fixed in: encoding/pem@go1.24.8
Vulnerable symbols found:
#1: pem.Decode
Vulnerability #8: GO-2025-4008
ALPN negotiation error contains attacker controlled information in
crypto/tls
More info: https://pkg.go.dev/vuln/GO-2025-4008
Standard library
Found in: crypto/tls@go1.24.6
Fixed in: crypto/tls@go1.24.8
Vulnerable symbols found:
#1: tls.Conn.Handshake
#2: tls.Conn.HandshakeContext
#3: tls.Conn.Read
#4: tls.Conn.Write
#5: tls.Dial
Use '-show traces' to see the other 4 found symbols
Vulnerability #9: GO-2025-4007
Quadratic complexity when checking name constraints in crypto/x509
More info: https://pkg.go.dev/vuln/GO-2025-4007
Standard library
Found in: crypto/x509@go1.24.6
Fixed in: crypto/x509@go1.24.9
Vulnerable symbols found:
#1: x509.CertPool.AppendCertsFromPEM
#2: x509.Certificate.CheckCRLSignature
#3: x509.Certificate.CheckSignature
#4: x509.Certificate.CheckSignatureFrom
#5: x509.Certificate.CreateCRL
Use '-show traces' to see the other 27 found symbols
Vulnerability #10: GO-2025-4006
Excessive CPU consumption in ParseAddress in net/mail
More info: https://pkg.go.dev/vuln/GO-2025-4006
Standard library
Found in: net/mail@go1.24.6
Fixed in: net/mail@go1.24.8
Vulnerable symbols found:
#1: mail.AddressParser.Parse
#2: mail.AddressParser.ParseList
#3: mail.Header.AddressList
#4: mail.ParseAddress
#5: mail.ParseAddressList
```
Signed-off-by: Fabiano Fidêncio <ffidencio@nvidia.com>
During the ${wait_time} for an expected condition, if
CreateContainerRequest was NOT expected to fail: detect possible
CreateContainerRequest failures early and abort the wait.
For example, before this change:
not ok 1 Successful replication controller with auto-generated policy in 123335ms
ok 2 Policy failure: unexpected container command in 14601ms
ok 3 Policy failure: unexpected volume mountPath in 14443ms
ok 4 Policy failure: unexpected host device mapping in 14515ms
ok 5 Policy failure: unexpected securityContext.allowPrivilegeEscalation in 14485ms
ok 6 Policy failure: unexpected capability in 14382ms
ok 7 Policy failure: unexpected UID = 1000 in 14578ms
After this change:
not ok 1 Successful replication controller with auto-generated policy in 17108ms
ok 2 Policy failure: unexpected container command in 14427ms
ok 3 Policy failure: unexpected volume mountPath in 14636ms
ok 4 Policy failure: unexpected host device mapping in 14493ms
ok 5 Policy failure: unexpected securityContext.allowPrivilegeEscalation in 14554ms
ok 6 Policy failure: unexpected capability in 15087ms
ok 7 Policy failure: unexpected UID = 1000 in 14371ms
Signed-off-by: Dan Mihai <dmihai@microsoft.com>
During the ${wait_time} for an expected condition, if
CreateContainerRequest was NOT expected to fail: detect possible
CreateContainerRequest failures early and abort the wait.
For example, before this change:
not ok 1 Successful pod with auto-generated policy in 94852ms
ok 2 Policy failure: unexpected device mount in 17807ms
After this change:
not ok 1 Successful pod with auto-generated policy in 35194ms
ok 2 Policy failure: unexpected device mount in 21355ms
Signed-off-by: Dan Mihai <dmihai@microsoft.com>
During the ${wait_time} for an expected condition, if
CreateContainerRequest was NOT expected to fail: detect possible
CreateContainerRequest failures early and abort the wait.
For example, before this change:
not ok 1 Logs empty when ReadStreamRequest is blocked in 102257ms
After this change:
not ok 1 Logs empty when ReadStreamRequest is blocked in 17339ms
Signed-off-by: Dan Mihai <dmihai@microsoft.com>
During the ${wait_time} for an expected condition, if
CreateContainerRequest was NOT expected to fail: detect possible
CreateContainerRequest failures early and abort the wait.
For example, before this change:
not ok 1 Successful job with auto-generated policy in 107111ms
ok 2 Policy failure: unexpected environment variable in 7920ms
ok 3 Policy failure: unexpected command line argument in 7874ms
ok 4 Policy failure: unexpected emptyDir volume in 7823ms
ok 5 Policy failure: unexpected projected volume in 7812ms
ok 6 Policy failure: unexpected readOnlyRootFilesystem in 7903ms
ok 7 Policy failure: unexpected UID = 222 in 7720ms
After this change:
not ok 1 Successful job with auto-generated policy in 10271ms
ok 2 Policy failure: unexpected environment variable in 8018ms
ok 3 Policy failure: unexpected command line argument in 7886ms
ok 4 Policy failure: unexpected emptyDir volume in 7621ms
ok 5 Policy failure: unexpected projected volume in 7843ms
ok 6 Policy failure: unexpected readOnlyRootFilesystem in 7632ms
ok 7 Policy failure: unexpected UID = 222 in 7619ms
Signed-off-by: Dan Mihai <dmihai@microsoft.com>
During the ${wait_time} for an expected condition, if
CreateContainerRequest was NOT expected to fail: detect possible
CreateContainerRequest failures early and abort the wait.
For example, before this change:
ok 1 Successful sc deployment with auto-generated policy and container image volumes in 14769ms
ok 2 Successful sc with fsGroup/supplementalGroup deployment with auto-generated policy and container image volumes in 8384ms
not ok 3 Successful sc deployment with security context choosing another valid user in 136149ms
ok 4 Successful layered sc deployment with auto-generated policy and container image volumes in 8862ms
ok 5 Policy failure: unexpected GID = 0 for layered securityContext deployment in 7941ms
ok 6 Policy failure: malicious root group added via supplementalGroups deployment in 11612ms
After:
ok 1 Successful sc deployment with auto-generated policy and container image volumes in 15230ms
ok 2 Successful sc with fsGroup/supplementalGroup deployment with auto-generated policy and container image volumes in 9364ms
not ok 3 Successful sc deployment with security context choosing another valid user in 11060ms
ok 4 Successful layered sc deployment with auto-generated policy and container image volumes in 9124ms
ok 5 Policy failure: unexpected GID = 0 for layered securityContext deployment in 7919ms
ok 6 Policy failure: malicious root group added via supplementalGroups deployment in 11666ms
Signed-off-by: Dan Mihai <dmihai@microsoft.com>
During the ${wait_time} for an expected condition, if
CreateContainerRequest was NOT expected to fail: detect possible
CreateContainerRequest failures early and abort the wait.
For example, before this change:
not ok 1 Successful pod with auto-generated policy in 110801ms
not ok 2 Able to read env variables sourced from configmap using envFrom in 94104ms
not ok 3 Successful pod with auto-generated policy and runtimeClassName filter in 95838ms
not ok 4 Successful pod with auto-generated policy and custom layers cache path in 110712ms
ok 5 Policy failure: unexpected container image in 8113ms
ok 6 Policy failure: unexpected privileged security context in 7943ms
ok 7 Policy failure: unexpected terminationMessagePath in 11530ms
ok 8 Policy failure: unexpected hostPath volume mount in 7970ms
ok 9 Policy failure: unexpected config map in 7933ms
not ok 10 Policy failure: unexpected lifecycle.postStart.exec.command in 112677ms
ok 11 RuntimeClassName filter: no policy in 2302ms
not ok 12 ExecProcessRequest tests in 93946ms
not ok 13 Successful pod: runAsUser having the same value as the UID from the container image in 94003ms
ok 14 Policy failure: unexpected UID = 0 in 8016ms
ok 15 Policy failure: unexpected UID = 1234 in 7850ms
After:
not ok 1 Successful pod with auto-generated policy in 12182ms
not ok 2 Able to read env variables sourced from configmap using envFrom in 10121ms
not ok 3 Successful pod with auto-generated policy and runtimeClassName filter in 11738ms
not ok 4 Successful pod with auto-generated policy and custom layers cache path in 26592ms
ok 5 Policy failure: unexpected container image in 7742ms
ok 6 Policy failure: unexpected privileged security context in 7949ms
ok 7 Policy failure: unexpected terminationMessagePath in 7789ms
ok 8 Policy failure: unexpected hostPath volume mount in 7887ms
ok 9 Policy failure: unexpected config map in 7818ms
not ok 10 Policy failure: unexpected lifecycle.postStart.exec.command in 9120ms
ok 11 RuntimeClassName filter: no policy in 2081ms
not ok 12 ExecProcessRequest tests in 9883ms
not ok 13 Successful pod: runAsUser having the same value as the UID from the container image in 9870ms
ok 14 Policy failure: unexpected UID = 0 in 11161ms
ok 15 Policy failure: unexpected UID = 1234 in 7814ms
Signed-off-by: Dan Mihai <dmihai@microsoft.com>
We've seen a few cases where we fail the test due to timeout and when we
print the pods we just see that they've been created.
With that in mind, let's just increase the timeout a little bit.
Example:
```
not ok 1 Parallel jobs in 6250ms
(in test file k8s-parallel.bats, line 41)
`kubectl wait --for=condition=Ready --timeout=$timeout pod -l jobgroup=${job_name}' failed
No resources found in kata-containers-k8s-tests namespace.
[bats-exec-test:71] INFO: k8s configured to use runtimeclass
job.batch/process-item-test1 created
job.batch/process-item-test2 created
job.batch/process-item-test3 created
NAME STATUS COMPLETIONS DURATION AGE
process-item-test1 Running 0/1 0s
process-item-test2 Running 0/1 0s
process-item-test3 Running 0/1 0s
error: no matching resources found
No resources found in kata-containers-k8s-tests namespace.
No resources found in kata-containers-k8s-tests namespace.
DEBUG: system logs of node 'aks-nodepool1-25989463-vmss000000' since test start time (2025-11-01 16:39:03)
-- No entries --
job.batch "process-item-test1" deleted
job.batch "process-item-test2" deleted
job.batch "process-item-test3" deleted
```
Signed-off-by: Fabiano Fidêncio <ffidencio@nvidia.com>
Otherwise we'll face issues like:
```
Error: found in Chart.yaml, but missing in charts/ directory: node-feature-discovery
```
Signed-off-by: Fabiano Fidêncio <ffidencio@nvidia.com>
As we're failing on the uninstall, which seems related to a bug on NFD
itself, but I don't have access to a s390x machine to debug, let's skip
the enablement for now and enable it back once we've experimented it
better on s390x.
Signed-off-by: Fabiano Fidêncio <ffidencio@nvidia.com>
As we're failing to install NFD on CBL Mariner, let's skip the
enablement there, and enable it once we've experimented it better there.
Signed-off-by: Fabiano Fidêncio <ffidencio@nvidia.com>
As we have the ability to deploy NFD as a sub-chart of our chart, let's
make sure we test it during our CI.
We had to increase the timeout values, where we had timeouts set, to
deploy / undeploy kata, as now NFD is also deployed / undeployed.
Signed-off-by: Fabiano Fidêncio <ffidencio@nvidia.com>
Let's ensure that we add NFD as a weak dependency of the kata-deploy
helm chart.
What we're doing for now is leaving it up to the user / admin to enable
it, and if enabled then we do a explicit check for virtualization
support (x86_64 only for now).
In case NFD is already deployed, we fail the installation (in case it's
enabled on the kata-deploy helm chart) with a clear error message to the
user.
While I know that kata-remote **DOES NOT** require virtualization, I've
left this out (with a comment for when we add a peer-pods dependency on
kata-deploy) in order to simplify things for now, as kata-remote is not
a deployed shim by default.
Signed-off-by: Fabiano Fidêncio <ffidencio@nvidia.com>
As Kata Containers can be consumed by other helm-charts, hard coding the
default runtime class name to `kata` is not optimal.
Signed-off-by: Fabiano Fidêncio <ffidencio@nvidia.com>
All the options that take a specific shim as an argument MUST have
specific per arch settings, as not all the shims are available for all
the arches, leading to issues when setting up multi-arch deployments.
Signed-off-by: Fabiano Fidêncio <ffidencio@nvidia.com>
Let's ensure that we consume NVRC releases straight from GitHub instead
of building the binaries ourselves.
Signed-off-by: Zvonko Kaiser <zkaiser@nvidia.com>
Signed-off-by: Fabiano Fidêncio <ffidencio@nvidia.com>
We have here either /dev/vfio/<num> or /dev/vfio/devices/vfio<num>,
for IOMMUFD format /dev/vfio/devices/vfio<num>, strip "vfio" prefix
/dev/vfio/123 - basename "123" - vfioNum = "123" - cdi.k8s.io/vfio123
/dev/vfio/devices/vfio123 - basename "vfio123" - strip - vfioNum = "123" - cdi.k8s.io/vfio123
Signed-off-by: Zvonko Kaiser <zkaiser@nvidia.com>
We have 2 tests running on GitHub provided runners:
* devmapper
* CRI-O
- devmapper situation
For devmapper, we're currently testing devmapper with s390x as part of
one of its jobs.
More than that, this test has been failing here due to a lack of space
in the machine for quite some time, and no-action was taken to bring it
back either via GARM or some other way.
With that said, let's rely on the s390x CI to test devmapper and avoid
one extra failure on our CI by removing this one.
- cri-o situation
CRI-O is being tested with a fixed version of kubernetes that's already
reached its EOL, and a CRI-O version that matches that k8s version.
There has been attempts to raise issues, and also to provide a PR that
does at least part of the work ... leaving the debugging part for the
maintainers of the CI. However, there was no action on those from the
maintainers.
Signed-off-by: Fabiano Fidêncio <ffidencio@nvidia.com>
- Explained the concept and benefits of VM templating
- Provided step-by-step instructions for enabling VM templating
- Detailed the setup for using snapshotter in place of VirtioFS for template-based VM creation
- Added performance test results comparing template-based and direct VM creation
Signed-off-by: ssc <741026400@qq.com>
- init: initialize the VM template factory
- status: check the current factory status
- destroy: clean up and remove factory resources
These commands provide basic lifecycle management for VM templates.
Signed-off-by: ssc <741026400@qq.com>
Use `ioctl_with_mut_ref` instead of `ioctl_with_ref` in the
`create_device` method as it needs to write to the `kvm_create_device`
struct passed to it, which was released in v0.12.1.
Signed-off-by: Siyu Tao <taosiyu2024@163.com>
Fix the cargo fmt issues and then we can make the libs tests required
again to avoid this regression happening again.
Signed-off-by: stevenhorsman <steven@uk.ibm.com>
kata-deploy helm chart is *THE* way to deploy kata-containers on
kubernetes environments, and kubernetes environments is basically the
only reliably tested deployment we have.
For now, let's just drop documentation that is outdated / incorrect, and
in the future let's ensure we update the linked docs, as we work on
update / upgrade for the helm chart.
Signed-off-by: Fabiano Fidêncio <ffidencio@nvidia.com>
The libs in question were added when moving to developer.nvidia.com
but switching back to ubuntu only based builds they are not needed.
Remove them to keep the rootfs as minimal as possible.
Signed-off-by: Zvonko Kaiser <zkaiser@nvidia.com>
In the case of CC we need additional libraries in the rootfs.
Add them conditionally if type == confidential.
Signed-off-by: Zvonko Kaiser <zkaiser@nvidia.com>
Add build_vm_from_template() that flips boot_from_template flag,
wires factory.template_path/{memory,state} into the hypervisor config,
and returns ready-to-use hypervisor & agent instances.
When factory.template is enabled, VirtContainer bypasses normal creation
and directly boots the VM by restoring the template through incoming migration,
completing the "create → save → clone" loop.
Fixes: #11413
Signed-off-by: ssc <741026400@qq.com>
Introduced factory::FactoryConfig with init/destroy/status commands to manage template pools.
Added template::Template to fetch, create and persist base VMs.
Introduced vm::{VM, VMConfig} exposing create, pause, save, resume, stop,
disconnect and migration helpers for sandbox integration.
Extended QemuInner to executes QMP incoming migration, pause/resume and status tracking.
Fixes: #11413
Signed-off-by: ssc <741026400@qq.com>
Added new fields in Hypervisor struct to support VM template creation,
template boot, memory and device state paths, shared path, and store
paths. Introduced a Factory struct in config to manage template path,
cache endpoint, cache number, and template enable flag. Integrated
Factory into TomlConfig for runtime configuration parsing.
Fixes: #11413
Signed-off-by: ssc <741026400@qq.com>
This change enables to run the Cloud Hypervisor VMM using a non-root user
when rootless flag is set true in configuration.
Fixes: #11414
Signed-off-by: stevenfryto <sunzitai_1832@bupt.edu.cn>
Pass the file descriptors of the tuntap device to the Cloud Hypervisor VMM process
so that the process could open the device without cap_net_admin
Signed-off-by: stevenfryto <sunzitai_1832@bupt.edu.cn>
There's no reason to keep the env var / input as it's never been used
and now kata-deploy detects automatically whether NFD is deployed or
not.
Signed-off-by: Fabiano Fidêncio <ffidencio@nvidia.com>
When the NodeFeatureRule CRD is detected kata-deploy will:
* Create the specific NodeFeatureRules for the x86_64 TEEs
* Adapt the TEEs runtime classes to take into account the amount of keys
available in the system when spawning the podsandbox.
Note, we still do not have NFD as sub-dependency of the helm chart, and
I'm not even sure if we will have. However, it's important to integrate
better with the scenarios where the NFD is already present.
Signed-off-by: Fabiano Fidêncio <ffidencio@nvidia.com>
Change NIM bats file logic to allow skipping test cases which
require multiple GPUs. This can be helpful for test clusters where
there is only one node with a single GPU, or for local test
environments with a single-node cluster with a single GPU.
Signed-off-by: Manuel Huber <manuelh@nvidia.com>
Temporarily disables the new runners for building artifacts jobs. Will be re-enabled once they are stable.
Signed-off-by: Amulyam24 <amulmek1@in.ibm.com>
This partially reverts 8dcd91c for the s390x because the
CI jobs are currently blocking the release. The new runners
will be re-introduced once they are stable and no longer
impact critical paths.
Signed-off-by: Hyounggyu Choi <Hyounggyu.Choi@ibm.com>
This allows us to do a full multi-arch deployment, as the user can
easily select which shim can be deployed per arch, as some of the VMMs
are not supported on all architectures, which would lead to a broken
installation.
Now, passing shims per arch we can easily have an heterogenous
deployment where, for instance, we can set qemu-se-runtime-rs for s390x,
qemu-cca for aarch64, and qemu-snp / qemu-tdx for x86_64 and call all of
those a default kata-confidential ... and have everything working with
the same deployment.
Signed-off-by: Fabiano Fidêncio <ffidencio@nvidia.com>
The previous procedure failed to reliably ensure that all unused Device
Cgroups were completely removed, a failure consistently verified by CI
tests.
This change introduces a more robust and thorough cleanup mechanism. The
goal is to prevent previous issues—likely stemming from improper use of
Rust mutable references—that caused the modifications to be ineffective
or incomplete.
This ensures a clean environment and reliable CI test execution.
Signed-off-by: Alex Lyn <alex.lyn@antgroup.com>
Build only from Ubuntu repositories do not mix with developer.nvidia.com
Signed-off-by: Zvonko Kaiser <zkaiser@nvidia.com>
Update tools/osbuilder/rootfs-builder/nvidia/nvidia_chroot.sh
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
Migrate the k8s job to a different runner and use a long running cluster
instead of creating the cluster on every run.
Signed-off-by: Amulyam24 <amulmek1@in.ibm.com>
This will help immensely projects consuming the kata-deploy helm chart
to use configuration options added during the development cycle that are
waiting for a release to be out ... allowing very early tests of the
stack.
Signed-off-by: Fabiano Fidêncio <ffidencio@nvidia.com>
By default, `kubectl exec` inherits some capabilities from the
container, which could pose a security risk in a confidential
environment.
This change modifies the agent policy to strictly enforce that any
process started via `ExecProcessRequest` has no Linux capabilities.
This prevents potential privilege escalation within an exec session,
adhering to the principle of least privilege.
Signed-off-by: Alex Lyn <alex.lyn@antgroup.com>
As in CoCo cases, the ApparmorProfile setting within runtime-go is set with None,
we should align it with runtime-go.
Signed-off-by: Alex Lyn <alex.lyn@antgroup.com>
Temporarily disable the auto-generated Agent Policy on Mariner hosts,
to workaround the new test failures on these hosts.
When re-enabling auto-generated policy in the future, that would be
better achieved with a tests/integration/kubernetes/gha-run.sh change.
Those changes are easier to test compared with GHA YAML changes.
Signed-off-by: Dan Mihai <dmihai@microsoft.com>
If a ConfigMap has more than 8 files it will not be mounted watchable
[1]. However, genpolicy assumes that ConfigMaps are always mounted at a
watchable path, so containers with large ConfigMap mounts fail
verification.
This commit allows mounting ConfigMaps from watchable and non-watchable
directories. ConfigMap mounts can't be meaningfully verified anyway, so
the exact location of the data does not matter, except that we stay in
the sandbox data dirs.
[1]: 0ce3f5fc6f/docs/design/inotify.md (L11-L21)Fixes: #11777
Signed-off-by: Markus Rudy <mr@edgeless.systems>
Every now and then, in case a failure happens, helm leaves the secret
behind without cleaning it up, leading to issues in the consecutive
runs.
Signed-off-by: Fabiano Fidêncio <ffidencio@nvidia.com>
Aurélien has moved to a reliable mirror for our tests, but we missed
that our tools Dockerfiles could benefit from the same change, which is
added now.
Signed-off-by: Fabiano Fidêncio <ffidencio@nvidia.com>
Although we saw this happening, we expected it to NOT happen ...
As the kernel is not signed, but we expect it to be (the cached
version), then we're bailing. :-/
Let's ensure a full rebuild of kernels happen and we'll be good from
that point onwards.
Signed-off-by: Fabiano Fidêncio <ffidencio@nvidia.com>
Add TDX QGS quote-generation-socket TDX QEMU object params for
attestation to work in NVGPU+TDX environment.
Signed-off-by: Mikko Ylinen <mikko.ylinen@intel.com>
By doing this we can ensure that more than one instance of
nydus-snapshotter can be running inside the cluster, which is super
useful for doing A-B "upgrades" (where we install a new version of
kata-containers + nydus on B, while A is still running, and then only
uninstall A after making sure that B is working as expected).
Signed-off-by: Fabiano Fidêncio <ffidencio@nvidia.com>
We've been wrongly trying to set up the `${shim}` (as the qemu-snp, for
instance) as the hypervisor name in the kata-containers configuration
file, leading to an `tomlq` breaking as all the .hypervisors.qemu* shims
are tied to the `qemu` hypervisor, and it happens regardless of the shim
having a different name, or the hypervisor being experimental or not.
```sh
$ grep "hypervisor.qemu*" src/runtime/config/configuration-*
src/runtime/config/configuration-qemu-cca.toml.in:[hypervisor.qemu]
src/runtime/config/configuration-qemu-coco-dev.toml.in:[hypervisor.qemu]
src/runtime/config/configuration-qemu-nvidia-gpu-snp.toml.in:[hypervisor.qemu]
src/runtime/config/configuration-qemu-nvidia-gpu-tdx.toml.in:[hypervisor.qemu]
src/runtime/config/configuration-qemu-nvidia-gpu.toml.in:[hypervisor.qemu]
src/runtime/config/configuration-qemu-se.toml.in:[hypervisor.qemu]
src/runtime/config/configuration-qemu-snp.toml.in:[hypervisor.qemu]
src/runtime/config/configuration-qemu-tdx.toml.in:[hypervisor.qemu]
src/runtime/config/configuration-qemu.toml.in:[hypervisor.qemu]
$ grep "hypervisor.qemu*" src/runtime-rs/config/configuration-*
src/runtime-rs/config/configuration-qemu-runtime-rs.toml.in:[hypervisor.qemu]
src/runtime-rs/config/configuration-qemu-se-runtime-rs.toml.in:[hypervisor.qemu]
```
Signed-off-by: Fabiano Fidêncio <ffidencio@nvidia.com>
`tests` module inside `memcg` module should be gated behind `test`, add
`[#cfg(test)]` to make those tests work properly.
Signed-off-by: Ruoqing He <heruoqing@iscas.ac.cn>
Some tests from mem-agent requires root privilege, use
`skip_if_not_root` to skip those tests if they were not executed under
root user.
Signed-off-by: Ruoqing He <heruoqing@iscas.ac.cn>
Prefixing with `#[allow(clippy::type_complexity)]` to silence this
warning, the return type is documented in comments.
```console
error: very complex type used. Consider factoring parts into `type` definitions
--> mem-agent/src/mglru.rs:184:6
|
184 | ) -> Result<HashMap<String, (usize, HashMap<usize, MGenLRU>)>> {
| ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
|
= help: for further information visit https://rust-lang.github.io/rust-clippy/master/index.html#type_complexity
= note: `-D clippy::type-complexity` implied by `-D warnings`
= help: to override `-D warnings` add `#[allow(clippy::type_complexity)]`
```
Signed-off-by: Ruoqing He <heruoqing@iscas.ac.cn>
Manually fix `redundant_field_names ` clippy warning by testing equality
against 0 as suggested by rust 1.85.1, since `mem-agent` is now a member
of `libs` workspace.
```console
error: this comparison involving the minimum or maximum element for this type contains a case that is always true or always false
--> mem-agent/src/psi.rs:62:8
|
62 | if reader
| ________^
63 | | .read_line(&mut first_line)
64 | | .map_err(|e| anyhow!("reader.read_line failed: {}", e))?
65 | | <= 0
| |____________^
|
= help: because `0` is the minimum value for this type, the case where the two sides are not equal never occurs, consider using `reader
.read_line(&mut first_line)
.map_err(|e| anyhow!("reader.read_line failed: {}", e))? == 0` instead
= help: for further information visit https://rust-lang.github.io/rust-clippy/master/index.html#absurd_extreme_comparisons
= note: `#[deny(clippy::absurd_extreme_comparisons)]` on by default
```
Signed-off-by: Ruoqing He <heruoqing@iscas.ac.cn>
Manually fix `redundant_field_names` clippy warning as suggested by rust
1.85.1, since `mem-agent` is now a member of `libs` workspace.
```console
error: redundant field names in struct initialization
--> mem-agent/src/memcg.rs:441:13
|
441 | numa_id: numa_id,
| ^^^^^^^^^^^^^^^^ help: replace it with: `numa_id`
|
= help: for further information visit https://rust-lang.github.io/rust-clippy/master/index.html#redundant_field_names
= note: `-D clippy::redundant-field-names` implied by `-D warnings`
= help: to override `-D warnings` add `#[allow(clippy::redundant_field_names)]`
```
Signed-off-by: Ruoqing He <heruoqing@iscas.ac.cn>
Manually fix `manual_strip` clippy warning as suggested by rust 1.85.1,
since `mem-agent` is now a member of `libs` workspace.
```console
error: stripping a prefix manually
--> mem-agent/src/mglru.rs:284:29
|
284 | u32::from_str_radix(&content[2..], 16)
| ^^^^^^^^^^^^^
|
note: the prefix was tested here
--> mem-agent/src/mglru.rs:283:13
|
283 | let r = if content.starts_with("0x") {
| ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
= help: for further information visit https://rust-lang.github.io/rust-clippy/master/index.html#manual_strip
= note: `-D clippy::manual-strip` implied by `-D warnings`
= help: to override `-D warnings` add `#[allow(clippy::manual_strip)]`
help: try using the `strip_prefix` method
|
283 ~ let r = if let Some(<stripped>) = content.strip_prefix("0x") {
284 ~ u32::from_str_radix(<stripped>, 16)
|
```
Signed-off-by: Ruoqing He <heruoqing@iscas.ac.cn>
Manually fix `field_reassign_with_default` clippy warning as suggested
by rust 1.85.1, since `mem-agent` is now a member of `libs` workspace.
```console
error: field assignment outside of initializer for an instance created with Default::default()
--> mem-agent/src/memcg.rs:874:21
|
874 | numa_cg.numa_id = numa;
| ^^^^^^^^^^^^^^^^^^^^^^^
|
note: consider initializing the variable with `memcg::CgroupConfig { numa_id: numa, ..Default::default() }` and removing relevant reassignments
--> mem-agent/src/memcg.rs:873:21
|
873 | let mut numa_cg = CgroupConfig::default();
| ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
= help: for further information visit https://rust-lang.github.io/rust-clippy/master/index.html#field_reassign_with_default
= note: `-D clippy::field-reassign-with-default` implied by `-D warnings`
= help: to override `-D warnings` add `#[allow(clippy::field_reassign_with_default)]`
```
Signed-off-by: Ruoqing He <heruoqing@iscas.ac.cn>
Fix `redundant_pattern_matching` clippy warning as suggested by rust
1.85.1, since `mem-agent` is now a member of `libs` workspace.
```console
error: redundant pattern matching, consider using `is_some()`
--> mem-agent/src/memcg.rs:595:40
|
595 | ... if let Some(_) = config_map.get_mut(path) {
| -------^^^^^^^--------------------------- help: try: `if config_map.get_mut(path).is_some()`
|
= help: for further information visit https://rust-lang.github.io/rust-clippy/master/index.html#redundant_pattern_matching
= note: `-D clippy::redundant-pattern-matching` implied by `-D warnings`
= help: to override `-D warnings` add `#[allow(clippy::redundant_pattern_matching)]`
```
Signed-off-by: Ruoqing He <heruoqing@iscas.ac.cn>
Fix `needless_bool` clippy warning as suggested by rust 1.85.1, since
`mem-agent` is now a member of `libs` workspace.
```console
error: this if-then-else expression returns a bool literal
--> mem-agent/src/memcg.rs:855:17
|
855 | / if configs.is_empty() {
856 | | true
857 | | } else {
858 | | false
859 | | }
| |_________________^ help: you can reduce it to: `configs.is_empty()`
|
= help: for further information visit https://rust-lang.github.io/rust-clippy/master/index.html#needless_bool
= note: `-D clippy::needless-bool` implied by `-D warnings`
= help: to override `-D warnings` add `#[allow(clippy::needless_bool)]`
```
Signed-off-by: Ruoqing He <heruoqing@iscas.ac.cn>
Fix `for_kv_map` clippy warning as suggested by rust 1.85.1, since
`mem-agent` is now a member of `libs` workspace.
```console
error: you seem to want to iterate on a map's keys
--> mem-agent/src/memcg.rs:822:43
|
822 | for (single_config, _) in &secs_map.cgs {
| ^^^^^^^^^^^^^
|
= help: for further information visit https://rust-lang.github.io/rust-clippy/master/index.html#for_kv_map
help: use the corresponding method
|
822 | for single_config in secs_map.cgs.keys() {
| ~~~~~~~~~~~~~ ~~~~~~~~~~~~~~~~~~~
```
Signed-off-by: Ruoqing He <heruoqing@iscas.ac.cn>
Fix `into_iter_on_ref` clippy warning as suggested by rust 1.85.1, since
`mem-agent` is now a member of `libs` workspace.
```console
error: this `.into_iter()` call is equivalent to `.iter_mut()` and will not consume the `Vec`
--> mem-agent/src/memcg.rs:1122:27
|
1122 | for info in infov.into_iter() {
| ^^^^^^^^^ help: call directly: `iter_mut`
|
= help: for further information visit https://rust-lang.github.io/rust-clippy/master/index.html#into_iter_on_ref
= note: `-D clippy::into-iter-on-ref` implied by `-D warnings`
= help: to override `-D warnings` add `#[allow(clippy::into_iter_on_ref)]`
```
Signed-off-by: Ruoqing He <heruoqing@iscas.ac.cn>
Fix `legacy_numeric_constants` clippy warning as suggested by rust
1.85.1, since `mem-agent` is now a member of `libs` workspace.
```console
error: usage of a legacy numeric constant
--> mem-agent/src/compact.rs:132:47
|
132 | if self.config.compact_force_times == std::u64::MAX {
| ^^^^^^^^^^^^^
|
= help: for further information visit https://rust-lang.github.io/rust-clippy/master/index.html#legacy_numeric_constants
help: use the associated constant instead
|
132 | if self.config.compact_force_times == u64::MAX {
| ~~~~~~~~
```
Signed-off-by: Ruoqing He <heruoqing@iscas.ac.cn>
Fix `single_component_path_imports` clippy warning as suggested by rust
1.85.1, since `mem-agent` is now a member of `libs` workspace.
```console
error: this import is redundant
--> mem-agent/src/mglru.rs:345:5
|
345 | use slog_term;
| ^^^^^^^^^^^^^^ help: remove it entirely
|
= help: for further information visit https://rust-lang.github.io/rust-clippy/master/index.html#single_component_path_imports
```
Signed-off-by: Ruoqing He <heruoqing@iscas.ac.cn>
Fix `from_str_radix_10` clippy warning as suggested by rust 1.85.1,
since `mem-agent` is now a member of `libs` workspace.
```console
error: this call to `from_str_radix` can be replaced with a call to `str::parse`
--> mem-agent/src/mglru.rs:29:14
|
29 | let id = usize::from_str_radix(words[1], 10)
| ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ help: try: `words[1].parse::<usize>()`
|
= help: for further information visit https://rust-lang.github.io/rust-clippy/master/index.html#from_str_radix_10
= note: `-D clippy::from-str-radix-10` implied by `-D warnings`
= help: to override `-D warnings` add `#[allow(clippy::from_str_radix_10)]`
```
Signed-off-by: Ruoqing He <heruoqing@iscas.ac.cn>
Fix `needless_borrow` clippy warning as suggested by rust 1.85.1, since
`mem-agent` is now a member of `libs` workspace.
```console
error: this expression creates a reference which is immediately dereferenced by the compiler
--> mem-agent/src/memcg.rs:1100:52
|
1100 | self.run_eviction_single_config(infov, &config)?;
| ^^^^^^^ help: change this to: `config`
|
= help: for further information visit https://rust-lang.github.io/rust-clippy/master/index.html#needless_borrow
```
Signed-off-by: Ruoqing He <heruoqing@iscas.ac.cn>
Fix `ptr_arg` clippy warning as suggested by rust 1.85.1, since
`mem-agent` is now a member of `libs` workspace.
```console
error: writing `&PathBuf` instead of `&Path` involves a new object where a slice will do
--> mem-agent/src/memcg.rs:367:19
|
367 | psi_path: &PathBuf,
| ^^^^^^^^ help: change this to: `&Path`
|
= help: for further information visit https://rust-lang.github.io/rust-clippy/master/index.html#ptr_arg
= note: requested on the command line with `-D clippy::ptr-arg`
```
Signed-off-by: Ruoqing He <heruoqing@iscas.ac.cn>
Fix `crate_in_macro_def` clippy warning as suggested by rust 1.85.1,
since `mem-agent` is now a member of `libs` workspace.
```console
error: `crate` references the macro call's crate
--> mem-agent/src/misc.rs:12:22
|
12 | slog::error!(crate::misc::sl(), "{}", format_args!($($arg)*))
| ^^^^^ help: to reference the macro definition's crate, use: `$crate`
|
= help: for further information visit https://rust-lang.github.io/rust-clippy/master/index.html#crate_in_macro_def
= note: `-D clippy::crate-in-macro-def` implied by `-D warnings`
= help: to override `-D warnings` add `#[allow(clippy::crate_in_macro_def)]`
```
Signed-off-by: Ruoqing He <heruoqing@iscas.ac.cn>
Fix `len_zero` clippy warning as suggested by rust 1.85.1, since
`mem-agent` is now a member of `libs` workspace.
```console
error: length comparison to zero
--> mem-agent/src/memcg.rs:225:61
|
225 | let (keep, moved) = vec.drain(..).partition(|c| c.numa_id.len() > 0);
| ^^^^^^^^^^^^^^^^^^^ help: using `!is_empty` is clearer and more explicit: `!c.numa_id.is_empty()`
|
= help: for further information visit https://rust-lang.github.io/rust-clippy/master/index.html#len_zero
```
Signed-off-by: Ruoqing He <heruoqing@iscas.ac.cn>
Fix `bool_assert_comparison` clippy warning as suggested by rust 1.85.1,
since `mem-agent` is now a member of `libs` workspace.
```console
error: used `assert_eq!` with a literal bool
--> mem-agent/src/memcg.rs:1378:9
|
1378 | assert_eq!(m.get_timeout_list().len() > 0, true);
| ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
|
= help: for further information visit https://rust-lang.github.io/rust-clippy/master/index.html#bool_assert_comparison
= note: `-D clippy::bool-assert-comparison` implied by `-D warnings`
= help: to override `-D warnings` add `#[allow(clippy::bool_assert_comparison)]`
help: replace it with `assert!(..)`
|
1378 - assert_eq!(m.get_timeout_list().len() > 0, true);
1378 + assert!(m.get_timeout_list().len() > 0);
|
```
Signed-off-by: Ruoqing He <heruoqing@iscas.ac.cn>
`mem-agent` now does not ship example binaries and serves as a library
for `agent` to reference, so we move it into `libs` to better manage it.
Signed-off-by: Ruoqing He <heruoqing@iscas.ac.cn>
Auto-generate policy in k8s-optional-empty-secret.bats, now that
genpolicy suppprts optional secret-based volumes.
Signed-off-by: Dan Mihai <dmihai@microsoft.com>
We've recently added support for:
* deploying and setting up a snapshotter, via
_experimentalSetupSnapshotter
* enabling experimental_force_guest_pull, via
_experimentalForceGuestPull
However, we never updated the documentation for those, thus let's do it
now.
Signed-off-by: Fabiano Fidêncio <ffidencio@nvidia.com>
Downloading Go from storage.googleapis.com fails intermittently with a 403
(see error below) so we switch to go.dev as referenced at
https://go.dev/dl/.
/tmp/install-go-tmp.Rw5Q4thEWr ~/work/kata-containers/kata-containers
/usr/bin/go
[install_go.sh:85] INFO: removing go version go1.24.9 linux/amd64
[install_go.sh:94] INFO: Download go version 1.24.6
% Total % Received % Xferd Average Speed Time Time Time Current
Dload Upload Total Spent Left Speed
0 0 0 0 0 0 0 0 --:--:-- --:--:-- --:--:-- 0
100 298 100 298 0 0 2610 0 --:--:-- --:--:-- --:--:-- 2614
[install_go.sh:97] INFO: Install go
gzip: stdin: not in gzip format
tar: Child returned status 1
tar: Error is not recoverable: exiting now
[install_go.sh:99] ERROR: sudo tar -C /usr/local/ -xzf go1.24.6.linux-amd64.tar.gz
https://github.com/kata-containers/kata-containers/actions/runs/18602801597/job/53045072109?pr=11947#step:5:17
Signed-off-by: Aurélien Bombo <abombo@microsoft.com>
This change ensures that the NVIDIA package repository for nvidia-imex
and libnvidia-nspc is being used as source.
The NVIDIA repository does not publish these packages with a -580
version suffix, which made us fall back to the packages from the
Ubuntu repository.
These two packages were recently updated by Ubuntu to depend on
nvidia-kernel-common-580-server (this happened from version
580.82.07-0ubuntu1 to version 580.95.05-0ubuntu1). This conflicts
with nvidia-kernel-common-580 which gets installed by
nvidia-headless-no-dkms-580-open, thus causing a build failure.
Signed-off-by: Manuel Huber <manuelh@nvidia.com>
in CI helm is not yet installed and we don't have root access. Let's use
the current dir, which should be writable, and --no-sudo option to
install it.
Note when helm is installed it should not change anything and simply use
the syste-wide installation.
Signed-off-by: Lukáš Doktor <ldoktor@redhat.com>
While the local-build's folder's Makefile dependencies for the
confidential nvidia rootfs targets already declare the pause image
and coco-guest-components dependencies, the actual rootfs
composition does not contain the pause image bundle and relevant
certificates for guest pull. This change ensure the rootfs gets
composed with the relevant files.
Signed-off-by: Manuel Huber <manuelh@nvidia.com>
TOML was chosen for initdata particularly for the ability to include
policy docs and other configuration files without mangling them. The
default TOML encoding renders string values as single-line,
double-quoted strings, effectively depriving us of this feature.
This commit changes the encoding to use `to_string_pretty`, and includes
a test that verifies the desirable aspect of encoding: newlines are kept
verbatim.
Fixes: #11943
Signed-off-by: Markus Rudy <mr@edgeless.systems>
After supporting the Arm CCA, it will rely on the kernel kvm.h headers to build the
runtime. The kernel-headers currently quite new with the traditional one, so that we
rely on build the kernel header first and then inject it to the shim-v2 build container.
Signed-off-by: Kevin Zhao <kevin.zhao@linaro.org>
Co-authored-by: Seunguk Shin <seunguk.shin@arm.com>
The new initdata variants of the tests are failing on the tdx
runner, so as discussed, skip them for now: Issue #11945
Signed-off-by: stevenhorsman <steven@uk.ibm.com>
Now we have wider coverage of initdata testing in
k8s-guest-pull-image-signature.bats then remove
the old testing.
Signed-off-by: stevenhorsman <steven@uk.ibm.com>
Our current set of authenticated registry tests involve setting
kernel_params to config the image pull process, but as of
kata-containers#11197
this approach is not the main way to set this configuration and the agent
config has been removed. Instead we should set the configuration in the
`cdh.toml` part of the initdata, so add new test cases for this. In future, when
we have been through the deprecation process, we should remove the old tests
Signed-off-by: stevenhorsman <steven@uk.ibm.com>
Our current set of signature tests involve setting kernel_parameters to
config the image pull process, but as of
https://github.com/kata-containers/kata-containers/pull/11197
this approach is not the main way to set this configuration and the agent
config has been removed. Instead we should set the configuration in the
`cdh.toml` part of the initdata, so add new test cases for this. In future, when
we have been through the deprecation process, we should remove the old tests
Signed-off-by: stevenhorsman <steven@uk.ibm.com>
Create a shared get_initdata method that injects a cdh image
section, so we don't duplicate the initdata structure everywhere
Signed-off-by: stevenhorsman <steven@uk.ibm.com>
One problem that we've been having for a reasonable amount of time, is
containerd not behaving very well when we have multiple snapshotters.
Although I'm adding this test with my "CoCo" hat in mind, the issue can
happen easily with any other case that requires a different snapshotter
(such as, for instance, firecracker + devmapper).
With this in mind, let's do some stability tests, checking every hour a
simple case of running a few pre-defined containers with runc, and then
running the same containers with kata.
This should be enough to put us in the situation where containerd gets
confused about which snapshotter owns the image layers, and break on us
(or not break and show us that this has been solved ...).
Signed-off-by: Fabiano Fidêncio <ffidencio@nvidia.com>
With this change we namespace the stage one rootfs tarball name
and use the same name across all uses. This will help overcome
several subtle local build problems.
Signed-off-by: Manuel Huber <manuelh@nvidia.com>
We fix the following error:
```
thread 'sandbox::tests::add_and_get_container' panicked at src/sandbox.rs:901:10:
called `Result::unwrap()` on an `Err` value: Create cgroupfs manager
Caused by:
0: fs error caused by: Os { code: 17, kind: AlreadyExists, message: "File exists" }
1: File exists (os error 17)
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace
```
by ensuring that the cgroup path is unique for tests run in the same millisecond.
Signed-off-by: Hyounggyu Choi <Hyounggyu.Choi@ibm.com>
Use CDI exclusively from crates.io and not from a GH repository.
Cargo can easily check if a new version is available and we can
far more easier bump it if needed.
Signed-off-by: Zvonko Kaiser <zkaiser@nvidia.com>
with the shellcheck fixes we accidentally quoted the "-n NAMESPACE"
argument where we should have used array instead, which lead to oc
considering this as a pod name and returning error.
Signed-off-by: Lukáš Doktor <ldoktor@redhat.com>
We need to ensure that any change on the Dockerfile (and its dir) leads
to the build being retriggered, rather than using the cached version.
Signed-off-by: Fabiano Fidêncio <ffidencio@nvidia.com>
We are seeing more protoc related failures on the new
runners, so try adding the protobuf-compiler dependency
to these steps to see if it helps.
Signed-off-by: stevenhorsman <steven@uk.ibm.com>
- copy default-initdata.toml in create_tmp_policy_settings_dir, so it can be modified by other tests if needed
- make auto_generate_policy use default-initdata.toml by default
- add auto_generate_policy_no_added_flags, so it may be used by tests that don't want to use default-initdata.toml by default
Signed-off-by: Saul Paredes <saulparedes@microsoft.com>
On commit 9602ba6ccc, from February this
year, we've introduced a check to ensure that the files needed for
signing the kernel build are present. However, we've noticed last week
that there were a reasonable amount of wrong assumptions with the
workflow. :-)
Zvonko fixed the majority of those, but this bit was left and it'd cause
breakages when using kernel that was cached ... although passing when
building new kernels.
Signed-off-by: Fabiano Fidêncio <ffidencio@nvidia.com>
This is needed to the kernel setup picks up the correct
config values from our fragments directories.
Signed-off-by: Zvonko Kaiser <zkaiser@nvidia.com>
We need to make sure that the kernel we're using has the
correct configs set, otherwise the module signing will not work.
Signed-off-by: Zvonko Kaiser <zkaiser@nvidia.com>
Now that we have added the ability to deploy kata-containers with
experimental_force_guest_pull configured, let's make sure we test it to
avoid any kind of regressions.
Signed-off-by: Fabiano Fidêncio <ffidencio@nvidia.com>
Otherwise we have no way to differentiate running tests on qemu-coco-dev
with different snapshotters.
Signed-off-by: Fabiano Fidêncio <ffidencio@nvidia.com>
What was done in the past, trying to set the env var on the same step
it'd be used, simply does not work.
Instead, we need to properly set it through the `env` set up, as done
now.
We're also bumping the kata_config_version to ensure we retrigger the
kernel builds.
Signed-off-by: Fabiano Fidêncio <ffidencio@nvidia.com>
We have some scalable s390x and ppc runners, so
start to use them for build and test, to improve
the throughput of our CI
Signed-off-by: stevenhorsman <steven@uk.ibm.com>
Co-authored-by: Hyounggyu Choi <Hyounggyu.Choi@ibm.com>
For some reason we didn't have the "Report tests" step as part of the
TEE jobs. This step immensely helps to check which tests are failing and
why, so let's add it while touching the workflow.
Signed-off-by: Fabiano Fidêncio <ffidencio@nvidia.com>
There's no reason to have the code duplication between the SNP / TDX
tests for CoCo, as those are basically using the same configuration
nowadays.
Note that for the TEEs case, as the nydus-snapshotter is deployed by the
admin, once, instead of deploying it on every run ... I'm actually
removing the nydus-snapshotter steps so we make it clear that those
steps are not performed by the CI.
Signed-off-by: Fabiano Fidêncio <ffidencio@nvidia.com>
As virtio-scsi has been set the default block device driver, the
runtime also need to correctly handle the virtio-scsi info, specially
the SCSI address required within kata-agent handling logic.
And getting and assigning the scsi_addr to kata agent device id
will be enough. This commit just do such work.
Signed-off-by: Alex Lyn <alex.lyn@antgroup.com>
Since runtime-rs support the block device hotplug with
creating new containers, and the device would also be
removed when the container stopped, thus add the block
device unplug for clh.
Signed-off-by: Fupan Li <fupan.lfp@antgroup.com>
This commit introduces support for selecting `virtio-scsi` as the
block device driver for QEMU during initial setup.
The primary goal is to resolve a conflict in non-TEE environments:
1. The global block device configuration defaults to `virtio-scsi`.
2. The `initdata` device driver was previously designed and hardcoded
to `virtio-blk-pci`.
3. This conflict prevented unified block device usage.
By allowing `virtio-scsi` to be configured at cold boot, the `initdata`
device can now correctly adhere to the global setting, eliminating the
need for a hardcoded driver and ensuring consistent block device
configuration across all supported devices (excluding rootfs).
Signed-off-by: Alex Lyn <alex.lyn@antgroup.com>
The implementation of the seccomp feature in Dragonball currently has a basic framework.
But the actual restriction rules are empty.
This pull request includes the following changes:
- Modifiy configuration files to relevant configuration files.
- Modifiy seccomp framework to support different restrictions for different threads.
- Add new seccomp rules for the modified framework.
This commit primarily implements the changes 1 and 3 for runtime-rs.
Fixes: #11673
Signed-off-by: wangxinge <wangxinge@bupt.edu.cn>
For DGX like systems we need additional binaries and libraries,
enable the Kata AND CoCo use-case.
Signed-off-by: Zvonko Kaiser <zkaiser@nvidia.com>
Update tools/osbuilder/rootfs-builder/nvidia/nvidia_rootfs.sh
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
Fix all instances of template injection by using environment variables as
recommended by Zizmor, instead of directly injecting values into the
commands.
Signed-off-by: Aurélien Bombo <abombo@microsoft.com>
The two ignored cases are strictly necessary for the CI to work today, and we
have various security mitigations in place.
Signed-off-by: Aurélien Bombo <abombo@microsoft.com>
There are 62 such warnings and addressing them would take quite a bit of
time so just disable them for now.
help[undocumented-permissions]: permissions without explanatory comments
--> ./.github/workflows/release.yaml:71:7
|
71 | packages: write
| ^^^^^^^^^^^^^^^ needs an explanatory comment
72 | id-token: write
| ^^^^^^^^^^^^^^^ needs an explanatory comment
73 | attestations: write
| ^^^^^^^^^^^^^^^^^^^ needs an explanatory comment
|
= note: audit confidence → High
Signed-off-by: Aurélien Bombo <abombo@microsoft.com>
We can't test this PR because the workflow needs this trigger, so adding
this will allow testing future PRs.
Signed-off-by: Aurélien Bombo <abombo@microsoft.com>
For those who are not willing to use the nydus-snapshotter for pulling
the image inside the guest, let's allow them setting the
experimetal_force_guest_pull, introduced by Edgeless, as part of our
helm-chart.
This option can be set as:
_experimentalForceGuestPull: "qemu-tdx,qemu-coco-dev"
Which would them ensure that the configuration for `qemu-tdx` and
`qemu-coco-dev` would have the option enabled.
Signed-off-by: Fabiano Fidêncio <ffidencio@nvidia.com>
As the kata-deploy helm chart has been the only way we've been testing
kata-containers deployment as part of our CI, it's time to finally get
rid of the kustomize yamls and avoid us having to maintain two different
methods (with one of those not being tested).
Here I removed:
* kata-deploy yamls and kustomize yamls
* kata-cleanup yamls and kustomize yamls
* kata-rbac yals and kustomize yamls
* README.md for the kustomize yamls was removed
Signed-off-by: Fabiano Fidêncio <ffidencio@nvidia.com>
erofs-snapshotter can be used to leverage sharing the image from the
host to the guest without the need of a shared filesystem (such as
virtio-fs or virtio-9p).
This case is ideal for Confidential Computing enabled on Kata
Containers, and we can immensely benefit from this snapshotter, thus
let's test it as soon as possible so we can find issues, report bugs,
and ask for enhancement requests.
There are at least a few things that we know for sure to be problematic
now:
* Policy has to be adjusted to the erofs-snapshotter
* There is no support for signed nor encrypted images
* Tests that use the KBS are disabled for now
Even with the limitations, I do believe we should be testing the
snapshoitter, so we can team up and get those limitations addressed.
Signed-off-by: Fabiano Fidêncio <ffidencio@nvidia.com>
As done in the previous commit, let's expand the vanilla k8s deployment
to also allow the erofs host side configuration.
Signed-off-by: Fabiano Fidêncio <ffidencio@nvidia.com>
We already have support for deploying a few flavours of k8s that are
required for different tests we perform.
Let's also add the ability to deploy vanilla k8s, as that will be very
useful in the next commits in this series.
Signed-off-by: Fabiano Fidêncio <ffidencio@nvidia.com>
The active version is 2.1.x, and the latest is 2.2.0-beta.0.
The latest is what we'll be using to test if the "to be released"
version of containerd works well for our use-cases.
Signed-off-by: Fabiano Fidêncio <ffidencio@nvidia.com>
Let's make sure that we can get non-official releases as well, otherwise
we won't be able to test a coming release of containerd, to know whether
it solves issues that we face or not, before it's actually released.
Signed-off-by: Fabiano Fidêncio <ffidencio@nvidia.com>
NVRC introduced the confidential feature flag and we
haven't updated the rootfs build to accomodate.
If rootfs_type==confidential user --feature=confidential
Signed-off-by: Zvonko Kaiser <zkaiser@nvidia.com>
Canonical TDX release is not needed for vanilla Ubuntu 25.10 but
GRUB_CMDLINE_LINUX_DEFAULT needs to contain `nohibernate` and
`kvm_intel.tdx=1`
Signed-off-by: Szymon Klimek <szymon.klimek@intel.com>
Use grep_pod_exec_output to retry possible failing "kubectl exec"
commands. Other tests have been hitting such errors during CI in
the past.
Signed-off-by: Dan Mihai <dmihai@microsoft.com>
This adds an integration test to verify that privileged containers work
properly when deploying Kata with kata-deploy.
This is a follow-up to #11878.
Signed-off-by: Aurélien Bombo <abombo@microsoft.com>
Let's rely on kata-deploy setting up the nydus snapshotter for us,
instead of doing this with external code.
Signed-off-by: Fabiano Fidêncio <ffidencio@nvidia.com>
This allows us to stop setting up the snapshotter ourselves, and just
rely con kata-deploy to do so.
Signed-off-by: Fabiano Fidêncio <ffidencio@nvidia.com>
Let's expose the EXPERIMENTAL_SETUP_SNAPSHOTTER script environment
variable to our chart, allowing then users of our helm chart to take
advantage of this experimental feature.
Signed-off-by: Fabiano Fidêncio <ffidencio@nvidia.com>
We may deploy in scenarios where we want to have both snapshotters set
up, sometimes even for simple test on which one behaves better.
With this in mind, let's allow EXTERNAL_SETUP_SNAPSHOTTER to receive a
comma separated list of snapshotters, such as:
```
EXPERIMENTAL_SETUP_SNAPSHOTTER="erofs,nydus"
```
Signed-off-by: Fabiano Fidêncio <ffidencio@nvidia.com>
Similarly to what's been done for the nydus-snapshotter, let's allow
users to have erofs-snapshotter set up by simply passing:
```
EXPERIMENTAL_SETUP_SNAPSHOTTER="erofs".
```
Mind that erofs, although a built-in containerd snapshotter, has system
depdencies that we will *NOT* install and it's up to the admin to do so.
These dependencies are:
* erofs-utils
* fsverity
* erofs module loaded
Signed-off-by: Fabiano Fidêncio <ffidencio@nvidia.com>
In the previous commit we added the assumption that the
nydus-snapshotter version should be the same in two different places.
Now, with this test, we ensure those will always be in sync.
Signed-off-by: Fabiano Fidêncio <ffidencio@nvidia.com>
Let's introduce a new EXPERIMENTAL_SETUP_SNAPSHOTTER environemnt
variable that, when set, allows kata-deploy to put the nydus snapshotter
in the correct place, and configure containerd accordingly.
Mind, this is a stop gap till the nydus-snapshotter helm chart is ready
to be used and behaving well enough to become a weak dependency of our
helm chart. When that happens this code can be deleted entirely.
Users can have nydus-snapshotter deployed and configured for the
guest-pull use case by simply passing:
```
EXPERIMENTAL_SETUP_SNAPSHOTTER="nydus"
```
Signed-off-by: Fabiano Fidêncio <ffidencio@nvidia.com>
Otherwise we'd end up adding a the file several times, which could lead
to problems when removing the entry, leading to containerd not being
able to start due to an import file not being present.
Signed-off-by: Fabiano Fidêncio <ffidencio@nvidia.com>
The code, how it was, would lead to the following broke command:
`--header "Authorization: Bearer: "`
Let's only expand that part of the command if ${GH_TOKEN} is passed,
otherwise we don't even bother adding it.
Signed-off-by: Fabiano Fidêncio <ffidencio@nvidia.com>
Document that privileged containers with
privileged_without_host_devices=false are not generally supported.
When you try the above, the runtime will pass all the host devices to Kata
in the OCI spec, and Kata will fail to create the container for various
reasons depending on the setup, e.g.:
- Attempting to hotplug uninitialized loop devices.
- Attempting to remount /dev devices on themselves when the agent had
already created them as default devices (e.g. /dev/full).
- "Conflicting device updates" errors.
- And more...
privileged_without_host_devices was originally created to support
Kata [1][2] and lots of people are having issues when it's set to
false [3].
[1] https://github.com/kata-containers/runtime/issues/1568
[2] https://github.com/containerd/cri/pull/1225
[3] https://github.com/kata-containers/kata-containers/issues?q=is%3Aissue%20%20in%3Atitle%20privileged
Signed-off-by: Aurélien Bombo <abombo@microsoft.com>
We have noticed in the CI that the `gen_init_cpio ...` was returning 255
and breaking the build. Why? I am not sure.
When chatting with Steve, he suggested to split the command, so it'd be
easier to see what's actually breaking. But guess what? There's no
breakage when we split the command.
So, let's try it out and see whether the CI passes after it.
If someone is willing to educate us on this one, please, that would be
helpful! :-)
Signed-off-by: Fabiano Fidêncio <ffidencio@nvidia.com>
Moving the CUDA repo to the top for all essential packages
and adding a repo priority favouring NVIDIA based repos.
Signed-off-by: Zvonko Kaiser <zkaiser@nvidia.com>
Introduce new test case which verifies that openvpn clients and servers
can run as Kata pods and can successfully establish a connection.
Volatile certificates and keys are generated by an initialization
container and injected into the client and server containers.
This scenario requires TUN/TAP support for the UVM kernel.
Signed-off-by: Manuel Huber <mahuber@microsoft.com>
Co-authored-by: Manuel Huber <manuelh@nvidia.com>
No need to die when a Kind that does not require a policy annotation is
found in a pod manifest. Print an informational message instead.
Signed-off-by: Manuel Huber <mahuber@microsoft.com>
Currently, use of openvpn clients/servers is not possible in Kata UVMs.
Following error message can be expected:
ERROR: Cannot open TUN/TAP dev /dev/net/tun: No such device (errno=19)
To support opevpn scenarios using bridging and TAP, we enable various
kernel networking config options.
Signed-off-by: Manuel Huber <mahuber@microsoft.com>
Manually added "hostPath" to main.txt then regenerated the dictionary
with `./kata-spell-check.sh make-dict`.
Signed-off-by: Aurélien Bombo <abombo@microsoft.com>
This change crystallizes and simplifies the current handling of /dev
hostPath mounts with virtually no functional change.
Before this change:
- If a mount DESTINATION is in /dev and it is a non-regular file on the HOST,
the shim passes the OCI bind mount as is to the guest (e.g.
/dev/kmsg:/dev/kmsg). The container rightfully sees the GUEST device.
- If the mount DESTINATION does not exist on the host, the shim relies on
k8s/containerd to automatically create a directory (ie. non-regular file) on
the HOST. The shim then also passes the OCI bind mount as is to the guest. The
container rightfully sees the GUEST device.
- For other /dev mounts, the shim passes the device major/minor to the guest
over virtio-fs. The container rightfully sees the GUEST device.
After this change:
- If a mount SOURCE is in /dev and it is a non-regular file on the HOST,
the shim passes the OCI bind mount as is to the guest. The container
rightfully sees the GUEST device.
- The shim does not anymore rely on k8s/containerd to create missing mount
directories. Instead it explicitely handles missing mount SOURCES, and
treats them like the previous bullet point.
- The shim no longer uses virtio-fs to pass /dev device major/minor to the
guest, instead it passes the OCI bind mount as is.
Signed-off-by: Aurélien Bombo <abombo@microsoft.com>
test_add_one_arp_neighbor modifies the root network namespace, so we
should ensure that it does not interfere with normal network setup.
Adding an IP to a device results in automatic routes, which may affect
routing to non-test endpoints. Thus, we change the addresses used in the
test to come from TEST-NET-1, which is designated for tests and usually
not routable.
Signed-off-by: Markus Rudy <mr@edgeless.systems>
list_routes and test_add_one_arp_neighbor have been flaky in the past
(#10856), but it's been hard to tell what exactly is going wrong.
This commit adds debug information for the most likely problem in
list_routes: devices being added/removed/modified concurrently.
Furthermore, it adds the exit code and stderr of the ip command, in case
it failed to list the ARP neighborhood.
Signed-off-by: Markus Rudy <mr@edgeless.systems>
The previous code only checked the result of with_nix_path(), not statfs(),
thus leading to an uninitialized memory read if statfs() failed.
No functional change otherwise.
Signed-off-by: Aurélien Bombo <abombo@microsoft.com>
85f3391bc added the support for TDX QGS port=0 but missed
defaultQgsPort in the default config. defaultQgsPort overrides
user provided tdx_quote_generation_service_socket_port=0.
After this change, defaultQgsPort is not needed anymore since
there's no default: any positive integer is OK and negative or
unset value becomes a parse error.
QEMUTDXQUOTEGENERATIONSERVICESOCKETPORT in the Makefile is used
to provide a sane default when tdx_quote_generation_service_socket_port
gets set in the configuration.
Signed-off-by: Mikko Ylinen <mikko.ylinen@intel.com>
genpolicy is a developer tool that should be usable on MacOS. Adding it
to the darwin CI job ensures that it can still be built after changes.
On an Apple M2, the output of `uname -m` is `arm64`, which is why a new
case is needed in the arch_to_* functions.
We're not going to cross-compile binaries on darwin, so don't install
any additional Rust targets.
Fixes: #11635
Signed-off-by: Markus Rudy <mr@edgeless.systems>
Most of the kata-types code is reusable across platforms. However, some
functions in the mount module require safe-path, which is Linux-specific
and can't be used on other platforms, notably darwin.
This commit adds a new feature `safe-path` to kata-types, which enables
the functions that use safe-path. The Linux-only callers kata-ctl and
runtime-rs enable this feature, whereas genpolicy only needs initdata
and does not need the functions from the mount module. Using a feature
instead of a target_os restriction ensures that the developer experience
for genpolicy remains the same.
Signed-off-by: Markus Rudy <mr@edgeless.systems>
This commit adds changes to enable fs sharing between host/guest
using virtio-fs when booting a pod VM for testing. This primarily
enables sharing container rootfs for testing container lifecycle
commands.
Summary of changes is as below:
- adds minimal virtiofsd code to start userspace daemon (based on
`runtime-rs/crates/resource/src/share_fs`)
- adds the virtiofs device to the test vm
- prepares and mounts the container rootfs on host
- modifies container storage & oci specs
Signed-off-by: Sumedh Alok Sharma <sumsharma@microsoft.com>
Fixing the shellcheck issues first so that they are not coupled to the
subsequent commit introducing Darwin support to the script.
Signed-off-by: Markus Rudy <mr@edgeless.systems>
Auto-generate policy for nginx-deployment pods, instead of hard-coding
the "allow all" policy.
Note that the `busybox_pod` - created using `kubectl run` - still
doesn't have an Init Data annotation, so it is using the default policy
built into the Kata Guest rootfs image file.
Signed-off-by: Dan Mihai <dmihai@microsoft.com>
Auto-generate agent policy in k8s-liveness-probes.bats, instead of using
the non-confidential "allow all" policy.
Signed-off-by: Dan Mihai <dmihai@microsoft.com>
Auto-generate the agent policy for pod-secret-env.yaml, using
"genpolicy -c inject_secret.yaml".
Support for passing Secret specification files as "-c" arguments of
genpolicy has been added when fixing #10033 with PR #10986.
Signed-off-by: Dan Mihai <dmihai@microsoft.com>
Since we cannot build all components with libc=musl and
static RUSTFLAG we still need to ship libcc for AA or other guest
components.
Without this change the guest components do not work and we see
/usr/local/bin/attestation-agent: error while loading shared
libraries: libgcc_s.so.1: cannot open shared object file: No such file or directory
Signed-off-by: Zvonko Kaiser <zkaiser@nvidia.com>
This fixes that error everywhere by adding a `name:` field to all jobs that
were missing it. We keep the same name as the job ID to ensure no
disturbance to the required job names.
Signed-off-by: Aurélien Bombo <abombo@microsoft.com>
-o pipefail in particular ensures that exec_host() returns the right exit
code.
-u is also added for good measure. Note that $BATS_TEST_DIRNAME is set by
bats so we move its usage inside the function.
Signed-off-by: Aurélien Bombo <abombo@microsoft.com>
As a consequence of moving away from Advanced Security for Zizmor, it now
checks the entire codebase and will error out on this PR and future.
To be reverted once we address all Zizmor findings in a future PR.
Signed-off-by: Aurélien Bombo <abombo@microsoft.com>
This PR fixes a test that failed on platforms like ppc64le due to a hardcoded mount option length.
* Test was failing on ppc64le due to larger system page size (e.g., 65536 bytes)
* Original test used a hardcoded 4097-byte string assuming 4KB page size
* Replaced with *MAX_MOUNT_PARAM_SIZE + 1 to reflect actual system limit
* Ensures test fails correctly across all architectures
Fixes: #11852
Signed-off-by: shwetha-s-poojary <shwetha.s-poojary@ibm.com>
The Hadolint warning DL3007 (pin the version explicitly) is no
longer applicable.
We have updated the base image to use a specific version
digest, which satisfies the linter's requirement for reproducible
builds. This commit removes the corresponding inline ignore comment.
Signed-off-by: Hyounggyu Choi <Hyounggyu.Choi@ibm.com>
This change enables to run the QEMU VMM using a non-root user when rootless flag is set true in the configuration.
Signed-off-by: stevenfryto <sunzitai_1832@bupt.edu.cn>
This commit introduces generic support for running the VMM in rootless mode in runtime-rs:
1.Detect whether the VMM is running in rootless mode.
2.Before starting the VMM process, create a non-root user and launch the VMM with that user’s UID and GID; also add the KVM user's group ID to the VMM process's supplementary groups so the VMM process can access /dev/kvm.
3.Add the setup of the rootless directory located in the dir /run/user/<uid> directory, and modify some path variables to be functions that return the path with the rootless directory prefix when running in rootless mode.
Fixes: #11414
Signed-off-by: stevenfryto <sunzitai_1832@bupt.edu.cn>
We recently hit the following error during build:
```
RUN ssh-keygen -t ed25519 -f /etc/ssh/ssh_host_ed25519_key -P ""
OpenSSL version mismatch. Built against 3050003f, you have 30500010
```
This happened because `alpine:latest` moved forward and the `ssh-keygen`
binary in the base image was compiled against a newer OpenSSL version
that is not available at runtime.
Pinning the base image to the stable release (3.20) avoids the mismatch
and ensures consistent builds.
Signed-off-by: Hyounggyu Choi <Hyounggyu.Choi@ibm.com>
This change fixes clean up logic when running tests
in a vm booted with qemu wrt to qmp.sock & console.sock
files, and no longer assumes any path for them.
Signed-off-by: Sumedh Alok Sharma <sumsharma@microsoft.com>
Change the default block driver to virtio-scsi.
Since the latest qemu's commit:
https://gitlab.com/qemu-project/qemu/-/commit/
984a32f17e8dab0dc3d2328c46cb3e0c0a472a73
brings a bug for virtio-blk-pci with io_uring mode at line:
https://gitlab.com/qemu-project/qemu/-/commit/
984a32f17e8dab0dc3d2328c46cb3e0c0a472a73#
ce8eeb01f8b84f8cb8d3c35684d473fe1ee670f9_345_352
In order to avoid this issue, change the default block driver
to virtio-scsi.
Signed-off-by: Fupan Li <fupan.lfp@antgroup.com>
As OCI Spec annotation has been updated with adding or remove items,
we should use the updated annotation as the passed argument.
Signed-off-by: Alex Lyn <alex.lyn@antgroup.com>
This commit removes the InitData annotation from the OCI Spec's
annotations.
Similar to the Policy annotation, InitData is now exclusively handled
and transmitted to the guest via the sandbox's init data mechanism.
Removing this redundant and potentially large annotation simplifies the
OCI Spec and streamlines the guest initialization process.
This change aligns the handling of InitData with existing practices
within runtime-go.
Signed-off-by: Alex Lyn <alex.lyn@antgroup.com>
The repro below would show this error in the logs (in debug mode only):
fatal runtime error: IO Safety violation: owned file descriptor already closed
The issue was that the `pseudo.slave` file descriptor was being owned by
multiple variables simultaneously. When any of those variables would go out
of scope, they would close the same file descriptor, which is undefined
behavior.
To fix this, we clone: we create a new file descriptOR that refers to the same
file descriptION as the original. When the cloned descriptor is closed, this
affect neither the original descriptor nor the description. Only when the last
descriptor is closed does the kernel cleans up the description.
Note that we purposely consume (not clone) the original descriptor with
`child_stdin` as `pseudo` is NOT dropped automatically.
Repro
-----
Prerequisites:
- Use Rust 1.80+.
- Build the agent in debug mode.
$ cat busybox.yaml
apiVersion: v1
kind: Pod
metadata:
name: busybox
spec:
containers:
- image: busybox:latest
name: busybox
runtimeClassName: kata
$ kubectl apply -f busyboox.yaml
pod/busybox created
$ kubectl exec -it busybox -- sh
error: Internal error occurred: Internal error occurred: error executing
command in container: failed to exec in container: failed to start exec
"e6c602352849647201860c1e1888d99ea3166512f1cc548b9d7f2533129508a9":
cannot enter container 76a499cbf747b9806689e51f6ba35e46d735064a3f176f9be034777e93a242d5,
with err ttrpc: closed
Fixes: #11054
Signed-off-by: Aurélien Bombo <abombo@microsoft.com>
Log how much time "kubectl get pods" and each test case are taking,
just in case that will reveal unusually slow test clusters, and/or
opportunities to improve tests.
Signed-off-by: Dan Mihai <dmihai@microsoft.com>
In the previous commit we've added some code that broke `cargo fmt --
--check` without even noticing, as the code didn't go through the CI
process (due to it being a security advisory).
Signed-off-by: Fabiano Fidêncio <fabiano@fidencio.org>
Trailing slash in DEFAULT_KATA_GUEST_SANDBOX_DIR caused double slashes
in mount_point (e.g. "/run/kata-containers/sandbox//shm"), which failed
OPA strict equality checks against policy mount_point. Removing it aligns
generated paths with policy and fixes CreateSandboxRequest denial.
Signed-off-by: Alex Lyn <alex.lyn@antgroup.com>
coco-guest-components tarball is used as is for both vanilla coco
rootfs and the nvidia enabled rootfs. nvidia-attester can be built
without nvml so make it globally enabled for coco-guest-components.
Signed-off-by: Mikko Ylinen <mikko.ylinen@intel.com>
`allow_interactive_exec` requires a sandbox-name annotation, however
this is only added for pods by genpolicy. Other pod-generating resources
have unpredictable sandbox names.
This patch instead uses a regex for the sandbox name in genpolicy, based
on the specified metadata and following Kubernetes' naming logic. The
generated regex is then used in the policy to correctly match the
sandbox name.
Fixes: #11823
Signed-off-by: Charlotte Hartmann Paludo <git@charlotteharludo.com>
Co-authored-by: Paul Meyer <katexochen0@gmail.com>
Co-authored-by: Markus Rudy <mr@edgeless.systems>
This commit addresses an issue where base64 output, when used with a
default configuration, would introduce newlines, causing decoding to
fail on the runtime.
The fix ensures base64 output is a single, continuous line using the -w0
flag. This guarantees the encoded string is a valid Base64 sequence,
preventing potential runtime errors caused by invalid characters.
Note that: When you use the base64 command without any parameters, it
typically automatically adds newlines to the output, usually every 76 chars.
In contrast, base64 -w0 explicitly tells the command not to add any
newlines (-w for wrap, and 0 for a width of zero), which results in a
continuous string with no whitespace.
This is a critical distinction because if you pass a Base64 string with
newlines to a runtime, it may be treated as an invalid string, causing
the decoding process to fail.
Signed-off-by: Alex Lyn <alex.lyn@antgroup.com>
Correctly set dir's permissions and mode. This update ensures:
The dir_mode field of CopyFileRequest is set to DIR_MODE_PERMS
(equivalent to Go's 0o750 | os.ModeDir), which is primarily used for the
top-level directory creation permissions.
The file_mode field now directly uses metadata.mode() (equivalent to
Go's st.Mode) for the target entry.
This change aims to resolve potential permission issues or inconsistencies
during directory and file creation within the guest environment by precisely
matching the expected mode propagation of the Kata agent.
Signed-off-by: Alex Lyn <alex.lyn@antgroup.com>
The core purpose of introducing volume_manager to VolumeResource is to
centralize the management of shared file system volumes. By creating a
single VolumeManager instance within VolumeResource, all shared file
volumes are managed by one central entity. This single volume_manager
can accurately track the references of all ShareFsVolume instances to
the shared volumes, ensuring correct reference counting, proper volume
lifecycle management, and preventing issues like volumes being
overwritten.
This new design ensures that all shared volumes are managed by a central
entity, which:
(1) Guarantees correct reference counting.
(2) Manages the volume lifecycle correctly, avoiding issues like volumes
being overwritten.
Signed-off-by: Alex Lyn <alex.lyn@antgroup.com>
This commit integrates the new `VolumeManager` into the `ShareFsVolume`
lifecycle. Instead of directly copying files, `ShareFsVolume::new` now
uses the `VolumeManager` to get a guest path and determine if the volume
needs to be copied. It also updates the `cleanup` function to release
the volume's reference count, allowing the `VolumeManager` to manage its
state and clean up resources when no longer in use.
Signed-off-by: Alex Lyn <alex.lyn@antgroup.com>
This commit introduces a new `VolumeManager` to track the state of shared
volumes, including their reference count and its corresponding container
ids.
The manager's goal is to handle the lifecycle of shared filesystem volumes,
including:
(1) Volume State Tracking: Tracks the mapping from host source paths to guest
destination paths.
(2) Reference Counting: Manages reference counts for each volume, preventing
premature cleanup when multiple containers share the same source.
(3) Deterministic guest paths: Generates unique guest paths using random string
to avoid naming conflicts.
(4) Improved Management: Provides a centralized way to handle volume creation,
copying, and release, including aborting file watchers when volumes are no longer
in use.
Signed-off-by: Alex Lyn <alex.lyn@antgroup.com>
This commit refactors the `CopyFile` related code to streamline the
logic for creating guest directories and make the code structure
clearer.
Its main goal is to improve the overall maintainability and facilitate
future feature extensions.
Signed-off-by: Alex Lyn <alex.lyn@antgroup.com>
This commit is designed to perform a full sync before starting monitoring
to ensure that files which exist before monitoring starts are also synced.
Signed-off-by: Alex Lyn <alex.lyn@antgroup.com>
This commit updates the configuration for the initdata block
device to use the BlockDeviceAio::Native mode.
Signed-off-by: Alex Lyn <alex.lyn@antgroup.com>
This commit enhances control over block device AIO modes via hotplug.
Previously, hotplugging block devices was set with default AIO mode (io_uring).
Even if users reset the AIO mode in the configuration file, the changes would
not be correctly applied to individual block devices.
With this update, users can now explicitly configure the AIO mode for hot-plugging
block devices via the configuration, and those settings will be correctly applied.
Signed-off-by: Alex Lyn <alex.lyn@antgroup.com>
We need more information about block device, just relapce the original
method get_block_driver with get_block_device_info and return its
BlockDeviceInfo.
Signed-off-by: Alex Lyn <alex.lyn@antgroup.com>
- Add disable_selinux and selinux_label fields to hypervisor for SELinux support.
- Implement related SELinux support functions.
Fixes: #9866
Signed-off-by: Caspian443 <scrisis843@gmail.com>
Add container_exec_with_retries(), useful for retrying if needed
commands similar to:
kubectl exec <pod_name> -c <container_name> -- <command>
Signed-off-by: Dan Mihai <dmihai@microsoft.com>
On s390x, QEMU fails if maxmem is set to 0:
```
invalid value of maxmem: maximum memory size (0x0) must be at least the initial memory size
```
This commit sets maxmem to the initial memory size for s390x when hotplug is disabled,
resolving the error while still ensuring that memory hotplug remains off.
Signed-off-by: Hyounggyu Choi <Hyounggyu.Choi@ibm.com>
The setting '-m xM,slots=y,maxmem=zM' where maxmem is from
the host's memory capacity is failing with confidential VMs
on hosts having 1T+ of RAM.
slots/maxmem are necessary for setups where the container
memory is hotplugged to the VM during container creation based
on createContainer info.
This is not the case with CoCo since StaticResourceManagement
is enabled and memory hotplug flows have not been checked.
To avoid unexpeted errors with maxmem, disable slots/maxmem
in case ConfidentialGuest is requested.
Signed-off-by: Mikko Ylinen <mikko.ylinen@intel.com>
Let's make sure that whenever we try to access the attestation agent
binariy, we only proceed the startup in case:
* the binary is found (CoCo case)
* the binary is not present (non-CoCo case)
In case any error that's not `NotFound`, we should simply abort as that
could mean a potential tampering with the binary (which would be
reported as an EIO).
Signed-off-by: Fabiano Fidêncio <fidencio@northflank.com>
Add the annotation of OCI bundle path to store its path.
As it'll be checked within agent policy, we need add them
to pass agent policy validations.
Signed-off-by: Alex Lyn <alex.lyn@antgroup.com>
With the help of `update_ocispec_annotations`, we'll add the contaienr
type key with "io.katacontainers.pkg.oci.container_type" and its
corresponding type "pod_sandbox" when it's pause container and
"pod_container" when it's an other containers.
Signed-off-by: Alex Lyn <alex.lyn@antgroup.com>
It'll updates OCI annotations by removing specified keys and adding
new ones. This function creates a new `HashMap` containing the updated
annotations, ensuring that the original map remains unchanged.
It is optimized for performance by pre-allocating the necessary capacity
and handling removals and additions efficiently.
Signed-off-by: Alex Lyn <alex.lyn@antgroup.com>
To enable access to the constants `POD_CONTAINER` and `POD_SANDBOX` from
other crates, their visibility has been updated to public. This change
addresses the previous limitation of restricted access and ensures these
values can be utilized across the codebase.
Signed-off-by: Alex Lyn <alex.lyn@antgroup.com>
Add the annotation of nerdctl network namespace to let nerdctl know which namespace
to use when calling the selected CNI plugin with "nerdctl/network-namespace".
As it'll be checked within agent policy, we need add them to pass agent policy validations.
Signed-off-by: Alex Lyn <alex.lyn@antgroup.com>
On 63f6dcdeb9 we added the support to
download either a .xz or a .zst tarball file. However, we missed adding
the code to properly unpack a .zst tarball file.
Signed-off-by: Fabiano Fidêncio <fabiano@fidencio.org>
agent-ctl's make check has been failing with:
```
Checking kata-agent-ctl v0.0.1 (/home/ubuntu/runner/_layout/_work/kata-containers/kata-containers/src/tools/agent-ctl)
error[E0432]: unresolved import `hypervisor::ch`
--> src/vm/vm_ops.rs:10:5
|
10 | ch::CloudHypervisor,
| ^^ could not find `ch` in `hypervisor`
|
note: found an item that was configured out
--> /home/ubuntu/runner/_layout/_work/kata-containers/kata-containers/src/runtime-rs/crates/hypervisor/src/lib.rs:30:9
|
30 | pub mod ch;
| ^^
note: the item is gated here
--> /home/ubuntu/runner/_layout/_work/kata-containers/kata-containers/src/runtime-rs/crates/hypervisor/src/lib.rs:26:1
|
26 | / #[cfg(all(
27 | | feature = "cloud-hypervisor",
28 | | any(target_arch = "x86_64", target_arch = "aarch64")
29 | | ))]
| |___^
```
Let's just make sure that we include ch conditionally as well.
Signed-off-by: Fabiano Fidêncio <fabiano@fidencio.org>
Docker containers support specifying the shm size using the --shm-size
option and support sandbox-level shm volumes, so we've added support for
shm volumes. Since Kubernetes doesn't support specifying the shm size,
it typically uses a memory-based emptydir as the container's shm, and
its size can be specified.
Signed-off-by: Fupan Li <fupan.lfp@antgroup.com>
Since the rate limiter would be shared by cloud-hypervisor
and firecracker etc, thus move it from clh's config to
hypervisor config crate which would be shared by other vmm.
Signed-off-by: Fupan Li <fupan.lfp@antgroup.com>
Given that Rust-based VMMs like cloud-hypervisor, Firecracker, and
Dragonball naturally offer user-level block I/O rate limiting, I/O
throttling has been implemented to leverage this capability for these
VMMs. This PR specifically introduces support for cloud-hypervisor.
Signed-off-by: Fupan Li <fupan.lfp@antgroup.com>
This commit introduces changes to parse the PciDeviceInfo received
in response payload when adding a network device to the VM with cloud
hypervisor. When hotplugging a network device for a given endpoint,
it rightly sets the PciPath of the plugged-in device in the endpoint.
In calls like virtcontainers/sandbox.go:AddInterface, the later call
to agent sends the pci info for uevents (instead of empty value) to
rightly update the interfaces instead of failing with `Link not found`
Signed-off-by: Sumedh Alok Sharma <sumsharma@microsoft.com>
Support for the share-rw=true parameter has been added. While this
parameter is essential for maintaining data consistency across multiple
QEMU instances sharing a backend disk image, its implementation also
serves to standardize parameters with the block device hotplug
functionality in kata-runtime/qemu.
Signed-off-by: Alex Lyn <alex.lyn@antgroup.com>
Docker tests have been broken for a while and should be removed if we
cannot maintain those.
For now, though, let's limit it to run only with one hypervisor and
avoid wasting resources for no reason.
Signed-off-by: Fabiano Fidêncio <fabiano@fidencio.org>
devmapper tests have been failing for a while. It's been breaking on the
kata-deploy deployment, which is most likely related to Disk Pressure.
Removing files was not enough to get the tests to run, so we'll just run
those with QEMU as a way to test fixes. Once we get the test working,
we can re-enable the other VMMs, but for now let's just not waste
resources for no reason.
Signed-off-by: Fabiano Fidêncio <fabiano@fidencio.org>
add_allow_all_policy_to_yaml now also sets the initdata annotation. So don't overwrite the
initdata annotation that was previously set by create_coco_pod_yaml_with_annotations.
Signed-off-by: Saul Paredes <saulparedes@microsoft.com>
Delete annotation from OCI spec and sandbox config. This is done after the optional initdata annotation value has been read.
Signed-off-by: Saul Paredes <saulparedes@microsoft.com>
Bump cri-o to 1.34.0 to try and remediate security advisories
CVE-2025-0750 and CVE-2025-4437.
Note: Running
```
go get github.com/cri-o/cri-o@v1.34.0
```
seems to bump a lot of other go modules, hence the size of the
vendor diff
Signed-off-by: stevenhorsman <steven@uk.ibm.com>
In google.golang.org/grpc v1.72.0, `DialContext`, is deprecated, so
switch to use `NewClient` instead.
`grpc.WithBlock()` is deprecated and not recommend, so remove this
Signed-off-by: stevenhorsman <steven@uk.ibm.com>
In `github.com/prometheus/common v0.62.0` expfmt.FmtText
is deprecated, so replace with `expfmt.NewFormat(expfmt.TypeTextPlain)`.
Signed-off-by: stevenhorsman <steven@uk.ibm.com>
In `github.com/prometheus/common v0.62.0` expfmt.FmtText
is deprecated, so replace with `expfmt.NewFormat(expfmt.TypeTextPlain)`.
Signed-off-by: stevenhorsman <steven@uk.ibm.com>
This is following Steve's suggestion, based on what's been done on
cloud-api-adaptor.
The reason we're doing it here is because we've seen pods being evicted
due to disk pressure.
Signed-off-by: Fabiano Fidêncio <fabiano@fidencio.org>
I've hit this when using a machine with slow internet connection, which
took ages to download the kata-cleanup image, and then helm timed out in
the middle of the cleanup, leading to the cleanup job being restarted
and then bailing with an error as the runtimeclasses that kata-deploy
tries to delete were already deleted.
Signed-off-by: Fabiano Fidêncio <fabiano@fidencio.org>
Since the cloud hypervisor's resize vCPU is an asynchronous operation,
it's possible that the previous resize operation hasn't completed when
the request is sent, causing the current call to return an error.
Therefore, several retries can be performed to avoid this error.
Signed-off-by: Fupan Li <fupan.lfp@antgroup.com>
A minor release of QEMU is out, so update to it for fixes and features.
QEMU changelog: https://wiki.qemu.org/ChangeLog/10.1
Notes:
* AVX support is not an option to be enabled / disabled anymore.
* Passt requires Glibc 2.40.+, which means a dependency on Ubuntu 25.04
or newer, thus we're disabling it.
Signed-off-by: Alex Tibbles <alex@bleg.org>
Although versions of slab prior to 0.4.10, don't have a security
vulnearability, we can bump them all to keep things in sync
Signed-off-by: stevenhorsman <steven@uk.ibm.com>
```
Experimental fw_cfg Device Support
This feature enables passing configuration data and files, such as VM
boot configurations (kernel, kernel cmdline, e820 memory map, and ACPI
tables), from the host to the guest. (#7117)
Experimental ivshmem Device Support
Support for inter-VM shared memory has been added. For more information,
please refer to the ivshmem documentation. (#6703)
Firmware Boot Support on riscv64
In addition to direct kernel boot, firmware boot support has been added
on riscv64 hosts. (#7249)
Increased vCPU Limit on x86_64/kvm
The maximum number of supported vCPUs on x86_64 hosts using KVM has been
raised from 254 to 8192. (#7299)
Improved Block Performance with Small Block Sizes
Performance for virtio-blk with small block sizes (16KB and below)
is enhanced via submitting async IO requests in batches. (#7146)
Faster VM Pause Operation
The VM pause operation now is significantly faster particularly for VMs
with a large number of vCPUs. (#7290)
Updated Documentation on Windows Guest Support
Our Windows documentation now includes instructions to run Windows 11
guests, in addition to Windows Server guests. (#7218)
Policy on AI Generated Code
We will decline any contributions known to contain contents generated or
derived from using Large Language Models (LLMs). Details can be found
in our contributing documentation. (#7162)
Removed SGX Support
The SGX support has been removed, as announced in the deprecation notice two
release cycles ago. (#7093)
Notable Bug Fixes
Seccomp filter fixes with glibc v2.42 (#7327)
Various fixes related to (#7331, #7334, #7335)
```
From https://github.com/cloud-hypervisor/cloud-hypervisor/releases/tag/v48.0
Signed-off-by: Fabiano Fidêncio <fabiano@fidencio.org>
Exclude 'cgroup' namespace from namespace checks during `allow_linux`
validation. This complements the existing exclusion of the 'network'
namespace.
As runtime-rs has specific cgroup namespace configurations, and excluding it from
policy validation ensures parity between runtime-rs and runtime-go implementations.
This allows focusing validation on critical namespaces like PID, IPC, and MNT, while
avoiding potential policy mismatches due to another cgroup namespace management by
the runtime-rs.
Signed-off-by: Alex Lyn <alex.lyn@antgroup.com>
Add `normalize_namespace_type()` function to map "mount"
(case-insensitive) to "mnt" while keeping other values unchanged.
This ensures namespace comparisons treat "mount" and "mnt" as
equivalent.
Signed-off-by: Alex Lyn <alex.lyn@antgroup.com>
- Use set comparison to ignore ordering differences when matching
capabilities.
- Add normalization to strip "CAP_" prefix to support both CAP_XXX and
XXX formats.
This makes capability matching more robust against different ordering
and naming formats.
Signed-off-by: Alex Lyn <alex.lyn@antgroup.com>
As prvious configure with overlayfs is incorrect, which causes the agent
policy validation failure. And it's also different with runtime-go's
configuration. In this patch, we'll correct its fstype with overlay and
align with runtime on this matter.
Signed-off-by: Alex Lyn <alex.lyn@antgroup.com>
When hot-removing a block device, the kernel must first unmount the
device and then destroy it on the VM. Therefore, a
prepare_remove_block_device procedure must be added to wait for the
kernel to unmount the device before destroying it on the VM.
Signed-off-by: Fupan Li <fupan.lfp@antgroup.com>
When hot-removing a block device, the kernel must first remove the
device and then destroy it on the VM. Therefore, a
prepare_remove_block_device procedure must be added to wait for the
kernel to unmount the device before destroying it on the VM.
Signed-off-by: Fupan Li <fupan.lfp@antgroup.com>
Since Dragonball's MMIO bus only supports legacy interrupts, while
the PCI bus supports MSIX interrupts, to improve performance for block
devices, virtio-blk devices are set to PCI bus mode by default.
We had tested the virtio-blk's performance using the fio with the
following commands:
fio -filename=./test -direct=1 -iodepth 32 -thread -rw=randrw
-rwmixread=50 -ioengine=libaio -bs=4k -size=10G -numjobs=4
-group_reporting -name=mytest
When used the legacy interrupt, the io test is as below:
read : io=20485MB, bw=195162KB/s, iops=48790, runt=107485msec
write: io=20475MB, bw=195061KB/s, iops=48765, runt=107485msec
Once switched to msix innterrupt, the io test is as below:
read : io=20485MB, bw=260862KB/s, iops=65215, runt= 80414msec
write: io=20475MB, bw=260727KB/s, iops=65181, runt= 80414msec
We can get 34% performance improvement.
Signed-off-by: Fupan Li <fupan.lfp@antgroup.com>
Added support for PCI buses for virtio-blk devices. This commit adds
support for PCI buses for both cold-plugged and hot-plugged
virtio-blk devices. Furthermore, during hot-plugging, support is added for
synchronous waiting for hot-plug completion. This ensures that multiple devices
can be hot-plugged successfully without causing upcall busy errors.
Signed-off-by: Fupan Li <fupan.lfp@antgroup.com>
In order to support the pci bus for virtio devices,
move the pci system manager from vfio manager to
device manager, thus it can be shared by both of
vfio and virtio pci devices.
Signed-off-by: Fupan Li <fupan.lfp@antgroup.com>
Use @DEFENABLEANNOTATIONS_COCO@ in configuration-qemu-snp.toml,
for consistency with the tdx and coco-dev configuration files.
k8s-initdata.bats was failing during CI on SNP without this change,
because the cc_init_data annotation was disabled.
Signed-off-by: Dan Mihai <dmihai@microsoft.com>
This runs Zizmor on pushes to any branch, not just main.
This is useful for:
1. Testing changes in feature branches with the manually-triggered CI.
2. Forked repos that may use a different name than "main" for their
default branch.
Signed-off-by: Aurélien Bombo <abombo@microsoft.com>
Test case for `get_uds_with_sid` with an empty run directory would not
hit the 0 match arm, i.e. "sandbox with the provided prefix {short_id:?}
is not found", because `get_uds_with_sid` will try to create the
directory with provided short id before detecting `target_id`.
Signed-off-by: Ruoqing He <heruoqing@iscas.ac.cn>
Preset directory `kata98654sandboxpath1` will produce more than one
`target_id` in `get_uds_with_sid`, which causes test to fail. Remove
that directory to make this test work.
Signed-off-by: Ruoqing He <heruoqing@iscas.ac.cn>
`test_arch_guest_protection_*` test cases get triggered simultaneously,
which is impossible for a single machine to pass. Modify tests to detect
protection file before preceding.
Signed-off-by: Ruoqing He <heruoqing@iscas.ac.cn>
Case 4 of `test_execute_hook` would fail because `args` could not be
empty, while by providing `build_oci_hook` with `vec![]` would result in
empty args at execution stage.
Modify `build_oci_hook` to set args as `None` when empty vector is
provided.
Signed-off-by: Ruoqing He <heruoqing@iscas.ac.cn>
The variable `addr` was used to store the log level string read from the
`LOG_LEVEL_ENV_VAR` environment variable. This name is misleading as it
implies a network address rather than a log level value.
This commit renames the variable to `level` to more accurately reflect
its purpose, improving the overall readability of the configuration code.
A minor whitespace formatting fix in a macro is also included.
Signed-off-by: Liang, Ma <liang3.ma@intel.com>
A new internal nightly test has been established for runtime-rs.
This commit adds a new entry `cc-se-e2e-tests-rs` to the existing
matrix and renames the existing entry `cc-se-e2e-tests` to
`cc-se-e2e-tests-go`.
Signed-off-by: Hyounggyu Choi <Hyounggyu.Choi@ibm.com>
Tests skipped because tests for `qemu-se` are skipped:
- k8s-empty-dirs.bats
- k8s-inotify.bats
- k8s-shared-volume.bats
Tests skipped because tests for `qemu-runtime-rs` are skipped:
- k8s-block-volume.bats
- k8s-cpu-ns.bats
- k8s-number-cpus.bats
Let's skip the tests above to run the nightly test
for runtime-rs on IBM SEL.
Signed-off-by: Hyounggyu Choi <Hyounggyu.Choi@ibm.com>
There are still some issues to be address before we can mark `make test`
for `libs` as required. Mark this case as not required temporarily.
Signed-off-by: Ruoqing He <heruoqing@iscas.ac.cn>
SNP launch was failing after the confidential guest kernel was upgraded to 6.16.1.
Added required module CONFIG_MTRR enabled.
Added required module CONFIG_X86_PAT enabled.
Fixes: #11779
Signed-off-by: Ryan Savino <ryan.savino@amd.com>
Bump the version of runtime-rs' hypervisor crate
to upgrade (indirectly) protobug and remediate vulnerability
RUSTSEC-2024-0437
Signed-off-by: stevenhorsman <steven@uk.ibm.com>
The previous document about the default of create_container_timeout
is 30,000 millseconds which not keep alignment with runtime-go.
In this commit, we'll change it as 30 seconds.
Signed-off-by: Alex Lyn <alex.lyn@antgroup.com>
Since it aligns with the create_container_timeout definition in
runtime-go, we need to set the value in configuration.toml in seconds,
not milliseconds. We must also convert it to milliseconds when the
configuration is loaded for request_timeout_ms.
Signed-off-by: Alex Lyn <alex.lyn@antgroup.com>
It's possible that tests take a long time to run and hence that the access
token expires before we delete the cluster. In this case `az cli` will try
to refresh the access token using the OIDC token (which will have
definitely also expired because its lifetime is ~5 minutes).
To address this we refresh the OIDC token manually instead. Automatic
refresh isn't supported per Azure/azure-cli#28708.
Fixes: #11758
Signed-off-by: Aurélien Bombo <abombo@microsoft.com>
Introduce new test case in k8s-iptables.bats which verifies that
workloads can configure iptables in the UVM.
Users discovered that they weren't able to do this for common usecases
such as istio. Proper support for this should be built into UVM
kernels. This test ensures that current and future kernel
configurations don't regress this functionality.
Signed-off-by: Cameron Baird <cameronbaird@microsoft.com>
Currently, the UVM kernel fails for istio deployments (at least with the
version we tested, 1.27.0). This is because the istio sidecar container
uses ip6tables and the required kernel configs are not built-in:
```
iptables binary ip6tables has no loaded kernel support and cannot be used, err: exit status 3 out: ip6tables v1.8.10 (legacy):
can't initialize ip6tables table `filter': Table does not exist (do you need to insmod?)
Perhaps ip6tables or your kernel needs to be upgraded.
```
Signed-off-by: Cameron Baird <cameronbaird@microsoft.com>
In certain scenarios, particularly under CoCo/Agent Policy enforcement,
the default initial value of `Linux.Resources.Devices` is considered
non-compliant, leading to container creation failures. To address this
issue and ensure consistency with the behavior in `runtime-go`, this
commit removes the default value of `Linux.Resources.Devices` from the
OCI Spec.
This cleanup ensures that the OCI Spec aligns with runtime expectations
and prevents policy violations during container creation.
Signed-off-by: Alex Lyn <alex.lyn@antgroup.com>
Only the StartContainer hook needs to be reserved for execution in the
guest, but we also make sure that the setting happens only when the OCI
Hooks does exist, otherwise we do nothing.
Signed-off-by: Alex Lyn <alex.lyn@antgroup.com>
In k8s-guest-pull-image.bats, `failed to pull image` is
not caught by assert_logs_contain() for runtime-rs.
To ensure consistency, this commit changes `failed` to
`Failed`, which works for both runtimes.
Signed-off-by: Hyounggyu Choi <Hyounggyu.Choi@ibm.com>
Mount validation for sealed secret requires the base path to start with
`/run/kata-containers/shared/containers`. Previously, it used
`/run/kata-containers/sandbox/passthrough`, which caused test
failures where volume mounts are used.
This commit renames the path to satisfy the validation check.
Signed-off-by: Hyounggyu Choi <Hyounggyu.Choi@ibm.com>
ef642fe890 added a special case to avoid
moving cgroups that are on the "default" slice in case of deletion.
However, this special check should be done in the Parent() method
instead, which ensures that the default resource controller ID is
returned, instead of ".".
Fixes: #11599
Signed-off-by: Fabiano Fidêncio <fidencio@northflank.com>
- Set guest Storage.options for block rootfs to empty (do not propagate host mount options).
- Align behavior with Go runtime: only add xfs nouuid when needed.
Signed-off-by: Caspian443 <scrisis843@gmail.com>
We moved to `.zst`, but users still use the upstream kata-manager to
download older versions of the project, thus we need to support both
suffixes.
Fixes: #11714
Signed-off-by: Fabiano Fidêncio <fidencio@northflank.com>
Similar to what we've done for Cloud Hypervisor in the commit
9f76467cb7, we're backporting a runtime-rs
feature that would be benificial to have as part of the go runtime.
This allows users to use virito-balloon for the hypervisor to reclaim
memory freed by the guest.
Signed-off-by: Fabiano Fidêncio <fidencio@northflank.com>
The default suggestion for top-level permissions was
`contents: read`, but scorecard notes anything other than empty,
so try updating it and see if there are any issues. I think it's
only needed if we run workflows from other repos.
Signed-off-by: stevenhorsman <steven@uk.ibm.com>
Since the previous tightening a few workflow updates have
gone in and the zizmor job isn't flagging them as issues,
so address this to remove potential attack vectors
Signed-off-by: stevenhorsman <steven@uk.ibm.com>
This reverts commit cb5f143b1b, as the
cached packages have been regenerated after the switch to using zstd.
Signed-off-by: Fabiano Fidêncio <fidencio@northflank.com>
As part of the go 1.24.6 bump there are errors about the incorrect
use of a errorf, so switch to the non-formatting version, or add
the format string as appropriate
Signed-off-by: stevenhorsman <steven@uk.ibm.com>
golang 1.25 has been released, so 1.23 is EoL,
so we should update to ensure we don't end up with security issues
Signed-off-by: stevenhorsman <steven@uk.ibm.com>
Update the two workflows that used setup-go to
instead call `install_go.sh` script, which handles
installing the correct version of golang
Signed-off-by: stevenhorsman <steven@uk.ibm.com>
`${kernel_name,,}` is bash 4.0 and not posix compliant, so doesn't
work on macos, so switch to `tr` which is more widely
supported
Signed-off-by: stevenhorsman <steven@uk.ibm.com>
In #11693 the cc_init_data annotation was changes to be hypervisor
scoped, so each hypervisor needs to explicitly allow it in order to
use it now, so add this to both the go and rust runtime's remote
configurations
Signed-off-by: stevenhorsman <steven@uk.ibm.com>
We need to get the root_hash.txt file from the image build, otherwise
there's no way to build the shim using those values for the
configuration files.
Signed-off-by: Fabiano Fidêncio <fidencio@northflank.com>
Although the compress ratio is not as optimal as using xz, it's way
faster to compress / uncompress, and it's "good enough".
This change is not small, but it's still self-contained, and has to get
in at once, in order to help bisects in the future.
Signed-off-by: Fabiano Fidêncio <fabiano@fidencio.org>
As 3.18 is already EOL.
We need to add `--break-system-packages` to enforce the install of the
installation of the yq version that we rely on. The tests have shown
that no breakage actually happens, fortunately.
Signed-off-by: Fabiano Fidêncio <fabiano@fidencio.org>
Currently, we change vm_rootfs_driver as the initdata device driver
with block_device_driver.
Fixes#11697
Signed-off-by: Alex Lyn <alex.lyn@antgroup.com>
we also need support initdat within nonprotection even though the
platform is detected as NonProtection or usually is called nontee
host. Within these cases, there's no need to validate the item of
`confidential_guest=true`, we believe the result of the method
`available_guest_protection()?`.
Fixes#11697
Signed-off-by: Alex Lyn <alex.lyn@antgroup.com>
The default `reconnect_timeout` (3 seconds) was found to be insufficient for
IBM SEL when using VSOCK. This commit updates the timeouts as follows:
- `dial_timeout_ms`: Set to 90ms to match the value used in go-runtime for IBM SEL
- `reconnect_timeout_ms`: Increased to 5000ms based on empirical testing
Signed-off-by: Hyounggyu Choi <Hyounggyu.Choi@ibm.com>
Add support for the `InitData` resource config on IBM SEL,
so that a corresponding block device is created and the
initdata is passed to the guest through this device.
Note that we skip passing the initdata hash via QEMU’s
object, since the hypervisor does not yet support this
mechanism for IBM SEL. It will be introduced separately
once QEMU adds the feature.
Signed-off-by: Hyounggyu Choi <Hyounggyu.Choi@ibm.com>
Linux v6.16 brings some useful features for the confidential guests.
Most importantly, it adds an ABI to extend runtime measurement registers
(RTMR) for the TEE platforms supporting it. This is currently enabled
on Intel TDX only.
The kernel version bump from v6.12.x to v6.16 forces some CONFIG_*
changes too:
MEMORY_HOTPLUG_DEFAULT_ONLINE was dropped in favor of more config
choices. The equivalent option is MHP_DEFAULT_ONLINE_TYPE_ONLINE_AUTO.
X86_5LEVEL was made unconditional. Since this was only a TDX
configuration, dropping it completely as part of v6.16 is fine.
CRYPTO_NULL2 was merged with CRYPTO_NULL. This was only added in
confidential guest fragments (cryptsetup) so we can drop it in this update.
CRYPTO_FIPS now depends on CRYPTO_SELFTESTS which further depends on
EXPERT which we don't have. Enable both in a separate config fragment
for confidential guests. This can be moved to a common setting once
other targets bump to post v6.16.
CRYPTO_SHA256_SSE3 arch optimizations were reworked and are now enabled
by default. Instead of adding it to whitelist.conf, just drop it completely
since it was only enabled as part of "measured boot" feature for
confidential guests. CONFIG_CRYPTO_CRC32_S390 was reworked the same way.
In this case, whitelist.conf is needed.
Signed-off-by: Mikko Ylinen <mikko.ylinen@intel.com>
This reverts commit ede773db17.
`cc_init_data` should be under a hypervisor category because
it is a hypervisor-specific feature. The annotation including
`runtime` also breaks a logic for `is_annotation_enabled()`.
Signed-off-by: Hyounggyu Choi <Hyounggyu.Choi@ibm.com>
We need to include `cc_init_data` in the enable_annotations
array to pass the data. Since initdata is a CoCo-specific
feature, this commit introduces a new array,
`DEFENABLEANNOTATIONS_COCO`, which contains the required
string and applies it to the relevant CoCo configuration.
Signed-off-by: Hyounggyu Choi <Hyounggyu.Choi@ibm.com>
Currently, there are 2 issues for the empty initdata annotation
test:
- Empty string handling
- "\[CDH\] \[ERROR\]: Get Resource failed" not appearing
`add_hypervisor_initdata_overrides()` does not handle
an empty string, which might lead to panic like:
```
called `Result::unwrap()` on an `Err` value: gz decoder failed
Caused by:
failed to fill whole buffer
```
This commit makes the function return an empty string
for a given empty input and updates the assertion string
to one that appears in both go-runtime and runtime-rs.
Signed-off-by: Alex Lyn <alex.lyn@antgroup.com>
Signed-off-by: Hyounggyu Choi <Hyounggyu.Choi@ibm.com>
Currently, runtime-rs related code within the libs directory lacks
sufficient CI protection. We frequently observe the following issues:
- Inconsistent Code Formatting: Code that has not been properly
formatted
is merged.
- Failing Tests: Code with failing unit or integration tests is merged.
To address these issues, we need introduce stricter CI checks for the
libs directory. This may specifically include:
- Code Formatting Checks
- Mandatory Test Runs
Fixes#11512
Signed-off-by: alex.lyn <alex.lyn@antgroup.com>
To make it aligned with the setting of runtime-go, we should keep
it as empty when users doesn't enable and set its specified path.
Signed-off-by: Alex Lyn <alex.lyn@antgroup.com>
We need to make sure that we use the latest kernel
and rebuild the initrd and image for the nvidia-gpu
use-cases otherwise the tests will fail since
the modules are not build against the new kernel and
they simply fail to load.
Signed-off-by: Zvonko Kaiser <zkaiser@nvidia.com>
OSV-Scanner highlights go.mod references to go stdlib 1.23.0 contrary to intention in versions.yaml, so synchronize them.
Make a converse comment for versions.yaml.
Fixes: #11700
Signed-off-by: Alex Tibbles <alex@bleg.org>
Let's rename the runtime-rs initdata annotation from
`io.katacontainers.config.runtime.cc_init_data` to
`io.katacontainers.config.hypervisor.cc_init_data`.
Rationale:
- initdata itself is a hypervisor-specific feature
- the new name aligns with the annotation handling logic:
c92bb1aa88/src/libs/kata-types/src/annotations/mod.rs (L514-L968)
This commit updates the annotation for go-runtime and tests accordingly.
Signed-off-by: Hyounggyu Choi <Hyounggyu.Choi@ibm.com>
Enable testing of initdata on the qemu-coco-dev and qemu-se
runtime classes, so we can validate the function on s390x
Signed-off-by: stevenhorsman <steven@uk.ibm.com>
This commit support the seccomp_sandbox option from the configuration.toml file
and add the logic for appending command-line arguments based on this new configuration parameter.
Fixes: #11524
Signed-off-by: wangxinge <wangxinge@bupt.edu.cn>
Previouly it is reusing the ovmf, which will enter some
issue for path checking, so move to aavmf as it should
be.
Signed-off-by: Kevin Zhao <kevin.zhao@linaro.org>
Read only the sealed secret prefix instead of the whole file.
Improves performance and reduces memory usage in I/O-heavy environments.
Fixes: #11643
Signed-off-by: Park.Jiyeon <jiyeonnn2@icloud.com>
Dependening on the platform configuration, users might want to
set a more secure policy than the QEMU default.
Signed-off-by: Paul Meyer <katexochen0@gmail.com>
This change introduces a new command line option `--vm`
to boot up a pod VM for testing. The tool connects with
kata agent running inside the VM to send the test commands.
The tool uses `hypervisor` crates from runtime-rs for VM
lifecycle management. Current implementation supports
Qemu & Cloud Hypervisor as VMMs.
In summary:
- tool parses the VMM specific runtime-rs kata config file in
/opt/kata/share/defaults/kata-containers/runtime-rs/*
- prepares and starts a VM using runtime-rs::hypervisor vm APIs
- retrieves agent's server address to setup connection
- tests the requested commands & shutdown the VM
Fixes#11566
Signed-off-by: Sumedh Alok Sharma <sumsharma@microsoft.com>
The seccomp feature for Cloud Hypervisor and Firecracker is enabled by default.
This commit introduces an option to disable seccomp for both and updates the built-in configuration.toml file accordingly.
Fixes: #11535
Signed-off-by: wangxinge <wangxinge@bupt.edu.cn>
Route kata-shim logs directly to systemd-journald under 'kata' identifier.
This refactoring enables `kata-shim` logs to be properly attributed to
'kata' in systemd-journald, instead of inheriting the 'containerd'
identifier.
Previously, `kata-shim` logs were challenging to filter and debug as
they
appeared under the `containerd.service` unit.
This commit resolves this by:
1. Introducing a `LogDestination` enum to explicitly define logging
targets (File or Journal).
2. Modifying logger creation to set `SYSLOG_IDENTIFIER=kata` when
logging
to Journald.
3. Ensuring type safety and correct ownership handling for different
logging backends.
This significantly enhances the observability and debuggability of Kata
Containers, making it easier to monitor and troubleshoot Kata-specific
events.
Fixes: #11590
Signed-off-by: Hyounggyu Choi <Hyounggyu.Choi@ibm.com>
Signed-off-by: Alex Lyn <alex.lyn@antgroup.com>
After moving Arm64 CI nodes to new one, we do faced an interesting
issue for timeout when it executes the command with crictl runp,
the error is usally: code = DeadlineExceeded
Fixes: #11662
Signed-off-by: Kevin Zhao <kevin.zhao@linaro.org>
This series should make runtime-rs's vcpu allocation behaviour match the
behaviour of runtime-go so we can now enable pertinent tests which were
skipped so far due the difference between both shims.
Signed-off-by: Pavel Mores <pmores@redhat.com>
Configuration information is adjusted after loading from file but so
far, there has been no similar check for configuration coming from
annotations. This commit introduces re-adjusting config after
annotations have been processed.
A small refactor was necessary as a prerequisite which introduces
function TomlConfig::adjust_config() to make it easier to invoke
the adjustment for a whole TomlConfig instance. This function is
analogous to the existing validate() function.
The immediate motivation for this change is to make sure that 0
in "default_vcpus" annotation will be properly adjusted to 1 as
is the case if 0 is loaded from a config file. This is required
to match the golang runtime behaviour.
Signed-off-by: Pavel Mores <pmores@redhat.com>
Also included (as commented out) is a test that does not pass although
it should. See source code comment for explanation why fixing this seems
beyond the scope of this PR.
Signed-off-by: Pavel Mores <pmores@redhat.com>
This commit focuses purely on the formal change of type. If any subsequent
changes in semantics are needed they are purposely avoided here so that the
commit can be reviewed as a 100% formal and 0% semantic change.
Signed-off-by: Pavel Mores <pmores@redhat.com>
This commit addresses a part of the same problem as PR #7623 did for the
golang runtime. So far we've been rounding up individual containers'
vCPU requests and then summing them up which can lead to allocation of
excess vCPUs as described in the mentioned PR's cover letter. We address
this by reversing the order of operations, we sum the (possibly fractional)
container requests and only then round up the total.
We also align runtime-rs's behaviour with runtime-go in that we now
include the default vcpu request from the config file ('default_vcpu')
in the total.
We diverge from PR #7623 in that `default_vcpu` is still treated as an
integer (this will be a topic of a separate commit), and that this
implementation avoids relying on 32-bit floating point arithmetic as there
are some potential problems with using f32. For instance, some numbers
commonly used in decimal, notably all of single-decimal-digit numbers
0.1, 0.2 .. 0.9 except 0.5, are periodic in binary and thus fundamentally
not representable exactly. Arithmetics performed on such numbers can lead
to surprising results, e.g. adding 0.1 ten times gives 1.0000001, not 1,
and taking a ceil() results in 2, clearly a wrong answer in vcpu
allocation.
So instead, we take advantage of the fact that container requests happen
to be expressed as a quota/period fraction so we can sum up quotas,
fundamentally integral numbers (possibly fractional only due to the need
to rewrite them with a common denominator) with much less danger of
precision loss.
Signed-off-by: Pavel Mores <pmores@redhat.com>
When hot-plugging CPUs on QEMU, we send a QMP command with JSON
arguments. QEMU 9.2 recently became more strict[1] enforcing the
JSON schema for QMP parameters. As a result, running Kata Containers
with QEMU 9.2 results in a message complaining that the core-id
parameter is expected to be an integer:
```
qmp hotplug cpu, cpuID=cpu-0 socketID=1, error:
QMP command failed:
Invalid parameter type for 'core-id', expected: integer
```
Fix that by changing the core-id, socket-id and thread-id to be
integer values.
[1]: be93fd5372Fixes: #11633
Signed-off-by: Christophe de Dinechin <dinechin@redhat.com>
As we have changed the initdata annotation definition, Accordingly, we also
need correct its const definition with KATA_ANNO_CFG_RUNTIME_INIT_DATA.
Signed-off-by: Alex Lyn <alex.lyn@antgroup.com>
This adds SECURITY.md to the list of GH-native files that should be excluded by
the reference checker.
Today this is useful for downstreams who already have a SECURITY.md file for
compliance reasons. When Kata onboards that file, this commit will also be
required.
Signed-off-by: Aurélien Bombo <abombo@microsoft.com>
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>
To make it work within CI, we do alignment with kata-runtime's definition
with "io.katacontainers.config.runtime.cc_init_data".
Signed-off-by: Alex Lyn <alex.lyn@antgroup.com>
Print more details about the behavior of "kubectl logs", trying to understand
errors like:
https://github.com/kata-containers/kata-containers/actions/runs/16662887973/job/47164791712
not ok 1 Check the number vcpus are correctly allocated to the sandbox
(in test file k8s-sandbox-vcpus-allocation.bats, line 37)
`[ `kubectl logs ${pods[$i]}` -eq ${expected_vcpus[$i]} ]' failed with status 2
No resources found in kata-containers-k8s-tests namespace.
...
k8s-sandbox-vcpus-allocation.bats: line 37: [: -eq: unary operator expected
Signed-off-by: Dan Mihai <dmihai@microsoft.com>
This auto-detects the repo by default (instead of having to specify
KATA_DEV_MODE=true) so that forked repos can leverage the static-checks.yaml CI
check without modification.
An alternative would have been to pass the repo in static-checks.yaml. However,
because of the matrix, this would've changed the check name, which is a pain to
handle in either the gatekeeper/GH UI.
Example fork failure:
https://github.com/microsoft/kata-containers/actions/runs/16656407213/job/47142421739#step:8:75
I've tested this change to work in a fork.
Signed-off-by: Aurélien Bombo <abombo@microsoft.com>
In order to have a reproducible code generation process, we need to pin
the versions of the tools used. This is accomplished easiest by
generating inside a container.
This commit adds a container image definition with fixed dependencies
for Golang proto/ttrpc code generation, and changes the agent Makefile
to invoke the update-generated-proto.sh script from within that
container.
Signed-off-by: Markus Rudy <mr@edgeless.systems>
The generated Go bindings for the agent are out of date. This commit
was produced by running
src/agent/src/libs/protocols/hack/update-generated-proto.sh with
protobuf compiler versions matching those of the last run, according to
the generated code comments.
Since there are new RPC methods, those needed to be added to the
HybridVSockTTRPCMockImp.
Signed-off-by: Markus Rudy <mr@edgeless.systems>
Updated versions.yaml to use Firecracker v1.12.1.
Replaced firecracker and jailer binaries under /opt/kata/bin.
Tested with kata-fc runtime on Kubernetes:
- Deployed pods using gitpod/openvscode-server
- Verified microVM startup, container access, and Firecracker usage
- Confirmed Firecracker and jailer versions via CLI
Signed-off-by: Kumar Mohit <68772712+itsmohitnarayan@users.noreply.github.com>
- "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>
Skip pulling container image layers when guest-pull=true. The contents
of these layers were ignored due to:
- #11162, and
- tarfs snapshotter support having been removed from genpolicy.
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>
`mem-agent` here is now a library and do not contain examples, ignore
Cargo.lock to get rid of untracked file noise produced by `cargo run` or
`cargo test`.
Signed-off-by: Ruoqing He <heruoqing@iscas.ac.cn>
Re-generates the client code against Cloud Hypervisor v47.0.
Note: The client code of cloud-hypervisor's OpenAPI is automatically
generated by openapi-generator.
Signed-off-by: Ruoqing He <heruoqing@iscas.ac.cn>
`MmapRegion` is only used while `virtio-fs` is enabled during testing
dragonball, gate the import behind `virtio-fs` feature.
Signed-off-by: Ruoqing He <heruoqing@iscas.ac.cn>
Some variables went unused if certain features are not enabled, use
`#[allow(unused)]` to suppress those warnings at the time being.
Signed-off-by: Ruoqing He <heruoqing@iscas.ac.cn>
`VcpuManagerError` is only needed when `host-device` feature is enabled,
gate the import behind that feature.
Signed-off-by: Ruoqing He <heruoqing@iscas.ac.cn>
Code inside `test_mac_addr_serialization_and_deserialization` test does
not actually require this `with-serde` feature to test, removing the
assertion here to enable this test.
Signed-off-by: Ruoqing He <heruoqing@iscas.ac.cn>
Add full cgroups support on host. Cgroups are managed by `FsManager` and
`SystemdManager`. As the names impies, the `FsManager` manages cgroups
through cgroupfs, while the `SystemdManager` manages cgroups through
systemd. The two manages support cgroup v1 and cgroup v2.
Two types of cgroups path are supported:
1. For colon paths, for example "foo.slice:bar:baz", the runtime manages
cgroups by `SystemdManager`;
2. For relative/absolute paths, the runtime manages cgroups by
`FsManager`.
vCPU threads are added into the sandbox cgroups in cgroup v1 + cgroupfs,
others, cgroup v1 + systemd, cgroup v2 + cgroupfs, cgroup v2 + systemd, VMM
process is added into the cgroups.
The systemd doesn't provide a way to add thread to a unit. `add_thread()`
in `SystemdManager` is equivalent to `add_process()`.
Cgroup v2 supports threaded mode. However, we should enable threaded mode
from leaf node to the root node (`/`) iteratively [1]. This means the
runtime needs to modify the cgroups created by container runtime (e.g.
containerd). Considering cgroupfs + cgroup v2 is not a common combination,
its behavior is aligned with systemd + cgroup v2, which is not allowed to
manage process at the thread level.
1: https://www.kernel.org/doc/html/v4.18/admin-guide/cgroup-v2.html#threadsFixes: #11356
Signed-off-by: Xuewei Niu <niuxuewei.nxw@antgroup.com>
As some reasons, it first should make it align with runtime-go, this
commit will do this work.
Fixes#11543
Signed-off-by: alex.lyn <alex.lyn@antgroup.com>
The actual memory usage on the host is equal to the hypervisor memory usage
plus the user memory usage. An OOM killer might kill the shim when the
memory limit on host is same with that of container and the container
consumes all available memory. In this case, the containerd will never
receive OOM event, but get "task exit" event. That makes the `k8s-oom.bats`
test fail.
The fix is to add a new container to increase the sandbox memory limit.
When the container "oom-test" is killed by OOM killer, there is still
available memory for the shim, so it will not be killed.
Signed-off-by: Xuewei Niu <niuxuewei.nxw@antgroup.com>
When enabling systemd cgroup driver and sandbox cgroup only, the shim is
under a systemd unit. When the unit is stopping, systemd sends SIGTERM to
the shim. The shim can't exit immediately, as there are some cleanups to
do. Therefore, ignoring SIGTERM is required here. The shim should complete
the work within a period (Kata sets it to 300s by default). Once a timeout
occurs, systemd will send SIGKILL.
Signed-off-by: Xuewei Niu <niuxuewei.nxw@antgroup.com>
Our CI keeps on getting
```
jq: error (at <stdin>:1): Cannot index string with string "tag_name"
```
during the install dependencies phase, which I suspect
might be due to github rate limits being reduced, so try
to pass through the `GH_TOKEN` env and use it in the auth header.
Signed-off-by: stevenhorsman <steven@uk.ibm.com>
It is important that we continue to support VirtIO-SCSI. While
VirtIO-BLK is a common choice, virtio-scsi offers significant
performance advantages in specific scenarios, particularly when
utilizing iothreads and with NVMe Fabrics.
Maintaining Flexibility and Choice by supporting both virtio-blk and
virtio-scsi, we provide greater flexibility for users to choose the
optimal storage（virtio-blk, virtio-scsi) interface based on their
specific workload requirements and hardware configurations.
As virtio-scsi controller has been created when qemu vm starts with
block device driver is set to `virtio-scsi`. This commit is for blockdev_add
the backend block device and device_add frondend virtio-scsi device via qmp.
Fixes#11516
Signed-off-by: alex.lyn <alex.lyn@antgroup.com>
As block device index is an very important unique id of a block device
and can indicate a block device which is equivalent to device_id.
In case of index is required in calculating scsi LUN and reduce
useless arguments within reusing `hotplug_block_device`, we'd better
change the device_id with block device index.
Signed-off-by: alex.lyn <alex.lyn@antgroup.com>
In this commit, block device aio are introduced within hotplug_block_device
within qemu via qmp and the "iouring" is set the default.
Signed-off-by: alex.lyn <alex.lyn@antgroup.com>
It should be correctly handled within the device manager when do
create_block_device if the driver_option is virtio-scsi.
Signed-off-by: alex.lyn <alex.lyn@antgroup.com>
It supports handling scsi device when block device driver is `scsi`.
And it will ensure a correct storage source with LUN.
Fixes#11516
Signed-off-by: alex.lyn <alex.lyn@antgroup.com>
It's used to help discover scsi devices inside guest and also add a
new const value `KATA_SCSI_DEV_TYPE` to help pass information.
Signed-off-by: alex.lyn <alex.lyn@antgroup.com>
AIO is the I/O mechanism used by qemu with options:
- threads
Pthread based disk I/O.
- native
Native Linux I/O.
- io_uring (default mode)
Linux io_uring API. This provides the fastest I/O operations on
Signed-off-by: alex.lyn <alex.lyn@antgroup.com>
Although Previous implementation of hotplugging block device via QMP
can successfully hot-plug the regular file based block device, but it
fails when the backend is /dev/xxx(e.g. /dev/loop0). With analysis about
it, we can know that it lacks the ablility to hotplug host block devices.
This commit will fill the gap, and make it work well for host block
devices.
Signed-off-by: alex.lyn <alex.lyn@antgroup.com>
As there were a few moderate security vulnerability fixes missed as part
of the 3.19.0 release.
Signed-off-by: Fabiano Fidêncio <fidencio@northflank.com>
The CoCo non-TEE job (run-k8s-tests-coco-nontee) used to be required but
we had to withdraw it to fix a problem (#11156). Now the job is back
running and stable, so time to make it required again.
Signed-off-by: Wainer dos Santos Moschetta <wainersm@redhat.com>
2025-07-10 12:19:19 +01:00
1660 changed files with 105299 additions and 56098 deletions
# Note: uses pushd to change into the clonned directory!
git_sparse_clone(){
localorigin="$1"
localrevision="$2"
shift2
localsparse_paths=("$@")
local repo
repo=$(basename -s .git "${origin}")
git init "${repo}"
pushd"${repo}"||exit1
git remote add origin "${origin}"
git fetch --depth 1 origin "${revision}"
git sparse-checkout init --cone
git sparse-checkout set"${sparse_paths[@]}"
git checkout FETCH_HEAD
}
###############################
# Disable security to allow e2e
###############################
@@ -116,33 +143,40 @@ az network vnet subnet update \
for NODE_NAME in $(kubectl get nodes -o jsonpath='{.items[*].metadata.name}');do[["${NODE_NAME}"=~ 'worker']]&& kubectl label node "${NODE_NAME}" node.kubernetes.io/worker=;done
@@ -450,7 +450,7 @@ You can build and install the guest kernel image as shown [here](../tools/packag
# Install a hypervisor
When setting up Kata using a [packaged installation method](install/README.md#installing-on-a-linux-system), the
`QEMU` VMM is installed automatically. Cloud-Hypervisor, Firecracker and StratoVirt VMMs are available from the [release tarballs](https://github.com/kata-containers/kata-containers/releases), as well as through [`kata-deploy`](../tools/packaging/kata-deploy/README.md).
`QEMU` VMM is installed automatically. Cloud-Hypervisor, Firecracker and StratoVirt VMMs are available from the [release tarballs](https://github.com/kata-containers/kata-containers/releases), as well as through [`kata-deploy`](../tools/packaging/kata-deploy/helm-chart/README.md).
You may choose to manually build your VMM/hypervisor.
To enhance security, Kata Containers supports using seccomp to restrict the hypervisor's system calls. Previously, this was only supported for a subset of hypervisors in runtime-go. Now, the runtime-rs also supports seccomp. This document describes how to enable/disable the seccomp feature for the corresponding hypervisor in runtime-rs.
## Pre-requisites
1. Ensure your system's kernel supports **seccomp**.
2. Confirm that each of the following virtual machines can run correctly on your system.
## Configure seccomp
With the exception of `qemu`, seccomp is enabled by default for all other supported hypervisors. Their corresponding built-in functionalities are also enabled by default.
### QEMU
As with runtime-go, you need to modify the following in your **configuration file**. These parameters will be passed directly to the `qemu` startup command line. For more details on the parameters, you can refer to: [https://www.qemu.org/docs/master/system/qemu-manpage.html](https://www.qemu.org/docs/master/system/qemu-manpage.html)
``` toml
# Qemu seccomp sandbox feature
# comma-separated list of seccomp sandbox features to control the syscall access.
# For example, `seccompsandbox= "on,obsolete=deny,spawn=deny,resourcecontrol=deny"`
# Note: "elevateprivileges=deny" doesn't work with daemonize option, so it's removed from the seccomp sandbox
# Another note: enabling this feature may reduce performance, you may enable
# /proc/sys/net/core/bpf_jit_enable to reduce the impact. see https://man7.org/linux/man-pages/man8/bpfc.8.html
The **seccomp** functionality is enabled by default for the following three hypervisors: `cloud hypervisor`, `firecracker`, and `dragonball`.
The seccomp rules for `cloud hypervisor` and `firecracker` are built directly into their executable files. For `dragonball`, the relevant configuration is currently located at `src/runtime-rs/crates/hypervisor/src/dragonball/seccomp.rs`.
To disable this functionality for these hypervisors, you can modify the following configuration options in your **configuration file**.
``` toml
# Disable the 'seccomp' feature from Cloud Hypervisor, firecracker or dragonball, default false
disable_seccomp = true
```
## Implementation details
For `qemu`, `cloud hypervisor`, and `firecracker`, their **seccomp** functionality is built into the respective executable files you are using. **runtime-rs** simply provides command-line arguments for their launch based on the configuration file.
For `dragonball`, a set of allowed system calls is currently provided for the entire **runtime-rs** process, and the process is prevented from using any system calls outside of this whitelist. As mentioned above, this set is located at `src/runtime-rs/crates/hypervisor/src/dragonball/seccomp.rs`.
VM templating is a Kata Containers feature that enables new VM creation using a cloning technique. When enabled, new VMs are created by cloning from a pre-created template VM, and they will share the same initramfs, kernel and agent memory in readonly mode. It is very much like a process fork done by the kernel but here we *fork* VMs.
For more details on VM templating, refer to the [What is VM templating and how do I use it](./what-is-vm-templating-and-how-do-I-use-it.md) article.
## How to Enable VM Templating
VM templating can be enabled by changing your Kata Containers config file (`/opt/kata/share/defaults/kata-containers/runtime-rs/configuration.toml`, overridden by `/etc/kata-containers/configuration.toml` if provided) such that:
-`qemu` version `v4.1.0` or above is specified in `hypervisor.qemu`->`path` section
-`enable_template = true`
-`template_path = "/run/vc/vm/template"` (default value, can be customized as needed)
-`initrd =` is set
-`image =` option is commented out or removed
-`shared_fs =` option is commented out or removed
-`default_memory =` should be set to more than 256MB
Then you can create a VM template for later usage by calling:
### Initialize and create the VM template
The `factory init` command creates a VM template by launching a new VM, initializing the Kata Agent, then pausing and saving its state (memory and device snapshots) to the template directory. This saved template is used to rapidly clone new VMs using QEMU's memory sharing capabilities.
```bash
sudo kata-ctl factory init
```
### Check the status of the VM template
The `factory status` command checks whether a VM template currently exists by verifying the presence of template files (memory snapshot and device state). It will output "VM factory is on" if the template exists, or "VM factory is off" otherwise.
```bash
sudo kata-ctl factory status
```
### Destroy and clean up the VM template
The `factory destroy` command removes the VM template by remove the `tmpfs` filesystem and deleting the template directory along with all its contents.
```bash
sudo kata-ctl factory destroy
```
## How to Create a New VM from VM Template
In the Go version of Kata Containers, the VM templating mechanism is implemented using virtio-9p (9pfs). However, 9pfs is not supported in runtime-rs due to its poor performance, limited cache coherence, and security risks. Instead, runtime-rs adopts `VirtioFS` as the default mechanism to provide rootfs for containers and VMs.
Yet, when enabling the VM template mechanism, `VirtioFS` introduces conflicts in memory sharing because its DAX-based shared memory mapping overlaps with the template's page-sharing design. To resolve these conflicts and ensure strict isolation between cloned VMs, runtime-rs replaces `VirtioFS` with the snapshotter approach — specifically, the `blockfile` snapshotter.
The `blockfile` snapshotter is used in runtime-rs because it provides each VM with an independent block-based root filesystem, ensuring strong isolation and full compatibility with the VM templating mechanism.
### Configure Snapshotter
#### Check if `Blockfile` Snapshotter is Available
```bash
ctr plugins ls | grep blockfile
```
If not available, continue with the following steps:
After modifying the configuration, restart containerd to apply changes:
```bash
sudo systemctl restart containerd
```
### Run Container with `blockfile` Snapshotter
After the VM template is created, you can pull an image and run a container using the `blockfile` snapshotter:
```bash
ctr run --rm -t --snapshotter blockfile docker.io/library/busybox:latest template sh
```
We can verify whether a VM was launched from a template or started normally by checking the launch parameters — if the parameters contain `incoming`, it indicates that the VM was started from a template rather than created directly.
## Performance Test
The comparative experiment between **template-based VM** creation and **direct VM** creation showed that the template-based approach achieved a ≈ **73.2%** reduction in startup latency (average launch time of **0.6s** vs. **0.82s**) and a ≈ **79.8%** reduction in memory usage (average memory usage of **178.2 MiB** vs. **223.2 MiB**), demonstrating significant improvements in VM startup efficiency and resource utilization.
The test script is as follows:
```bash
# Clear the page cache, dentries, and inodes to free up memory
echo 3 | sudo tee /proc/sys/vm/drop_caches
# Display the current memory usage
free -h
# Create 100 normal VMs and template-based VMs, and track the time
time for I in $(seq 100); do
echo -n " ${I}th" # Display the iteration number
ctr run -d --runtime io.containerd.kata.v2 --snapshotter blockfile docker.io/library/busybox:latest normal/template${I}
done
# Display the memory usage again after running the test
@@ -32,11 +32,24 @@ Kubernetes users can encode in `base64` format their Policy documents, and add t
### Encode a Policy file
For example, the [`allow-all-except-exec-process.rego`](../../src/kata-opa/allow-all-except-exec-process.rego) sample policy file is different from the [default Policy](../../src/kata-opa/allow-all.rego) because it rejects any `ExecProcess` requests. You can encode this policy file:
For example, the [`allow-all-except-exec-process.rego`](../../src/kata-opa/allow-all-except-exec-process.rego) sample policy file is different from the [default Policy](../../src/kata-opa/allow-all.rego) because it rejects any `ExecProcess` requests. To encode this policy file, you need to:
While creating the Pod sandbox, the Kata Shim will notice the `io.katacontainers.config.agent.policy` annotation and will send the Policy document to the Kata Agent - by sending a `SetPolicy` request. Note that this request will fail if the default Policy, included in the Guest image, doesn't allow this `SetPolicy` request. If the `SetPolicy` request is rejected by the Guest, the Kata Shim will fail to start the Pod sandbox.
While creating the Pod sandbox, the Kata Shim will notice the `io.katacontainers.config.hypervisor.cc_init_data` annotation and will create the init data device on the host and mount it on the guest as a block device. The agent then reads the init data struct from this device and sets the policy if present.
@@ -6,4 +6,4 @@ Container deployments utilize explicit or implicit file sharing between host fil
As of the 2.0 release of Kata Containers, [virtio-fs](https://virtio-fs.gitlab.io/) is the default filesystem sharing mechanism.
virtio-fs support works out of the box for `cloud-hypervisor` and `qemu`, when Kata Containers is deployed using `kata-deploy`. Learn more about `kata-deploy` and how to use `kata-deploy` in Kubernetes [here](../../tools/packaging/kata-deploy/README.md#kubernetes-quick-start).
virtio-fs support works out of the box for `cloud-hypervisor` and `qemu`, when Kata Containers is deployed using `kata-deploy`. Learn more about `kata-deploy` and how to use `kata-deploy` in Kubernetes [here](../../tools/packaging/kata-deploy/helm-chart/README.md).
| [Using official distro packages](#official-packages) | Kata packages provided by Linux distributions official repositories | yes | Recommended for most users. |
| [Automatic](#automatic-installation) | Run a single command to install a full system | **No!** | For those wanting the latest release quickly. |
| [Using kata-deploy Helm chart](#kata-deploy-helm-chart) | The preferred way to deploy the Kata Containers distributed binaries on a Kubernetes cluster | **No!** | Best way to give it a try on kata-containers on an already up and running Kubernetes cluster. |
### Kata Deploy Helm Chart
The Kata Deploy Helm chart is a convenient way to install all of the binaries and
The Kata Deploy Helm chart is the preferred way to install all of the binaries and
artifacts required to run Kata Containers on Kubernetes.
[Use Kata Deploy Helm Chart](/tools/packaging/kata-deploy/helm-chart/README.md) to install Kata Containers on a Kubernetes Cluster.
### Official packages
Kata packages are provided by official distribution repositories for:
| Distribution (link to installation guide) | Minimum versions |
Kata Containers on Amazon Web Services (AWS) makes use of [i3.metal](https://aws.amazon.com/ec2/instance-types/i3/) instances. Most of the installation procedure is identical to that for Kata on your preferred distribution, except that you have to run it on bare metal instances since AWS doesn't support nested virtualization yet. This guide walks you through creating an i3.metal instance.
region = <your-aws-region-for-your-i3-metal-instance>
EOF
```
For more information on how to get AWS credentials please refer to [this guide](https://docs.aws.amazon.com/IAM/latest/UserGuide/id_credentials_access-keys.html). Alternatively, you can ask the administrator of your AWS account to issue one with the AWS CLI:
```sh
$ aws_username="myusername"
$ aws iam create-access-key --user-name="$aws_username"
```
More general AWS CLI guidelines can be found [here](https://docs.aws.amazon.com/cli/latest/userguide/installing.html).
Refer to [this guide](https://docs.aws.amazon.com/cli/latest/userguide/cli-ec2-launch.html) for more details on how to launch instances with the AWS CLI.
SSH into the machine
```bash
$ ssh -i MyKeyPair.pem ubuntu@${IP}
```
Go onto the next step.
## Install Kata
The process for installing Kata itself on bare metal is identical to that of a virtualization-enabled VM.
For detailed information to install Kata on your distribution of choice, see the [Kata Containers installation user guides](../install/README.md).
# Install Kata Containers on Google Compute Engine
Kata Containers on Google Compute Engine (GCE) makes use of [nested virtualization](https://cloud.google.com/compute/docs/instances/enable-nested-virtualization-vm-instances). Most of the installation procedure is identical to that for Kata on your preferred distribution, but enabling nested virtualization currently requires extra steps on GCE. This guide walks you through creating an image and instance with nested virtualization enabled. Note that `kata-runtime check` checks for nested virtualization, but does not fail if support is not found.
As a pre-requisite this guide assumes an installed and configured instance of the [Google Cloud SDK](https://cloud.google.com/sdk/downloads). For a zero-configuration option, all of the commands below were been tested under [Google Cloud Shell](https://cloud.google.com/shell/) (as of Jun 2018). Verify your `gcloud` installation and configuration:
```bash
$ gcloud info ||{echo"ERROR: no Google Cloud SDK";exit 1;}
```
## Create an Image with Nested Virtualization Enabled
VM images on GCE are grouped into families under projects. Officially supported images are automatically discoverable with `gcloud compute images list`. That command produces a list similar to the following (likely with different image names):
Each distribution has its own project, and each project can host images for multiple versions of the distribution, typically grouped into families. We recommend you select images by project and family, rather than by name. This ensures any scripts or other automation always works with a non-deprecated image, including security updates, updates to GCE-specific scripts, etc.
### Create the Image
The following example (substitute your preferred distribution project and image family) produces an image with nested virtualization enabled in your currently active GCE project:
If successful, `gcloud` reports that the image was created. Verify that the image has the nested virtualization license with `gcloud compute images describe $IMAGE_NAME`. This produces output like the following (some fields have been removed for clarity and to redact personal info):
The primary criterion of interest here is the presence of the `enable-vmx` license. Without that licence Kata will not work. Without that license Kata does not work. The presence of that license instructs the Google Compute Engine hypervisor to enable Intel's VT-x instructions in virtual machines created from the image. Note that nested virtualization is only available in VMs running on Intel Haswell or later CPU micro-architectures.
### Verify VMX is Available
Assuming you created a nested-enabled image using the previous instructions, verify that VMs created from this image are VMX-enabled with the following:
1. Create a VM from the image created previously:
```bash
$ gcloud compute instances create \
--image $IMAGE_NAME \
--machine-type n1-standard-2 \
--min-cpu-platform "Intel Broadwell" \
kata-testing
```
> **NOTE**: In most zones the `--min-cpu-platform` argument can be omitted. It is only necessary in GCE Zones that include hosts based on Intel's Ivybridge platform.
2. Verify that the VMX CPUID flag is set:
```bash
$ gcloud compute ssh kata-testing
# While ssh'd into the VM:
$ [ -z "$(lscpu|grep GenuineIntel)" ] && { echo "ERROR: Need an Intel CPU"; exit 1; }
```
If this fails, ensure you created your instance from the correct image and that the previously listed `enable-vmx` license is included.
## Install Kata
The process for installing Kata itself on a virtualization-enabled VM is identical to that for bare metal.
For detailed information to install Kata on your distribution of choice, see the [Kata Containers installation user guides](../install/README.md).
## Create a Kata-enabled Image
Optionally, after installing Kata, create an image to preserve the fruits of your labor:
```bash
$ gcloud compute instances stop kata-testing
$ gcloud compute images create \
--source-disk kata-testing \
kata-base
```
The result is an image that includes any changes made to the `kata-testing` instance as well as the `enable-vmx` flag. Verify this with `gcloud compute images describe kata-base`. The result, which omits some fields for clarity, should be similar to the following:
Some files were not shown because too many files have changed in this diff
Show More
Reference in New Issue
Block a user
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
