diff options
Diffstat (limited to 'pkg/specgen')
-rw-r--r-- | pkg/specgen/config_linux_cgo.go | 62 | ||||
-rw-r--r-- | pkg/specgen/config_linux_nocgo.go | 11 | ||||
-rw-r--r-- | pkg/specgen/config_unsupported.go | 13 | ||||
-rw-r--r-- | pkg/specgen/container_create.go | 189 | ||||
-rw-r--r-- | pkg/specgen/container_validate.go | 167 | ||||
-rw-r--r-- | pkg/specgen/namespaces.go | 474 | ||||
-rw-r--r-- | pkg/specgen/oci.go | 260 | ||||
-rw-r--r-- | pkg/specgen/pod_create.go | 83 | ||||
-rw-r--r-- | pkg/specgen/pod_validate.go | 104 | ||||
-rw-r--r-- | pkg/specgen/podspecgen.go | 153 | ||||
-rw-r--r-- | pkg/specgen/specgen.go | 416 |
11 files changed, 1932 insertions, 0 deletions
diff --git a/pkg/specgen/config_linux_cgo.go b/pkg/specgen/config_linux_cgo.go new file mode 100644 index 000000000..6f547a40d --- /dev/null +++ b/pkg/specgen/config_linux_cgo.go @@ -0,0 +1,62 @@ +// +build linux,cgo + +package specgen + +import ( + "context" + "io/ioutil" + + "github.com/containers/libpod/libpod/image" + "github.com/containers/libpod/pkg/seccomp" + spec "github.com/opencontainers/runtime-spec/specs-go" + "github.com/pkg/errors" + goSeccomp "github.com/seccomp/containers-golang" + "github.com/sirupsen/logrus" +) + +func (s *SpecGenerator) getSeccompConfig(configSpec *spec.Spec, img *image.Image) (*spec.LinuxSeccomp, error) { + var seccompConfig *spec.LinuxSeccomp + var err error + + scp, err := seccomp.LookupPolicy(s.SeccompPolicy) + if err != nil { + return nil, err + } + + if scp == seccomp.PolicyImage { + labels, err := img.Labels(context.Background()) + if err != nil { + return nil, err + } + imagePolicy := labels[seccomp.ContainerImageLabel] + if len(imagePolicy) < 1 { + return nil, errors.New("no seccomp policy defined by image") + } + logrus.Debug("Loading seccomp profile from the security config") + seccompConfig, err = goSeccomp.LoadProfile(imagePolicy, configSpec) + if err != nil { + return nil, errors.Wrap(err, "loading seccomp profile failed") + } + return seccompConfig, nil + } + + if s.SeccompProfilePath != "" { + logrus.Debugf("Loading seccomp profile from %q", s.SeccompProfilePath) + seccompProfile, err := ioutil.ReadFile(s.SeccompProfilePath) + if err != nil { + return nil, errors.Wrapf(err, "opening seccomp profile (%s) failed", s.SeccompProfilePath) + } + seccompConfig, err = goSeccomp.LoadProfile(string(seccompProfile), configSpec) + if err != nil { + return nil, errors.Wrapf(err, "loading seccomp profile (%s) failed", s.SeccompProfilePath) + } + } else { + logrus.Debug("Loading default seccomp profile") + seccompConfig, err = goSeccomp.GetDefaultProfile(configSpec) + if err != nil { + return nil, errors.Wrapf(err, "loading seccomp profile (%s) failed", s.SeccompProfilePath) + } + } + + return seccompConfig, nil +} diff --git a/pkg/specgen/config_linux_nocgo.go b/pkg/specgen/config_linux_nocgo.go new file mode 100644 index 000000000..fc0c58c37 --- /dev/null +++ b/pkg/specgen/config_linux_nocgo.go @@ -0,0 +1,11 @@ +// +build linux,!cgo + +package specgen + +import ( + spec "github.com/opencontainers/runtime-spec/specs-go" +) + +func (s *SpecGenerator) getSeccompConfig(configSpec *spec.Spec) (*spec.LinuxSeccomp, error) { + return nil, nil +} diff --git a/pkg/specgen/config_unsupported.go b/pkg/specgen/config_unsupported.go new file mode 100644 index 000000000..c2d3257c9 --- /dev/null +++ b/pkg/specgen/config_unsupported.go @@ -0,0 +1,13 @@ +// +build !linux + +package specgen + +import ( + "github.com/containers/libpod/libpod/image" + spec "github.com/opencontainers/runtime-spec/specs-go" + "github.com/pkg/errors" +) + +func (s *SpecGenerator) getSeccompConfig(configSpec *spec.Spec, img *image.Image) (*spec.LinuxSeccomp, error) { + return nil, errors.New("function not supported on non-linux OS's") +} diff --git a/pkg/specgen/container_create.go b/pkg/specgen/container_create.go new file mode 100644 index 000000000..b4039bb91 --- /dev/null +++ b/pkg/specgen/container_create.go @@ -0,0 +1,189 @@ +package specgen + +import ( + "context" + "os" + + "github.com/containers/common/pkg/config" + "github.com/containers/libpod/libpod" + "github.com/containers/libpod/libpod/define" + "github.com/containers/storage" + "github.com/pkg/errors" + "github.com/sirupsen/logrus" +) + +// MakeContainer creates a container based on the SpecGenerator +func (s *SpecGenerator) MakeContainer(rt *libpod.Runtime) (*libpod.Container, error) { + if err := s.validate(); err != nil { + return nil, errors.Wrap(err, "invalid config provided") + } + rtc, err := rt.GetConfig() + if err != nil { + return nil, err + } + + options, err := s.createContainerOptions(rt) + if err != nil { + return nil, err + } + + podmanPath, err := os.Executable() + if err != nil { + return nil, err + } + options = append(options, s.createExitCommandOption(rt.StorageConfig(), rtc, podmanPath)) + newImage, err := rt.ImageRuntime().NewFromLocal(s.Image) + if err != nil { + return nil, err + } + + options = append(options, libpod.WithRootFSFromImage(newImage.ID(), s.Image, s.RawImageName)) + + runtimeSpec, err := s.toOCISpec(rt, newImage) + if err != nil { + return nil, err + } + return rt.NewContainer(context.Background(), runtimeSpec, options...) +} + +func (s *SpecGenerator) createContainerOptions(rt *libpod.Runtime) ([]libpod.CtrCreateOption, error) { + var options []libpod.CtrCreateOption + var err error + + if s.Stdin { + options = append(options, libpod.WithStdin()) + } + if len(s.Systemd) > 0 { + options = append(options, libpod.WithSystemd()) + } + if len(s.Name) > 0 { + logrus.Debugf("setting container name %s", s.Name) + options = append(options, libpod.WithName(s.Name)) + } + if s.Pod != "" { + pod, err := rt.LookupPod(s.Pod) + if err != nil { + return nil, err + } + logrus.Debugf("adding container to pod %s", s.Pod) + options = append(options, rt.WithPod(pod)) + } + destinations := []string{} + // // Take all mount and named volume destinations. + for _, mount := range s.Mounts { + destinations = append(destinations, mount.Destination) + } + for _, volume := range s.Volumes { + destinations = append(destinations, volume.Dest) + } + options = append(options, libpod.WithUserVolumes(destinations)) + + if len(s.Volumes) != 0 { + options = append(options, libpod.WithNamedVolumes(s.Volumes)) + } + + if len(s.Command) != 0 { + options = append(options, libpod.WithCommand(s.Command)) + } + + options = append(options, libpod.WithEntrypoint(s.Entrypoint)) + if s.StopSignal != nil { + options = append(options, libpod.WithStopSignal(*s.StopSignal)) + } + if s.StopTimeout != nil { + options = append(options, libpod.WithStopTimeout(*s.StopTimeout)) + } + if s.LogConfiguration != nil { + if len(s.LogConfiguration.Path) > 0 { + options = append(options, libpod.WithLogPath(s.LogConfiguration.Path)) + } + if len(s.LogConfiguration.Options) > 0 && s.LogConfiguration.Options["tag"] != "" { + // Note: I'm really guessing here. + options = append(options, libpod.WithLogTag(s.LogConfiguration.Options["tag"])) + } + + if len(s.LogConfiguration.Driver) > 0 { + options = append(options, libpod.WithLogDriver(s.LogConfiguration.Driver)) + } + } + + // Security options + if len(s.SelinuxOpts) > 0 { + options = append(options, libpod.WithSecLabels(s.SelinuxOpts)) + } + options = append(options, libpod.WithPrivileged(s.Privileged)) + + // Get namespace related options + namespaceOptions, err := s.generateNamespaceContainerOpts(rt) + if err != nil { + return nil, err + } + options = append(options, namespaceOptions...) + + if len(s.ConmonPidFile) > 0 { + options = append(options, libpod.WithConmonPidFile(s.ConmonPidFile)) + } + options = append(options, libpod.WithLabels(s.Labels)) + if s.ShmSize != nil { + options = append(options, libpod.WithShmSize(*s.ShmSize)) + } + if s.Rootfs != "" { + options = append(options, libpod.WithRootFS(s.Rootfs)) + } + // Default used if not overridden on command line + + if s.RestartPolicy != "" { + if s.RestartPolicy == "unless-stopped" { + return nil, errors.Wrapf(define.ErrInvalidArg, "the unless-stopped restart policy is not supported") + } + if s.RestartRetries != nil { + options = append(options, libpod.WithRestartRetries(*s.RestartRetries)) + } + options = append(options, libpod.WithRestartPolicy(s.RestartPolicy)) + } + + if s.ContainerHealthCheckConfig.HealthConfig != nil { + options = append(options, libpod.WithHealthCheck(s.ContainerHealthCheckConfig.HealthConfig)) + logrus.Debugf("New container has a health check") + } + return options, nil +} + +func (s *SpecGenerator) createExitCommandOption(storageConfig storage.StoreOptions, config *config.Config, podmanPath string) libpod.CtrCreateOption { + // We need a cleanup process for containers in the current model. + // But we can't assume that the caller is Podman - it could be another + // user of the API. + // As such, provide a way to specify a path to Podman, so we can + // still invoke a cleanup process. + + command := []string{podmanPath, + "--root", storageConfig.GraphRoot, + "--runroot", storageConfig.RunRoot, + "--log-level", logrus.GetLevel().String(), + "--cgroup-manager", config.Engine.CgroupManager, + "--tmpdir", config.Engine.TmpDir, + } + if config.Engine.OCIRuntime != "" { + command = append(command, []string{"--runtime", config.Engine.OCIRuntime}...) + } + if storageConfig.GraphDriverName != "" { + command = append(command, []string{"--storage-driver", storageConfig.GraphDriverName}...) + } + for _, opt := range storageConfig.GraphDriverOptions { + command = append(command, []string{"--storage-opt", opt}...) + } + if config.Engine.EventsLogger != "" { + command = append(command, []string{"--events-backend", config.Engine.EventsLogger}...) + } + + // TODO Mheon wants to leave this for now + //if s.sys { + // command = append(command, "--syslog", "true") + //} + command = append(command, []string{"container", "cleanup"}...) + + if s.Remove { + command = append(command, "--rm") + } + return libpod.WithExitCommand(command) +} diff --git a/pkg/specgen/container_validate.go b/pkg/specgen/container_validate.go new file mode 100644 index 000000000..b27659f5f --- /dev/null +++ b/pkg/specgen/container_validate.go @@ -0,0 +1,167 @@ +package specgen + +import ( + "strings" + + "github.com/containers/libpod/pkg/rootless" + "github.com/containers/libpod/pkg/util" + "github.com/pkg/errors" +) + +var ( + // ErrInvalidSpecConfig describes an error that the given SpecGenerator is invalid + ErrInvalidSpecConfig error = errors.New("invalid configuration") + // SystemDValues describes the only values that SystemD can be + SystemDValues = []string{"true", "false", "always"} + // ImageVolumeModeValues describes the only values that ImageVolumeMode can be + ImageVolumeModeValues = []string{"ignore", "tmpfs", "anonymous"} +) + +func exclusiveOptions(opt1, opt2 string) error { + return errors.Errorf("%s and %s are mutually exclusive options", opt1, opt2) +} + +// Validate verifies that the given SpecGenerator is valid and satisfies required +// input for creating a container. +func (s *SpecGenerator) validate() error { + + // + // ContainerBasicConfig + // + // Rootfs and Image cannot both populated + if len(s.ContainerStorageConfig.Image) > 0 && len(s.ContainerStorageConfig.Rootfs) > 0 { + return errors.Wrap(ErrInvalidSpecConfig, "both image and rootfs cannot be simultaneously") + } + // Cannot set hostname and utsns + if len(s.ContainerBasicConfig.Hostname) > 0 && !s.ContainerBasicConfig.UtsNS.IsPrivate() { + return errors.Wrap(ErrInvalidSpecConfig, "cannot set hostname when creating an UTS namespace") + } + // systemd values must be true, false, or always + if len(s.ContainerBasicConfig.Systemd) > 0 && !util.StringInSlice(strings.ToLower(s.ContainerBasicConfig.Systemd), SystemDValues) { + return errors.Wrapf(ErrInvalidSpecConfig, "SystemD values must be one of %s", strings.Join(SystemDValues, ",")) + } + + // + // ContainerStorageConfig + // + // rootfs and image cannot both be set + if len(s.ContainerStorageConfig.Image) > 0 && len(s.ContainerStorageConfig.Rootfs) > 0 { + return exclusiveOptions("rootfs", "image") + } + // imagevolumemode must be one of ignore, tmpfs, or anonymous if given + if len(s.ContainerStorageConfig.ImageVolumeMode) > 0 && !util.StringInSlice(strings.ToLower(s.ContainerStorageConfig.ImageVolumeMode), ImageVolumeModeValues) { + return errors.Errorf("ImageVolumeMode values must be one of %s", strings.Join(ImageVolumeModeValues, ",")) + } + // shmsize conflicts with IPC namespace + if s.ContainerStorageConfig.ShmSize != nil && !s.ContainerStorageConfig.IpcNS.IsPrivate() { + return errors.New("cannot set shmsize when creating an IPC namespace") + } + + // + // ContainerSecurityConfig + // + // groups and privileged are exclusive + if len(s.Groups) > 0 && s.Privileged { + return exclusiveOptions("Groups", "privileged") + } + // capadd and privileged are exclusive + if len(s.CapAdd) > 0 && s.Privileged { + return exclusiveOptions("CapAdd", "privileged") + } + // selinuxprocesslabel and privileged are exclusive + if len(s.SelinuxProcessLabel) > 0 && s.Privileged { + return exclusiveOptions("SelinuxProcessLabel", "privileged") + } + // selinuxmounmtlabel and privileged are exclusive + if len(s.SelinuxMountLabel) > 0 && s.Privileged { + return exclusiveOptions("SelinuxMountLabel", "privileged") + } + // selinuxopts and privileged are exclusive + if len(s.SelinuxOpts) > 0 && s.Privileged { + return exclusiveOptions("SelinuxOpts", "privileged") + } + // apparmor and privileged are exclusive + if len(s.ApparmorProfile) > 0 && s.Privileged { + return exclusiveOptions("AppArmorProfile", "privileged") + } + // userns and idmappings conflict + if s.UserNS.IsPrivate() && s.IDMappings == nil { + return errors.Wrap(ErrInvalidSpecConfig, "IDMappings are required when not creating a User namespace") + } + + // + // ContainerCgroupConfig + // + // + // None for now + + // + // ContainerNetworkConfig + // + if !s.NetNS.IsPrivate() && s.ConfigureNetNS { + return errors.New("can only configure network namespace when creating a network a network namespace") + } + // useimageresolveconf conflicts with dnsserver, dnssearch, dnsoption + if s.UseImageResolvConf { + if len(s.DNSServer) > 0 { + return exclusiveOptions("UseImageResolvConf", "DNSServer") + } + if len(s.DNSSearch) > 0 { + return exclusiveOptions("UseImageResolvConf", "DNSSearch") + } + if len(s.DNSOption) > 0 { + return exclusiveOptions("UseImageResolvConf", "DNSOption") + } + } + // UseImageHosts and HostAdd are exclusive + if s.UseImageHosts && len(s.HostAdd) > 0 { + return exclusiveOptions("UseImageHosts", "HostAdd") + } + + // TODO the specgen does not appear to handle this? Should it + //switch config.Cgroup.Cgroups { + //case "disabled": + // if addedResources { + // return errors.New("cannot specify resource limits when cgroups are disabled is specified") + // } + // configSpec.Linux.Resources = &spec.LinuxResources{} + //case "enabled", "no-conmon", "": + // // Do nothing + //default: + // return errors.New("unrecognized option for cgroups; supported are 'default', 'disabled', 'no-conmon'") + //} + + // Namespaces + if err := s.UtsNS.validate(); err != nil { + return err + } + if err := s.IpcNS.validate(); err != nil { + return err + } + if err := s.PidNS.validate(); err != nil { + return err + } + if err := s.CgroupNS.validate(); err != nil { + return err + } + if err := s.UserNS.validate(); err != nil { + return err + } + + // The following are defaults as needed by container creation + if len(s.WorkDir) < 1 { + s.WorkDir = "/" + } + + // Set defaults if network info is not provided + if s.NetNS.NSMode == "" { + s.NetNS.NSMode = Bridge + if rootless.IsRootless() { + s.NetNS.NSMode = Slirp + } + } + if err := validateNetNS(&s.NetNS); err != nil { + return err + } + return nil +} diff --git a/pkg/specgen/namespaces.go b/pkg/specgen/namespaces.go new file mode 100644 index 000000000..fa2dee77d --- /dev/null +++ b/pkg/specgen/namespaces.go @@ -0,0 +1,474 @@ +package specgen + +import ( + "os" + + "github.com/containers/common/pkg/capabilities" + "github.com/containers/libpod/libpod" + "github.com/containers/libpod/libpod/image" + "github.com/cri-o/ocicni/pkg/ocicni" + spec "github.com/opencontainers/runtime-spec/specs-go" + "github.com/opencontainers/runtime-tools/generate" + "github.com/pkg/errors" + "github.com/sirupsen/logrus" +) + +type NamespaceMode string + +const ( + // Host means the the namespace is derived from + // the host + Host NamespaceMode = "host" + // Path is the path to a namespace + Path NamespaceMode = "path" + // FromContainer means namespace is derived from a + // different container + FromContainer NamespaceMode = "container" + // FromPod indicates the namespace is derived from a pod + FromPod NamespaceMode = "pod" + // Private indicates the namespace is private + Private NamespaceMode = "private" + // NoNetwork indicates no network namespace should + // be joined. loopback should still exists + NoNetwork NamespaceMode = "none" + // Bridge indicates that a CNI network stack + // should be used + Bridge NamespaceMode = "bridge" + // Slirp indicates that a slirp4ns network stack should + // be used + Slirp NamespaceMode = "slirp4ns" +) + +// Namespace describes the namespace +type Namespace struct { + NSMode NamespaceMode `json:"nsmode,omitempty"` + Value string `json:"string,omitempty"` +} + +// IsHost returns a bool if the namespace is host based +func (n *Namespace) IsHost() bool { + return n.NSMode == Host +} + +// IsPath indicates via bool if the namespace is based on a path +func (n *Namespace) IsPath() bool { + return n.NSMode == Path +} + +// IsContainer indicates via bool if the namespace is based on a container +func (n *Namespace) IsContainer() bool { + return n.NSMode == FromContainer +} + +// IsPod indicates via bool if the namespace is based on a pod +func (n *Namespace) IsPod() bool { + return n.NSMode == FromPod +} + +// IsPrivate indicates the namespace is private +func (n *Namespace) IsPrivate() bool { + return n.NSMode == Private +} + +func validateNetNS(n *Namespace) error { + if n == nil { + return nil + } + switch n.NSMode { + case Host, Path, FromContainer, FromPod, Private, NoNetwork, Bridge, Slirp: + break + default: + return errors.Errorf("invalid network %q", n.NSMode) + } + return nil +} + +// validate perform simple validation on the namespace to make sure it is not +// invalid from the get-go +func (n *Namespace) validate() error { + if n == nil { + return nil + } + // Path and From Container MUST have a string value set + if n.NSMode == Path || n.NSMode == FromContainer { + if len(n.Value) < 1 { + return errors.Errorf("namespace mode %s requires a value", n.NSMode) + } + } else { + // All others must NOT set a string value + if len(n.Value) > 0 { + return errors.Errorf("namespace value %s cannot be provided with namespace mode %s", n.Value, n.NSMode) + } + } + return nil +} + +func (s *SpecGenerator) generateNamespaceContainerOpts(rt *libpod.Runtime) ([]libpod.CtrCreateOption, error) { + var portBindings []ocicni.PortMapping + options := make([]libpod.CtrCreateOption, 0) + + // Cgroups + switch { + case s.CgroupNS.IsPrivate(): + ns := s.CgroupNS.Value + if _, err := os.Stat(ns); err != nil { + return nil, err + } + case s.CgroupNS.IsContainer(): + connectedCtr, err := rt.LookupContainer(s.CgroupNS.Value) + if err != nil { + return nil, errors.Wrapf(err, "container %q not found", s.CgroupNS.Value) + } + options = append(options, libpod.WithCgroupNSFrom(connectedCtr)) + // TODO + //default: + // return nil, errors.New("cgroup name only supports private and container") + } + + if s.CgroupParent != "" { + options = append(options, libpod.WithCgroupParent(s.CgroupParent)) + } + + if s.CgroupsMode != "" { + options = append(options, libpod.WithCgroupsMode(s.CgroupsMode)) + } + + // ipc + switch { + case s.IpcNS.IsHost(): + options = append(options, libpod.WithShmDir("/dev/shm")) + case s.IpcNS.IsContainer(): + connectedCtr, err := rt.LookupContainer(s.IpcNS.Value) + if err != nil { + return nil, errors.Wrapf(err, "container %q not found", s.IpcNS.Value) + } + options = append(options, libpod.WithIPCNSFrom(connectedCtr)) + options = append(options, libpod.WithShmDir(connectedCtr.ShmDir())) + } + + // pid + if s.PidNS.IsContainer() { + connectedCtr, err := rt.LookupContainer(s.PidNS.Value) + if err != nil { + return nil, errors.Wrapf(err, "container %q not found", s.PidNS.Value) + } + options = append(options, libpod.WithPIDNSFrom(connectedCtr)) + } + + // uts + switch { + case s.UtsNS.IsPod(): + connectedPod, err := rt.LookupPod(s.UtsNS.Value) + if err != nil { + return nil, errors.Wrapf(err, "pod %q not found", s.UtsNS.Value) + } + options = append(options, libpod.WithUTSNSFromPod(connectedPod)) + case s.UtsNS.IsContainer(): + connectedCtr, err := rt.LookupContainer(s.UtsNS.Value) + if err != nil { + return nil, errors.Wrapf(err, "container %q not found", s.UtsNS.Value) + } + + options = append(options, libpod.WithUTSNSFrom(connectedCtr)) + } + + if s.UseImageHosts { + options = append(options, libpod.WithUseImageHosts()) + } else if len(s.HostAdd) > 0 { + options = append(options, libpod.WithHosts(s.HostAdd)) + } + + // User + + switch { + case s.UserNS.IsPath(): + ns := s.UserNS.Value + if ns == "" { + return nil, errors.Errorf("invalid empty user-defined user namespace") + } + _, err := os.Stat(ns) + if err != nil { + return nil, err + } + if s.IDMappings != nil { + options = append(options, libpod.WithIDMappings(*s.IDMappings)) + } + case s.UserNS.IsContainer(): + connectedCtr, err := rt.LookupContainer(s.UserNS.Value) + if err != nil { + return nil, errors.Wrapf(err, "container %q not found", s.UserNS.Value) + } + options = append(options, libpod.WithUserNSFrom(connectedCtr)) + default: + if s.IDMappings != nil { + options = append(options, libpod.WithIDMappings(*s.IDMappings)) + } + } + + options = append(options, libpod.WithUser(s.User)) + options = append(options, libpod.WithGroups(s.Groups)) + + if len(s.PortMappings) > 0 { + portBindings = s.PortMappings + } + + switch { + case s.NetNS.IsPath(): + ns := s.NetNS.Value + if ns == "" { + return nil, errors.Errorf("invalid empty user-defined network namespace") + } + _, err := os.Stat(ns) + if err != nil { + return nil, err + } + case s.NetNS.IsContainer(): + connectedCtr, err := rt.LookupContainer(s.NetNS.Value) + if err != nil { + return nil, errors.Wrapf(err, "container %q not found", s.NetNS.Value) + } + options = append(options, libpod.WithNetNSFrom(connectedCtr)) + case !s.NetNS.IsHost() && s.NetNS.NSMode != NoNetwork: + postConfigureNetNS := !s.UserNS.IsHost() + options = append(options, libpod.WithNetNS(portBindings, postConfigureNetNS, string(s.NetNS.NSMode), s.CNINetworks)) + } + + if len(s.DNSSearch) > 0 { + options = append(options, libpod.WithDNSSearch(s.DNSSearch)) + } + if len(s.DNSServer) > 0 { + // TODO I'm not sure how we are going to handle this given the input + if len(s.DNSServer) == 1 { //&& strings.ToLower(s.DNSServer[0].) == "none" { + options = append(options, libpod.WithUseImageResolvConf()) + } else { + var dnsServers []string + for _, d := range s.DNSServer { + dnsServers = append(dnsServers, d.String()) + } + options = append(options, libpod.WithDNS(dnsServers)) + } + } + if len(s.DNSOption) > 0 { + options = append(options, libpod.WithDNSOption(s.DNSOption)) + } + if s.StaticIP != nil { + options = append(options, libpod.WithStaticIP(*s.StaticIP)) + } + + if s.StaticMAC != nil { + options = append(options, libpod.WithStaticMAC(*s.StaticMAC)) + } + return options, nil +} + +func (s *SpecGenerator) pidConfigureGenerator(g *generate.Generator) error { + if s.PidNS.IsPath() { + return g.AddOrReplaceLinuxNamespace(string(spec.PIDNamespace), s.PidNS.Value) + } + if s.PidNS.IsHost() { + return g.RemoveLinuxNamespace(string(spec.PIDNamespace)) + } + if s.PidNS.IsContainer() { + logrus.Debugf("using container %s pidmode", s.PidNS.Value) + } + if s.PidNS.IsPod() { + logrus.Debug("using pod pidmode") + } + return nil +} + +func (s *SpecGenerator) utsConfigureGenerator(g *generate.Generator, runtime *libpod.Runtime) error { + hostname := s.Hostname + var err error + if hostname == "" { + switch { + case s.UtsNS.IsContainer(): + utsCtr, err := runtime.LookupContainer(s.UtsNS.Value) + if err != nil { + return errors.Wrapf(err, "unable to retrieve hostname from dependency container %s", s.UtsNS.Value) + } + hostname = utsCtr.Hostname() + case s.NetNS.IsHost() || s.UtsNS.IsHost(): + hostname, err = os.Hostname() + if err != nil { + return errors.Wrap(err, "unable to retrieve hostname of the host") + } + default: + logrus.Debug("No hostname set; container's hostname will default to runtime default") + } + } + g.RemoveHostname() + if s.Hostname != "" || !s.UtsNS.IsHost() { + // Set the hostname in the OCI configuration only + // if specified by the user or if we are creating + // a new UTS namespace. + g.SetHostname(hostname) + } + g.AddProcessEnv("HOSTNAME", hostname) + + if s.UtsNS.IsPath() { + return g.AddOrReplaceLinuxNamespace(string(spec.UTSNamespace), s.UtsNS.Value) + } + if s.UtsNS.IsHost() { + return g.RemoveLinuxNamespace(string(spec.UTSNamespace)) + } + if s.UtsNS.IsContainer() { + logrus.Debugf("using container %s utsmode", s.UtsNS.Value) + } + return nil +} + +func (s *SpecGenerator) ipcConfigureGenerator(g *generate.Generator) error { + if s.IpcNS.IsPath() { + return g.AddOrReplaceLinuxNamespace(string(spec.IPCNamespace), s.IpcNS.Value) + } + if s.IpcNS.IsHost() { + return g.RemoveLinuxNamespace(s.IpcNS.Value) + } + if s.IpcNS.IsContainer() { + logrus.Debugf("Using container %s ipcmode", s.IpcNS.Value) + } + return nil +} + +func (s *SpecGenerator) cgroupConfigureGenerator(g *generate.Generator) error { + if s.CgroupNS.IsPath() { + return g.AddOrReplaceLinuxNamespace(string(spec.CgroupNamespace), s.CgroupNS.Value) + } + if s.CgroupNS.IsHost() { + return g.RemoveLinuxNamespace(s.CgroupNS.Value) + } + if s.CgroupNS.IsPrivate() { + return g.AddOrReplaceLinuxNamespace(string(spec.CgroupNamespace), "") + } + if s.CgroupNS.IsContainer() { + logrus.Debugf("Using container %s cgroup mode", s.CgroupNS.Value) + } + return nil +} + +func (s *SpecGenerator) networkConfigureGenerator(g *generate.Generator) error { + switch { + case s.NetNS.IsHost(): + logrus.Debug("Using host netmode") + if err := g.RemoveLinuxNamespace(string(spec.NetworkNamespace)); err != nil { + return err + } + + case s.NetNS.NSMode == NoNetwork: + logrus.Debug("Using none netmode") + case s.NetNS.NSMode == Bridge: + logrus.Debug("Using bridge netmode") + case s.NetNS.IsContainer(): + logrus.Debugf("using container %s netmode", s.NetNS.Value) + case s.NetNS.IsPath(): + logrus.Debug("Using ns netmode") + if err := g.AddOrReplaceLinuxNamespace(string(spec.NetworkNamespace), s.NetNS.Value); err != nil { + return err + } + case s.NetNS.IsPod(): + logrus.Debug("Using pod netmode, unless pod is not sharing") + case s.NetNS.NSMode == Slirp: + logrus.Debug("Using slirp4netns netmode") + default: + return errors.Errorf("unknown network mode") + } + + if g.Config.Annotations == nil { + g.Config.Annotations = make(map[string]string) + } + + if s.PublishImagePorts { + g.Config.Annotations[libpod.InspectAnnotationPublishAll] = libpod.InspectResponseTrue + } else { + g.Config.Annotations[libpod.InspectAnnotationPublishAll] = libpod.InspectResponseFalse + } + + return nil +} + +func (s *SpecGenerator) userConfigureGenerator(g *generate.Generator) error { + if s.UserNS.IsPath() { + if err := g.AddOrReplaceLinuxNamespace(string(spec.UserNamespace), s.UserNS.Value); err != nil { + return err + } + // runc complains if no mapping is specified, even if we join another ns. So provide a dummy mapping + g.AddLinuxUIDMapping(uint32(0), uint32(0), uint32(1)) + g.AddLinuxGIDMapping(uint32(0), uint32(0), uint32(1)) + } + + if s.IDMappings != nil { + if (len(s.IDMappings.UIDMap) > 0 || len(s.IDMappings.GIDMap) > 0) && !s.UserNS.IsHost() { + if err := g.AddOrReplaceLinuxNamespace(string(spec.UserNamespace), ""); err != nil { + return err + } + } + for _, uidmap := range s.IDMappings.UIDMap { + g.AddLinuxUIDMapping(uint32(uidmap.HostID), uint32(uidmap.ContainerID), uint32(uidmap.Size)) + } + for _, gidmap := range s.IDMappings.GIDMap { + g.AddLinuxGIDMapping(uint32(gidmap.HostID), uint32(gidmap.ContainerID), uint32(gidmap.Size)) + } + } + return nil +} + +func (s *SpecGenerator) securityConfigureGenerator(g *generate.Generator, newImage *image.Image) error { + // HANDLE CAPABILITIES + // NOTE: Must happen before SECCOMP + if s.Privileged { + g.SetupPrivileged(true) + } + + useNotRoot := func(user string) bool { + if user == "" || user == "root" || user == "0" { + return false + } + return true + } + configSpec := g.Config + var err error + var caplist []string + bounding := configSpec.Process.Capabilities.Bounding + if useNotRoot(s.User) { + configSpec.Process.Capabilities.Bounding = caplist + } + caplist, err = capabilities.MergeCapabilities(configSpec.Process.Capabilities.Bounding, s.CapAdd, s.CapDrop) + if err != nil { + return err + } + + configSpec.Process.Capabilities.Bounding = caplist + configSpec.Process.Capabilities.Permitted = caplist + configSpec.Process.Capabilities.Inheritable = caplist + configSpec.Process.Capabilities.Effective = caplist + configSpec.Process.Capabilities.Ambient = caplist + if useNotRoot(s.User) { + caplist, err = capabilities.MergeCapabilities(bounding, s.CapAdd, s.CapDrop) + if err != nil { + return err + } + } + configSpec.Process.Capabilities.Bounding = caplist + + // HANDLE SECCOMP + if s.SeccompProfilePath != "unconfined" { + seccompConfig, err := s.getSeccompConfig(configSpec, newImage) + if err != nil { + return err + } + configSpec.Linux.Seccomp = seccompConfig + } + + // Clear default Seccomp profile from Generator for privileged containers + if s.SeccompProfilePath == "unconfined" || s.Privileged { + configSpec.Linux.Seccomp = nil + } + + g.SetRootReadonly(s.ReadOnlyFilesystem) + for sysctlKey, sysctlVal := range s.Sysctl { + g.AddLinuxSysctl(sysctlKey, sysctlVal) + } + + return nil +} diff --git a/pkg/specgen/oci.go b/pkg/specgen/oci.go new file mode 100644 index 000000000..2523f21b3 --- /dev/null +++ b/pkg/specgen/oci.go @@ -0,0 +1,260 @@ +package specgen + +import ( + "strings" + + "github.com/containers/libpod/libpod" + "github.com/containers/libpod/libpod/image" + "github.com/containers/libpod/pkg/rootless" + createconfig "github.com/containers/libpod/pkg/spec" + spec "github.com/opencontainers/runtime-spec/specs-go" + "github.com/opencontainers/runtime-tools/generate" +) + +func (s *SpecGenerator) toOCISpec(rt *libpod.Runtime, newImage *image.Image) (*spec.Spec, error) { + var ( + inUserNS bool + ) + cgroupPerm := "ro" + g, err := generate.New("linux") + if err != nil { + return nil, err + } + // Remove the default /dev/shm mount to ensure we overwrite it + g.RemoveMount("/dev/shm") + g.HostSpecific = true + addCgroup := true + canMountSys := true + + isRootless := rootless.IsRootless() + if isRootless { + inUserNS = true + } + if !s.UserNS.IsHost() { + if s.UserNS.IsContainer() || s.UserNS.IsPath() { + inUserNS = true + } + if s.UserNS.IsPrivate() { + inUserNS = true + } + } + if inUserNS && s.NetNS.IsHost() { + canMountSys = false + } + + if s.Privileged && canMountSys { + cgroupPerm = "rw" + g.RemoveMount("/sys") + sysMnt := spec.Mount{ + Destination: "/sys", + Type: "sysfs", + Source: "sysfs", + Options: []string{"rprivate", "nosuid", "noexec", "nodev", "rw"}, + } + g.AddMount(sysMnt) + } else if !canMountSys { + addCgroup = false + g.RemoveMount("/sys") + r := "ro" + if s.Privileged { + r = "rw" + } + sysMnt := spec.Mount{ + Destination: "/sys", + Type: "bind", // should we use a constant for this, like createconfig? + Source: "/sys", + Options: []string{"rprivate", "nosuid", "noexec", "nodev", r, "rbind"}, + } + g.AddMount(sysMnt) + if !s.Privileged && isRootless { + g.AddLinuxMaskedPaths("/sys/kernel") + } + } + gid5Available := true + if isRootless { + nGids, err := createconfig.GetAvailableGids() + if err != nil { + return nil, err + } + gid5Available = nGids >= 5 + } + // When using a different user namespace, check that the GID 5 is mapped inside + // the container. + if gid5Available && (s.IDMappings != nil && len(s.IDMappings.GIDMap) > 0) { + mappingFound := false + for _, r := range s.IDMappings.GIDMap { + if r.ContainerID <= 5 && 5 < r.ContainerID+r.Size { + mappingFound = true + break + } + } + if !mappingFound { + gid5Available = false + } + + } + if !gid5Available { + // If we have no GID mappings, the gid=5 default option would fail, so drop it. + g.RemoveMount("/dev/pts") + devPts := spec.Mount{ + Destination: "/dev/pts", + Type: "devpts", + Source: "devpts", + Options: []string{"rprivate", "nosuid", "noexec", "newinstance", "ptmxmode=0666", "mode=0620"}, + } + g.AddMount(devPts) + } + + if inUserNS && s.IpcNS.IsHost() { + g.RemoveMount("/dev/mqueue") + devMqueue := spec.Mount{ + Destination: "/dev/mqueue", + Type: "bind", // constant ? + Source: "/dev/mqueue", + Options: []string{"bind", "nosuid", "noexec", "nodev"}, + } + g.AddMount(devMqueue) + } + if inUserNS && s.PidNS.IsHost() { + g.RemoveMount("/proc") + procMount := spec.Mount{ + Destination: "/proc", + Type: createconfig.TypeBind, + Source: "/proc", + Options: []string{"rbind", "nosuid", "noexec", "nodev"}, + } + g.AddMount(procMount) + } + + if addCgroup { + cgroupMnt := spec.Mount{ + Destination: "/sys/fs/cgroup", + Type: "cgroup", + Source: "cgroup", + Options: []string{"rprivate", "nosuid", "noexec", "nodev", "relatime", cgroupPerm}, + } + g.AddMount(cgroupMnt) + } + g.SetProcessCwd(s.WorkDir) + g.SetProcessArgs(s.Command) + g.SetProcessTerminal(s.Terminal) + + for key, val := range s.Annotations { + g.AddAnnotation(key, val) + } + g.AddProcessEnv("container", "podman") + + g.Config.Linux.Resources = s.ResourceLimits + + // Devices + if s.Privileged { + // If privileged, we need to add all the host devices to the + // spec. We do not add the user provided ones because we are + // already adding them all. + if err := createconfig.AddPrivilegedDevices(&g); err != nil { + return nil, err + } + } else { + for _, device := range s.Devices { + if err := createconfig.DevicesFromPath(&g, device.Path); err != nil { + return nil, err + } + } + } + + // SECURITY OPTS + g.SetProcessNoNewPrivileges(s.NoNewPrivileges) + + if !s.Privileged { + g.SetProcessApparmorProfile(s.ApparmorProfile) + } + + createconfig.BlockAccessToKernelFilesystems(s.Privileged, s.PidNS.IsHost(), &g) + + for name, val := range s.Env { + g.AddProcessEnv(name, val) + } + + // TODO rlimits and ulimits needs further refinement by someone more + // familiar with the code. + //if err := addRlimits(config, &g); err != nil { + // return nil, err + //} + + // NAMESPACES + + if err := s.pidConfigureGenerator(&g); err != nil { + return nil, err + } + + if err := s.userConfigureGenerator(&g); err != nil { + return nil, err + } + + if err := s.networkConfigureGenerator(&g); err != nil { + return nil, err + } + + if err := s.utsConfigureGenerator(&g, rt); err != nil { + return nil, err + } + + if err := s.ipcConfigureGenerator(&g); err != nil { + return nil, err + } + + if err := s.cgroupConfigureGenerator(&g); err != nil { + return nil, err + } + configSpec := g.Config + + if err := s.securityConfigureGenerator(&g, newImage); err != nil { + return nil, err + } + + // BIND MOUNTS + configSpec.Mounts = createconfig.SupercedeUserMounts(s.Mounts, configSpec.Mounts) + // Process mounts to ensure correct options + finalMounts, err := createconfig.InitFSMounts(configSpec.Mounts) + if err != nil { + return nil, err + } + configSpec.Mounts = finalMounts + + // Add annotations + if configSpec.Annotations == nil { + configSpec.Annotations = make(map[string]string) + } + + // TODO cidfile is not in specgen; when wiring up cli, we will need to move this out of here + // leaving as a reminder + //if config.CidFile != "" { + // configSpec.Annotations[libpod.InspectAnnotationCIDFile] = config.CidFile + //} + + if s.Remove { + configSpec.Annotations[libpod.InspectAnnotationAutoremove] = libpod.InspectResponseTrue + } else { + configSpec.Annotations[libpod.InspectAnnotationAutoremove] = libpod.InspectResponseFalse + } + + if len(s.VolumesFrom) > 0 { + configSpec.Annotations[libpod.InspectAnnotationVolumesFrom] = strings.Join(s.VolumesFrom, ",") + } + + if s.Privileged { + configSpec.Annotations[libpod.InspectAnnotationPrivileged] = libpod.InspectResponseTrue + } else { + configSpec.Annotations[libpod.InspectAnnotationPrivileged] = libpod.InspectResponseFalse + } + + // TODO Init might not make it into the specgen and therefore is not available here. We should deal + // with this when we wire up the CLI; leaving as a reminder + //if s.Init { + // configSpec.Annotations[libpod.InspectAnnotationInit] = libpod.InspectResponseTrue + //} else { + // configSpec.Annotations[libpod.InspectAnnotationInit] = libpod.InspectResponseFalse + //} + + return configSpec, nil +} diff --git a/pkg/specgen/pod_create.go b/pkg/specgen/pod_create.go new file mode 100644 index 000000000..06aa24e22 --- /dev/null +++ b/pkg/specgen/pod_create.go @@ -0,0 +1,83 @@ +package specgen + +import ( + "context" + + "github.com/containers/libpod/cmd/podman/shared" + "github.com/containers/libpod/libpod" + "github.com/sirupsen/logrus" +) + +func (p *PodSpecGenerator) MakePod(rt *libpod.Runtime) (*libpod.Pod, error) { + if err := p.validate(); err != nil { + return nil, err + } + options, err := p.createPodOptions() + if err != nil { + return nil, err + } + return rt.NewPod(context.Background(), options...) +} + +func (p *PodSpecGenerator) createPodOptions() ([]libpod.PodCreateOption, error) { + var ( + options []libpod.PodCreateOption + ) + if !p.NoInfra { + options = append(options, libpod.WithInfraContainer()) + nsOptions, err := shared.GetNamespaceOptions(p.SharedNamespaces) + if err != nil { + return nil, err + } + options = append(options, nsOptions...) + } + if len(p.CgroupParent) > 0 { + options = append(options, libpod.WithPodCgroupParent(p.CgroupParent)) + } + if len(p.Labels) > 0 { + options = append(options, libpod.WithPodLabels(p.Labels)) + } + if len(p.Name) > 0 { + options = append(options, libpod.WithPodName(p.Name)) + } + if len(p.Hostname) > 0 { + options = append(options, libpod.WithPodHostname(p.Hostname)) + } + if len(p.HostAdd) > 0 { + options = append(options, libpod.WithPodHosts(p.HostAdd)) + } + if len(p.DNSOption) > 0 { + options = append(options, libpod.WithPodDNSOption(p.DNSOption)) + } + if len(p.DNSSearch) > 0 { + options = append(options, libpod.WithPodDNSSearch(p.DNSSearch)) + } + if p.StaticIP != nil { + options = append(options, libpod.WithPodStaticIP(*p.StaticIP)) + } + if p.StaticMAC != nil { + options = append(options, libpod.WithPodStaticMAC(*p.StaticMAC)) + } + if p.NoManageResolvConf { + options = append(options, libpod.WithPodUseImageResolvConf()) + } + switch p.NetNS.NSMode { + case Bridge: + logrus.Debugf("Pod using default network mode") + case Host: + logrus.Debugf("Pod will use host networking") + options = append(options, libpod.WithPodHostNetwork()) + default: + logrus.Debugf("Pod joining CNI networks: %v", p.CNINetworks) + options = append(options, libpod.WithPodNetworks(p.CNINetworks)) + } + + if p.NoManageHosts { + options = append(options, libpod.WithPodUseImageHosts()) + } + if len(p.PortMappings) > 0 { + options = append(options, libpod.WithInfraContainerPorts(p.PortMappings)) + } + options = append(options, libpod.WithPodCgroups()) + return options, nil +} diff --git a/pkg/specgen/pod_validate.go b/pkg/specgen/pod_validate.go new file mode 100644 index 000000000..50309f096 --- /dev/null +++ b/pkg/specgen/pod_validate.go @@ -0,0 +1,104 @@ +package specgen + +import ( + "github.com/containers/libpod/libpod/define" + "github.com/containers/libpod/pkg/rootless" + "github.com/pkg/errors" +) + +var ( + // ErrInvalidPodSpecConfig describes an error given when the podspecgenerator is invalid + ErrInvalidPodSpecConfig error = errors.New("invalid pod spec") +) + +func exclusivePodOptions(opt1, opt2 string) error { + return errors.Wrapf(ErrInvalidPodSpecConfig, "%s and %s are mutually exclusive pod options", opt1, opt2) +} + +func (p *PodSpecGenerator) validate() error { + // PodBasicConfig + if p.NoInfra { + if len(p.InfraCommand) > 0 { + return exclusivePodOptions("NoInfra", "InfraCommand") + } + if len(p.InfraImage) > 0 { + return exclusivePodOptions("NoInfra", "InfraImage") + } + if len(p.SharedNamespaces) > 0 { + return exclusivePodOptions("NoInfo", "SharedNamespaces") + } + } + + // PodNetworkConfig + if err := p.NetNS.validate(); err != nil { + return err + } + if p.NoInfra { + if p.NetNS.NSMode == NoNetwork { + return errors.New("NoInfra and a none network cannot be used toegther") + } + if p.StaticIP != nil { + return exclusivePodOptions("NoInfra", "StaticIP") + } + if p.StaticMAC != nil { + return exclusivePodOptions("NoInfra", "StaticMAC") + } + if len(p.DNSOption) > 0 { + return exclusivePodOptions("NoInfra", "DNSOption") + } + if len(p.DNSSearch) > 0 { + return exclusivePodOptions("NoInfo", "DNSSearch") + } + if len(p.DNSServer) > 0 { + return exclusivePodOptions("NoInfra", "DNSServer") + } + if len(p.HostAdd) > 0 { + return exclusivePodOptions("NoInfra", "HostAdd") + } + if p.NoManageResolvConf { + return exclusivePodOptions("NoInfra", "NoManageResolvConf") + } + } + if p.NetNS.NSMode != Bridge { + if len(p.PortMappings) > 0 { + return errors.New("PortMappings can only be used with Bridge mode networking") + } + if len(p.CNINetworks) > 0 { + return errors.New("CNINetworks can only be used with Bridge mode networking") + } + } + if p.NoManageResolvConf { + if len(p.DNSServer) > 0 { + return exclusivePodOptions("NoManageResolvConf", "DNSServer") + } + if len(p.DNSSearch) > 0 { + return exclusivePodOptions("NoManageResolvConf", "DNSSearch") + } + if len(p.DNSOption) > 0 { + return exclusivePodOptions("NoManageResolvConf", "DNSOption") + } + } + if p.NoManageHosts && len(p.HostAdd) > 0 { + return exclusivePodOptions("NoManageHosts", "HostAdd") + } + + if err := p.NetNS.validate(); err != nil { + return err + } + + // Set Defaults + if p.NetNS.Value == "" { + if rootless.IsRootless() { + p.NetNS.NSMode = Slirp + } else { + p.NetNS.NSMode = Bridge + } + } + if len(p.InfraImage) < 1 { + p.InfraImage = define.DefaultInfraImage + } + if len(p.InfraCommand) < 1 { + p.InfraCommand = []string{define.DefaultInfraCommand} + } + return nil +} diff --git a/pkg/specgen/podspecgen.go b/pkg/specgen/podspecgen.go new file mode 100644 index 000000000..3f830014d --- /dev/null +++ b/pkg/specgen/podspecgen.go @@ -0,0 +1,153 @@ +package specgen + +import ( + "net" + + "github.com/cri-o/ocicni/pkg/ocicni" +) + +// PodBasicConfig contains basic configuration options for pods. +type PodBasicConfig struct { + // Name is the name of the pod. + // If not provided, a name will be generated when the pod is created. + // Optional. + Name string `json:"name,omitempty"` + // Hostname is the pod's hostname. If not set, the name of the pod will + // be used (if a name was not provided here, the name auto-generated for + // the pod will be used). This will be used by the infra container and + // all containers in the pod as long as the UTS namespace is shared. + // Optional. + Hostname string `json:"hostname,omitempty"` + // Labels are key-value pairs that are used to add metadata to pods. + // Optional. + Labels map[string]string `json:"labels,omitempty"` + // NoInfra tells the pod not to create an infra container. If this is + // done, many networking-related options will become unavailable. + // Conflicts with setting any options in PodNetworkConfig, and the + // InfraCommand and InfraImages in this struct. + // Optional. + NoInfra bool `json:"no_infra,omitempty"` + // InfraCommand sets the command that will be used to start the infra + // container. + // If not set, the default set in the Libpod configuration file will be + // used. + // Conflicts with NoInfra=true. + // Optional. + InfraCommand []string `json:"infra_command,omitempty"` + // InfraImage is the image that will be used for the infra container. + // If not set, the default set in the Libpod configuration file will be + // used. + // Conflicts with NoInfra=true. + // Optional. + InfraImage string `json:"infra_image,omitempty"` + // SharedNamespaces instructs the pod to share a set of namespaces. + // Shared namespaces will be joined (by default) by every container + // which joins the pod. + // If not set and NoInfra is false, the pod will set a default set of + // namespaces to share. + // Conflicts with NoInfra=true. + // Optional. + SharedNamespaces []string `json:"shared_namespaces,omitempty"` +} + +// PodNetworkConfig contains networking configuration for a pod. +type PodNetworkConfig struct { + // NetNS is the configuration to use for the infra container's network + // namespace. This network will, by default, be shared with all + // containers in the pod. + // Cannot be set to FromContainer and FromPod. + // Setting this to anything except "" conflicts with NoInfra=true. + // Defaults to Bridge as root and Slirp as rootless. + // Mandatory. + NetNS Namespace `json:"netns,omitempty"` + // StaticIP sets a static IP for the infra container. As the infra + // container's network is used for the entire pod by default, this will + // thus be a static IP for the whole pod. + // Only available if NetNS is set to Bridge (the default for root). + // As such, conflicts with NoInfra=true by proxy. + // Optional. + StaticIP *net.IP `json:"static_ip,omitempty"` + // StaticMAC sets a static MAC for the infra container. As the infra + // container's network is used for the entire pod by default, this will + // thus be a static MAC for the entire pod. + // Only available if NetNS is set to Bridge (the default for root). + // As such, conflicts with NoInfra=true by proxy. + // Optional. + StaticMAC *net.HardwareAddr `json:"static_mac,omitempty"` + // PortMappings is a set of ports to map into the infra container. + // As, by default, containers share their network with the infra + // container, this will forward the ports to the entire pod. + // Only available if NetNS is set to Bridge or Slirp. + // Optional. + PortMappings []ocicni.PortMapping `json:"portmappings,omitempty"` + // CNINetworks is a list of CNI networks that the infra container will + // join. As, by default, containers share their network with the infra + // container, these networks will effectively be joined by the + // entire pod. + // Only available when NetNS is set to Bridge, the default for root. + // Optional. + CNINetworks []string `json:"cni_networks,omitempty"` + // NoManageResolvConf indicates that /etc/resolv.conf should not be + // managed by the pod. Instead, each container will create and manage a + // separate resolv.conf as if they had not joined a pod. + // Conflicts with NoInfra=true and DNSServer, DNSSearch, DNSOption. + // Optional. + NoManageResolvConf bool `json:"no_manage_resolv_conf,omitempty"` + // DNSServer is a set of DNS servers that will be used in the infra + // container's resolv.conf, which will, by default, be shared with all + // containers in the pod. + // If not provided, the host's DNS servers will be used, unless the only + // server set is a localhost address. As the container cannot connect to + // the host's localhost, a default server will instead be set. + // Conflicts with NoInfra=true. + // Optional. + DNSServer []net.IP `json:"dns_server,omitempty"` + // DNSSearch is a set of DNS search domains that will be used in the + // infra container's resolv.conf, which will, by default, be shared with + // all containers in the pod. + // If not provided, DNS search domains from the host's resolv.conf will + // be used. + // Conflicts with NoInfra=true. + // Optional. + DNSSearch []string `json:"dns_search,omitempty"` + // DNSOption is a set of DNS options that will be used in the infra + // container's resolv.conf, which will, by default, be shared with all + // containers in the pod. + // Conflicts with NoInfra=true. + // Optional. + DNSOption []string `json:"dns_option,omitempty"` + // NoManageHosts indicates that /etc/hosts should not be managed by the + // pod. Instead, each container will create a separate /etc/hosts as + // they would if not in a pod. + // Conflicts with HostAdd. + NoManageHosts bool `json:"no_manage_hosts,omitempty"` + // HostAdd is a set of hosts that will be added to the infra container's + // /etc/hosts that will, by default, be shared with all containers in + // the pod. + // Conflicts with NoInfra=true and NoManageHosts. + // Optional. + HostAdd []string `json:"hostadd,omitempty"` +} + +// PodCgroupConfig contains configuration options about a pod's cgroups. +// This will be expanded in future updates to pods. +type PodCgroupConfig struct { + // CgroupParent is the parent for the CGroup that the pod will create. + // This pod cgroup will, in turn, be the default cgroup parent for all + // containers in the pod. + // Optional. + CgroupParent string `json:"cgroup_parent,omitempty"` +} + +// PodSpecGenerator describes options to create a pod +// swagger:model PodSpecGenerator +type PodSpecGenerator struct { + PodBasicConfig + PodNetworkConfig + PodCgroupConfig +} + +// NewPodSpecGenerator creates a new pod spec +func NewPodSpecGenerator() *PodSpecGenerator { + return &PodSpecGenerator{} +} diff --git a/pkg/specgen/specgen.go b/pkg/specgen/specgen.go new file mode 100644 index 000000000..89c76c273 --- /dev/null +++ b/pkg/specgen/specgen.go @@ -0,0 +1,416 @@ +package specgen + +import ( + "net" + "syscall" + + "github.com/containers/image/v5/manifest" + "github.com/containers/libpod/libpod" + "github.com/containers/libpod/pkg/rootless" + "github.com/containers/storage" + "github.com/cri-o/ocicni/pkg/ocicni" + spec "github.com/opencontainers/runtime-spec/specs-go" +) + +// LogConfig describes the logging characteristics for a container +type LogConfig struct { + // LogDriver is the container's log driver. + // Optional. + Driver string `json:"driver,omitempty"` + // LogPath is the path the container's logs will be stored at. + // Only available if LogDriver is set to "json-file" or "k8s-file". + // Optional. + Path string `json:"path,omitempty"` + // A set of options to accompany the log driver. + // Optional. + Options map[string]string `json:"options,omitempty"` +} + +// ContainerBasicConfig contains the basic parts of a container. +type ContainerBasicConfig struct { + // Name is the name the container will be given. + // If no name is provided, one will be randomly generated. + // Optional. + Name string `json:"name,omitempty"` + // Pod is the ID of the pod the container will join. + // Optional. + Pod string `json:"pod,omitempty"` + // Entrypoint is the container's entrypoint. + // If not given and Image is specified, this will be populated by the + // image's configuration. + // Optional. + Entrypoint []string `json:"entrypoint,omitempty"` + // Command is the container's command. + // If not given and Image is specified, this will be populated by the + // image's configuration. + // Optional. + Command []string `json:"command,omitempty"` + // Env is a set of environment variables that will be set in the + // container. + // Optional. + Env map[string]string `json:"env,omitempty"` + // Terminal is whether the container will create a PTY. + Terminal bool `json:"terminal,omitempty"` + // Stdin is whether the container will keep its STDIN open. + Stdin bool `json:"stdin,omitempty"` + // Labels are key-value pairs that are used to add metadata to + // containers. + // Optional. + Labels map[string]string `json:"labels,omitempty"` + // Annotations are key-value options passed into the container runtime + // that can be used to trigger special behavior. + // Optional. + Annotations map[string]string `json:"annotations,omitempty"` + // StopSignal is the signal that will be used to stop the container. + // Must be a non-zero integer below SIGRTMAX. + // If not provided, the default, SIGTERM, will be used. + // Will conflict with Systemd if Systemd is set to "true" or "always". + // Optional. + StopSignal *syscall.Signal `json:"stop_signal,omitempty"` + // StopTimeout is a timeout between the container's stop signal being + // sent and SIGKILL being sent. + // If not provided, the default will be used. + // If 0 is used, stop signal will not be sent, and SIGKILL will be sent + // instead. + // Optional. + StopTimeout *uint `json:"stop_timeout,omitempty"` + // LogConfiguration describes the logging for a container including + // driver, path, and options. + // Optional + LogConfiguration *LogConfig `json:"log_configuration,omitempty"` + // ConmonPidFile is a path at which a PID file for Conmon will be + // placed. + // If not given, a default location will be used. + // Optional. + ConmonPidFile string `json:"conmon_pid_file,omitempty"` + // RestartPolicy is the container's restart policy - an action which + // will be taken when the container exits. + // If not given, the default policy, which does nothing, will be used. + // Optional. + RestartPolicy string `json:"restart_policy,omitempty"` + // RestartRetries is the number of attempts that will be made to restart + // the container. + // Only available when RestartPolicy is set to "on-failure". + // Optional. + RestartRetries *uint `json:"restart_tries,omitempty"` + // OCIRuntime is the name of the OCI runtime that will be used to create + // the container. + // If not specified, the default will be used. + // Optional. + OCIRuntime string `json:"oci_runtime,omitempty"` + // Systemd is whether the container will be started in systemd mode. + // Valid options are "true", "false", and "always". + // "true" enables this mode only if the binary run in the container is + // /sbin/init or systemd. "always" unconditionally enables systemd mode. + // "false" unconditionally disables systemd mode. + // If enabled, mounts and stop signal will be modified. + // If set to "always" or set to "true" and conditionally triggered, + // conflicts with StopSignal. + // If not specified, "false" will be assumed. + // Optional. + Systemd string `json:"systemd,omitempty"` + // Namespace is the libpod namespace the container will be placed in. + // Optional. + Namespace string `json:"namespace,omitempty"` + // PidNS is the container's PID namespace. + // It defaults to private. + // Mandatory. + PidNS Namespace `json:"pidns,omitempty"` + // UtsNS is the container's UTS namespace. + // It defaults to private. + // Must be set to Private to set Hostname. + // Mandatory. + UtsNS Namespace `json:"utsns,omitempty"` + // Hostname is the container's hostname. If not set, the hostname will + // not be modified (if UtsNS is not private) or will be set to the + // container ID (if UtsNS is private). + // Conflicts with UtsNS if UtsNS is not set to private. + // Optional. + Hostname string `json:"hostname,omitempty"` + // Sysctl sets kernel parameters for the container + Sysctl map[string]string `json:"sysctl,omitempty"` + // Remove indicates if the container should be removed once it has been started + // and exits + Remove bool `json:"remove"` +} + +// ContainerStorageConfig contains information on the storage configuration of a +// container. +type ContainerStorageConfig struct { + // Image is the image the container will be based on. The image will be + // used as the container's root filesystem, and its environment vars, + // volumes, and other configuration will be applied to the container. + // Conflicts with Rootfs. + // At least one of Image or Rootfs must be specified. + Image string `json:"image"` + // RawImageName is the unprocessed and not-normalized user-specified image + // name. One use case for having this data at hand are auto-updates where + // the _exact_ user input is needed in order to look-up the correct image. + RawImageName string `json:"raw_image_name,omitempty"` + // Rootfs is the path to a directory that will be used as the + // container's root filesystem. No modification will be made to the + // directory, it will be directly mounted into the container as root. + // Conflicts with Image. + // At least one of Image or Rootfs must be specified. + Rootfs string `json:"rootfs,omitempty"` + // ImageVolumeMode indicates how image volumes will be created. + // Supported modes are "ignore" (do not create), "tmpfs" (create as + // tmpfs), and "anonymous" (create as anonymous volumes). + // The default is anonymous. + // Optional. + ImageVolumeMode string `json:"image_volume_mode,omitempty"` + // VolumesFrom is a list of containers whose volumes will be added to + // this container. Supported mount options may be added after the + // container name with a : and include "ro" and "rw". + // Optional. + VolumesFrom []string `json:"volumes_from,omitempty"` + // Mounts are mounts that will be added to the container. + // These will supersede Image Volumes and VolumesFrom volumes where + // there are conflicts. + // Optional. + Mounts []spec.Mount `json:"mounts,omitempty"` + // Volumes are named volumes that will be added to the container. + // These will supersede Image Volumes and VolumesFrom volumes where + // there are conflicts. + // Optional. + Volumes []*libpod.ContainerNamedVolume `json:"volumes,omitempty"` + // Devices are devices that will be added to the container. + // Optional. + Devices []spec.LinuxDevice `json:"devices,omitempty"` + // IpcNS is the container's IPC namespace. + // Default is private. + // Conflicts with ShmSize if not set to private. + // Mandatory. + IpcNS Namespace `json:"ipcns,omitempty"` + // ShmSize is the size of the tmpfs to mount in at /dev/shm, in bytes. + // Conflicts with ShmSize if IpcNS is not private. + // Optional. + ShmSize *int64 `json:"shm_size,omitempty"` + // WorkDir is the container's working directory. + // If unset, the default, /, will be used. + // Optional. + WorkDir string `json:"work_dir,omitempty"` + // RootfsPropagation is the rootfs propagation mode for the container. + // If not set, the default of rslave will be used. + // Optional. + RootfsPropagation string `json:"rootfs_propagation,omitempty"` +} + +// ContainerSecurityConfig is a container's security features, including +// SELinux, Apparmor, and Seccomp. +type ContainerSecurityConfig struct { + // Privileged is whether the container is privileged. + // Privileged does the following: + // - Adds all devices on the system to the container. + // - Adds all capabilities to the container. + // - Disables Seccomp, SELinux, and Apparmor confinement. + // TODO: this conflicts with things. + // TODO: this does more. + Privileged bool `json:"privileged,omitempty"` + // User is the user the container will be run as. + // Can be given as a UID or a username; if a username, it will be + // resolved within the container, using the container's /etc/passwd. + // If unset, the container will be run as root. + // Optional. + User string `json:"user,omitempty"` + // Groups are a list of supplemental groups the container's user will + // be granted access to. + // Optional. + Groups []string `json:"groups,omitempty"` + // CapAdd are capabilities which will be added to the container. + // Conflicts with Privileged. + // Optional. + CapAdd []string `json:"cap_add,omitempty"` + // CapDrop are capabilities which will be removed from the container. + // Conflicts with Privileged. + // Optional. + CapDrop []string `json:"cap_drop,omitempty"` + // SelinuxProcessLabel is the process label the container will use. + // If SELinux is enabled and this is not specified, a label will be + // automatically generated if not specified. + // Optional. + SelinuxProcessLabel string `json:"selinux_process_label,omitempty"` + // SelinuxMountLabel is the mount label the container will use. + // If SELinux is enabled and this is not specified, a label will be + // automatically generated if not specified. + // Optional. + SelinuxMountLabel string `json:"selinux_mount_label,omitempty"` + // SelinuxOpts are options for configuring SELinux. + // Optional. + SelinuxOpts []string `json:"selinux_opts,omitempty"` + // ApparmorProfile is the name of the Apparmor profile the container + // will use. + // Optional. + ApparmorProfile string `json:"apparmor_profile,omitempty"` + // SeccompPolicy determines which seccomp profile gets applied + // the container. valid values: empty,default,image + SeccompPolicy string `json:"seccomp_policy,omitempty"` + // SeccompProfilePath is the path to a JSON file containing the + // container's Seccomp profile. + // If not specified, no Seccomp profile will be used. + // Optional. + SeccompProfilePath string `json:"seccomp_profile_path,omitempty"` + // NoNewPrivileges is whether the container will set the no new + // privileges flag on create, which disables gaining additional + // privileges (e.g. via setuid) in the container. + NoNewPrivileges bool `json:"no_new_privileges,omitempty"` + // UserNS is the container's user namespace. + // It defaults to host, indicating that no user namespace will be + // created. + // If set to private, IDMappings must be set. + // Mandatory. + UserNS Namespace `json:"userns,omitempty"` + // IDMappings are UID and GID mappings that will be used by user + // namespaces. + // Required if UserNS is private. + IDMappings *storage.IDMappingOptions `json:"idmappings,omitempty"` + // ReadOnlyFilesystem indicates that everything will be mounted + // as read-only + ReadOnlyFilesystem bool `json:"read_only_filesystem,omittempty"` +} + +// ContainerCgroupConfig contains configuration information about a container's +// cgroups. +type ContainerCgroupConfig struct { + // CgroupNS is the container's cgroup namespace. + // It defaults to private. + // Mandatory. + CgroupNS Namespace `json:"cgroupns,omitempty"` + // CgroupsMode sets a policy for how cgroups will be created in the + // container, including the ability to disable creation entirely. + CgroupsMode string `json:"cgroups_mode,omitempty"` + // CgroupParent is the container's CGroup parent. + // If not set, the default for the current cgroup driver will be used. + // Optional. + CgroupParent string `json:"cgroup_parent,omitempty"` +} + +// ContainerNetworkConfig contains information on a container's network +// configuration. +type ContainerNetworkConfig struct { + // NetNS is the configuration to use for the container's network + // namespace. + // Mandatory. + NetNS Namespace `json:"netns,omitempty"` + // ConfigureNetNS is whether Libpod will configure the container's + // network namespace to send and receive traffic. + // Only available is NetNS is private - conflicts with other NetNS + // modes. + ConfigureNetNS bool `json:"configure_netns,omitempty"` + // StaticIP is the a IPv4 address of the container. + // Only available if ConfigureNetNS is true. + // Optional. + StaticIP *net.IP `json:"static_ip,omitempty"` + // StaticIPv6 is a static IPv6 address to set in the container. + // Only available if ConfigureNetNS is true. + // Optional. + StaticIPv6 *net.IP `json:"static_ipv6,omitempty"` + // StaticMAC is a static MAC address to set in the container. + // Only available if ConfigureNetNS is true. + // Optional. + StaticMAC *net.HardwareAddr `json:"static_mac,omitempty"` + // PortBindings is a set of ports to map into the container. + // Only available if ConfigureNetNS is true. + // Optional. + PortMappings []ocicni.PortMapping `json:"portmappings,omitempty"` + // PublishImagePorts will publish ports specified in the image to random + // ports outside. + // Requires Image to be set. + PublishImagePorts bool `json:"publish_image_ports,omitempty"` + // CNINetworks is a list of CNI networks to join the container to. + // If this list is empty, the default CNI network will be joined + // instead. If at least one entry is present, we will not join the + // default network (unless it is part of this list). + // Only available if ConfigureNetNS is true. + // Optional. + CNINetworks []string `json:"cni_networks,omitempty"` + // UseImageResolvConf indicates that resolv.conf should not be managed + // by Podman, but instead sourced from the image. + // Conflicts with DNSServer, DNSSearch, DNSOption. + UseImageResolvConf bool `json:"use_image_resolve_conf,omitempty"` + // DNSServer is a set of DNS servers that will be used in the + // container's resolv.conf, replacing the host's DNS Servers which are + // used by default. + // Conflicts with UseImageResolvConf. + // Optional. + DNSServer []net.IP `json:"dns_server,omitempty"` + // DNSSearch is a set of DNS search domains that will be used in the + // container's resolv.conf, replacing the host's DNS search domains + // which are used by default. + // Conflicts with UseImageResolvConf. + // Optional. + DNSSearch []string `json:"dns_search,omitempty"` + // DNSOption is a set of DNS options that will be used in the + // container's resolv.conf, replacing the host's DNS options which are + // used by default. + // Conflicts with UseImageResolvConf. + // Optional. + DNSOption []string `json:"dns_option,omitempty"` + // UseImageHosts indicates that /etc/hosts should not be managed by + // Podman, and instead sourced from the image. + // Conflicts with HostAdd. + UseImageHosts bool `json:"use_image_hosts,omitempty"` + // HostAdd is a set of hosts which will be added to the container's + // /etc/hosts file. + // Conflicts with UseImageHosts. + // Optional. + HostAdd []string `json:"hostadd,omitempty"` +} + +// ContainerResourceConfig contains information on container resource limits. +type ContainerResourceConfig struct { + // ResourceLimits are resource limits to apply to the container., + // Can only be set as root on cgroups v1 systems, but can be set as + // rootless as well for cgroups v2. + // Optional. + ResourceLimits *spec.LinuxResources `json:"resource_limits,omitempty"` + // Rlimits are POSIX rlimits to apply to the container. + // Optional. + Rlimits []spec.POSIXRlimit `json:"r_limits,omitempty"` + // OOMScoreAdj adjusts the score used by the OOM killer to determine + // processes to kill for the container's process. + // Optional. + OOMScoreAdj *int `json:"oom_score_adj,omitempty"` +} + +// ContainerHealthCheckConfig describes a container healthcheck with attributes +// like command, retries, interval, start period, and timeout. +type ContainerHealthCheckConfig struct { + HealthConfig *manifest.Schema2HealthConfig `json:"healthconfig,omitempty"` +} + +// SpecGenerator creates an OCI spec and Libpod configuration options to create +// a container based on the given configuration. +// swagger:model SpecGenerator +type SpecGenerator struct { + ContainerBasicConfig + ContainerStorageConfig + ContainerSecurityConfig + ContainerCgroupConfig + ContainerNetworkConfig + ContainerResourceConfig + ContainerHealthCheckConfig +} + +// NewSpecGenerator returns a SpecGenerator struct given one of two mandatory inputs +func NewSpecGenerator(image string) *SpecGenerator { + networkConfig := ContainerNetworkConfig{ + NetNS: Namespace{ + NSMode: Bridge, + }, + } + csc := ContainerStorageConfig{Image: image} + if rootless.IsRootless() { + networkConfig.NetNS.NSMode = Slirp + } + return &SpecGenerator{ + ContainerStorageConfig: csc, + ContainerNetworkConfig: networkConfig, + } +} + +// NewSpecGenerator returns a SpecGenerator struct given one of two mandatory inputs +func NewSpecGeneratorWithRootfs(rootfs string) *SpecGenerator { + csc := ContainerStorageConfig{Rootfs: rootfs} + return &SpecGenerator{ContainerStorageConfig: csc} +} |