podman: add option --cgroup-conf

it allows to manually tweak the configuration for cgroup v2.

we will expose some of the options in future as single
options (e.g. the new memory knobs), but for now add the more generic
--cgroup-conf mechanism for maximum control on the cgroup
configuration.

OCI specs change: https://github.com/opencontainers/runtime-spec/pull/1040

Requires: https://github.com/containers/crun/pull/459

Signed-off-by: Giuseppe Scrivano <gscrivan@redhat.com>

11 files changed, 50 insertions, 7 deletions

diff --git a/cmd/podman/common/create.go b/cmd/podman/common/create.go
index 403a1065b..d0bc8d466 100644
--- a/cmd/podman/common/create.go
+++ b/cmd/podman/common/create.go
@@ -516,5 +516,10 @@ func GetCreateFlags(cf *ContainerCLIOpts) *pflag.FlagSet {
 		"seccomp-policy", "default",
 		"Policy for selecting a seccomp profile (experimental)",
 	)
+	createFlags.StringSliceVar(
+		&cf.CgroupConf,
+		"cgroup-conf", []string{},
+		"Configure cgroup v2 (key=value)",
+	)
 	return &createFlags
 }
diff --git a/cmd/podman/common/create_opts.go b/cmd/podman/common/create_opts.go
index f9e4d7ca5..16d41988f 100644
--- a/cmd/podman/common/create_opts.go
+++ b/cmd/podman/common/create_opts.go
@@ -106,4 +106,6 @@ type ContainerCLIOpts struct {
 	SeccompPolicy     string
 
 	Net *entities.NetOptions
+
+	CgroupConf []string
 }
diff --git a/cmd/podman/common/specgen.go b/cmd/podman/common/specgen.go
index bf50bb56b..4de622916 100644
--- a/cmd/podman/common/specgen.go
+++ b/cmd/podman/common/specgen.go
@@ -450,7 +450,20 @@ func FillOutSpecGen(s *specgen.SpecGenerator, c *ContainerCLIOpts, args []string
 		s.ResourceLimits.Pids = &pids
 	}
 	s.ResourceLimits.CPU = getCPULimits(c)
-	if s.ResourceLimits.CPU == nil && s.ResourceLimits.Pids == nil && s.ResourceLimits.BlockIO == nil && s.ResourceLimits.Memory == nil {
+
+	unifieds := make(map[string]string)
+	for _, unified := range c.CgroupConf {
+		splitUnified := strings.SplitN(unified, "=", 2)
+		if len(splitUnified) < 2 {
+			return errors.Errorf("--cgroup-conf must be formatted KEY=VALUE")
+		}
+		unifieds[splitUnified[0]] = splitUnified[1]
+	}
+	if len(unifieds) > 0 {
+		s.ResourceLimits.Unified = unifieds
+	}
+
+	if s.ResourceLimits.CPU == nil && s.ResourceLimits.Pids == nil && s.ResourceLimits.BlockIO == nil && s.ResourceLimits.Memory == nil && s.ResourceLimits.Unified == nil {
 		s.ResourceLimits = nil
 	}
 
diff --git a/completions/bash/podman b/completions/bash/podman
index 379ba7fc8..af222bf65 100644
--- a/completions/bash/podman
+++ b/completions/bash/podman
@@ -2107,6 +2107,7 @@ _podman_container_run() {
 		--cap-add
 		--cap-drop
 		--cgroup-parent
+		--cgroup-conf
 		--cidfile
 		--conmon-pidfile
 		--cpu-period
diff --git a/docs/source/markdown/podman-create.1.md b/docs/source/markdown/podman-create.1.md
index 976a1e681..2b5e86637 100644
--- a/docs/source/markdown/podman-create.1.md
+++ b/docs/source/markdown/podman-create.1.md
@@ -89,6 +89,10 @@ The *split* option splits the current cgroup in two sub-cgroups: one for conmon
 
 Path to cgroups under which the cgroup for the container will be created. If the path is not absolute, the path is considered to be relative to the cgroups path of the init process. Cgroups will be created if they do not already exist.
 
+**--cgroup-conf**=*KEY=VALUE*
+
+When running on cgroup v2, specify the cgroup file to write to and its value.  For example **--cgroup-conf=memory.high=1073741824** sets the memory.high limit to 1GB.
+
 **--cidfile**=*id*
 
 Write the container ID to the file
diff --git a/docs/source/markdown/podman-run.1.md b/docs/source/markdown/podman-run.1.md
index b6c1fab17..4449b53fc 100644
--- a/docs/source/markdown/podman-run.1.md
+++ b/docs/source/markdown/podman-run.1.md
@@ -104,6 +104,10 @@ The **split** option splits the current cgroup in two sub-cgroups: one for conmo
 
 Path to cgroups under which the cgroup for the container will be created. If the path is not absolute, the path is considered to be relative to the cgroups path of the init process. Cgroups will be created if they do not already exist.
 
+**--cgroup-conf**=*KEY=VALUE*
+
+When running on cgroup v2, specify the cgroup file to write to and its value.  For example **--cgroup-conf=memory.high=1073741824** sets the memory.high limit to 1GB.
+
 **--cidfile**=*file*
 
 Write the container ID to *file*.
diff --git a/libpod/container_inspect.go b/libpod/container_inspect.go
index 437729c2d..835dccd71 100644
--- a/libpod/container_inspect.go
+++ b/libpod/container_inspect.go
@@ -465,6 +465,7 @@ func (c *Container) generateInspectContainerHostConfig(ctrSpec *spec.Spec, named
 			if ctrSpec.Linux.Resources.Pids != nil {
 				hostConfig.PidsLimit = ctrSpec.Linux.Resources.Pids.Limit
 			}
+			hostConfig.CgroupConf = ctrSpec.Linux.Resources.Unified
 			if ctrSpec.Linux.Resources.BlockIO != nil {
 				if ctrSpec.Linux.Resources.BlockIO.Weight != nil {
 					hostConfig.BlkioWeight = *ctrSpec.Linux.Resources.BlockIO.Weight
diff --git a/libpod/define/container_inspect.go b/libpod/define/container_inspect.go
index 8adf3c077..44c3d515b 100644
--- a/libpod/define/container_inspect.go
+++ b/libpod/define/container_inspect.go
@@ -518,6 +518,8 @@ type InspectContainerHostConfig struct {
 	IOMaximumIOps uint64 `json:"IOMaximumIOps"`
 	// IOMaximumBandwidth is Windows-only and not presently implemented.
 	IOMaximumBandwidth uint64 `json:"IOMaximumBandwidth"`
+	// CgroupConf is the configuration for cgroup v2.
+	CgroupConf map[string]string `json:"CgroupConf"`
 }
 
 // InspectBasicNetworkConfig holds basic configuration information (e.g. IP
diff --git a/pkg/spec/createconfig.go b/pkg/spec/createconfig.go
index c49d51fc5..e0c875fe9 100644
--- a/pkg/spec/createconfig.go
+++ b/pkg/spec/createconfig.go
@@ -31,12 +31,13 @@ const (
 type CreateResourceConfig struct {
 	BlkioWeight       uint16   // blkio-weight
 	BlkioWeightDevice []string // blkio-weight-device
-	CPUPeriod         uint64   // cpu-period
-	CPUQuota          int64    // cpu-quota
-	CPURtPeriod       uint64   // cpu-rt-period
-	CPURtRuntime      int64    // cpu-rt-runtime
-	CPUShares         uint64   // cpu-shares
-	CPUs              float64  // cpus
+	CgroupConf        map[string]string
+	CPUPeriod         uint64  // cpu-period
+	CPUQuota          int64   // cpu-quota
+	CPURtPeriod       uint64  // cpu-rt-period
+	CPURtRuntime      int64   // cpu-rt-runtime
+	CPUShares         uint64  // cpu-shares
+	CPUs              float64 // cpus
 	CPUsetCPUs        string
 	CPUsetMems        string   // cpuset-mems
 	DeviceCgroupRules []string //device-cgroup-rule
diff --git a/pkg/specgen/generate/validate.go b/pkg/specgen/generate/validate.go
index dca45cc0e..ed337321b 100644
--- a/pkg/specgen/generate/validate.go
+++ b/pkg/specgen/generate/validate.go
@@ -23,6 +23,12 @@ func verifyContainerResources(s *specgen.SpecGenerator) ([]string, error) {
 		return warnings, nil
 	}
 
+	if s.ResourceLimits.Unified != nil {
+		if !cgroup2 {
+			return nil, errors.New("Cannot use --cgroup-conf without cgroup v2")
+		}
+	}
+
 	// Memory checks
 	if s.ResourceLimits.Memory != nil {
 		memory := s.ResourceLimits.Memory
diff --git a/pkg/specgen/specgen.go b/pkg/specgen/specgen.go
index a9161071b..a52225f87 100644
--- a/pkg/specgen/specgen.go
+++ b/pkg/specgen/specgen.go
@@ -415,6 +415,10 @@ type ContainerResourceConfig struct {
 	ThrottleReadIOPSDevice map[string]spec.LinuxThrottleDevice `json:"throttleReadIOPSDevice,omitempty"`
 	// IO write rate limit per cgroup per device, IO per second
 	ThrottleWriteIOPSDevice map[string]spec.LinuxThrottleDevice `json:"throttleWriteIOPSDevice,omitempty"`
+	// CgroupConf are key-value options passed into the container runtime
+	// that are used to configure cgroup v2.
+	// Optional.
+	CgroupConf map[string]string `json:"unified,omitempty"`
 }
 
 // ContainerHealthCheckConfig describes a container healthcheck with attributes