host_system.go

/*
Copyright (c) 2017 VMware, Inc. All Rights Reserved.

Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at

    http://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.
*/

package simulator

import (
	"fmt"
	"net"
	"os"
	"sync"
	"time"

	"github.com/vmware/govmomi/simulator/esx"
	"github.com/vmware/govmomi/vim25/methods"
	"github.com/vmware/govmomi/vim25/mo"
	"github.com/vmware/govmomi/vim25/soap"
	"github.com/vmware/govmomi/vim25/types"
)

var (
	hostPortUnique = os.Getenv("VCSIM_HOST_PORT_UNIQUE") == "true"

	globalLock sync.Mutex
	// globalHostCount is used to construct unique hostnames. Should be consumed under globalLock.
	globalHostCount = 0
)

type HostSystem struct {
	mo.HostSystem

	sh *simHost
}

func asHostSystemMO(obj mo.Reference) (*mo.HostSystem, bool) {
	h, ok := getManagedObject(obj).Addr().Interface().(*mo.HostSystem)
	return h, ok
}

func NewHostSystem(host mo.HostSystem) *HostSystem {
	if hostPortUnique { // configure unique port for each host
		port := &esx.HostSystem.Summary.Config.Port
		*port++
		host.Summary.Config.Port = *port
	}

	now := time.Now()

	hs := &HostSystem{
		HostSystem: host,
	}

	hs.Name = hs.Summary.Config.Name
	hs.Summary.Runtime = &hs.Runtime
	hs.Summary.Runtime.BootTime = &now

	// shallow copy Summary.Hardware, as each host will be assigned its own .Uuid
	hardware := *host.Summary.Hardware
	hs.Summary.Hardware = &hardware

	if hs.Hardware == nil {
		// shallow copy Hardware, as each host will be assigned its own .Uuid
		info := *esx.HostHardwareInfo
		hs.Hardware = &info
	}

	cfg := new(types.HostConfigInfo)
	deepCopy(hs.Config, cfg)
	hs.Config = cfg

	// copy over the reference advanced options so each host can have it's own, allowing hosts to be configured for
	// container backing individually
	deepCopy(esx.AdvancedOptions, &cfg.Option)

	// add a supported option to the AdvancedOption manager
	simOption := types.OptionDef{ElementDescription: types.ElementDescription{Key: advOptContainerBackingImage}}
	// TODO: how do we enter patterns here? Or should we stick to a list in the value?
	// patterns become necessary if we want to enforce correctness on options for RUN.underlay.<pnic> or allow RUN.port.xxx
	hs.Config.OptionDef = append(hs.Config.OptionDef, simOption)

	config := []struct {
		ref **types.ManagedObjectReference
		obj mo.Reference
	}{
		{&hs.ConfigManager.DatastoreSystem, &HostDatastoreSystem{Host: &hs.HostSystem}},
		{&hs.ConfigManager.NetworkSystem, NewHostNetworkSystem(&hs.HostSystem)},
		{&hs.ConfigManager.AdvancedOption, NewOptionManager(nil, nil, &hs.Config.Option)},
		{&hs.ConfigManager.FirewallSystem, NewHostFirewallSystem(&hs.HostSystem)},
		{&hs.ConfigManager.StorageSystem, NewHostStorageSystem(&hs.HostSystem)},
	}

	for _, c := range config {
		ref := Map.Put(c.obj).Reference()

		*c.ref = &ref
	}

	return hs
}

func (h *HostSystem) configure(ctx *Context, spec types.HostConnectSpec, connected bool) {
	h.Runtime.ConnectionState = types.HostSystemConnectionStateDisconnected
	if connected {
		h.Runtime.ConnectionState = types.HostSystemConnectionStateConnected
	}

	// lets us construct non-conflicting hostname automatically if omitted
	// does not use the unique port instead to avoid constraints on port, such as >1024

	globalLock.Lock()
	instanceID := globalHostCount
	globalHostCount++
	globalLock.Unlock()

	if spec.HostName == "" {
		spec.HostName = fmt.Sprintf("esx-%d", instanceID)
	} else if net.ParseIP(spec.HostName) != nil {
		h.Config.Network.Vnic[0].Spec.Ip.IpAddress = spec.HostName
	}

	h.Summary.Config.Name = spec.HostName
	h.Name = h.Summary.Config.Name
	id := newUUID(h.Name)
	h.Summary.Hardware.Uuid = id
	h.Hardware.SystemInfo.Uuid = id

	var err error
	h.sh, err = createSimulationHost(ctx, h)
	if err != nil {
		panic("failed to create simulation host and no path to return error: " + err.Error())
	}
}

// configureContainerBacking sets up _this_ host for simulation using a container backing.
// Args:
//
//		image - the container image with which to simulate the host
//		mounts - array of mount info that should be translated into /vmfs/volumes/... mounts backed by container volumes
//	 	networks - names of bridges to use for underlays. Will create a pNIC for each. The first will be treated as the management network.
//
// Restrictions adopted from createSimulationHost:
// * no mock of VLAN connectivity
// * only a single vmknic, used for "the management IP"
// * pNIC connectivity does not directly impact VMs/vmks using it as uplink
//
// The pnics will be named using standard pattern, ie. vmnic0, vmnic1, ...
// This will sanity check the NetConfig for "management" nicType to ensure that it maps through PortGroup->vSwitch->pNIC to vmnic0.
func (h *HostSystem) configureContainerBacking(ctx *Context, image string, mounts []types.HostFileSystemMountInfo, networks ...string) error {
	option := &types.OptionValue{
		Key:   advOptContainerBackingImage,
		Value: image,
	}

	advOpts := ctx.Map.Get(h.ConfigManager.AdvancedOption.Reference()).(*OptionManager)
	fault := advOpts.UpdateOptions(&types.UpdateOptions{ChangedValue: []types.BaseOptionValue{option}}).Fault()
	if fault != nil {
		panic(fault)
	}

	h.Config.FileSystemVolume = nil
	if mounts != nil {
		h.Config.FileSystemVolume = &types.HostFileSystemVolumeInfo{
			VolumeTypeList: []string{"VMFS", "OTHER"},
			MountInfo:      mounts,
		}
	}

	// force at least a management network
	if len(networks) == 0 {
		networks = []string{defaultUnderlayBridgeName}
	}

	// purge pNICs from the template - it makes no sense to keep them for a sim host
	h.Config.Network.Pnic = make([]types.PhysicalNic, len(networks))

	// purge any IPs and MACs associated with existing NetConfigs for the host
	for cfgIdx := range h.Config.VirtualNicManagerInfo.NetConfig {
		config := &h.Config.VirtualNicManagerInfo.NetConfig[cfgIdx]
		for candidateIdx := range config.CandidateVnic {
			candidate := &config.CandidateVnic[candidateIdx]
			candidate.Spec.Ip.IpAddress = "0.0.0.0"
			candidate.Spec.Ip.SubnetMask = "0.0.0.0"
			candidate.Spec.Mac = "00:00:00:00:00:00"
		}
	}

	// The presence of a pNIC is used to indicate connectivity to a specific underlay. We construct an empty pNIC entry and specify the underly via
	// host.ConfigManager.AdvancedOptions. The pNIC will be populated with the MAC (accurate) and IP (divergence - we need to stash it somewhere) for the veth.
	// We create a NetConfig "management" entry for the first pNIC - this will be populated with the IP of the "host" container.

	// create a pNIC for each underlay
	for i, net := range networks {
		name := fmt.Sprintf("vmnic%d", i)

		// we don't have a natural field for annotating which pNIC is connected to which network, so stash it in an adv option.
		option := &types.OptionValue{
			Key:   advOptPrefixPnicToUnderlayPrefix + name,
			Value: net,
		}
		fault = advOpts.UpdateOptions(&types.UpdateOptions{ChangedValue: []types.BaseOptionValue{option}}).Fault()
		if fault != nil {
			panic(fault)
		}

		h.Config.Network.Pnic[i] = types.PhysicalNic{
			Key:             "key-vim.host.PhysicalNic-" + name,
			Device:          name,
			Pci:             fmt.Sprintf("0000:%2d:00.0", i+1),
			Driver:          "vcsim-bridge",
			DriverVersion:   "1.2.10.0",
			FirmwareVersion: "1.57, 0x80000185",
			LinkSpeed: &types.PhysicalNicLinkInfo{
				SpeedMb: 10000,
				Duplex:  true,
			},
			ValidLinkSpecification: []types.PhysicalNicLinkInfo{
				{
					SpeedMb: 10000,
					Duplex:  true,
				},
			},
			Spec: types.PhysicalNicSpec{
				Ip:                            &types.HostIpConfig{},
				LinkSpeed:                     (*types.PhysicalNicLinkInfo)(nil),
				EnableEnhancedNetworkingStack: types.NewBool(false),
				EnsInterruptEnabled:           types.NewBool(false),
			},
			WakeOnLanSupported: false,
			Mac:                "00:00:00:00:00:00",
			FcoeConfiguration: &types.FcoeConfig{
				PriorityClass: 3,
				SourceMac:     "00:00:00:00:00:00",
				VlanRange: []types.FcoeConfigVlanRange{
					{},
				},
				Capabilities: types.FcoeConfigFcoeCapabilities{},
				FcoeActive:   false,
			},
			VmDirectPathGen2Supported:             types.NewBool(false),
			VmDirectPathGen2SupportedMode:         "",
			ResourcePoolSchedulerAllowed:          types.NewBool(false),
			ResourcePoolSchedulerDisallowedReason: nil,
			AutoNegotiateSupported:                types.NewBool(true),
			EnhancedNetworkingStackSupported:      types.NewBool(false),
			EnsInterruptSupported:                 types.NewBool(false),
			RdmaDevice:                            "",
			DpuId:                                 "",
		}
	}

	// sanity check that everything's hung together sufficiently well
	details, err := h.getNetConfigInterface(ctx, "management")
	if err != nil {
		return err
	}

	if details.uplink == nil || details.uplink.Device != "vmnic0" {
		return fmt.Errorf("Config provided for host %s does not result in a consistent 'management' NetConfig that's bound to 'vmnic0'", h.Name)
	}

	return nil
}

// netConfigDetails is used to packaged up all the related network entities associated with a NetConfig binding
type netConfigDetails struct {
	nicType   string
	netconfig *types.VirtualNicManagerNetConfig
	vmk       *types.HostVirtualNic
	netstack  *types.HostNetStackInstance
	portgroup *types.HostPortGroup
	vswitch   *types.HostVirtualSwitch
	uplink    *types.PhysicalNic
}

// getNetConfigInterface returns the set of constructs active for a given nicType (eg. "management", "vmotion")
// This method is provided because the Config structure held by HostSystem is heavily interconnected but serialized and not cross-linked with pointers.
// As such there's a _lot_ of cross-referencing that needs to be done to navigate.
// The pNIC returned is the uplink associated with the vSwitch for the netconfig
func (h *HostSystem) getNetConfigInterface(ctx *Context, nicType string) (*netConfigDetails, error) {
	details := &netConfigDetails{
		nicType: nicType,
	}

	for i := range h.Config.VirtualNicManagerInfo.NetConfig {
		if h.Config.VirtualNicManagerInfo.NetConfig[i].NicType == nicType {
			details.netconfig = &h.Config.VirtualNicManagerInfo.NetConfig[i]
			break
		}
	}
	if details.netconfig == nil {
		return nil, fmt.Errorf("no matching NetConfig for NicType=%s", nicType)
	}

	if details.netconfig.SelectedVnic == nil {
		return details, nil
	}

	vnicKey := details.netconfig.SelectedVnic[0]
	for i := range details.netconfig.CandidateVnic {
		if details.netconfig.CandidateVnic[i].Key == vnicKey {
			details.vmk = &details.netconfig.CandidateVnic[i]
			break
		}
	}
	if details.vmk == nil {
		panic(fmt.Sprintf("NetConfig for host %s references non-existant vNIC key %s for %s nicType", h.Name, vnicKey, nicType))
	}

	portgroupName := details.vmk.Portgroup
	netstackKey := details.vmk.Spec.NetStackInstanceKey

	for i := range h.Config.Network.NetStackInstance {
		if h.Config.Network.NetStackInstance[i].Key == netstackKey {
			details.netstack = &h.Config.Network.NetStackInstance[i]
			break
		}
	}
	if details.netstack == nil {
		panic(fmt.Sprintf("NetConfig for host %s references non-existant NetStack key %s for %s nicType", h.Name, netstackKey, nicType))
	}

	for i := range h.Config.Network.Portgroup {
		// TODO: confirm correctness of this - seems weird it references the Spec.Name instead of the key like everything else.
		if h.Config.Network.Portgroup[i].Spec.Name == portgroupName {
			details.portgroup = &h.Config.Network.Portgroup[i]
			break
		}
	}
	if details.portgroup == nil {
		panic(fmt.Sprintf("NetConfig for host %s references non-existant PortGroup name %s for %s nicType", h.Name, portgroupName, nicType))
	}

	vswitchKey := details.portgroup.Vswitch
	for i := range h.Config.Network.Vswitch {
		if h.Config.Network.Vswitch[i].Key == vswitchKey {
			details.vswitch = &h.Config.Network.Vswitch[i]
			break
		}
	}
	if details.vswitch == nil {
		panic(fmt.Sprintf("NetConfig for host %s references non-existant vSwitch key %s for %s nicType", h.Name, vswitchKey, nicType))
	}

	if len(details.vswitch.Pnic) != 1 {
		// to change this, look at the Active NIC in the NicTeamingPolicy, but for now not worth it
		panic(fmt.Sprintf("vSwitch %s for host %s has multiple pNICs associated which is not supported.", vswitchKey, h.Name))
	}

	pnicKey := details.vswitch.Pnic[0]
	for i := range h.Config.Network.Pnic {
		if h.Config.Network.Pnic[i].Key == pnicKey {
			details.uplink = &h.Config.Network.Pnic[i]
			break
		}
	}
	if details.uplink == nil {
		panic(fmt.Sprintf("NetConfig for host %s references non-existant pNIC key %s for %s nicType", h.Name, pnicKey, nicType))
	}

	return details, nil
}

func (h *HostSystem) event() types.HostEvent {
	return types.HostEvent{
		Event: types.Event{
			Datacenter:      datacenterEventArgument(h),
			ComputeResource: h.eventArgumentParent(),
			Host:            h.eventArgument(),
		},
	}
}

func (h *HostSystem) eventArgument() *types.HostEventArgument {
	return &types.HostEventArgument{
		Host:                h.Self,
		EntityEventArgument: types.EntityEventArgument{Name: h.Name},
	}
}

func (h *HostSystem) eventArgumentParent() *types.ComputeResourceEventArgument {
	parent := hostParent(&h.HostSystem)

	return &types.ComputeResourceEventArgument{
		ComputeResource:     parent.Self,
		EntityEventArgument: types.EntityEventArgument{Name: parent.Name},
	}
}

func hostParent(host *mo.HostSystem) *mo.ComputeResource {
	switch parent := Map.Get(*host.Parent).(type) {
	case *mo.ComputeResource:
		return parent
	case *ClusterComputeResource:
		return &parent.ComputeResource
	default:
		return nil
	}
}

func addComputeResource(s *types.ComputeResourceSummary, h *HostSystem) {
	s.TotalCpu += h.Summary.Hardware.CpuMhz
	s.TotalMemory += h.Summary.Hardware.MemorySize
	s.NumCpuCores += h.Summary.Hardware.NumCpuCores
	s.NumCpuThreads += h.Summary.Hardware.NumCpuThreads
	s.EffectiveCpu += h.Summary.Hardware.CpuMhz
	s.EffectiveMemory += h.Summary.Hardware.MemorySize
	s.NumHosts++
	s.NumEffectiveHosts++
	s.OverallStatus = types.ManagedEntityStatusGreen
}

// CreateDefaultESX creates a standalone ESX
// Adds objects of type: Datacenter, Network, ComputeResource, ResourcePool and HostSystem
func CreateDefaultESX(ctx *Context, f *Folder) {
	dc := NewDatacenter(ctx, &f.Folder)

	host := NewHostSystem(esx.HostSystem)

	summary := new(types.ComputeResourceSummary)
	addComputeResource(summary, host)

	cr := &mo.ComputeResource{
		Summary: summary,
		Network: esx.Datacenter.Network,
	}
	cr.Self = *host.Parent
	cr.Name = host.Name
	cr.Host = append(cr.Host, host.Reference())
	host.Network = cr.Network
	ctx.Map.PutEntity(cr, host)
	cr.EnvironmentBrowser = newEnvironmentBrowser(ctx, host.Reference())

	pool := NewResourcePool()
	cr.ResourcePool = &pool.Self
	ctx.Map.PutEntity(cr, pool)
	pool.Owner = cr.Self

	folderPutChild(ctx, &ctx.Map.Get(dc.HostFolder).(*Folder).Folder, cr)
}

// CreateStandaloneHost uses esx.HostSystem as a template, applying the given spec
// and creating the ComputeResource parent and ResourcePool sibling.
func CreateStandaloneHost(ctx *Context, f *Folder, spec types.HostConnectSpec) (*HostSystem, types.BaseMethodFault) {
	if spec.HostName == "" {
		return nil, &types.NoHost{}
	}

	template := esx.HostSystem
	network := ctx.Map.getEntityDatacenter(f).defaultNetwork()

	if p := ctx.Map.FindByName(spec.UserName, f.ChildEntity); p != nil {
		cr := p.(*mo.ComputeResource)
		h := ctx.Map.Get(cr.Host[0])
		// "clone" an existing host from the inventory
		template = h.(*HostSystem).HostSystem
		template.Vm = nil
		network = cr.Network
	}

	pool := NewResourcePool()
	host := NewHostSystem(template)
	host.configure(ctx, spec, false)

	summary := new(types.ComputeResourceSummary)
	addComputeResource(summary, host)

	cr := &mo.ComputeResource{
		ConfigurationEx: &types.ComputeResourceConfigInfo{
			VmSwapPlacement: string(types.VirtualMachineConfigInfoSwapPlacementTypeVmDirectory),
		},
		Summary: summary,
	}

	ctx.Map.PutEntity(cr, ctx.Map.NewEntity(host))
	cr.EnvironmentBrowser = newEnvironmentBrowser(ctx, host.Reference())

	host.Summary.Host = &host.Self
	host.Config.Host = host.Self

	ctx.Map.PutEntity(cr, ctx.Map.NewEntity(pool))

	cr.Name = host.Name
	cr.Network = network
	cr.Host = append(cr.Host, host.Reference())
	cr.ResourcePool = &pool.Self

	folderPutChild(ctx, &f.Folder, cr)
	pool.Owner = cr.Self
	host.Network = cr.Network

	return host, nil
}

func (h *HostSystem) DestroyTask(ctx *Context, req *types.Destroy_Task) soap.HasFault {
	task := CreateTask(h, "destroy", func(t *Task) (types.AnyType, types.BaseMethodFault) {
		if len(h.Vm) > 0 {
			return nil, &types.ResourceInUse{}
		}

		ctx.postEvent(&types.HostRemovedEvent{HostEvent: h.event()})

		f := ctx.Map.getEntityParent(h, "Folder").(*Folder)
		folderRemoveChild(ctx, &f.Folder, h.Reference())
		err := h.sh.remove(ctx)

		if err != nil {
			return nil, &types.RuntimeFault{
				MethodFault: types.MethodFault{
					FaultCause: &types.LocalizedMethodFault{
						Fault:            &types.SystemErrorFault{Reason: err.Error()},
						LocalizedMessage: err.Error()}}}
		}

		// TODO: should there be events on lifecycle operations as with VMs?

		return nil, nil
	})

	return &methods.Destroy_TaskBody{
		Res: &types.Destroy_TaskResponse{
			Returnval: task.Run(ctx),
		},
	}
}

func (h *HostSystem) EnterMaintenanceModeTask(ctx *Context, spec *types.EnterMaintenanceMode_Task) soap.HasFault {
	task := CreateTask(h, "enterMaintenanceMode", func(t *Task) (types.AnyType, types.BaseMethodFault) {
		h.Runtime.InMaintenanceMode = true
		return nil, nil
	})

	return &methods.EnterMaintenanceMode_TaskBody{
		Res: &types.EnterMaintenanceMode_TaskResponse{
			Returnval: task.Run(ctx),
		},
	}
}

func (h *HostSystem) ExitMaintenanceModeTask(ctx *Context, spec *types.ExitMaintenanceMode_Task) soap.HasFault {
	task := CreateTask(h, "exitMaintenanceMode", func(t *Task) (types.AnyType, types.BaseMethodFault) {
		h.Runtime.InMaintenanceMode = false
		return nil, nil
	})

	return &methods.ExitMaintenanceMode_TaskBody{
		Res: &types.ExitMaintenanceMode_TaskResponse{
			Returnval: task.Run(ctx),
		},
	}
}

func (h *HostSystem) DisconnectHostTask(ctx *Context, spec *types.DisconnectHost_Task) soap.HasFault {
	task := CreateTask(h, "disconnectHost", func(t *Task) (types.AnyType, types.BaseMethodFault) {
		h.Runtime.ConnectionState = types.HostSystemConnectionStateDisconnected
		return nil, nil
	})

	return &methods.DisconnectHost_TaskBody{
		Res: &types.DisconnectHost_TaskResponse{
			Returnval: task.Run(ctx),
		},
	}
}

func (h *HostSystem) ReconnectHostTask(ctx *Context, spec *types.ReconnectHost_Task) soap.HasFault {
	task := CreateTask(h, "reconnectHost", func(t *Task) (types.AnyType, types.BaseMethodFault) {
		h.Runtime.ConnectionState = types.HostSystemConnectionStateConnected
		return nil, nil
	})

	return &methods.ReconnectHost_TaskBody{
		Res: &types.ReconnectHost_TaskResponse{
			Returnval: task.Run(ctx),
		},
	}
}