diff options
Diffstat (limited to 'pkg/spec')
-rw-r--r-- | pkg/spec/config_linux.go | 55 | ||||
-rw-r--r-- | pkg/spec/config_linux_cgo.go | 11 | ||||
-rw-r--r-- | pkg/spec/config_unsupported.go | 8 | ||||
-rw-r--r-- | pkg/spec/createconfig.go | 85 | ||||
-rw-r--r-- | pkg/spec/namespaces.go | 2 | ||||
-rw-r--r-- | pkg/spec/parse.go | 32 | ||||
-rw-r--r-- | pkg/spec/security.go | 45 | ||||
-rw-r--r-- | pkg/spec/spec.go | 109 | ||||
-rw-r--r-- | pkg/spec/storage.go | 11 |
9 files changed, 234 insertions, 124 deletions
diff --git a/pkg/spec/config_linux.go b/pkg/spec/config_linux.go index 32d8cb4de..544c0020d 100644 --- a/pkg/spec/config_linux.go +++ b/pkg/spec/config_linux.go @@ -7,6 +7,7 @@ import ( "io/ioutil" "os" "path/filepath" + "strconv" "strings" "github.com/containers/libpod/pkg/rootless" @@ -31,8 +32,8 @@ func Device(d *configs.Device) spec.LinuxDevice { } } -// devicesFromPath computes a list of devices -func devicesFromPath(g *generate.Generator, devicePath string) error { +// DevicesFromPath computes a list of devices +func DevicesFromPath(g *generate.Generator, devicePath string) error { devs := strings.Split(devicePath, ":") resolvedDevicePath := devs[0] // check if it is a symbolic link @@ -90,6 +91,42 @@ func devicesFromPath(g *generate.Generator, devicePath string) error { return addDevice(g, strings.Join(append([]string{resolvedDevicePath}, devs[1:]...), ":")) } +func deviceCgroupRules(g *generate.Generator, deviceCgroupRules []string) error { + for _, deviceCgroupRule := range deviceCgroupRules { + if err := validateDeviceCgroupRule(deviceCgroupRule); err != nil { + return err + } + ss := parseDeviceCgroupRule(deviceCgroupRule) + if len(ss[0]) != 5 { + return errors.Errorf("invalid device cgroup rule format: '%s'", deviceCgroupRule) + } + matches := ss[0] + var major, minor *int64 + if matches[2] == "*" { + majorDev := int64(-1) + major = &majorDev + } else { + majorDev, err := strconv.ParseInt(matches[2], 10, 64) + if err != nil { + return errors.Errorf("invalid major value in device cgroup rule format: '%s'", deviceCgroupRule) + } + major = &majorDev + } + if matches[3] == "*" { + minorDev := int64(-1) + minor = &minorDev + } else { + minorDev, err := strconv.ParseInt(matches[2], 10, 64) + if err != nil { + return errors.Errorf("invalid major value in device cgroup rule format: '%s'", deviceCgroupRule) + } + minor = &minorDev + } + g.AddLinuxResourcesDevice(true, matches[1], major, minor, matches[4]) + } + return nil +} + func addDevice(g *generate.Generator, device string) error { src, dst, permissions, err := ParseDevice(device) if err != nil { @@ -179,7 +216,7 @@ func getDevices(path string) ([]*configs.Device, error) { return out, nil } -func (c *CreateConfig) addPrivilegedDevices(g *generate.Generator) error { +func addPrivilegedDevices(g *generate.Generator) error { hostDevices, err := getDevices("/dev") if err != nil { return err @@ -243,16 +280,16 @@ func (c *CreateConfig) createBlockIO() (*spec.LinuxBlockIO, error) { var lwds []spec.LinuxWeightDevice ret = bio for _, i := range c.Resources.BlkioWeightDevice { - wd, err := validateweightDevice(i) + wd, err := ValidateweightDevice(i) if err != nil { return ret, errors.Wrapf(err, "invalid values for blkio-weight-device") } - wdStat, err := getStatFromPath(wd.path) + wdStat, err := GetStatFromPath(wd.Path) if err != nil { - return ret, errors.Wrapf(err, "error getting stat from path %q", wd.path) + return ret, errors.Wrapf(err, "error getting stat from path %q", wd.Path) } lwd := spec.LinuxWeightDevice{ - Weight: &wd.weight, + Weight: &wd.Weight, } lwd.Major = int64(unix.Major(wdStat.Rdev)) lwd.Minor = int64(unix.Minor(wdStat.Rdev)) @@ -310,7 +347,7 @@ func makeThrottleArray(throttleInput []string, rateType int) ([]spec.LinuxThrott if err != nil { return []spec.LinuxThrottleDevice{}, err } - ltdStat, err := getStatFromPath(t.path) + ltdStat, err := GetStatFromPath(t.path) if err != nil { return ltds, errors.Wrapf(err, "error getting stat from path %q", t.path) } @@ -324,7 +361,7 @@ func makeThrottleArray(throttleInput []string, rateType int) ([]spec.LinuxThrott return ltds, nil } -func getStatFromPath(path string) (unix.Stat_t, error) { +func GetStatFromPath(path string) (unix.Stat_t, error) { s := unix.Stat_t{} err := unix.Stat(path, &s) return s, err diff --git a/pkg/spec/config_linux_cgo.go b/pkg/spec/config_linux_cgo.go index ae83c9d52..05f42c4da 100644 --- a/pkg/spec/config_linux_cgo.go +++ b/pkg/spec/config_linux_cgo.go @@ -5,9 +5,10 @@ package createconfig import ( "io/ioutil" + "github.com/containers/libpod/pkg/seccomp" spec "github.com/opencontainers/runtime-spec/specs-go" "github.com/pkg/errors" - seccomp "github.com/seccomp/containers-golang" + goSeccomp "github.com/seccomp/containers-golang" "github.com/sirupsen/logrus" ) @@ -15,9 +16,9 @@ func getSeccompConfig(config *SecurityConfig, configSpec *spec.Spec) (*spec.Linu var seccompConfig *spec.LinuxSeccomp var err error - if config.SeccompPolicy == SeccompPolicyImage && config.SeccompProfileFromImage != "" { + if config.SeccompPolicy == seccomp.PolicyImage && config.SeccompProfileFromImage != "" { logrus.Debug("Loading seccomp profile from the security config") - seccompConfig, err = seccomp.LoadProfile(config.SeccompProfileFromImage, configSpec) + seccompConfig, err = goSeccomp.LoadProfile(config.SeccompProfileFromImage, configSpec) if err != nil { return nil, errors.Wrap(err, "loading seccomp profile failed") } @@ -30,13 +31,13 @@ func getSeccompConfig(config *SecurityConfig, configSpec *spec.Spec) (*spec.Linu if err != nil { return nil, errors.Wrapf(err, "opening seccomp profile (%s) failed", config.SeccompProfilePath) } - seccompConfig, err = seccomp.LoadProfile(string(seccompProfile), configSpec) + seccompConfig, err = goSeccomp.LoadProfile(string(seccompProfile), configSpec) if err != nil { return nil, errors.Wrapf(err, "loading seccomp profile (%s) failed", config.SeccompProfilePath) } } else { logrus.Debug("Loading default seccomp profile") - seccompConfig, err = seccomp.GetDefaultProfile(configSpec) + seccompConfig, err = goSeccomp.GetDefaultProfile(configSpec) if err != nil { return nil, errors.Wrapf(err, "loading seccomp profile (%s) failed", config.SeccompProfilePath) } diff --git a/pkg/spec/config_unsupported.go b/pkg/spec/config_unsupported.go index a2c7f4416..568afde55 100644 --- a/pkg/spec/config_unsupported.go +++ b/pkg/spec/config_unsupported.go @@ -15,7 +15,7 @@ func addDevice(g *generate.Generator, device string) error { return errors.New("function not implemented") } -func (c *CreateConfig) addPrivilegedDevices(g *generate.Generator) error { +func addPrivilegedDevices(g *generate.Generator) error { return errors.New("function not implemented") } @@ -27,6 +27,10 @@ func makeThrottleArray(throttleInput []string, rateType int) ([]spec.LinuxThrott return nil, errors.New("function not implemented") } -func devicesFromPath(g *generate.Generator, devicePath string) error { +func DevicesFromPath(g *generate.Generator, devicePath string) error { + return errors.New("function not implemented") +} + +func deviceCgroupRules(g *generate.Generator, deviceCgroupRules []string) error { return errors.New("function not implemented") } diff --git a/pkg/spec/createconfig.go b/pkg/spec/createconfig.go index fb222083b..daa997104 100644 --- a/pkg/spec/createconfig.go +++ b/pkg/spec/createconfig.go @@ -2,7 +2,6 @@ package createconfig import ( "os" - "sort" "strconv" "strings" "syscall" @@ -11,6 +10,7 @@ import ( "github.com/containers/libpod/libpod" "github.com/containers/libpod/libpod/define" "github.com/containers/libpod/pkg/namespaces" + "github.com/containers/libpod/pkg/seccomp" "github.com/containers/storage" "github.com/docker/go-connections/nat" spec "github.com/opencontainers/runtime-spec/specs-go" @@ -38,6 +38,7 @@ type CreateResourceConfig struct { CPUs float64 // cpus CPUsetCPUs string CPUsetMems string // cpuset-mems + DeviceCgroupRules []string //device-cgroup-rule DeviceReadBps []string // device-read-bps DeviceReadIOps []string // device-read-iops DeviceWriteBps []string // device-write-bps @@ -107,58 +108,17 @@ type NetworkConfig struct { PublishAll bool //publish-all } -// SeccompPolicy determines which seccomp profile gets applied to the container. -type SeccompPolicy int - -const ( - // SeccompPolicyDefault - if set use SecurityConfig.SeccompProfilePath, - // otherwise use the default profile. The SeccompProfilePath might be - // explicitly set by the user. - SeccompPolicyDefault SeccompPolicy = iota - // SeccompPolicyImage - if set use SecurityConfig.SeccompProfileFromImage, - // otherwise follow SeccompPolicyDefault. - SeccompPolicyImage -) - -// Map for easy lookups of supported policies. -var supportedSeccompPolicies = map[string]SeccompPolicy{ - "": SeccompPolicyDefault, - "default": SeccompPolicyDefault, - "image": SeccompPolicyImage, -} - -// LookupSeccompPolicy looksup the corresponding SeccompPolicy for the specified -// string. If none is found, an errors is returned including the list of -// supported policies. -// Note that an empty string resolved to SeccompPolicyDefault. -func LookupSeccompPolicy(s string) (SeccompPolicy, error) { - policy, exists := supportedSeccompPolicies[s] - if exists { - return policy, nil - } - - // Sort the keys first as maps are non-deterministic. - keys := []string{} - for k := range supportedSeccompPolicies { - if k != "" { - keys = append(keys, k) - } - } - sort.Strings(keys) - - return -1, errors.Errorf("invalid seccomp policy %q: valid policies are %+q", s, keys) -} - // SecurityConfig configures the security features for the container type SecurityConfig struct { CapAdd []string // cap-add CapDrop []string // cap-drop + CapRequired []string // cap-required LabelOpts []string //SecurityOpts NoNewPrivs bool //SecurityOpts ApparmorProfile string //SecurityOpts SeccompProfilePath string //SecurityOpts SeccompProfileFromImage string // seccomp profile from the container image - SeccompPolicy SeccompPolicy + SeccompPolicy seccomp.Policy SecurityOpts []string Privileged bool //privileged ReadOnlyRootfs bool //read-only @@ -167,6 +127,7 @@ type SecurityConfig struct { } // CreateConfig is a pre OCI spec structure. It represents user input from varlink or the CLI +// swagger:model CreateConfig type CreateConfig struct { Annotations map[string]string Args []string @@ -183,6 +144,7 @@ type CreateConfig struct { InitPath string //init-path Image string ImageID string + RawImageName string BuiltinImgVolumes map[string]struct{} // volumes defined in the image config ImageVolumeType string // how to handle the image volume, either bind, tmpfs, or ignore Interactive bool //interactive @@ -196,6 +158,7 @@ type CreateConfig struct { Resources CreateResourceConfig RestartPolicy string Rm bool //rm + Rmi bool //rmi StopSignal syscall.Signal // stop-signal StopTimeout uint // stop-timeout Systemd bool @@ -233,6 +196,7 @@ func (c *CreateConfig) createExitCommand(runtime *libpod.Runtime) ([]string, err if err != nil { return nil, err } + storageConfig := runtime.StorageConfig() // 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 @@ -245,23 +209,23 @@ func (c *CreateConfig) createExitCommand(runtime *libpod.Runtime) ([]string, err } command := []string{cmd, - "--root", config.StorageConfig.GraphRoot, - "--runroot", config.StorageConfig.RunRoot, + "--root", storageConfig.GraphRoot, + "--runroot", storageConfig.RunRoot, "--log-level", logrus.GetLevel().String(), - "--cgroup-manager", config.CgroupManager, - "--tmpdir", config.TmpDir, + "--cgroup-manager", config.Engine.CgroupManager, + "--tmpdir", config.Engine.TmpDir, } - if config.OCIRuntime != "" { - command = append(command, []string{"--runtime", config.OCIRuntime}...) + if config.Engine.OCIRuntime != "" { + command = append(command, []string{"--runtime", config.Engine.OCIRuntime}...) } - if config.StorageConfig.GraphDriverName != "" { - command = append(command, []string{"--storage-driver", config.StorageConfig.GraphDriverName}...) + if storageConfig.GraphDriverName != "" { + command = append(command, []string{"--storage-driver", storageConfig.GraphDriverName}...) } - for _, opt := range config.StorageConfig.GraphDriverOptions { + for _, opt := range storageConfig.GraphDriverOptions { command = append(command, []string{"--storage-opt", opt}...) } - if config.EventsLogger != "" { - command = append(command, []string{"--events-backend", config.EventsLogger}...) + if config.Engine.EventsLogger != "" { + command = append(command, []string{"--events-backend", config.Engine.EventsLogger}...) } if c.Syslog { @@ -273,6 +237,10 @@ func (c *CreateConfig) createExitCommand(runtime *libpod.Runtime) ([]string, err command = append(command, "--rm") } + if c.Rmi { + command = append(command, "--rmi") + } + return command, nil } @@ -381,9 +349,8 @@ func (c *CreateConfig) getContainerCreateOptions(runtime *libpod.Runtime, pod *l } options = append(options, nsOpts...) - useImageVolumes := c.ImageVolumeType == TypeBind // Gather up the options for NewContainer which consist of With... funcs - options = append(options, libpod.WithRootFSFromImage(c.ImageID, c.Image, useImageVolumes)) + options = append(options, libpod.WithRootFSFromImage(c.ImageID, c.Image, c.RawImageName)) options = append(options, libpod.WithConmonPidFile(c.ConmonPidFile)) options = append(options, libpod.WithLabels(c.Labels)) options = append(options, libpod.WithShmSize(c.Resources.ShmSize)) @@ -427,6 +394,6 @@ func (c *CreateConfig) getContainerCreateOptions(runtime *libpod.Runtime, pod *l // AddPrivilegedDevices iterates through host devices and adds all // host devices to the spec -func (c *CreateConfig) AddPrivilegedDevices(g *generate.Generator) error { - return c.addPrivilegedDevices(g) +func AddPrivilegedDevices(g *generate.Generator) error { + return addPrivilegedDevices(g) } diff --git a/pkg/spec/namespaces.go b/pkg/spec/namespaces.go index 1f98e6e25..838d95c54 100644 --- a/pkg/spec/namespaces.go +++ b/pkg/spec/namespaces.go @@ -422,7 +422,7 @@ func (c *UtsConfig) ConfigureGenerator(g *generate.Generator, net *NetworkConfig if hostname == "" { switch { case utsCtrID != "": - utsCtr, err := runtime.GetContainer(utsCtrID) + utsCtr, err := runtime.LookupContainer(utsCtrID) if err != nil { return errors.Wrapf(err, "unable to retrieve hostname from dependency container %s", utsCtrID) } diff --git a/pkg/spec/parse.go b/pkg/spec/parse.go index 6fa0b0636..38d93b87f 100644 --- a/pkg/spec/parse.go +++ b/pkg/spec/parse.go @@ -2,24 +2,29 @@ package createconfig import ( "fmt" + "regexp" "strconv" "strings" "github.com/docker/go-units" + "github.com/pkg/errors" ) +// deviceCgroupRulegex defines the valid format of device-cgroup-rule +var deviceCgroupRuleRegex = regexp.MustCompile(`^([acb]) ([0-9]+|\*):([0-9]+|\*) ([rwm]{1,3})$`) + // Pod signifies a kernel namespace is being shared // by a container with the pod it is associated with const Pod = "pod" // weightDevice is a structure that holds device:weight pair type weightDevice struct { - path string - weight uint16 + Path string + Weight uint16 } func (w *weightDevice) String() string { - return fmt.Sprintf("%s:%d", w.path, w.weight) + return fmt.Sprintf("%s:%d", w.Path, w.Weight) } // LinuxNS is a struct that contains namespace information @@ -54,9 +59,9 @@ func NS(s string) string { return "" } -// validateweightDevice validates that the specified string has a valid device-weight format +// ValidateweightDevice validates that the specified string has a valid device-weight format // for blkio-weight-device flag -func validateweightDevice(val string) (*weightDevice, error) { +func ValidateweightDevice(val string) (*weightDevice, error) { split := strings.SplitN(val, ":", 2) if len(split) != 2 { return nil, fmt.Errorf("bad format: %s", val) @@ -73,8 +78,8 @@ func validateweightDevice(val string) (*weightDevice, error) { } return &weightDevice{ - path: split[0], - weight: uint16(weight), + Path: split[0], + Weight: uint16(weight), }, nil } @@ -205,3 +210,16 @@ func IsValidDeviceMode(mode string) bool { } return true } + +// validateDeviceCgroupRule validates the format of deviceCgroupRule +func validateDeviceCgroupRule(deviceCgroupRule string) error { + if !deviceCgroupRuleRegex.MatchString(deviceCgroupRule) { + return errors.Errorf("invalid device cgroup rule format: '%s'", deviceCgroupRule) + } + return nil +} + +// parseDeviceCgroupRule matches and parses the deviceCgroupRule into slice +func parseDeviceCgroupRule(deviceCgroupRule string) [][]string { + return deviceCgroupRuleRegex.FindAllStringSubmatch(deviceCgroupRule, -1) +} diff --git a/pkg/spec/security.go b/pkg/spec/security.go index 372fe87c6..0f8d36f00 100644 --- a/pkg/spec/security.go +++ b/pkg/spec/security.go @@ -4,11 +4,13 @@ import ( "fmt" "strings" + "github.com/containers/common/pkg/capabilities" "github.com/containers/libpod/libpod" - "github.com/docker/docker/oci/caps" + "github.com/containers/libpod/pkg/util" "github.com/opencontainers/runtime-tools/generate" "github.com/opencontainers/selinux/go-selinux/label" "github.com/pkg/errors" + "github.com/sirupsen/logrus" ) // ToCreateOptions convert the SecurityConfig to a slice of container create @@ -113,28 +115,49 @@ func (c *SecurityConfig) ConfigureGenerator(g *generate.Generator, user *UserCon configSpec := g.Config var err error - var caplist []string + var defaultCaplist []string bounding := configSpec.Process.Capabilities.Bounding if useNotRoot(user.User) { - configSpec.Process.Capabilities.Bounding = caplist + configSpec.Process.Capabilities.Bounding = defaultCaplist } - caplist, err = caps.TweakCapabilities(configSpec.Process.Capabilities.Bounding, c.CapAdd, c.CapDrop, nil, false) + defaultCaplist, err = capabilities.MergeCapabilities(configSpec.Process.Capabilities.Bounding, c.CapAdd, c.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 + privCapRequired := []string{} + + if !c.Privileged && len(c.CapRequired) > 0 { + // Pass CapRequired in CapAdd field to normalize capabilities names + capRequired, err := capabilities.MergeCapabilities(nil, c.CapRequired, nil) + if err != nil { + logrus.Errorf("capabilities requested by user or image are not valid: %q", strings.Join(c.CapRequired, ",")) + } else { + // Verify all capRequiered are in the defaultCapList + for _, cap := range capRequired { + if !util.StringInSlice(cap, defaultCaplist) { + privCapRequired = append(privCapRequired, cap) + } + } + } + if len(privCapRequired) == 0 { + defaultCaplist = capRequired + } else { + logrus.Errorf("capabilities requested by user or image are not allowed by default: %q", strings.Join(privCapRequired, ",")) + } + } + configSpec.Process.Capabilities.Bounding = defaultCaplist + configSpec.Process.Capabilities.Permitted = defaultCaplist + configSpec.Process.Capabilities.Inheritable = defaultCaplist + configSpec.Process.Capabilities.Effective = defaultCaplist + configSpec.Process.Capabilities.Ambient = defaultCaplist if useNotRoot(user.User) { - caplist, err = caps.TweakCapabilities(bounding, c.CapAdd, c.CapDrop, nil, false) + defaultCaplist, err = capabilities.MergeCapabilities(bounding, c.CapAdd, c.CapDrop) if err != nil { return err } } - configSpec.Process.Capabilities.Bounding = caplist + configSpec.Process.Capabilities.Bounding = defaultCaplist // HANDLE SECCOMP if c.SeccompProfilePath != "unconfined" { diff --git a/pkg/spec/spec.go b/pkg/spec/spec.go index cae055bb0..d4fd5976f 100644 --- a/pkg/spec/spec.go +++ b/pkg/spec/spec.go @@ -3,12 +3,14 @@ package createconfig import ( "strings" + "github.com/containers/common/pkg/capabilities" + cconfig "github.com/containers/common/pkg/config" "github.com/containers/libpod/libpod" - libpodconfig "github.com/containers/libpod/libpod/config" - "github.com/containers/libpod/libpod/define" "github.com/containers/libpod/pkg/cgroups" + "github.com/containers/libpod/pkg/env" "github.com/containers/libpod/pkg/rootless" "github.com/containers/libpod/pkg/sysinfo" + "github.com/containers/libpod/pkg/util" "github.com/docker/go-units" "github.com/opencontainers/runc/libcontainer/user" spec "github.com/opencontainers/runtime-spec/specs-go" @@ -16,9 +18,9 @@ import ( "github.com/pkg/errors" ) -const cpuPeriod = 100000 +const CpuPeriod = 100000 -func getAvailableGids() (int64, error) { +func GetAvailableGids() (int64, error) { idMap, err := user.ParseIDMapFile("/proc/self/gid_map") if err != nil { return 0, err @@ -78,9 +80,40 @@ func (config *CreateConfig) createConfigToOCISpec(runtime *libpod.Runtime, userM g.AddLinuxMaskedPaths("/sys/kernel") } } + var runtimeConfig *cconfig.Config + + if runtime != nil { + runtimeConfig, err = runtime.GetConfig() + if err != nil { + return nil, err + } + g.Config.Process.Capabilities.Bounding = runtimeConfig.Containers.DefaultCapabilities + sysctls, err := util.ValidateSysctls(runtimeConfig.Containers.DefaultSysctls) + if err != nil { + return nil, err + } + + for name, val := range config.Security.Sysctl { + sysctls[name] = val + } + config.Security.Sysctl = sysctls + if !util.StringInSlice("host", config.Resources.Ulimit) { + config.Resources.Ulimit = append(runtimeConfig.Containers.DefaultUlimits, config.Resources.Ulimit...) + } + if config.Resources.PidsLimit < 0 && !config.cgroupDisabled() { + config.Resources.PidsLimit = runtimeConfig.Containers.PidsLimit + } + + } else { + g.Config.Process.Capabilities.Bounding = cconfig.DefaultCapabilities + if config.Resources.PidsLimit < 0 && !config.cgroupDisabled() { + config.Resources.PidsLimit = cconfig.DefaultPidsLimit + } + } + gid5Available := true if isRootless { - nGids, err := getAvailableGids() + nGids, err := GetAvailableGids() if err != nil { return nil, err } @@ -150,7 +183,6 @@ func (config *CreateConfig) createConfigToOCISpec(runtime *libpod.Runtime, userM for key, val := range config.Annotations { g.AddAnnotation(key, val) } - g.AddProcessEnv("container", "podman") addedResources := false @@ -197,8 +229,8 @@ func (config *CreateConfig) createConfigToOCISpec(runtime *libpod.Runtime, userM addedResources = true } if config.Resources.CPUs != 0 { - g.SetLinuxResourcesCPUPeriod(cpuPeriod) - g.SetLinuxResourcesCPUQuota(int64(config.Resources.CPUs * cpuPeriod)) + g.SetLinuxResourcesCPUPeriod(CpuPeriod) + g.SetLinuxResourcesCPUQuota(int64(config.Resources.CPUs * CpuPeriod)) addedResources = true } if config.Resources.CPURtRuntime != 0 { @@ -223,35 +255,43 @@ func (config *CreateConfig) createConfigToOCISpec(runtime *libpod.Runtime, userM // 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 := config.AddPrivilegedDevices(&g); err != nil { + if err := AddPrivilegedDevices(&g); err != nil { return nil, err } } else { for _, devicePath := range config.Devices { - if err := devicesFromPath(&g, devicePath); err != nil { + if err := DevicesFromPath(&g, devicePath); err != nil { return nil, err } } + if len(config.Resources.DeviceCgroupRules) != 0 { + if err := deviceCgroupRules(&g, config.Resources.DeviceCgroupRules); err != nil { + return nil, err + } + addedResources = true + } } - // SECURITY OPTS g.SetProcessNoNewPrivileges(config.Security.NoNewPrivs) if !config.Security.Privileged { g.SetProcessApparmorProfile(config.Security.ApparmorProfile) } - blockAccessToKernelFilesystems(config, &g) - - var runtimeConfig *libpodconfig.Config - - if runtime != nil { - runtimeConfig, err = runtime.GetConfig() - if err != nil { + // Unless already set via the CLI, check if we need to disable process + // labels or set the defaults. + if len(config.Security.LabelOpts) == 0 && runtimeConfig != nil { + if !runtimeConfig.Containers.EnableLabeling { + // Disabled in the config. + config.Security.LabelOpts = append(config.Security.LabelOpts, "disable") + } else if err := config.Security.SetLabelOpts(runtime, &config.Pid, &config.Ipc); err != nil { + // Defaults! return nil, err } } + BlockAccessToKernelFilesystems(config.Security.Privileged, config.Pid.PidMode.IsHost(), &g) + // RESOURCES - PIDS if config.Resources.PidsLimit > 0 { // if running on rootless on a cgroupv1 machine or using the cgroupfs manager, pids @@ -264,7 +304,7 @@ func (config *CreateConfig) createConfigToOCISpec(runtime *libpod.Runtime, userM if err != nil { return nil, err } - if (!cgroup2 || (runtimeConfig != nil && runtimeConfig.CgroupManager != define.SystemdCgroupsManager)) && config.Resources.PidsLimit == sysinfo.GetDefaultPidsLimit() { + if (!cgroup2 || (runtimeConfig != nil && runtimeConfig.Engine.CgroupManager != cconfig.SystemdCgroupsManager)) && config.Resources.PidsLimit == sysinfo.GetDefaultPidsLimit() { setPidLimit = false } } @@ -274,6 +314,9 @@ func (config *CreateConfig) createConfigToOCISpec(runtime *libpod.Runtime, userM } } + // Make sure to always set the default variables unless overridden in the + // config. + config.Env = env.Join(env.DefaultEnvVariables, config.Env) for name, val := range config.Env { g.AddProcessEnv(name, val) } @@ -309,14 +352,26 @@ func (config *CreateConfig) createConfigToOCISpec(runtime *libpod.Runtime, userM } configSpec := g.Config + // If the container image specifies an label with a + // capabilities.ContainerImageLabel then split the comma separated list + // of capabilities and record them. This list indicates the only + // capabilities, required to run the container. + var capRequired []string + for key, val := range config.Labels { + if util.StringInSlice(key, capabilities.ContainerImageLabels) { + capRequired = strings.Split(val, ",") + } + } + config.Security.CapRequired = capRequired + if err := config.Security.ConfigureGenerator(&g, &config.User); err != nil { return nil, err } // BIND MOUNTS - configSpec.Mounts = supercedeUserMounts(userMounts, configSpec.Mounts) + configSpec.Mounts = SupercedeUserMounts(userMounts, configSpec.Mounts) // Process mounts to ensure correct options - finalMounts, err := initFSMounts(configSpec.Mounts) + finalMounts, err := InitFSMounts(configSpec.Mounts) if err != nil { return nil, err } @@ -341,7 +396,7 @@ func (config *CreateConfig) createConfigToOCISpec(runtime *libpod.Runtime, userM configSpec.Linux.Resources = &spec.LinuxResources{} } - canUseResources := cgroup2 && runtimeConfig != nil && (runtimeConfig.CgroupManager == define.SystemdCgroupsManager) + canUseResources := cgroup2 && runtimeConfig != nil && (runtimeConfig.Engine.CgroupManager == cconfig.SystemdCgroupsManager) if addedResources && !canUseResources { return nil, errors.New("invalid configuration, cannot specify resource limits without cgroups v2 and --cgroup-manager=systemd") @@ -398,8 +453,12 @@ func (config *CreateConfig) createConfigToOCISpec(runtime *libpod.Runtime, userM return configSpec, nil } -func blockAccessToKernelFilesystems(config *CreateConfig, g *generate.Generator) { - if !config.Security.Privileged { +func (config *CreateConfig) cgroupDisabled() bool { + return config.Cgroup.Cgroups == "disabled" +} + +func BlockAccessToKernelFilesystems(privileged, pidModeIsHost bool, g *generate.Generator) { + if !privileged { for _, mp := range []string{ "/proc/acpi", "/proc/kcore", @@ -415,7 +474,7 @@ func blockAccessToKernelFilesystems(config *CreateConfig, g *generate.Generator) g.AddLinuxMaskedPaths(mp) } - if config.Pid.PidMode.IsHost() && rootless.IsRootless() { + if pidModeIsHost && rootless.IsRootless() { return } diff --git a/pkg/spec/storage.go b/pkg/spec/storage.go index 0e2098c1d..b0687b4c2 100644 --- a/pkg/spec/storage.go +++ b/pkg/spec/storage.go @@ -124,7 +124,7 @@ func (config *CreateConfig) parseVolumes(runtime *libpod.Runtime) ([]spec.Mount, if err != nil { return nil, nil, err } - initPath = rtc.InitPath + initPath = rtc.Engine.InitPath } initMount, err := config.addContainerInitBinary(initPath) if err != nil { @@ -739,6 +739,7 @@ func (config *CreateConfig) getImageVolumes() (map[string]spec.Mount, map[string for vol := range config.BuiltinImgVolumes { cleanDest := filepath.Clean(vol) + logrus.Debugf("Adding image volume at %s", cleanDest) if config.ImageVolumeType == "tmpfs" { // Tmpfs image volumes are handled as mounts mount := spec.Mount{ @@ -747,13 +748,13 @@ func (config *CreateConfig) getImageVolumes() (map[string]spec.Mount, map[string Type: TypeTmpfs, Options: []string{"rprivate", "rw", "nodev", "exec"}, } - mounts[vol] = mount + mounts[cleanDest] = mount } else { // Anonymous volumes have no name. namedVolume := new(libpod.ContainerNamedVolume) namedVolume.Options = []string{"rprivate", "rw", "nodev", "exec"} namedVolume.Dest = cleanDest - volumes[vol] = namedVolume + volumes[cleanDest] = namedVolume } } @@ -825,7 +826,7 @@ func (config *CreateConfig) addContainerInitBinary(path string) (spec.Mount, err // TODO: Should we unmount subtree mounts? E.g., if /tmp/ is mounted by // one mount, and we already have /tmp/a and /tmp/b, should we remove // the /tmp/a and /tmp/b mounts in favor of the more general /tmp? -func supercedeUserMounts(mounts []spec.Mount, configMount []spec.Mount) []spec.Mount { +func SupercedeUserMounts(mounts []spec.Mount, configMount []spec.Mount) []spec.Mount { if len(mounts) > 0 { // If we have overlappings mounts, remove them from the spec in favor of // the user-added volume mounts @@ -854,7 +855,7 @@ func supercedeUserMounts(mounts []spec.Mount, configMount []spec.Mount) []spec.M } // Ensure mount options on all mounts are correct -func initFSMounts(inputMounts []spec.Mount) ([]spec.Mount, error) { +func InitFSMounts(inputMounts []spec.Mount) ([]spec.Mount, error) { // We need to look up mounts so we can figure out the proper mount flags // to apply. systemMounts, err := pmount.GetMounts() |