Signed-off-by: Sebastian Sch <sebassch@gmail.com>

Signed-off-by: Zenghui Shi <zshi@redhat.com>

Bug 2021151: Sync master 27 12 21

It is found that VF netdev disappears after it was successfully
created (observed via ip link show). While investigating the root
cause, setting addtional dependency in NetworkManager.service
as temporary fix to the issue.

This commits also changes service and script names to lower case

Signed-off-by: Zenghui Shi <zshi@redhat.com>
(cherry picked from commit 6325355230a16d06426810b8ab1c76cd17a42787)

Bug 2041734: ensure switchdev service dependencies

Reconciling with https://github.com/openshift/ocp-build-data/tree/0059e470a61eba28225165168b22d9e77c33c926/images/sriov-network-webhook.yml

Reconciling with https://github.com/openshift/ocp-build-data/tree/0059e470a61eba28225165168b22d9e77c33c926/images/sriov-network-config-daemon.yml

Signed-off-by: Sebastian Sch <sebassch@gmail.com>

…4.11-sriov-network-webhook

Updating sriov-network-webhook images to be consistent with ART

…4.11-sriov-network-config-daemon

Updating sriov-network-config-daemon images to be consistent with ART

Reconciling with https://github.com/openshift/ocp-build-data/tree/0059e470a61eba28225165168b22d9e77c33c926/images/sriov-network-must-gather.yml

…4.11-sriov-network-must-gather

Updating sriov-network-must-gather images to be consistent with ART

Bump sriov operator image to ocp 4.11

there is no oc command inside the base container so we need to use the cli container

Signed-off-by: Sebastian Sch <sebassch@gmail.com>

Fix must-gather container

Substitute SriovNetworkNodeConfigPolicy name with
SriovNetworkNodePolicy as it is the name of the struct
CR type.

Signed-off-by: Andrea Panattoni <apanatto@redhat.com>

This patch fixes image build target to tag image
'ghcr.io/k8snetworkplumbingwg/sriov-network-operator:latest'
by default.

Signed-off-by: Ivan Kolodyazhny <ikolodiazhny@nvidia.com>

added openshift api module for ease of use with infra topology

Signed-off-by: ehila <ehila@redhat.com>

leader election config will query the cluster for the topology and if we are in an SNO toplogy we use one set of configs otherwise we use conventional defaults for HA cluster config.

Signed-off-by: ehila <ehila@redhat.com>

upkeep: ran go mod tidy/vendor/verify

Signed-off-by: ehila <ehila@redhat.com>

upkeep: spelling fix

Signed-off-by: ehila <ehila@redhat.com>

feat: moved sno logic to utils

moved sno logic to use the cluster.go file
added check for k8s or openshift environment
upkeep re-organize imports

Signed-off-by: ehila <ehila@redhat.com>

feat: added rbac for operator

Signed-off-by: ehila <ehila@redhat.com>

refactor: updated to use client-go leader election

updated to use client-go leader election struct instead of openshift/api
updated wording to remove SNO acronym

Signed-off-by: ehila <ehila@redhat.com>

This commit removes the pod that is created to validate if the node has
secure boot enable or not

Before this PR after every test running there are test pods in the sriov
operator namespace that does not get clean

Signed-off-by: Sebastian Sch <sebassch@gmail.com>

This patch introduces following changes:
* explicitly set go version to v1.17 in GitHub actions
* updates go.mod file and vendored packages
* updates go version in Dockerfiles

'go install' instead of 'go get' [1] will be used once #231
is fixed.

[1] https://go.dev/doc/go-get-install-deprecation

Since drain operation started we don't need to requires
drain lock for this node because node already has required
annotation.

It's safe to continue node drain procedure without lock.

Closes: #230

Signed-off-by: Ivan Kolodyazhny <ikolodiazhny@nvidia.com>

Signed-off-by: Ivan Kolodyazhny <ikolodiazhny@nvidia.com>

PR #223 updated base images to CentOS 8 which is EOL now so we can't
build new images using it.

Closes: #239

Signed-off-by: Ivan Kolodyazhny <ikolodiazhny@nvidia.com>

updated to attach scheme to client for infra call
updated error message output to bubble up error from single node call

Signed-off-by: ehila <ehila@redhat.com>

…ver only

Signed-off-by: Sebastian Sch <sebassch@gmail.com>

Signed-off-by: Sebastian Sch <sebassch@gmail.com>

Bug 2045087: Sync master 10 02 22

4.11 sync upstream: 2022-02-07

Rename the manifest folder as requested by Automated
Release Tooling (ART)

Signed-off-by: Andrea Panattoni <apanatto@redhat.com>

Signed-off-by: Andrea Panattoni <apanatto@redhat.com>

Move the manifest folder from 4.10 to `stable` and bump-4.11

This change fixes VFGroups assigned to nodes based
on the policies priorities: highest priority (lowest
value of priority) should be the only one present in the
SriovNetworkNodeState.spec. Exception is made for policies
with non-overlapping VFGroups, which will be merged.

For same priority policies we discard overlapping VF ranges,
only the highest priority is present.

Added description of this behaviour to README.

This patch fixes current behaviour for BF2 NICs to configure
SR-IOV VFs in vanilla Kubernetes cluster.

PR #201 broke BF2 NICs configration in Kubernetes. It assumes
that BF2 will be configured via systemd service in OpenShift
using MachineConfigPool object.

Signed-off-by: Ivan Kolodyazhny <ikolodiazhny@nvidia.com>

Machines are immutable, so there is no need to fetch for metadata every
30s.

This patch will move out the functions to read the metadata file and do
it in the daemon writer.

Old behavior:
Every 30 seconds, for each network device found, fetch, read and process
OpenStack metadata.

New behavior:
daemon process does a first run then runs a routine that executes every
30 seconds. So we'll only do fetch metadata two times in total.

Also, in case the API contract is broken, we might encounter some issues so
let's switch to the latest stable, 2018-08-27, which has all the data
that we need and is currently in the same format as latest for our
needs.

* Patch Makefile to allow overriding the APP_NAME, so any image can be
  built using `make image`.
* Add some docs about how to build & use an image.

As well known and well documented, in virtual deployments of Kubernetes such as running on top
of OpenStack where the underlying virtualization platform is KVM, it does not support a virtualized iommu,
the VFIO PCI driver needs to be loaded with a special flag.

This patch aims to do it by default in the virtual_plugin, since this
plugin is only used by OpenStack for now.

Note: it also changes `utils.LoadKernelModule` function to support
kernel module arguments. It also adds a note in `load-kmod.sh` about the
fact that a module won't be reloaded with potential new options if the
module is already loaded on the system.

It can be really useful to provide logs when a device (PF or VF) is not
supported and give the vendorId + deviceId; so the deployer can check
what device is being selected and compare with the ConfigMap which
describes the supported nics.

ctrl.Log var contains noop logger until logger instance is
explicitly set with ctrl.SetLogger function

If error happen during k8s client creation it will be not logged
because ctrl.SetLogger is called after.

Move logger initialization step before k8s client
creation to be able to log client initialization errors.

Signed-off-by: Yury Kulazhenkov <ykulazhenkov@nvidia.com>

Update Readme and Yamls with device information

Co-authored-by: Edwin Joseph <edwin.ant.joseph@hpe.com>
Co-authored-by: Eoghan Russell <eoghanrussell1212@gmail.com>

This commit includes a new dependency that
provides a HTTP client interface with automatic
retries and exponential backoff.

In most cases, OpenStack metadata service is available in the cloud,
which removes the requirement of config-drive.

This patch will change the logic to try reading Metadata and Network data from
the config-drive directory but if it's not present, it'll try to get
it by safely fetching the URL.

This commit adds a dedicated document for supported hardware
which lists the current hardware supported by sriov-network-operator
as well as defines a process to add support for new hardware
in sriov-network-operator

Signed-off-by: Adrian Chiris <adrianc@nvidia.com>

Sync u/s 2022-03-18

Don't overwrite the supported NICs downstream. We maintain a different
list of NICs than upstream does.

Bug 2072455: fix make bundle

Before creating the InterfaceExt in utils.DiscoverSriovDevices,
check that the discovered NIC is a supported model as configured
in the `supported-nic-ids` config map.

With this logic, the plugins will not get unsupported NIC to configure.

It is possible to get all the NICs discovered by using an env parameter:
DEV_MODE with value TRUE in the pod of the sriov-network-config-daemon.

Also, if DEV_MODE is set to TRUE, the webhook will allow unsupported NICS
in `SriovNetworkNodePolicy`.

Signed-off-by: Fred Rolland <frolland@nvidia.com>

SriovNetworkNodeState is outdated. IsRdma field was introduced
in the PR #207.

This patch syncs CRDs between config/crd/bases and helm chart.

Remove "yum update" command from
Dockerfile.sriov-network-config-daemon.

For config-daemon, centos:stream8 is used as a base image.
This image is actively maintained and should receive
security updates with minimal delay.

Signed-off-by: Yury Kulazhenkov <ykulazhenkov@nvidia.com>

Closes: #274

As the Makefile uses deprecated Go features, GH actions
should use a fixed 1.17.* version of Golang.

Signed-off-by: Andrea Panattoni <apanatto@redhat.com>

As a maintainer of Mellanox Plugin I would like to have permissions
to run e2e vendor CI tests.

After PR #252 is merged there is a race condition between NicIdMap
init and accec to it.
This patch moves InitNicIdMap function call before NicIdMap will be
used in node writer and config daemon run function.

Closes-Bug: #279

4.11 update 2022 04 14

This commit add the support for virtio interfaces like vhostuser for openstack virtual workers.

implementation details:
* on first run (after a reboot) we get all the information we need when the devices are visible to the kernel
* we match the mac address to the openstack network ID
* if the sriov-network-config-daemon gets reboot it will use the initial file on the node so even if the nics are in vfio the node state will be right

move the operator initial state file to /tmp on the host so it will get deleted on every reboot to support nic changes both for virtual and BM environments

Signed-off-by: Sebastian Sch <sebassch@gmail.com>
(cherry picked from commit 2c52beb8ba6fd751b0eb52dfd0e494e2c54eeea6)

Signed-off-by: Sebastian Sch <sebassch@gmail.com>
(cherry picked from commit 04a3f698003cfea381af9a3def9c0c495fcde0a2)

we can't run the udev rule on virtual environments because all the nics in that type of deployment
will be presented as VFs. this means that we are going to disable nics on the machine itself

Signed-off-by: Sebastian Sch <sebassch@gmail.com>
(cherry picked from commit 346de677b14696b3495291b922b868dba205d79c)

This is required for vhostuser ports support in OpenShift.

Bug 2036948: improve the virtual plugin support

Add support for Pensando DSC card

If `/run/ostree-booted` is present, then the system is based on rpm-ostree
and kernel arguments can be updated via `rpm-ostree` command. File
`/host/etc/redhat-release` may not be available because it can be an
absolute link (see openshift/os#713).

Signed-off-by: Andrea Panattoni <apanatto@redhat.com>

Add test harness to allow running `bindata/scripts/*.sh` files in
a mocked filesystem via `make test-bindata-scripts`. Harness includes
shUnit2 files and a mocked `rpm-ostree` implementation.

Bug 2070888: Use /run/ostree-booted to determine RHCOS

This reverts commit 7316cd0.

Bug 2087084: Revert "Add support for Pensando DSC card"

Bug 2083459: Make numvfs > totalvfs error message more verbose

`must-gather` logic has been moved to
openshift/must-gather repository with
openshift/must-gather#297

Signed-off-by: Andrea Panattoni <apanatto@redhat.com>

Bug 2099346: Remove must-gather references

…orted card listing

- Update docs with newly supported hardware
- Update supported device ids in config map

Update sysctl to a safe one in metapluging chaining test

Reconciling with https://github.com/openshift/ocp-build-data/tree/2e5eba0108299087402c119ba0dc979a0a6c0bd1/images/sriov-network-webhook.yml

Bug 2101525: add Intel E810 NIC model (1591) support in SR-IOV Operator

Fixes issue #249

Signed-off-by: Fred Rolland <frolland@nvidia.com>

Signed-off-by: Fred Rolland <frolland@nvidia.com>

This PR fix issues found with:
golangci-lint run --disable-all -E unused

Signed-off-by: Fred Rolland <frolland@nvidia.com>

This commit fix the issues found by:
`golangci-lint run --disable-all -E whitespace`

Signed-off-by: Fred Rolland <frolland@nvidia.com>

This commit fix the issues found by:
`golangci-lint run --disable-all -E deadcode`
and `golangci-lint run --disable-all -E goimports`

Signed-off-by: Fred Rolland <frolland@nvidia.com>

This commit fixes issues found by
`golangci-lint run --disable-all -E misspell`
and `golangci-lint run --disable-all -E staticcheck`

Signed-off-by: Fred Rolland <frolland@nvidia.com>

Signed-off-by: Adrian Chiris <adrianc@nvidia.com>

If error occures during sriov configuration in
generic plugin, config daemon will log the error
and requeue the request to reconcile the node.

On the next run, when generic plugin Apply() method
is called, interface configuration may be skipped,
leaving the interface partially configured.

To ensure generic plugin re-configures the interface
on the next iteration, we reset the interface
before returning an error.

This will also ensure that a drain will be requested
as well.

Signed-off-by: Adrian Chiris <adrianc@nvidia.com>

In case of error, it is useful to know the reason why it
was not possible to get the IP address of SR-IOV interface.

Add description to each returned error and return an error
in case the requested interface is not found.

Add "github.com/stretchr/testify/assert" dependency adn
unit tests for GetSriovNicIPs

Signed-off-by: Andrea Panattoni <apanatto@redhat.com>

This commit fixes issues found with:
`golangci-lint run --disable-all -E gosimple`

Signed-off-by: Fred Rolland <frolland@nvidia.com>

Fix issues found by:
`golangci-lint run --disable-all -E ineffassign`

Signed-off-by: Fred Rolland <frolland@nvidia.com>

This commit fixes issues found by
`golangci-lint run --disable-all -E stylecheck`

Signed-off-by: Fred Rolland <frolland@nvidia.com>

With Kubernetes 1.24, master nodes are now
labeld and tainted as `node-role.kubernetes.io/control-plane`

Support this in sriov network operator by:

- extending tolerations in relevant yaml files
- adjusting node affinity in relevant yaml files
- modify helm chart to deploy operator based on node affinity
  and not node selector

Signed-off-by: Adrian Chiris <adrianc@nvidia.com>

Signed-off-by: Sebastian Sch <sebassch@gmail.com>

Signed-off-by: Sebastian Sch <sebassch@gmail.com>