docker-compose/pkg/compose/create.go

/*
   Copyright 2020 Docker Compose CLI authors

   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 compose

import (
	"bytes"
	"context"
	"encoding/json"
	"errors"
	"fmt"
	"io/fs"
	"os"
	"path/filepath"
	"sort"
	"strconv"
	"strings"

	"github.com/compose-spec/compose-go/v2/types"
	"github.com/docker/compose/v2/internal/desktop"
	pathutil "github.com/docker/compose/v2/internal/paths"
	"github.com/docker/compose/v2/pkg/api"
	"github.com/docker/compose/v2/pkg/progress"
	"github.com/docker/compose/v2/pkg/utils"
	moby "github.com/docker/docker/api/types"
	"github.com/docker/docker/api/types/blkiodev"
	"github.com/docker/docker/api/types/container"
	"github.com/docker/docker/api/types/filters"
	"github.com/docker/docker/api/types/mount"
	"github.com/docker/docker/api/types/network"
	"github.com/docker/docker/api/types/strslice"
	"github.com/docker/docker/api/types/versions"
	volumetypes "github.com/docker/docker/api/types/volume"
	"github.com/docker/docker/errdefs"
	"github.com/docker/go-connections/nat"
	"github.com/sirupsen/logrus"
	cdi "tags.cncf.io/container-device-interface/pkg/parser"
)

type createOptions struct {
	AutoRemove        bool
	AttachStdin       bool
	UseNetworkAliases bool
	Labels            types.Labels
}

type createConfigs struct {
	Container *container.Config
	Host      *container.HostConfig
	Network   *network.NetworkingConfig
	Links     []string
}

func (s *composeService) Create(ctx context.Context, project *types.Project, createOpts api.CreateOptions) error {
	return progress.RunWithTitle(ctx, func(ctx context.Context) error {
		return s.create(ctx, project, createOpts)
	}, s.stdinfo(), "Creating")
}

func (s *composeService) create(ctx context.Context, project *types.Project, options api.CreateOptions) error {
	if len(options.Services) == 0 {
		options.Services = project.ServiceNames()
	}

	err := project.CheckContainerNameUnicity()
	if err != nil {
		return err
	}

	var observedState Containers
	observedState, err = s.getContainers(ctx, project.Name, oneOffInclude, true)
	if err != nil {
		return err
	}

	err = s.ensureImagesExists(ctx, project, options.Build, options.QuietPull)
	if err != nil {
		return err
	}

	prepareNetworks(project)

	if err := s.ensureNetworks(ctx, project.Networks); err != nil {
		return err
	}

	if err := s.ensureProjectVolumes(ctx, project); err != nil {
		return err
	}

	orphans := observedState.filter(isOrphaned(project))
	if len(orphans) > 0 && !options.IgnoreOrphans {
		if options.RemoveOrphans {
			err := s.removeContainers(ctx, orphans, nil, nil, false)
			if err != nil {
				return err
			}
		} else {
			logrus.Warnf("Found orphan containers (%s) for this project. If "+
				"you removed or renamed this service in your compose "+
				"file, you can run this command with the "+
				"--remove-orphans flag to clean it up.", orphans.names())
		}
	}
	return newConvergence(options.Services, observedState, s).apply(ctx, project, options)
}

func prepareNetworks(project *types.Project) {
	for k, nw := range project.Networks {
		nw.Labels = nw.Labels.Add(api.NetworkLabel, k)
		nw.Labels = nw.Labels.Add(api.ProjectLabel, project.Name)
		nw.Labels = nw.Labels.Add(api.VersionLabel, api.ComposeVersion)
		project.Networks[k] = nw
	}
}

func (s *composeService) ensureNetworks(ctx context.Context, networks types.Networks) error {
	for i, nw := range networks {
		err := s.ensureNetwork(ctx, &nw)
		if err != nil {
			return err
		}
		networks[i] = nw
	}
	return nil
}

func (s *composeService) ensureProjectVolumes(ctx context.Context, project *types.Project) error {
	for k, volume := range project.Volumes {
		volume.Labels = volume.Labels.Add(api.VolumeLabel, k)
		volume.Labels = volume.Labels.Add(api.ProjectLabel, project.Name)
		volume.Labels = volume.Labels.Add(api.VersionLabel, api.ComposeVersion)
		err := s.ensureVolume(ctx, volume, project.Name)
		if err != nil {
			return err
		}
	}

	err := func() error {
		if s.manageDesktopFileSharesEnabled(ctx) {
			// collect all the bind mount paths and try to set up file shares in
			// Docker Desktop for them
			var paths []string
			for _, svcName := range project.ServiceNames() {
				svc := project.Services[svcName]
				for _, vol := range svc.Volumes {
					if vol.Type != string(mount.TypeBind) {
						continue
					}
					p := filepath.Clean(vol.Source)
					if !filepath.IsAbs(p) {
						return fmt.Errorf("file share path is not absolute: %s", p)
					}
					if fi, err := os.Stat(p); errors.Is(err, fs.ErrNotExist) {
						// actual directory will be implicitly created when the
						// file share is initialized if it doesn't exist, so
						// need to filter out any that should not be auto-created
						if vol.Bind != nil && !vol.Bind.CreateHostPath {
							logrus.Debugf("Skipping creating file share for %q: does not exist and `create_host_path` is false", p)
							continue
						}
					} else if err != nil {
						// if we can't read the path, we won't be able to make
						// a file share for it
						logrus.Debugf("Skipping creating file share for %q: %v", p, err)
						continue
					} else if !fi.IsDir() {
						// ignore files & special types (e.g. Unix sockets)
						logrus.Debugf("Skipping creating file share for %q: not a directory", p)
						continue
					}

					paths = append(paths, p)
				}
			}

			// remove duplicate/unnecessary child paths and sort them for predictability
			paths = pathutil.EncompassingPaths(paths)
			sort.Strings(paths)

			fileShareManager := desktop.NewFileShareManager(s.desktopCli, project.Name, paths)
			if err := fileShareManager.EnsureExists(ctx); err != nil {
				return fmt.Errorf("initializing file shares: %w", err)
			}
		}
		return nil
	}()

	if err != nil {
		progress.ContextWriter(ctx).TailMsgf("Failed to prepare Synchronized file shares: %v", err)
	}
	return nil
}

func (s *composeService) getCreateConfigs(ctx context.Context,
	p *types.Project,
	service types.ServiceConfig,
	number int,
	inherit *moby.Container,
	opts createOptions,
) (createConfigs, error) {
	labels, err := s.prepareLabels(opts.Labels, service, number)
	if err != nil {
		return createConfigs{}, err
	}

	var (
		runCmd     strslice.StrSlice
		entrypoint strslice.StrSlice
	)
	if service.Command != nil {
		runCmd = strslice.StrSlice(service.Command)
	}
	if service.Entrypoint != nil {
		entrypoint = strslice.StrSlice(service.Entrypoint)
	}

	var (
		tty       = service.Tty
		stdinOpen = service.StdinOpen
	)

	proxyConfig := types.MappingWithEquals(s.configFile().ParseProxyConfig(s.apiClient().DaemonHost(), nil))
	env := proxyConfig.OverrideBy(service.Environment)

	var mainNwName string
	var mainNw *types.ServiceNetworkConfig
	if len(service.Networks) > 0 {
		mainNwName = service.NetworksByPriority()[0]
		mainNw = service.Networks[mainNwName]
	}

	macAddress, err := s.prepareContainerMACAddress(ctx, service, mainNw, mainNwName)
	if err != nil {
		return createConfigs{}, err
	}

	healthcheck, err := s.ToMobyHealthCheck(ctx, service.HealthCheck)
	if err != nil {
		return createConfigs{}, err
	}
	var containerConfig = container.Config{
		Hostname:        service.Hostname,
		Domainname:      service.DomainName,
		User:            service.User,
		ExposedPorts:    buildContainerPorts(service),
		Tty:             tty,
		OpenStdin:       stdinOpen,
		StdinOnce:       opts.AttachStdin && stdinOpen,
		AttachStdin:     opts.AttachStdin,
		AttachStderr:    true,
		AttachStdout:    true,
		Cmd:             runCmd,
		Image:           api.GetImageNameOrDefault(service, p.Name),
		WorkingDir:      service.WorkingDir,
		Entrypoint:      entrypoint,
		NetworkDisabled: service.NetworkMode == "disabled",
		MacAddress:      macAddress,
		Labels:          labels,
		StopSignal:      service.StopSignal,
		Env:             ToMobyEnv(env),
		Healthcheck:     healthcheck,
		StopTimeout:     ToSeconds(service.StopGracePeriod),
	} // VOLUMES/MOUNTS/FILESYSTEMS
	tmpfs := map[string]string{}
	for _, t := range service.Tmpfs {
		if arr := strings.SplitN(t, ":", 2); len(arr) > 1 {
			tmpfs[arr[0]] = arr[1]
		} else {
			tmpfs[arr[0]] = ""
		}
	}
	binds, mounts, err := s.buildContainerVolumes(ctx, *p, service, inherit)
	if err != nil {
		return createConfigs{}, err
	}

	// NETWORKING
	links, err := s.getLinks(ctx, p.Name, service, number)
	if err != nil {
		return createConfigs{}, err
	}
	apiVersion, err := s.RuntimeVersion(ctx)
	if err != nil {
		return createConfigs{}, err
	}
	networkMode, networkingConfig := defaultNetworkSettings(p, service, number, links, opts.UseNetworkAliases, apiVersion)
	portBindings := buildContainerPortBindingOptions(service)

	// MISC
	resources := getDeployResources(service)
	var logConfig container.LogConfig
	if service.Logging != nil {
		logConfig = container.LogConfig{
			Type:   service.Logging.Driver,
			Config: service.Logging.Options,
		}
	}
	securityOpts, unconfined, err := parseSecurityOpts(p, service.SecurityOpt)
	if err != nil {
		return createConfigs{}, err
	}

	hostConfig := container.HostConfig{
		AutoRemove:     opts.AutoRemove,
		Annotations:    service.Annotations,
		Binds:          binds,
		Mounts:         mounts,
		CapAdd:         strslice.StrSlice(service.CapAdd),
		CapDrop:        strslice.StrSlice(service.CapDrop),
		NetworkMode:    networkMode,
		Init:           service.Init,
		IpcMode:        container.IpcMode(service.Ipc),
		CgroupnsMode:   container.CgroupnsMode(service.Cgroup),
		ReadonlyRootfs: service.ReadOnly,
		RestartPolicy:  getRestartPolicy(service),
		ShmSize:        int64(service.ShmSize),
		Sysctls:        service.Sysctls,
		PortBindings:   portBindings,
		Resources:      resources,
		VolumeDriver:   service.VolumeDriver,
		VolumesFrom:    service.VolumesFrom,
		DNS:            service.DNS,
		DNSSearch:      service.DNSSearch,
		DNSOptions:     service.DNSOpts,
		ExtraHosts:     service.ExtraHosts.AsList(":"),
		SecurityOpt:    securityOpts,
		StorageOpt:     service.StorageOpt,
		UsernsMode:     container.UsernsMode(service.UserNSMode),
		UTSMode:        container.UTSMode(service.Uts),
		Privileged:     service.Privileged,
		PidMode:        container.PidMode(service.Pid),
		Tmpfs:          tmpfs,
		Isolation:      container.Isolation(service.Isolation),
		Runtime:        service.Runtime,
		LogConfig:      logConfig,
		GroupAdd:       service.GroupAdd,
		Links:          links,
		OomScoreAdj:    int(service.OomScoreAdj),
	}

	if unconfined {
		hostConfig.MaskedPaths = []string{}
		hostConfig.ReadonlyPaths = []string{}
	}

	cfgs := createConfigs{
		Container: &containerConfig,
		Host:      &hostConfig,
		Network:   networkingConfig,
		Links:     links,
	}
	return cfgs, nil
}

// prepareContainerMACAddress handles the service-level mac_address field and the newer mac_address field added to service
// network config. This newer field is only compatible with the Engine API v1.44 (and onwards), and this API version
// also deprecates the container-wide mac_address field. Thus, this method will validate service config and mutate the
// passed mainNw to provide backward-compatibility whenever possible.
//
// It returns the container-wide MAC address, but this value will be kept empty for newer API versions.
func (s *composeService) prepareContainerMACAddress(ctx context.Context, service types.ServiceConfig, mainNw *types.ServiceNetworkConfig, nwName string) (string, error) {
	version, err := s.RuntimeVersion(ctx)
	if err != nil {
		return "", err
	}

	// Engine API 1.44 added support for endpoint-specific MAC address and now returns a warning when a MAC address is
	// set in container.Config. Thus, we have to jump through a number of hoops:
	//
	// 1. Top-level mac_address and main endpoint's MAC address should be the same ;
	// 2. If supported by the API, top-level mac_address should be migrated to the main endpoint and container.Config
	//    should be kept empty ;
	// 3. Otherwise, the endpoint mac_address should be set in container.Config and no other endpoint-specific
	//    mac_address can be specified. If that's the case, use top-level mac_address ;
	//
	// After that, if an endpoint mac_address is set, it's either user-defined or migrated by the code below, so
	// there's no need to check for API version in defaultNetworkSettings.
	macAddress := service.MacAddress
	if macAddress != "" && mainNw != nil && mainNw.MacAddress != "" && mainNw.MacAddress != macAddress {
		return "", fmt.Errorf("the service-level mac_address should have the same value as network %s", nwName)
	}
	if versions.GreaterThanOrEqualTo(version, "1.44") {
		if mainNw != nil && mainNw.MacAddress == "" {
			mainNw.MacAddress = macAddress
		}
		macAddress = ""
	} else if len(service.Networks) > 0 {
		var withMacAddress []string
		for nwName, nw := range service.Networks {
			if nw != nil && nw.MacAddress != "" {
				withMacAddress = append(withMacAddress, nwName)
			}
		}

		if len(withMacAddress) > 1 {
			return "", fmt.Errorf("a MAC address is specified for multiple networks (%s), but this feature requires Docker Engine 1.44 or later (currently: %s)", strings.Join(withMacAddress, ", "), version)
		}

		if mainNw != nil {
			macAddress = mainNw.MacAddress
		}
	}

	return macAddress, nil
}

func getAliases(project *types.Project, service types.ServiceConfig, serviceIndex int, networkKey string, useNetworkAliases bool) []string {
	aliases := []string{getContainerName(project.Name, service, serviceIndex)}
	if useNetworkAliases {
		aliases = append(aliases, service.Name)
		if cfg := service.Networks[networkKey]; cfg != nil {
			aliases = append(aliases, cfg.Aliases...)
		}
	}
	return aliases
}

func createEndpointSettings(p *types.Project, service types.ServiceConfig, serviceIndex int, networkKey string, links []string, useNetworkAliases bool) *network.EndpointSettings {
	config := service.Networks[networkKey]
	var ipam *network.EndpointIPAMConfig
	var (
		ipv4Address string
		ipv6Address string
		macAddress  string
		driverOpts  types.Options
	)
	if config != nil {
		ipv4Address = config.Ipv4Address
		ipv6Address = config.Ipv6Address
		ipam = &network.EndpointIPAMConfig{
			IPv4Address:  ipv4Address,
			IPv6Address:  ipv6Address,
			LinkLocalIPs: config.LinkLocalIPs,
		}
		macAddress = config.MacAddress
		driverOpts = config.DriverOpts
	}
	return &network.EndpointSettings{
		Aliases:     getAliases(p, service, serviceIndex, networkKey, useNetworkAliases),
		Links:       links,
		IPAddress:   ipv4Address,
		IPv6Gateway: ipv6Address,
		IPAMConfig:  ipam,
		MacAddress:  macAddress,
		DriverOpts:  driverOpts,
	}
}

// copy/pasted from https://github.com/docker/cli/blob/9de1b162f/cli/command/container/opts.go#L673-L697 + RelativePath
// TODO find so way to share this code with docker/cli
func parseSecurityOpts(p *types.Project, securityOpts []string) ([]string, bool, error) {
	var (
		unconfined bool
		parsed     []string
	)
	for _, opt := range securityOpts {
		if opt == "systempaths=unconfined" {
			unconfined = true
			continue
		}
		con := strings.SplitN(opt, "=", 2)
		if len(con) == 1 && con[0] != "no-new-privileges" {
			if strings.Contains(opt, ":") {
				con = strings.SplitN(opt, ":", 2)
			} else {
				return securityOpts, false, fmt.Errorf("Invalid security-opt: %q", opt)
			}
		}
		if con[0] == "seccomp" && con[1] != "unconfined" {
			f, err := os.ReadFile(p.RelativePath(con[1]))
			if err != nil {
				return securityOpts, false, fmt.Errorf("opening seccomp profile (%s) failed: %w", con[1], err)
			}
			b := bytes.NewBuffer(nil)
			if err := json.Compact(b, f); err != nil {
				return securityOpts, false, fmt.Errorf("compacting json for seccomp profile (%s) failed: %w", con[1], err)
			}
			parsed = append(parsed, fmt.Sprintf("seccomp=%s", b.Bytes()))
		} else {
			parsed = append(parsed, opt)
		}
	}

	return parsed, unconfined, nil
}

func (s *composeService) prepareLabels(labels types.Labels, service types.ServiceConfig, number int) (map[string]string, error) {
	hash, err := ServiceHash(service)
	if err != nil {
		return nil, err
	}
	labels[api.ConfigHashLabel] = hash

	if number > 0 {
		// One-off containers are not indexed
		labels[api.ContainerNumberLabel] = strconv.Itoa(number)
	}

	var dependencies []string
	for s, d := range service.DependsOn {
		dependencies = append(dependencies, fmt.Sprintf("%s:%s:%t", s, d.Condition, d.Restart))
	}
	labels[api.DependenciesLabel] = strings.Join(dependencies, ",")
	return labels, nil
}

// defaultNetworkSettings determines the container.NetworkMode and corresponding network.NetworkingConfig (nil if not applicable).
func defaultNetworkSettings(
	project *types.Project,
	service types.ServiceConfig,
	serviceIndex int,
	links []string,
	useNetworkAliases bool,
	version string,
) (container.NetworkMode, *network.NetworkingConfig) {
	if service.NetworkMode != "" {
		return container.NetworkMode(service.NetworkMode), nil
	}

	if len(project.Networks) == 0 {
		return "none", nil
	}

	var primaryNetworkKey string
	if len(service.Networks) > 0 {
		primaryNetworkKey = service.NetworksByPriority()[0]
	} else {
		primaryNetworkKey = "default"
	}
	primaryNetworkMobyNetworkName := project.Networks[primaryNetworkKey].Name
	endpointsConfig := map[string]*network.EndpointSettings{
		primaryNetworkMobyNetworkName: createEndpointSettings(project, service, serviceIndex, primaryNetworkKey, links, useNetworkAliases),
	}

	// Starting from API version 1.44, the Engine will take several EndpointsConfigs
	// so we can pass all the extra networks we want the container to be connected to
	// in the network configuration instead of connecting the container to each extra
	// network individually after creation.
	if versions.GreaterThanOrEqualTo(version, "1.44") && len(service.Networks) > 1 {
		serviceNetworks := service.NetworksByPriority()
		for _, networkKey := range serviceNetworks[1:] {
			mobyNetworkName := project.Networks[networkKey].Name
			epSettings := createEndpointSettings(project, service, serviceIndex, networkKey, links, useNetworkAliases)
			endpointsConfig[mobyNetworkName] = epSettings
		}
	}

	networkConfig := &network.NetworkingConfig{
		EndpointsConfig: endpointsConfig,
	}

	// From the Engine API docs:
	// > Supported standard values are: bridge, host, none, and container:<name|id>.
	// > Any other value is taken as a custom network's name to which this container should connect to.
	return container.NetworkMode(primaryNetworkMobyNetworkName), networkConfig
}

func getRestartPolicy(service types.ServiceConfig) container.RestartPolicy {
	var restart container.RestartPolicy
	if service.Restart != "" {
		split := strings.Split(service.Restart, ":")
		var attempts int
		if len(split) > 1 {
			attempts, _ = strconv.Atoi(split[1])
		}
		restart = container.RestartPolicy{
			Name:              mapRestartPolicyCondition(split[0]),
			MaximumRetryCount: attempts,
		}
	}
	if service.Deploy != nil && service.Deploy.RestartPolicy != nil {
		policy := *service.Deploy.RestartPolicy
		var attempts int
		if policy.MaxAttempts != nil {
			attempts = int(*policy.MaxAttempts)
		}
		restart = container.RestartPolicy{
			Name:              mapRestartPolicyCondition(policy.Condition),
			MaximumRetryCount: attempts,
		}
	}
	return restart
}

func mapRestartPolicyCondition(condition string) container.RestartPolicyMode {
	// map definitions of deploy.restart_policy to engine definitions
	switch condition {
	case "none", "no":
		return container.RestartPolicyDisabled
	case "on-failure":
		return container.RestartPolicyOnFailure
	case "unless-stopped":
		return container.RestartPolicyUnlessStopped
	case "any", "always":
		return container.RestartPolicyAlways
	default:
		return container.RestartPolicyMode(condition)
	}
}

func getDeployResources(s types.ServiceConfig) container.Resources {
	var swappiness *int64
	if s.MemSwappiness != 0 {
		val := int64(s.MemSwappiness)
		swappiness = &val
	}
	resources := container.Resources{
		CgroupParent:       s.CgroupParent,
		Memory:             int64(s.MemLimit),
		MemorySwap:         int64(s.MemSwapLimit),
		MemorySwappiness:   swappiness,
		MemoryReservation:  int64(s.MemReservation),
		OomKillDisable:     &s.OomKillDisable,
		CPUCount:           s.CPUCount,
		CPUPeriod:          s.CPUPeriod,
		CPUQuota:           s.CPUQuota,
		CPURealtimePeriod:  s.CPURTPeriod,
		CPURealtimeRuntime: s.CPURTRuntime,
		CPUShares:          s.CPUShares,
		NanoCPUs:           int64(s.CPUS * 1e9),
		CPUPercent:         int64(s.CPUPercent * 100),
		CpusetCpus:         s.CPUSet,
		DeviceCgroupRules:  s.DeviceCgroupRules,
	}

	if s.PidsLimit != 0 {
		resources.PidsLimit = &s.PidsLimit
	}

	setBlkio(s.BlkioConfig, &resources)

	if s.Deploy != nil {
		setLimits(s.Deploy.Resources.Limits, &resources)
		setReservations(s.Deploy.Resources.Reservations, &resources)
	}

	var cdiDeviceNames []string
	for _, device := range s.Devices {

		if device.Source == device.Target && cdi.IsQualifiedName(device.Source) {
			cdiDeviceNames = append(cdiDeviceNames, device.Source)
			continue
		}

		resources.Devices = append(resources.Devices, container.DeviceMapping{
			PathOnHost:        device.Source,
			PathInContainer:   device.Target,
			CgroupPermissions: device.Permissions,
		})
	}

	if len(cdiDeviceNames) > 0 {
		resources.DeviceRequests = append(resources.DeviceRequests, container.DeviceRequest{
			Driver:    "cdi",
			DeviceIDs: cdiDeviceNames,
		})
	}

	for _, gpus := range s.Gpus {
		resources.DeviceRequests = append(resources.DeviceRequests, container.DeviceRequest{
			Driver:       gpus.Driver,
			Count:        int(gpus.Count),
			DeviceIDs:    gpus.IDs,
			Capabilities: [][]string{append(gpus.Capabilities, "gpu")},
			Options:      gpus.Options,
		})
	}

	ulimits := toUlimits(s.Ulimits)
	resources.Ulimits = ulimits
	return resources
}

func toUlimits(m map[string]*types.UlimitsConfig) []*container.Ulimit {
	var ulimits []*container.Ulimit
	for name, u := range m {
		soft := u.Single
		if u.Soft != 0 {
			soft = u.Soft
		}
		hard := u.Single
		if u.Hard != 0 {
			hard = u.Hard
		}
		ulimits = append(ulimits, &container.Ulimit{
			Name: name,
			Hard: int64(hard),
			Soft: int64(soft),
		})
	}
	return ulimits
}

func setReservations(reservations *types.Resource, resources *container.Resources) {
	if reservations == nil {
		return
	}
	// Cpu reservation is a swarm option and PIDs is only a limit
	// So we only need to map memory reservation and devices
	if reservations.MemoryBytes != 0 {
		resources.MemoryReservation = int64(reservations.MemoryBytes)
	}

	for _, device := range reservations.Devices {
		resources.DeviceRequests = append(resources.DeviceRequests, container.DeviceRequest{
			Capabilities: [][]string{device.Capabilities},
			Count:        int(device.Count),
			DeviceIDs:    device.IDs,
			Driver:       device.Driver,
			Options:      device.Options,
		})
	}
}

func setLimits(limits *types.Resource, resources *container.Resources) {
	if limits == nil {
		return
	}
	if limits.MemoryBytes != 0 {
		resources.Memory = int64(limits.MemoryBytes)
	}
	if limits.NanoCPUs != 0 {
		resources.NanoCPUs = int64(limits.NanoCPUs * 1e9)
	}
	if limits.Pids > 0 {
		resources.PidsLimit = &limits.Pids
	}
}

func setBlkio(blkio *types.BlkioConfig, resources *container.Resources) {
	if blkio == nil {
		return
	}
	resources.BlkioWeight = blkio.Weight
	for _, b := range blkio.WeightDevice {
		resources.BlkioWeightDevice = append(resources.BlkioWeightDevice, &blkiodev.WeightDevice{
			Path:   b.Path,
			Weight: b.Weight,
		})
	}
	for _, b := range blkio.DeviceReadBps {
		resources.BlkioDeviceReadBps = append(resources.BlkioDeviceReadBps, &blkiodev.ThrottleDevice{
			Path: b.Path,
			Rate: uint64(b.Rate),
		})
	}
	for _, b := range blkio.DeviceReadIOps {
		resources.BlkioDeviceReadIOps = append(resources.BlkioDeviceReadIOps, &blkiodev.ThrottleDevice{
			Path: b.Path,
			Rate: uint64(b.Rate),
		})
	}
	for _, b := range blkio.DeviceWriteBps {
		resources.BlkioDeviceWriteBps = append(resources.BlkioDeviceWriteBps, &blkiodev.ThrottleDevice{
			Path: b.Path,
			Rate: uint64(b.Rate),
		})
	}
	for _, b := range blkio.DeviceWriteIOps {
		resources.BlkioDeviceWriteIOps = append(resources.BlkioDeviceWriteIOps, &blkiodev.ThrottleDevice{
			Path: b.Path,
			Rate: uint64(b.Rate),
		})
	}
}

func buildContainerPorts(s types.ServiceConfig) nat.PortSet {
	ports := nat.PortSet{}
	for _, s := range s.Expose {
		p := nat.Port(s)
		ports[p] = struct{}{}
	}
	for _, p := range s.Ports {
		p := nat.Port(fmt.Sprintf("%d/%s", p.Target, p.Protocol))
		ports[p] = struct{}{}
	}
	return ports
}

func buildContainerPortBindingOptions(s types.ServiceConfig) nat.PortMap {
	bindings := nat.PortMap{}
	for _, port := range s.Ports {
		p := nat.Port(fmt.Sprintf("%d/%s", port.Target, port.Protocol))
		binding := nat.PortBinding{
			HostIP:   port.HostIP,
			HostPort: port.Published,
		}
		bindings[p] = append(bindings[p], binding)
	}
	return bindings
}

func getDependentServiceFromMode(mode string) string {
	if strings.HasPrefix(
		mode,
		types.NetworkModeServicePrefix,
	) {
		return mode[len(types.NetworkModeServicePrefix):]
	}
	return ""
}

func (s *composeService) buildContainerVolumes(
	ctx context.Context,
	p types.Project,
	service types.ServiceConfig,
	inherit *moby.Container,
) ([]string, []mount.Mount, error) {
	var mounts []mount.Mount
	var binds []string

	image := api.GetImageNameOrDefault(service, p.Name)
	imgInspect, _, err := s.apiClient().ImageInspectWithRaw(ctx, image)
	if err != nil {
		return nil, nil, err
	}

	mountOptions, err := buildContainerMountOptions(p, service, imgInspect, inherit)
	if err != nil {
		return nil, nil, err
	}

MOUNTS:
	for _, m := range mountOptions {
		if m.Type == mount.TypeNamedPipe {
			mounts = append(mounts, m)
			continue
		}
		if m.Type == mount.TypeBind {
			// `Mount` is preferred but does not offer option to created host path if missing
			// so `Bind` API is used here with raw volume string
			// see https://github.com/moby/moby/issues/43483
			for _, v := range service.Volumes {
				if v.Target == m.Target {
					switch {
					case string(m.Type) != v.Type:
						v.Source = m.Source
						fallthrough
					case !requireMountAPI(v.Bind):
						binds = append(binds, v.String())
						continue MOUNTS
					}
				}
			}
		}
		mounts = append(mounts, m)
	}
	return binds, mounts, nil
}

// requireMountAPI check if Bind declaration can be implemented by the plain old Bind API or uses any of the advanced
// options which require use of Mount API
func requireMountAPI(bind *types.ServiceVolumeBind) bool {
	switch {
	case bind == nil:
		return false
	case !bind.CreateHostPath:
		return true
	case bind.Propagation != "":
		return true
	case bind.Recursive != "":
		return true
	default:
		return false
	}
}

func buildContainerMountOptions(p types.Project, s types.ServiceConfig, img moby.ImageInspect, inherit *moby.Container) ([]mount.Mount, error) {
	var mounts = map[string]mount.Mount{}
	if inherit != nil {
		for _, m := range inherit.Mounts {
			if m.Type == "tmpfs" {
				continue
			}
			src := m.Source
			if m.Type == "volume" {
				src = m.Name
			}

			if img.Config != nil {
				if _, ok := img.Config.Volumes[m.Destination]; ok {
					// inherit previous container's anonymous volume
					mounts[m.Destination] = mount.Mount{
						Type:     m.Type,
						Source:   src,
						Target:   m.Destination,
						ReadOnly: !m.RW,
					}
				}
			}
			volumes := []types.ServiceVolumeConfig{}
			for _, v := range s.Volumes {
				if v.Target != m.Destination || v.Source != "" {
					volumes = append(volumes, v)
					continue
				}
				// inherit previous container's anonymous volume
				mounts[m.Destination] = mount.Mount{
					Type:     m.Type,
					Source:   src,
					Target:   m.Destination,
					ReadOnly: !m.RW,
				}
			}
			s.Volumes = volumes
		}
	}

	mounts, err := fillBindMounts(p, s, mounts)
	if err != nil {
		return nil, err
	}

	values := make([]mount.Mount, 0, len(mounts))
	for _, v := range mounts {
		values = append(values, v)
	}
	return values, nil
}

func fillBindMounts(p types.Project, s types.ServiceConfig, m map[string]mount.Mount) (map[string]mount.Mount, error) {
	for _, v := range s.Volumes {
		bindMount, err := buildMount(p, v)
		if err != nil {
			return nil, err
		}
		m[bindMount.Target] = bindMount
	}

	secrets, err := buildContainerSecretMounts(p, s)
	if err != nil {
		return nil, err
	}
	for _, s := range secrets {
		if _, found := m[s.Target]; found {
			continue
		}
		m[s.Target] = s
	}

	configs, err := buildContainerConfigMounts(p, s)
	if err != nil {
		return nil, err
	}
	for _, c := range configs {
		if _, found := m[c.Target]; found {
			continue
		}
		m[c.Target] = c
	}
	return m, nil
}

func buildContainerConfigMounts(p types.Project, s types.ServiceConfig) ([]mount.Mount, error) {
	var mounts = map[string]mount.Mount{}

	configsBaseDir := "/"
	for _, config := range s.Configs {
		target := config.Target
		if config.Target == "" {
			target = configsBaseDir + config.Source
		} else if !isAbsTarget(config.Target) {
			target = configsBaseDir + config.Target
		}

		if config.UID != "" || config.GID != "" || config.Mode != nil {
			logrus.Warn("config `uid`, `gid` and `mode` are not supported, they will be ignored")
		}

		definedConfig := p.Configs[config.Source]
		if definedConfig.External {
			return nil, fmt.Errorf("unsupported external config %s", definedConfig.Name)
		}

		if definedConfig.Driver != "" {
			return nil, errors.New("Docker Compose does not support configs.*.driver")
		}
		if definedConfig.TemplateDriver != "" {
			return nil, errors.New("Docker Compose does not support configs.*.template_driver")
		}

		if definedConfig.Environment != "" || definedConfig.Content != "" {
			continue
		}

		bindMount, err := buildMount(p, types.ServiceVolumeConfig{
			Type:     types.VolumeTypeBind,
			Source:   definedConfig.File,
			Target:   target,
			ReadOnly: true,
		})
		if err != nil {
			return nil, err
		}
		mounts[target] = bindMount
	}
	values := make([]mount.Mount, 0, len(mounts))
	for _, v := range mounts {
		values = append(values, v)
	}
	return values, nil
}

func buildContainerSecretMounts(p types.Project, s types.ServiceConfig) ([]mount.Mount, error) {
	var mounts = map[string]mount.Mount{}

	secretsDir := "/run/secrets/"
	for _, secret := range s.Secrets {
		target := secret.Target
		if secret.Target == "" {
			target = secretsDir + secret.Source
		} else if !isAbsTarget(secret.Target) {
			target = secretsDir + secret.Target
		}

		if secret.UID != "" || secret.GID != "" || secret.Mode != nil {
			logrus.Warn("secrets `uid`, `gid` and `mode` are not supported, they will be ignored")
		}

		definedSecret := p.Secrets[secret.Source]
		if definedSecret.External {
			return nil, fmt.Errorf("unsupported external secret %s", definedSecret.Name)
		}

		if definedSecret.Driver != "" {
			return nil, errors.New("Docker Compose does not support secrets.*.driver")
		}
		if definedSecret.TemplateDriver != "" {
			return nil, errors.New("Docker Compose does not support secrets.*.template_driver")
		}

		if definedSecret.Environment != "" {
			continue
		}

		if _, err := os.Stat(definedSecret.File); os.IsNotExist(err) {
			logrus.Warnf("secret file %s does not exist", definedSecret.Name)
		}

		mnt, err := buildMount(p, types.ServiceVolumeConfig{
			Type:     types.VolumeTypeBind,
			Source:   definedSecret.File,
			Target:   target,
			ReadOnly: true,
			Bind: &types.ServiceVolumeBind{
				CreateHostPath: false,
			},
		})
		if err != nil {
			return nil, err
		}
		mounts[target] = mnt
	}
	values := make([]mount.Mount, 0, len(mounts))
	for _, v := range mounts {
		values = append(values, v)
	}
	return values, nil
}

func isAbsTarget(p string) bool {
	return isUnixAbs(p) || isWindowsAbs(p)
}

func isUnixAbs(p string) bool {
	return strings.HasPrefix(p, "/")
}

func isWindowsAbs(p string) bool {
	if strings.HasPrefix(p, "\\\\") {
		return true
	}
	if len(p) > 2 && p[1] == ':' {
		return p[2] == '\\'
	}
	return false
}

func buildMount(project types.Project, volume types.ServiceVolumeConfig) (mount.Mount, error) {
	source := volume.Source
	// on windows, filepath.IsAbs(source) is false for unix style abs path like /var/run/docker.sock.
	// do not replace these with  filepath.Abs(source) that will include a default drive.
	if volume.Type == types.VolumeTypeBind && !filepath.IsAbs(source) && !strings.HasPrefix(source, "/") {
		// volume source has already been prefixed with workdir if required, by compose-go project loader
		var err error
		source, err = filepath.Abs(source)
		if err != nil {
			return mount.Mount{}, err
		}
	}
	if volume.Type == types.VolumeTypeVolume {
		if volume.Source != "" {
			pVolume, ok := project.Volumes[volume.Source]
			if ok {
				source = pVolume.Name
			}
		}
	}

	bind, vol, tmpfs := buildMountOptions(volume)

	if bind != nil {
		volume.Type = types.VolumeTypeBind
	}

	return mount.Mount{
		Type:          mount.Type(volume.Type),
		Source:        source,
		Target:        volume.Target,
		ReadOnly:      volume.ReadOnly,
		Consistency:   mount.Consistency(volume.Consistency),
		BindOptions:   bind,
		VolumeOptions: vol,
		TmpfsOptions:  tmpfs,
	}, nil
}

func buildMountOptions(volume types.ServiceVolumeConfig) (*mount.BindOptions, *mount.VolumeOptions, *mount.TmpfsOptions) {
	switch volume.Type {
	case "bind":
		if volume.Volume != nil {
			logrus.Warnf("mount of type `bind` should not define `volume` option")
		}
		if volume.Tmpfs != nil {
			logrus.Warnf("mount of type `bind` should not define `tmpfs` option")
		}
		return buildBindOption(volume.Bind), nil, nil
	case "volume":
		if volume.Bind != nil {
			logrus.Warnf("mount of type `volume` should not define `bind` option")
		}
		if volume.Tmpfs != nil {
			logrus.Warnf("mount of type `volume` should not define `tmpfs` option")
		}
		return nil, buildVolumeOptions(volume.Volume), nil
	case "tmpfs":
		if volume.Bind != nil {
			logrus.Warnf("mount of type `tmpfs` should not define `bind` option")
		}
		if volume.Volume != nil {
			logrus.Warnf("mount of type `tmpfs` should not define `volume` option")
		}
		return nil, nil, buildTmpfsOptions(volume.Tmpfs)
	}
	return nil, nil, nil
}

func buildBindOption(bind *types.ServiceVolumeBind) *mount.BindOptions {
	if bind == nil {
		return nil
	}
	opts := &mount.BindOptions{
		Propagation:      mount.Propagation(bind.Propagation),
		CreateMountpoint: bind.CreateHostPath,
	}
	switch bind.Recursive {
	case "disabled":
		opts.NonRecursive = true
	case "writable":
		opts.ReadOnlyNonRecursive = true
	case "readonly":
		opts.ReadOnlyForceRecursive = true
	}
	return opts
}

func buildVolumeOptions(vol *types.ServiceVolumeVolume) *mount.VolumeOptions {
	if vol == nil {
		return nil
	}
	return &mount.VolumeOptions{
		NoCopy:  vol.NoCopy,
		Subpath: vol.Subpath,
		// Labels:       , // FIXME missing from model ?
		// DriverConfig: , // FIXME missing from model ?
	}
}

func buildTmpfsOptions(tmpfs *types.ServiceVolumeTmpfs) *mount.TmpfsOptions {
	if tmpfs == nil {
		return nil
	}
	return &mount.TmpfsOptions{
		SizeBytes: int64(tmpfs.Size),
		Mode:      os.FileMode(tmpfs.Mode),
	}
}

func (s *composeService) ensureNetwork(ctx context.Context, n *types.NetworkConfig) error {
	if n.External {
		return s.resolveExternalNetwork(ctx, n)
	}

	err := s.resolveOrCreateNetwork(ctx, n)
	if errdefs.IsConflict(err) {
		// Maybe another execution of `docker compose up|run` created same network
		// let's retry once
		return s.resolveOrCreateNetwork(ctx, n)
	}
	return err
}

func (s *composeService) resolveOrCreateNetwork(ctx context.Context, n *types.NetworkConfig) error { //nolint:gocyclo
	expectedNetworkLabel := n.Labels[api.NetworkLabel]
	expectedProjectLabel := n.Labels[api.ProjectLabel]

	// First, try to find a unique network matching by name or ID
	inspect, err := s.apiClient().NetworkInspect(ctx, n.Name, network.InspectOptions{})
	if err == nil {
		// NetworkInspect will match on ID prefix, so double check we get the expected one
		// as looking for network named `db` we could erroneously match network ID `db9086999caf`
		if inspect.Name == n.Name || inspect.ID == n.Name {
			p, ok := inspect.Labels[api.ProjectLabel]
			if !ok {
				logrus.Warnf("a network with name %s exists but was not created by compose.\n"+
					"Set `external: true` to use an existing network", n.Name)
			} else if p != expectedProjectLabel {
				logrus.Warnf("a network with name %s exists but was not created for project %q.\n"+
					"Set `external: true` to use an existing network", n.Name, expectedProjectLabel)
			}
			if inspect.Labels[api.NetworkLabel] != expectedNetworkLabel {
				return fmt.Errorf(
					"network %s was found but has incorrect label %s set to %q (expected: %q)",
					n.Name,
					api.NetworkLabel,
					inspect.Labels[api.NetworkLabel],
					expectedNetworkLabel,
				)
			}
			return nil
		}
	}
	// ignore other errors. Typically, an ambiguous request by name results in some generic `invalidParameter` error

	// Either not found, or name is ambiguous - use NetworkList to list by name
	networks, err := s.apiClient().NetworkList(ctx, network.ListOptions{
		Filters: filters.NewArgs(filters.Arg("name", n.Name)),
	})
	if err != nil {
		return err
	}

	// NetworkList Matches all or part of a network name, so we have to filter for a strict match
	networks = utils.Filter(networks, func(net network.Summary) bool {
		return net.Name == n.Name
	})

	for _, net := range networks {
		if net.Labels[api.ProjectLabel] == expectedProjectLabel &&
			net.Labels[api.NetworkLabel] == expectedNetworkLabel {
			return nil
		}
	}

	// we could have set NetworkList with a projectFilter and networkFilter but not doing so allows to catch this
	// scenario were a network with same name exists but doesn't have label, and use of `CheckDuplicate: true`
	// prevents to create another one.
	if len(networks) > 0 {
		logrus.Warnf("a network with name %s exists but was not created by compose.\n"+
			"Set `external: true` to use an existing network", n.Name)
		return nil
	}

	var ipam *network.IPAM
	if n.Ipam.Config != nil {
		var config []network.IPAMConfig
		for _, pool := range n.Ipam.Config {
			config = append(config, network.IPAMConfig{
				Subnet:     pool.Subnet,
				IPRange:    pool.IPRange,
				Gateway:    pool.Gateway,
				AuxAddress: pool.AuxiliaryAddresses,
			})
		}
		ipam = &network.IPAM{
			Driver: n.Ipam.Driver,
			Config: config,
		}
	}
	createOpts := network.CreateOptions{
		Labels:     n.Labels,
		Driver:     n.Driver,
		Options:    n.DriverOpts,
		Internal:   n.Internal,
		Attachable: n.Attachable,
		IPAM:       ipam,
		EnableIPv6: n.EnableIPv6,
	}

	if n.Ipam.Driver != "" || len(n.Ipam.Config) > 0 {
		createOpts.IPAM = &network.IPAM{}
	}

	if n.Ipam.Driver != "" {
		createOpts.IPAM.Driver = n.Ipam.Driver
	}

	for _, ipamConfig := range n.Ipam.Config {
		config := network.IPAMConfig{
			Subnet:     ipamConfig.Subnet,
			IPRange:    ipamConfig.IPRange,
			Gateway:    ipamConfig.Gateway,
			AuxAddress: ipamConfig.AuxiliaryAddresses,
		}
		createOpts.IPAM.Config = append(createOpts.IPAM.Config, config)
	}
	networkEventName := fmt.Sprintf("Network %s", n.Name)
	w := progress.ContextWriter(ctx)
	w.Event(progress.CreatingEvent(networkEventName))

	_, err = s.apiClient().NetworkCreate(ctx, n.Name, createOpts)
	if err != nil {
		w.Event(progress.ErrorEvent(networkEventName))
		return fmt.Errorf("failed to create network %s: %w", n.Name, err)
	}
	w.Event(progress.CreatedEvent(networkEventName))
	return nil
}

func (s *composeService) resolveExternalNetwork(ctx context.Context, n *types.NetworkConfig) error {
	// NetworkInspect will match on ID prefix, so NetworkList with a name
	// filter is used to look for an exact match to prevent e.g. a network
	// named `db` from getting erroneously matched to a network with an ID
	// like `db9086999caf`
	networks, err := s.apiClient().NetworkList(ctx, network.ListOptions{
		Filters: filters.NewArgs(filters.Arg("name", n.Name)),
	})

	if err != nil {
		return err
	}

	if len(networks) == 0 {
		// in this instance, n.Name is really an ID
		sn, err := s.apiClient().NetworkInspect(ctx, n.Name, network.InspectOptions{})
		if err != nil && !errdefs.IsNotFound(err) {
			return err
		}
		networks = append(networks, sn)
	}

	// NetworkList API doesn't return the exact name match, so we can retrieve more than one network with a request
	networks = utils.Filter(networks, func(net network.Inspect) bool {
		// later in this function, the name is changed the to ID.
		// this function is called during the rebuild stage of `compose watch`.
		// we still require just one network back, but we need to run the search on the ID
		return net.Name == n.Name || net.ID == n.Name
	})

	switch len(networks) {
	case 1:
		n.Name = networks[0].ID
		return nil
	case 0:
		enabled, err := s.isSWarmEnabled(ctx)
		if err != nil {
			return err
		}
		if enabled {
			// Swarm nodes do not register overlay networks that were
			// created on a different node unless they're in use.
			// So we can't preemptively check network exists, but
			// networkAttach will later fail anyway if network actually doesn't exist
			return nil
		}
		return fmt.Errorf("network %s declared as external, but could not be found", n.Name)
	default:
		return fmt.Errorf("multiple networks with name %q were found. Use network ID as `name` to avoid ambiguity", n.Name)
	}
}

func (s *composeService) ensureVolume(ctx context.Context, volume types.VolumeConfig, project string) error {
	inspected, err := s.apiClient().VolumeInspect(ctx, volume.Name)
	if err != nil {
		if !errdefs.IsNotFound(err) {
			return err
		}
		if volume.External {
			return fmt.Errorf("external volume %q not found", volume.Name)
		}
		err := s.createVolume(ctx, volume)
		return err
	}

	if volume.External {
		return nil
	}

	// Volume exists with name, but let's double-check this is the expected one
	p, ok := inspected.Labels[api.ProjectLabel]
	if !ok {
		logrus.Warnf("volume %q already exists but was not created by Docker Compose. Use `external: true` to use an existing volume", volume.Name)
	}
	if ok && p != project {
		logrus.Warnf("volume %q already exists but was created for project %q (expected %q). Use `external: true` to use an existing volume", volume.Name, p, project)
	}
	return nil
}

func (s *composeService) createVolume(ctx context.Context, volume types.VolumeConfig) error {
	eventName := fmt.Sprintf("Volume %q", volume.Name)
	w := progress.ContextWriter(ctx)
	w.Event(progress.CreatingEvent(eventName))
	_, err := s.apiClient().VolumeCreate(ctx, volumetypes.CreateOptions{
		Labels:     volume.Labels,
		Name:       volume.Name,
		Driver:     volume.Driver,
		DriverOpts: volume.DriverOpts,
	})
	if err != nil {
		w.Event(progress.ErrorEvent(eventName))
		return err
	}
	w.Event(progress.CreatedEvent(eventName))
	return nil
}