diff options
Diffstat (limited to 'pkg/specgen')
| -rw-r--r-- | pkg/specgen/config_linux.go | 93 | ||||
| -rw-r--r-- | pkg/specgen/config_linux_nocgo.go | 11 | ||||
| -rw-r--r-- | pkg/specgen/container_validate.go | 16 | ||||
| -rw-r--r-- | pkg/specgen/generate/config_linux.go | 323 | ||||
| -rw-r--r-- | pkg/specgen/generate/config_linux_cgo.go (renamed from pkg/specgen/config_linux_cgo.go) | 6 | ||||
| -rw-r--r-- | pkg/specgen/generate/config_linux_nocgo.go | 14 | ||||
| -rw-r--r-- | pkg/specgen/generate/container.go | 176 | ||||
| -rw-r--r-- | pkg/specgen/generate/container_create.go (renamed from pkg/specgen/container_create.go) | 29 | ||||
| -rw-r--r-- | pkg/specgen/generate/namespaces.go | 417 | ||||
| -rw-r--r-- | pkg/specgen/generate/oci.go (renamed from pkg/specgen/oci.go) | 49 | ||||
| -rw-r--r-- | pkg/specgen/generate/pod_create.go (renamed from pkg/specgen/pod_create.go) | 18 | ||||
| -rw-r--r-- | pkg/specgen/generate/security.go | 154 | ||||
| -rw-r--r-- | pkg/specgen/generate/storage.go | 885 | ||||
| -rw-r--r-- | pkg/specgen/namespaces.go | 388 | ||||
| -rw-r--r-- | pkg/specgen/pod_validate.go | 5 | ||||
| -rw-r--r-- | pkg/specgen/specgen.go | 28 |
16 files changed, 2152 insertions, 460 deletions
diff --git a/pkg/specgen/config_linux.go b/pkg/specgen/config_linux.go new file mode 100644 index 000000000..82a371492 --- /dev/null +++ b/pkg/specgen/config_linux.go @@ -0,0 +1,93 @@ +package specgen + +//func createBlockIO() (*spec.LinuxBlockIO, error) { +// var ret *spec.LinuxBlockIO +// bio := &spec.LinuxBlockIO{} +// if c.Resources.BlkioWeight > 0 { +// ret = bio +// bio.Weight = &c.Resources.BlkioWeight +// } +// if len(c.Resources.BlkioWeightDevice) > 0 { +// var lwds []spec.LinuxWeightDevice +// ret = bio +// for _, i := range c.Resources.BlkioWeightDevice { +// wd, err := ValidateweightDevice(i) +// if err != nil { +// return ret, errors.Wrapf(err, "invalid values for blkio-weight-device") +// } +// wdStat, err := GetStatFromPath(wd.Path) +// if err != nil { +// return ret, errors.Wrapf(err, "error getting stat from path %q", wd.Path) +// } +// lwd := spec.LinuxWeightDevice{ +// Weight: &wd.Weight, +// } +// lwd.Major = int64(unix.Major(wdStat.Rdev)) +// lwd.Minor = int64(unix.Minor(wdStat.Rdev)) +// lwds = append(lwds, lwd) +// } +// bio.WeightDevice = lwds +// } +// if len(c.Resources.DeviceReadBps) > 0 { +// ret = bio +// readBps, err := makeThrottleArray(c.Resources.DeviceReadBps, bps) +// if err != nil { +// return ret, err +// } +// bio.ThrottleReadBpsDevice = readBps +// } +// if len(c.Resources.DeviceWriteBps) > 0 { +// ret = bio +// writeBpds, err := makeThrottleArray(c.Resources.DeviceWriteBps, bps) +// if err != nil { +// return ret, err +// } +// bio.ThrottleWriteBpsDevice = writeBpds +// } +// if len(c.Resources.DeviceReadIOps) > 0 { +// ret = bio +// readIOps, err := makeThrottleArray(c.Resources.DeviceReadIOps, iops) +// if err != nil { +// return ret, err +// } +// bio.ThrottleReadIOPSDevice = readIOps +// } +// if len(c.Resources.DeviceWriteIOps) > 0 { +// ret = bio +// writeIOps, err := makeThrottleArray(c.Resources.DeviceWriteIOps, iops) +// if err != nil { +// return ret, err +// } +// bio.ThrottleWriteIOPSDevice = writeIOps +// } +// return ret, nil +//} + +//func makeThrottleArray(throttleInput []string, rateType int) ([]spec.LinuxThrottleDevice, error) { +// var ( +// ltds []spec.LinuxThrottleDevice +// t *throttleDevice +// err error +// ) +// for _, i := range throttleInput { +// if rateType == bps { +// t, err = validateBpsDevice(i) +// } else { +// t, err = validateIOpsDevice(i) +// } +// if err != nil { +// return []spec.LinuxThrottleDevice{}, err +// } +// ltdStat, err := GetStatFromPath(t.path) +// if err != nil { +// return ltds, errors.Wrapf(err, "error getting stat from path %q", t.path) +// } +// ltd := spec.LinuxThrottleDevice{ +// Rate: t.rate, +// } +// ltd.Major = int64(unix.Major(ltdStat.Rdev)) +// ltd.Minor = int64(unix.Minor(ltdStat.Rdev)) +// ltds = append(ltds, ltd) +// } +// return ltds, nil +//} diff --git a/pkg/specgen/config_linux_nocgo.go b/pkg/specgen/config_linux_nocgo.go deleted file mode 100644 index fc0c58c37..000000000 --- a/pkg/specgen/config_linux_nocgo.go +++ /dev/null @@ -1,11 +0,0 @@ -// +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/container_validate.go b/pkg/specgen/container_validate.go index b27659f5f..9152e7ee7 100644 --- a/pkg/specgen/container_validate.go +++ b/pkg/specgen/container_validate.go @@ -14,7 +14,7 @@ var ( // 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"} + ImageVolumeModeValues = []string{"ignore", "tmpfs", "bind"} ) func exclusiveOptions(opt1, opt2 string) error { @@ -23,7 +23,7 @@ func exclusiveOptions(opt1, opt2 string) error { // Validate verifies that the given SpecGenerator is valid and satisfies required // input for creating a container. -func (s *SpecGenerator) validate() error { +func (s *SpecGenerator) Validate() error { // // ContainerBasicConfig @@ -68,18 +68,6 @@ func (s *SpecGenerator) validate() error { 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") diff --git a/pkg/specgen/generate/config_linux.go b/pkg/specgen/generate/config_linux.go new file mode 100644 index 000000000..1b2a2ac32 --- /dev/null +++ b/pkg/specgen/generate/config_linux.go @@ -0,0 +1,323 @@ +package generate + +import ( + "fmt" + "io/ioutil" + "os" + "path/filepath" + "strings" + + "github.com/containers/libpod/pkg/rootless" + "github.com/opencontainers/runc/libcontainer/configs" + "github.com/opencontainers/runc/libcontainer/devices" + spec "github.com/opencontainers/runtime-spec/specs-go" + "github.com/opencontainers/runtime-tools/generate" + "github.com/pkg/errors" + "golang.org/x/sys/unix" +) + +func u32Ptr(i int64) *uint32 { u := uint32(i); return &u } +func fmPtr(i int64) *os.FileMode { fm := os.FileMode(i); return &fm } + +// Device transforms a libcontainer configs.Device to a specs.LinuxDevice object. +func Device(d *configs.Device) spec.LinuxDevice { + return spec.LinuxDevice{ + Type: string(d.Type), + Path: d.Path, + Major: d.Major, + Minor: d.Minor, + FileMode: fmPtr(int64(d.FileMode)), + UID: u32Ptr(int64(d.Uid)), + GID: u32Ptr(int64(d.Gid)), + } +} + +func addPrivilegedDevices(g *generate.Generator) error { + hostDevices, err := getDevices("/dev") + if err != nil { + return err + } + g.ClearLinuxDevices() + + if rootless.IsRootless() { + mounts := make(map[string]interface{}) + for _, m := range g.Mounts() { + mounts[m.Destination] = true + } + newMounts := []spec.Mount{} + for _, d := range hostDevices { + devMnt := spec.Mount{ + Destination: d.Path, + Type: TypeBind, + Source: d.Path, + Options: []string{"slave", "nosuid", "noexec", "rw", "rbind"}, + } + if d.Path == "/dev/ptmx" || strings.HasPrefix(d.Path, "/dev/tty") { + continue + } + if _, found := mounts[d.Path]; found { + continue + } + st, err := os.Stat(d.Path) + if err != nil { + if err == unix.EPERM { + continue + } + return errors.Wrapf(err, "stat %s", d.Path) + } + // Skip devices that the user has not access to. + if st.Mode()&0007 == 0 { + continue + } + newMounts = append(newMounts, devMnt) + } + g.Config.Mounts = append(newMounts, g.Config.Mounts...) + g.Config.Linux.Resources.Devices = nil + } else { + for _, d := range hostDevices { + g.AddDevice(Device(d)) + } + // Add resources device - need to clear the existing one first. + g.Config.Linux.Resources.Devices = nil + g.AddLinuxResourcesDevice(true, "", nil, nil, "rwm") + } + + return nil +} + +// 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 + if src, err := os.Lstat(resolvedDevicePath); err == nil && src.Mode()&os.ModeSymlink == os.ModeSymlink { + if linkedPathOnHost, err := filepath.EvalSymlinks(resolvedDevicePath); err == nil { + resolvedDevicePath = linkedPathOnHost + } + } + st, err := os.Stat(resolvedDevicePath) + if err != nil { + return errors.Wrapf(err, "cannot stat %s", devicePath) + } + if st.IsDir() { + found := false + src := resolvedDevicePath + dest := src + var devmode string + if len(devs) > 1 { + if len(devs[1]) > 0 && devs[1][0] == '/' { + dest = devs[1] + } else { + devmode = devs[1] + } + } + if len(devs) > 2 { + if devmode != "" { + return errors.Wrapf(unix.EINVAL, "invalid device specification %s", devicePath) + } + devmode = devs[2] + } + + // mount the internal devices recursively + if err := filepath.Walk(resolvedDevicePath, func(dpath string, f os.FileInfo, e error) error { + + if f.Mode()&os.ModeDevice == os.ModeDevice { + found = true + device := fmt.Sprintf("%s:%s", dpath, filepath.Join(dest, strings.TrimPrefix(dpath, src))) + if devmode != "" { + device = fmt.Sprintf("%s:%s", device, devmode) + } + if err := addDevice(g, device); err != nil { + return errors.Wrapf(err, "failed to add %s device", dpath) + } + } + return nil + }); err != nil { + return err + } + if !found { + return errors.Wrapf(unix.EINVAL, "no devices found in %s", devicePath) + } + return nil + } + + return addDevice(g, strings.Join(append([]string{resolvedDevicePath}, devs[1:]...), ":")) +} + +func BlockAccessToKernelFilesystems(privileged, pidModeIsHost bool, g *generate.Generator) { + if !privileged { + for _, mp := range []string{ + "/proc/acpi", + "/proc/kcore", + "/proc/keys", + "/proc/latency_stats", + "/proc/timer_list", + "/proc/timer_stats", + "/proc/sched_debug", + "/proc/scsi", + "/sys/firmware", + "/sys/fs/selinux", + } { + g.AddLinuxMaskedPaths(mp) + } + + if pidModeIsHost && rootless.IsRootless() { + return + } + + for _, rp := range []string{ + "/proc/asound", + "/proc/bus", + "/proc/fs", + "/proc/irq", + "/proc/sys", + "/proc/sysrq-trigger", + } { + g.AddLinuxReadonlyPaths(rp) + } + } +} + +// based on getDevices from runc (libcontainer/devices/devices.go) +func getDevices(path string) ([]*configs.Device, error) { + files, err := ioutil.ReadDir(path) + if err != nil { + if rootless.IsRootless() && os.IsPermission(err) { + return nil, nil + } + return nil, err + } + out := []*configs.Device{} + for _, f := range files { + switch { + case f.IsDir(): + switch f.Name() { + // ".lxc" & ".lxd-mounts" added to address https://github.com/lxc/lxd/issues/2825 + case "pts", "shm", "fd", "mqueue", ".lxc", ".lxd-mounts": + continue + default: + sub, err := getDevices(filepath.Join(path, f.Name())) + if err != nil { + return nil, err + } + if sub != nil { + out = append(out, sub...) + } + continue + } + case f.Name() == "console": + continue + } + device, err := devices.DeviceFromPath(filepath.Join(path, f.Name()), "rwm") + if err != nil { + if err == devices.ErrNotADevice { + continue + } + if os.IsNotExist(err) { + continue + } + return nil, err + } + out = append(out, device) + } + return out, nil +} + +func addDevice(g *generate.Generator, device string) error { + src, dst, permissions, err := ParseDevice(device) + if err != nil { + return err + } + dev, err := devices.DeviceFromPath(src, permissions) + if err != nil { + return errors.Wrapf(err, "%s is not a valid device", src) + } + if rootless.IsRootless() { + if _, err := os.Stat(src); err != nil { + if os.IsNotExist(err) { + return errors.Wrapf(err, "the specified device %s doesn't exist", src) + } + return errors.Wrapf(err, "stat device %s exist", src) + } + perm := "ro" + if strings.Contains(permissions, "w") { + perm = "rw" + } + devMnt := spec.Mount{ + Destination: dst, + Type: TypeBind, + Source: src, + Options: []string{"slave", "nosuid", "noexec", perm, "rbind"}, + } + g.Config.Mounts = append(g.Config.Mounts, devMnt) + return nil + } + dev.Path = dst + linuxdev := spec.LinuxDevice{ + Path: dev.Path, + Type: string(dev.Type), + Major: dev.Major, + Minor: dev.Minor, + FileMode: &dev.FileMode, + UID: &dev.Uid, + GID: &dev.Gid, + } + g.AddDevice(linuxdev) + g.AddLinuxResourcesDevice(true, string(dev.Type), &dev.Major, &dev.Minor, dev.Permissions) + return nil +} + +// ParseDevice parses device mapping string to a src, dest & permissions string +func ParseDevice(device string) (string, string, string, error) { //nolint + src := "" + dst := "" + permissions := "rwm" + arr := strings.Split(device, ":") + switch len(arr) { + case 3: + if !IsValidDeviceMode(arr[2]) { + return "", "", "", fmt.Errorf("invalid device mode: %s", arr[2]) + } + permissions = arr[2] + fallthrough + case 2: + if IsValidDeviceMode(arr[1]) { + permissions = arr[1] + } else { + if arr[1][0] != '/' { + return "", "", "", fmt.Errorf("invalid device mode: %s", arr[1]) + } + dst = arr[1] + } + fallthrough + case 1: + src = arr[0] + default: + return "", "", "", fmt.Errorf("invalid device specification: %s", device) + } + + if dst == "" { + dst = src + } + return src, dst, permissions, nil +} + +// IsValidDeviceMode checks if the mode for device is valid or not. +// IsValid mode is a composition of r (read), w (write), and m (mknod). +func IsValidDeviceMode(mode string) bool { + var legalDeviceMode = map[rune]bool{ + 'r': true, + 'w': true, + 'm': true, + } + if mode == "" { + return false + } + for _, c := range mode { + if !legalDeviceMode[c] { + return false + } + legalDeviceMode[c] = false + } + return true +} diff --git a/pkg/specgen/config_linux_cgo.go b/pkg/specgen/generate/config_linux_cgo.go index 6f547a40d..b06ef5c9a 100644 --- a/pkg/specgen/config_linux_cgo.go +++ b/pkg/specgen/generate/config_linux_cgo.go @@ -1,6 +1,6 @@ // +build linux,cgo -package specgen +package generate import ( "context" @@ -8,16 +8,16 @@ import ( "github.com/containers/libpod/libpod/image" "github.com/containers/libpod/pkg/seccomp" + "github.com/containers/libpod/pkg/specgen" 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) { +func getSeccompConfig(s *specgen.SpecGenerator, 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 diff --git a/pkg/specgen/generate/config_linux_nocgo.go b/pkg/specgen/generate/config_linux_nocgo.go new file mode 100644 index 000000000..fc8ed206d --- /dev/null +++ b/pkg/specgen/generate/config_linux_nocgo.go @@ -0,0 +1,14 @@ +// +build linux,!cgo + +package generate + +import ( + "errors" + + "github.com/containers/libpod/pkg/specgen" + spec "github.com/opencontainers/runtime-spec/specs-go" +) + +func (s *specgen.SpecGenerator) getSeccompConfig(configSpec *spec.Spec) (*spec.LinuxSeccomp, error) { + return nil, errors.New("not implemented") +} diff --git a/pkg/specgen/generate/container.go b/pkg/specgen/generate/container.go new file mode 100644 index 000000000..7233acb8a --- /dev/null +++ b/pkg/specgen/generate/container.go @@ -0,0 +1,176 @@ +package generate + +import ( + "context" + + "github.com/containers/libpod/libpod" + ann "github.com/containers/libpod/pkg/annotations" + envLib "github.com/containers/libpod/pkg/env" + "github.com/containers/libpod/pkg/signal" + "github.com/containers/libpod/pkg/specgen" + "github.com/pkg/errors" + "golang.org/x/sys/unix" +) + +func CompleteSpec(ctx context.Context, r *libpod.Runtime, s *specgen.SpecGenerator) error { + + newImage, err := r.ImageRuntime().NewFromLocal(s.Image) + if err != nil { + return err + } + + // Image stop signal + if s.StopSignal == nil && newImage.Config != nil { + sig, err := signal.ParseSignalNameOrNumber(newImage.Config.StopSignal) + if err != nil { + return err + } + s.StopSignal = &sig + } + // Image envs from the image if they don't exist + // already + if newImage.Config != nil && len(newImage.Config.Env) > 0 { + envs, err := envLib.ParseSlice(newImage.Config.Env) + if err != nil { + return err + } + for k, v := range envs { + if _, exists := s.Env[k]; !exists { + s.Env[v] = k + } + } + } + + // labels from the image that dont exist already + if config := newImage.Config; config != nil { + for k, v := range config.Labels { + if _, exists := s.Labels[k]; !exists { + s.Labels[k] = v + } + } + } + + // annotations + // in the event this container is in a pod, and the pod has an infra container + // we will want to configure it as a type "container" instead defaulting to + // the behavior of a "sandbox" container + // In Kata containers: + // - "sandbox" is the annotation that denotes the container should use its own + // VM, which is the default behavior + // - "container" denotes the container should join the VM of the SandboxID + // (the infra container) + s.Annotations = make(map[string]string) + if len(s.Pod) > 0 { + s.Annotations[ann.SandboxID] = s.Pod + s.Annotations[ann.ContainerType] = ann.ContainerTypeContainer + } + // + // Next, add annotations from the image + annotations, err := newImage.Annotations(ctx) + if err != nil { + return err + } + for k, v := range annotations { + annotations[k] = v + } + + // entrypoint + if config := newImage.Config; config != nil { + if len(s.Entrypoint) < 1 && len(config.Entrypoint) > 0 { + s.Entrypoint = config.Entrypoint + } + if len(s.Command) < 1 && len(config.Cmd) > 0 { + s.Command = config.Cmd + } + if len(s.Command) < 1 && len(s.Entrypoint) < 1 { + return errors.Errorf("No command provided or as CMD or ENTRYPOINT in this image") + } + // workdir + if len(s.WorkDir) < 1 && len(config.WorkingDir) > 1 { + s.WorkDir = config.WorkingDir + } + } + + if len(s.SeccompProfilePath) < 1 { + p, err := libpod.DefaultSeccompPath() + if err != nil { + return err + } + s.SeccompProfilePath = p + } + + if user := s.User; len(user) == 0 { + switch { + // TODO This should be enabled when namespaces actually work + //case usernsMode.IsKeepID(): + // user = fmt.Sprintf("%d:%d", rootless.GetRootlessUID(), rootless.GetRootlessGID()) + case newImage.Config == nil || (newImage.Config != nil && len(newImage.Config.User) == 0): + s.User = "0" + default: + s.User = newImage.Config.User + } + } + if err := finishThrottleDevices(s); err != nil { + return err + } + // Unless already set via the CLI, check if we need to disable process + // labels or set the defaults. + if len(s.SelinuxOpts) == 0 { + if err := SetLabelOpts(s, r, s.PidNS, s.IpcNS); err != nil { + return err + } + } + + return nil +} + +// finishThrottleDevices takes the temporary representation of the throttle +// devices in the specgen and looks up the major and major minors. it then +// sets the throttle devices proper in the specgen +func finishThrottleDevices(s *specgen.SpecGenerator) error { + if bps := s.ThrottleReadBpsDevice; len(bps) > 0 { + for k, v := range bps { + statT := unix.Stat_t{} + if err := unix.Stat(k, &statT); err != nil { + return err + } + v.Major = (int64(unix.Major(statT.Rdev))) + v.Minor = (int64(unix.Minor(statT.Rdev))) + s.ResourceLimits.BlockIO.ThrottleReadBpsDevice = append(s.ResourceLimits.BlockIO.ThrottleReadBpsDevice, v) + } + } + if bps := s.ThrottleWriteBpsDevice; len(bps) > 0 { + for k, v := range bps { + statT := unix.Stat_t{} + if err := unix.Stat(k, &statT); err != nil { + return err + } + v.Major = (int64(unix.Major(statT.Rdev))) + v.Minor = (int64(unix.Minor(statT.Rdev))) + s.ResourceLimits.BlockIO.ThrottleWriteBpsDevice = append(s.ResourceLimits.BlockIO.ThrottleWriteBpsDevice, v) + } + } + if iops := s.ThrottleReadIOPSDevice; len(iops) > 0 { + for k, v := range iops { + statT := unix.Stat_t{} + if err := unix.Stat(k, &statT); err != nil { + return err + } + v.Major = (int64(unix.Major(statT.Rdev))) + v.Minor = (int64(unix.Minor(statT.Rdev))) + s.ResourceLimits.BlockIO.ThrottleReadIOPSDevice = append(s.ResourceLimits.BlockIO.ThrottleReadIOPSDevice, v) + } + } + if iops := s.ThrottleWriteBpsDevice; len(iops) > 0 { + for k, v := range iops { + statT := unix.Stat_t{} + if err := unix.Stat(k, &statT); err != nil { + return err + } + v.Major = (int64(unix.Major(statT.Rdev))) + v.Minor = (int64(unix.Minor(statT.Rdev))) + s.ResourceLimits.BlockIO.ThrottleWriteIOPSDevice = append(s.ResourceLimits.BlockIO.ThrottleWriteIOPSDevice, v) + } + } + return nil +} diff --git a/pkg/specgen/container_create.go b/pkg/specgen/generate/container_create.go index b4039bb91..264e0ff8e 100644 --- a/pkg/specgen/container_create.go +++ b/pkg/specgen/generate/container_create.go @@ -1,4 +1,4 @@ -package specgen +package generate import ( "context" @@ -7,14 +7,15 @@ import ( "github.com/containers/common/pkg/config" "github.com/containers/libpod/libpod" "github.com/containers/libpod/libpod/define" + "github.com/containers/libpod/pkg/specgen" "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 { +func MakeContainer(rt *libpod.Runtime, s *specgen.SpecGenerator) (*libpod.Container, error) { + if err := s.Validate(); err != nil { return nil, errors.Wrap(err, "invalid config provided") } rtc, err := rt.GetConfig() @@ -22,7 +23,7 @@ func (s *SpecGenerator) MakeContainer(rt *libpod.Runtime) (*libpod.Container, er return nil, err } - options, err := s.createContainerOptions(rt) + options, err := createContainerOptions(rt, s) if err != nil { return nil, err } @@ -31,7 +32,7 @@ func (s *SpecGenerator) MakeContainer(rt *libpod.Runtime) (*libpod.Container, er if err != nil { return nil, err } - options = append(options, s.createExitCommandOption(rt.StorageConfig(), rtc, podmanPath)) + options = append(options, createExitCommandOption(s, rt.StorageConfig(), rtc, podmanPath)) newImage, err := rt.ImageRuntime().NewFromLocal(s.Image) if err != nil { return nil, err @@ -39,14 +40,14 @@ func (s *SpecGenerator) MakeContainer(rt *libpod.Runtime) (*libpod.Container, er options = append(options, libpod.WithRootFSFromImage(newImage.ID(), s.Image, s.RawImageName)) - runtimeSpec, err := s.toOCISpec(rt, newImage) + runtimeSpec, err := SpecGenToOCI(s, 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) { +func createContainerOptions(rt *libpod.Runtime, s *specgen.SpecGenerator) ([]libpod.CtrCreateOption, error) { var options []libpod.CtrCreateOption var err error @@ -79,7 +80,15 @@ func (s *SpecGenerator) createContainerOptions(rt *libpod.Runtime) ([]libpod.Ctr options = append(options, libpod.WithUserVolumes(destinations)) if len(s.Volumes) != 0 { - options = append(options, libpod.WithNamedVolumes(s.Volumes)) + var volumes []*libpod.ContainerNamedVolume + for _, v := range s.Volumes { + volumes = append(volumes, &libpod.ContainerNamedVolume{ + Name: v.Name, + Dest: v.Dest, + Options: v.Options, + }) + } + options = append(options, libpod.WithNamedVolumes(volumes)) } if len(s.Command) != 0 { @@ -114,7 +123,7 @@ func (s *SpecGenerator) createContainerOptions(rt *libpod.Runtime) ([]libpod.Ctr options = append(options, libpod.WithPrivileged(s.Privileged)) // Get namespace related options - namespaceOptions, err := s.generateNamespaceContainerOpts(rt) + namespaceOptions, err := GenerateNamespaceContainerOpts(s, rt) if err != nil { return nil, err } @@ -149,7 +158,7 @@ func (s *SpecGenerator) createContainerOptions(rt *libpod.Runtime) ([]libpod.Ctr return options, nil } -func (s *SpecGenerator) createExitCommandOption(storageConfig storage.StoreOptions, config *config.Config, podmanPath string) libpod.CtrCreateOption { +func createExitCommandOption(s *specgen.SpecGenerator, 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. diff --git a/pkg/specgen/generate/namespaces.go b/pkg/specgen/generate/namespaces.go new file mode 100644 index 000000000..cdd7d86da --- /dev/null +++ b/pkg/specgen/generate/namespaces.go @@ -0,0 +1,417 @@ +package generate + +import ( + "os" + + "github.com/containers/common/pkg/capabilities" + "github.com/containers/libpod/libpod" + "github.com/containers/libpod/libpod/image" + "github.com/containers/libpod/pkg/specgen" + "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" +) + +func GenerateNamespaceContainerOpts(s *specgen.SpecGenerator, 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 != specgen.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 pidConfigureGenerator(s *specgen.SpecGenerator, 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 utsConfigureGenerator(s *specgen.SpecGenerator, 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 ipcConfigureGenerator(s *specgen.SpecGenerator, 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 cgroupConfigureGenerator(s *specgen.SpecGenerator, 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 networkConfigureGenerator(s *specgen.SpecGenerator, 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 == specgen.NoNetwork: + logrus.Debug("Using none netmode") + case s.NetNS.NSMode == specgen.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 == specgen.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 userConfigureGenerator(s *specgen.SpecGenerator, 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 securityConfigureGenerator(s *specgen.SpecGenerator, 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 := getSeccompConfig(s, 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 +} + +// GetNamespaceOptions transforms a slice of kernel namespaces +// into a slice of pod create options. Currently, not all +// kernel namespaces are supported, and they will be returned in an error +func GetNamespaceOptions(ns []string) ([]libpod.PodCreateOption, error) { + var options []libpod.PodCreateOption + var erroredOptions []libpod.PodCreateOption + for _, toShare := range ns { + switch toShare { + case "cgroup": + options = append(options, libpod.WithPodCgroups()) + case "net": + options = append(options, libpod.WithPodNet()) + case "mnt": + return erroredOptions, errors.Errorf("Mount sharing functionality not supported on pod level") + case "pid": + options = append(options, libpod.WithPodPID()) + case "user": + return erroredOptions, errors.Errorf("User sharing functionality not supported on pod level") + case "ipc": + options = append(options, libpod.WithPodIPC()) + case "uts": + options = append(options, libpod.WithPodUTS()) + case "": + case "none": + return erroredOptions, nil + default: + return erroredOptions, errors.Errorf("Invalid kernel namespace to share: %s. Options are: net, pid, ipc, uts or none", toShare) + } + } + return options, nil +} diff --git a/pkg/specgen/oci.go b/pkg/specgen/generate/oci.go index 2523f21b3..0ed091f9a 100644 --- a/pkg/specgen/oci.go +++ b/pkg/specgen/generate/oci.go @@ -1,4 +1,4 @@ -package specgen +package generate import ( "strings" @@ -6,12 +6,13 @@ import ( "github.com/containers/libpod/libpod" "github.com/containers/libpod/libpod/image" "github.com/containers/libpod/pkg/rootless" - createconfig "github.com/containers/libpod/pkg/spec" + "github.com/containers/libpod/pkg/specgen" + "github.com/opencontainers/runc/libcontainer/user" 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) { +func SpecGenToOCI(s *specgen.SpecGenerator, rt *libpod.Runtime, newImage *image.Image) (*spec.Spec, error) { var ( inUserNS bool ) @@ -72,7 +73,7 @@ func (s *SpecGenerator) toOCISpec(rt *libpod.Runtime, newImage *image.Image) (*s } gid5Available := true if isRootless { - nGids, err := createconfig.GetAvailableGids() + nGids, err := GetAvailableGids() if err != nil { return nil, err } @@ -119,7 +120,7 @@ func (s *SpecGenerator) toOCISpec(rt *libpod.Runtime, newImage *image.Image) (*s g.RemoveMount("/proc") procMount := spec.Mount{ Destination: "/proc", - Type: createconfig.TypeBind, + Type: TypeBind, Source: "/proc", Options: []string{"rbind", "nosuid", "noexec", "nodev"}, } @@ -151,12 +152,12 @@ func (s *SpecGenerator) toOCISpec(rt *libpod.Runtime, newImage *image.Image) (*s // 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 { + if err := addPrivilegedDevices(&g); err != nil { return nil, err } } else { for _, device := range s.Devices { - if err := createconfig.DevicesFromPath(&g, device.Path); err != nil { + if err := DevicesFromPath(&g, device.Path); err != nil { return nil, err } } @@ -169,7 +170,7 @@ func (s *SpecGenerator) toOCISpec(rt *libpod.Runtime, newImage *image.Image) (*s g.SetProcessApparmorProfile(s.ApparmorProfile) } - createconfig.BlockAccessToKernelFilesystems(s.Privileged, s.PidNS.IsHost(), &g) + BlockAccessToKernelFilesystems(s.Privileged, s.PidNS.IsHost(), &g) for name, val := range s.Env { g.AddProcessEnv(name, val) @@ -183,43 +184,41 @@ func (s *SpecGenerator) toOCISpec(rt *libpod.Runtime, newImage *image.Image) (*s // NAMESPACES - if err := s.pidConfigureGenerator(&g); err != nil { + if err := pidConfigureGenerator(s, &g); err != nil { return nil, err } - if err := s.userConfigureGenerator(&g); err != nil { + if err := userConfigureGenerator(s, &g); err != nil { return nil, err } - if err := s.networkConfigureGenerator(&g); err != nil { + if err := networkConfigureGenerator(s, &g); err != nil { return nil, err } - if err := s.utsConfigureGenerator(&g, rt); err != nil { + if err := utsConfigureGenerator(s, &g, rt); err != nil { return nil, err } - if err := s.ipcConfigureGenerator(&g); err != nil { + if err := ipcConfigureGenerator(s, &g); err != nil { return nil, err } - if err := s.cgroupConfigureGenerator(&g); err != nil { + if err := cgroupConfigureGenerator(s, &g); err != nil { return nil, err } configSpec := g.Config - if err := s.securityConfigureGenerator(&g, newImage); err != nil { + if err := securityConfigureGenerator(s, &g, newImage); err != nil { return nil, err } // BIND MOUNTS - configSpec.Mounts = createconfig.SupercedeUserMounts(s.Mounts, configSpec.Mounts) + configSpec.Mounts = SupercedeUserMounts(s.Mounts, configSpec.Mounts) // Process mounts to ensure correct options - finalMounts, err := createconfig.InitFSMounts(configSpec.Mounts) - if err != nil { + if err := InitFSMounts(configSpec.Mounts); err != nil { return nil, err } - configSpec.Mounts = finalMounts // Add annotations if configSpec.Annotations == nil { @@ -258,3 +257,15 @@ func (s *SpecGenerator) toOCISpec(rt *libpod.Runtime, newImage *image.Image) (*s return configSpec, nil } + +func GetAvailableGids() (int64, error) { + idMap, err := user.ParseIDMapFile("/proc/self/gid_map") + if err != nil { + return 0, err + } + count := int64(0) + for _, r := range idMap { + count += r.Count + } + return count, nil +} diff --git a/pkg/specgen/pod_create.go b/pkg/specgen/generate/pod_create.go index 06aa24e22..292f9b155 100644 --- a/pkg/specgen/pod_create.go +++ b/pkg/specgen/generate/pod_create.go @@ -1,31 +1,31 @@ -package specgen +package generate import ( "context" - "github.com/containers/libpod/cmd/podman/shared" "github.com/containers/libpod/libpod" + "github.com/containers/libpod/pkg/specgen" "github.com/sirupsen/logrus" ) -func (p *PodSpecGenerator) MakePod(rt *libpod.Runtime) (*libpod.Pod, error) { - if err := p.validate(); err != nil { +func MakePod(p *specgen.PodSpecGenerator, rt *libpod.Runtime) (*libpod.Pod, error) { + if err := p.Validate(); err != nil { return nil, err } - options, err := p.createPodOptions() + options, err := createPodOptions(p) if err != nil { return nil, err } return rt.NewPod(context.Background(), options...) } -func (p *PodSpecGenerator) createPodOptions() ([]libpod.PodCreateOption, error) { +func createPodOptions(p *specgen.PodSpecGenerator) ([]libpod.PodCreateOption, error) { var ( options []libpod.PodCreateOption ) if !p.NoInfra { options = append(options, libpod.WithInfraContainer()) - nsOptions, err := shared.GetNamespaceOptions(p.SharedNamespaces) + nsOptions, err := GetNamespaceOptions(p.SharedNamespaces) if err != nil { return nil, err } @@ -62,9 +62,9 @@ func (p *PodSpecGenerator) createPodOptions() ([]libpod.PodCreateOption, error) options = append(options, libpod.WithPodUseImageResolvConf()) } switch p.NetNS.NSMode { - case Bridge: + case specgen.Bridge: logrus.Debugf("Pod using default network mode") - case Host: + case specgen.Host: logrus.Debugf("Pod will use host networking") options = append(options, libpod.WithPodHostNetwork()) default: diff --git a/pkg/specgen/generate/security.go b/pkg/specgen/generate/security.go new file mode 100644 index 000000000..ef4b3b47a --- /dev/null +++ b/pkg/specgen/generate/security.go @@ -0,0 +1,154 @@ +package generate + +import ( + "github.com/containers/libpod/libpod" + "github.com/containers/libpod/pkg/specgen" + "github.com/opencontainers/selinux/go-selinux/label" + "github.com/pkg/errors" +) + +// SetLabelOpts sets the label options of the SecurityConfig according to the +// input. +func SetLabelOpts(s *specgen.SpecGenerator, runtime *libpod.Runtime, pidConfig specgen.Namespace, ipcConfig specgen.Namespace) error { + if !runtime.EnableLabeling() || s.Privileged { + s.SelinuxOpts = label.DisableSecOpt() + return nil + } + + var labelOpts []string + if pidConfig.IsHost() { + labelOpts = append(labelOpts, label.DisableSecOpt()...) + } else if pidConfig.IsContainer() { + ctr, err := runtime.LookupContainer(pidConfig.Value) + if err != nil { + return errors.Wrapf(err, "container %q not found", pidConfig.Value) + } + secopts, err := label.DupSecOpt(ctr.ProcessLabel()) + if err != nil { + return errors.Wrapf(err, "failed to duplicate label %q ", ctr.ProcessLabel()) + } + labelOpts = append(labelOpts, secopts...) + } + + if ipcConfig.IsHost() { + labelOpts = append(labelOpts, label.DisableSecOpt()...) + } else if ipcConfig.IsContainer() { + ctr, err := runtime.LookupContainer(ipcConfig.Value) + if err != nil { + return errors.Wrapf(err, "container %q not found", ipcConfig.Value) + } + secopts, err := label.DupSecOpt(ctr.ProcessLabel()) + if err != nil { + return errors.Wrapf(err, "failed to duplicate label %q ", ctr.ProcessLabel()) + } + labelOpts = append(labelOpts, secopts...) + } + + s.SelinuxOpts = append(s.SelinuxOpts, labelOpts...) + return nil +} + +// ConfigureGenerator configures the generator according to the input. +/* +func (c *SecurityConfig) ConfigureGenerator(g *generate.Generator, user *UserConfig) error { + // HANDLE CAPABILITIES + // NOTE: Must happen before SECCOMP + if c.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 defaultCaplist []string + bounding := configSpec.Process.Capabilities.Bounding + if useNotRoot(user.User) { + configSpec.Process.Capabilities.Bounding = defaultCaplist + } + defaultCaplist, err = capabilities.MergeCapabilities(configSpec.Process.Capabilities.Bounding, c.CapAdd, c.CapDrop) + if err != nil { + return err + } + + 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) { + defaultCaplist, err = capabilities.MergeCapabilities(bounding, c.CapAdd, c.CapDrop) + if err != nil { + return err + } + } + configSpec.Process.Capabilities.Bounding = defaultCaplist + + // HANDLE SECCOMP + if c.SeccompProfilePath != "unconfined" { + seccompConfig, err := getSeccompConfig(c, configSpec) + if err != nil { + return err + } + configSpec.Linux.Seccomp = seccompConfig + } + + // Clear default Seccomp profile from Generator for privileged containers + if c.SeccompProfilePath == "unconfined" || c.Privileged { + configSpec.Linux.Seccomp = nil + } + + for _, opt := range c.SecurityOpts { + // Split on both : and = + splitOpt := strings.Split(opt, "=") + if len(splitOpt) == 1 { + splitOpt = strings.Split(opt, ":") + } + if len(splitOpt) < 2 { + continue + } + switch splitOpt[0] { + case "label": + configSpec.Annotations[libpod.InspectAnnotationLabel] = splitOpt[1] + case "seccomp": + configSpec.Annotations[libpod.InspectAnnotationSeccomp] = splitOpt[1] + case "apparmor": + configSpec.Annotations[libpod.InspectAnnotationApparmor] = splitOpt[1] + } + } + + g.SetRootReadonly(c.ReadOnlyRootfs) + for sysctlKey, sysctlVal := range c.Sysctl { + g.AddLinuxSysctl(sysctlKey, sysctlVal) + } + + return nil +} + +*/ diff --git a/pkg/specgen/generate/storage.go b/pkg/specgen/generate/storage.go new file mode 100644 index 000000000..c9a36ed46 --- /dev/null +++ b/pkg/specgen/generate/storage.go @@ -0,0 +1,885 @@ +package generate + +//nolint + +import ( + "fmt" + "path" + "path/filepath" + "strings" + + "github.com/containers/buildah/pkg/parse" + "github.com/containers/libpod/libpod" + "github.com/containers/libpod/pkg/specgen" + "github.com/containers/libpod/pkg/util" + spec "github.com/opencontainers/runtime-spec/specs-go" + "github.com/pkg/errors" + "github.com/sirupsen/logrus" +) + +const ( + // TypeBind is the type for mounting host dir + TypeBind = "bind" + // TypeVolume is the type for named volumes + TypeVolume = "volume" + // TypeTmpfs is the type for mounting tmpfs + TypeTmpfs = "tmpfs" +) + +var ( + errDuplicateDest = errors.Errorf("duplicate mount destination") //nolint + optionArgError = errors.Errorf("must provide an argument for option") //nolint + noDestError = errors.Errorf("must set volume destination") //nolint +) + +// Parse all volume-related options in the create config into a set of mounts +// and named volumes to add to the container. +// Handles --volumes-from, --volumes, --tmpfs, --init, and --init-path flags. +// TODO: Named volume options - should we default to rprivate? It bakes into a +// bind mount under the hood... +// TODO: handle options parsing/processing via containers/storage/pkg/mount +func parseVolumes(s *specgen.SpecGenerator, mounts, volMounts, tmpMounts []string) error { //nolint + + // TODO this needs to come from the image and erquires a runtime + + // Add image volumes. + //baseMounts, baseVolumes, err := config.getImageVolumes() + //if err != nil { + // return nil, nil, err + //} + + // Add --volumes-from. + // Overrides image volumes unconditionally. + //vFromMounts, vFromVolumes, err := config.getVolumesFrom(runtime) + //if err != nil { + // return nil, nil, err + //} + //for dest, mount := range vFromMounts { + // baseMounts[dest] = mount + //} + //for dest, volume := range vFromVolumes { + // baseVolumes[dest] = volume + //} + + // Next mounts from the --mounts flag. + // Do not override yet. + //unifiedMounts, _, err := getMounts(mounts) + //if err != nil { + // return err + //} + // + //// Next --volumes flag. + //// Do not override yet. + //volumeMounts, _ , err := getVolumeMounts(volMounts) + //if err != nil { + // return err + //} + // + //// Next --tmpfs flag. + //// Do not override yet. + //tmpfsMounts, err := getTmpfsMounts(tmpMounts) + //if err != nil { + // return err + //} + + //// Unify mounts from --mount, --volume, --tmpfs. + //// Also add mounts + volumes directly from createconfig. + //// Start with --volume. + //for dest, mount := range volumeMounts { + // if _, ok := unifiedMounts[dest]; ok { + // return nil, nil, errors.Wrapf(errDuplicateDest, dest) + // } + // unifiedMounts[dest] = mount + //} + //for dest, volume := range volumeVolumes { + // if _, ok := unifiedVolumes[dest]; ok { + // return nil, nil, errors.Wrapf(errDuplicateDest, dest) + // } + // unifiedVolumes[dest] = volume + //} + //// Now --tmpfs + //for dest, tmpfs := range tmpfsMounts { + // if _, ok := unifiedMounts[dest]; ok { + // return nil, nil, errors.Wrapf(errDuplicateDest, dest) + // } + // unifiedMounts[dest] = tmpfs + //} + //// Now spec mounts and volumes + //for _, mount := range config.Mounts { + // dest := mount.Destination + // if _, ok := unifiedMounts[dest]; ok { + // return nil, nil, errors.Wrapf(errDuplicateDest, dest) + // } + // unifiedMounts[dest] = mount + //} + //for _, volume := range config.NamedVolumes { + // dest := volume.Dest + // if _, ok := unifiedVolumes[dest]; ok { + // return nil, nil, errors.Wrapf(errDuplicateDest, dest) + // } + // unifiedVolumes[dest] = volume + //} + // + //// If requested, add container init binary + //if config.Init { + // initPath := config.InitPath + // if initPath == "" { + // rtc, err := runtime.GetConfig() + // if err != nil { + // return nil, nil, err + // } + // initPath = rtc.Engine.InitPath + // } + // initMount, err := config.addContainerInitBinary(initPath) + // if err != nil { + // return nil, nil, err + // } + // if _, ok := unifiedMounts[initMount.Destination]; ok { + // return nil, nil, errors.Wrapf(errDuplicateDest, "conflict with mount added by --init to %q", initMount.Destination) + // } + // unifiedMounts[initMount.Destination] = initMount + //} + // + //// Before superseding, we need to find volume mounts which conflict with + //// named volumes, and vice versa. + //// We'll delete the conflicts here as we supersede. + //for dest := range unifiedMounts { + // if _, ok := baseVolumes[dest]; ok { + // delete(baseVolumes, dest) + // } + //} + //for dest := range unifiedVolumes { + // if _, ok := baseMounts[dest]; ok { + // delete(baseMounts, dest) + // } + //} + // + //// Supersede volumes-from/image volumes with unified volumes from above. + //// This is an unconditional replacement. + //for dest, mount := range unifiedMounts { + // baseMounts[dest] = mount + //} + //for dest, volume := range unifiedVolumes { + // baseVolumes[dest] = volume + //} + // + //// If requested, add tmpfs filesystems for read-only containers. + //if config.Security.ReadOnlyRootfs && config.Security.ReadOnlyTmpfs { + // readonlyTmpfs := []string{"/tmp", "/var/tmp", "/run"} + // options := []string{"rw", "rprivate", "nosuid", "nodev", "tmpcopyup"} + // for _, dest := range readonlyTmpfs { + // if _, ok := baseMounts[dest]; ok { + // continue + // } + // if _, ok := baseVolumes[dest]; ok { + // continue + // } + // localOpts := options + // if dest == "/run" { + // localOpts = append(localOpts, "noexec", "size=65536k") + // } else { + // localOpts = append(localOpts, "exec") + // } + // baseMounts[dest] = spec.Mount{ + // Destination: dest, + // Type: "tmpfs", + // Source: "tmpfs", + // Options: localOpts, + // } + // } + //} + // + //// Check for conflicts between named volumes and mounts + //for dest := range baseMounts { + // if _, ok := baseVolumes[dest]; ok { + // return nil, nil, errors.Wrapf(errDuplicateDest, "conflict at mount destination %v", dest) + // } + //} + //for dest := range baseVolumes { + // if _, ok := baseMounts[dest]; ok { + // return nil, nil, errors.Wrapf(errDuplicateDest, "conflict at mount destination %v", dest) + // } + //} + // + //// Final step: maps to arrays + //finalMounts := make([]spec.Mount, 0, len(baseMounts)) + //for _, mount := range baseMounts { + // if mount.Type == TypeBind { + // absSrc, err := filepath.Abs(mount.Source) + // if err != nil { + // return nil, nil, errors.Wrapf(err, "error getting absolute path of %s", mount.Source) + // } + // mount.Source = absSrc + // } + // finalMounts = append(finalMounts, mount) + //} + //finalVolumes := make([]*define.ContainerNamedVolume, 0, len(baseVolumes)) + //for _, volume := range baseVolumes { + // finalVolumes = append(finalVolumes, volume) + //} + + //return finalMounts, finalVolumes, nil + return nil +} + +// Parse volumes from - a set of containers whose volumes we will mount in. +// Grab the containers, retrieve any user-created spec mounts and all named +// volumes, and return a list of them. +// Conflicts are resolved simply - the last container specified wins. +// Container names may be suffixed by mount options after a colon. +// TODO: We should clean these paths if possible +// TODO deferred baude +func getVolumesFrom() (map[string]spec.Mount, map[string]*libpod.ContainerNamedVolume, error) { //nolint + // Both of these are maps of mount destination to mount type. + // We ensure that each destination is only mounted to once in this way. + //finalMounts := make(map[string]spec.Mount) + //finalNamedVolumes := make(map[string]*define.ContainerNamedVolume) + // + //for _, vol := range config.VolumesFrom { + // var ( + // options = []string{} + // err error + // splitVol = strings.SplitN(vol, ":", 2) + // ) + // if len(splitVol) == 2 { + // splitOpts := strings.Split(splitVol[1], ",") + // for _, checkOpt := range splitOpts { + // switch checkOpt { + // case "z", "ro", "rw": + // // Do nothing, these are valid options + // default: + // return nil, nil, errors.Errorf("invalid options %q, can only specify 'ro', 'rw', and 'z'", splitVol[1]) + // } + // } + // + // if options, err = parse.ValidateVolumeOpts(splitOpts); err != nil { + // return nil, nil, err + // } + // } + // ctr, err := runtime.LookupContainer(splitVol[0]) + // if err != nil { + // return nil, nil, errors.Wrapf(err, "error looking up container %q for volumes-from", splitVol[0]) + // } + // + // logrus.Debugf("Adding volumes from container %s", ctr.ID()) + // + // // Look up the container's user volumes. This gets us the + // // destinations of all mounts the user added to the container. + // userVolumesArr := ctr.UserVolumes() + // + // // We're going to need to access them a lot, so convert to a map + // // to reduce looping. + // // We'll also use the map to indicate if we missed any volumes along the way. + // userVolumes := make(map[string]bool) + // for _, dest := range userVolumesArr { + // userVolumes[dest] = false + // } + // + // // Now we get the container's spec and loop through its volumes + // // and append them in if we can find them. + // spec := ctr.Spec() + // if spec == nil { + // return nil, nil, errors.Errorf("error retrieving container %s spec for volumes-from", ctr.ID()) + // } + // for _, mnt := range spec.Mounts { + // if mnt.Type != TypeBind { + // continue + // } + // if _, exists := userVolumes[mnt.Destination]; exists { + // userVolumes[mnt.Destination] = true + // + // if len(options) != 0 { + // mnt.Options = options + // } + // + // if _, ok := finalMounts[mnt.Destination]; ok { + // logrus.Debugf("Overriding mount to %s with new mount from container %s", mnt.Destination, ctr.ID()) + // } + // finalMounts[mnt.Destination] = mnt + // } + // } + // + // // We're done with the spec mounts. Add named volumes. + // // Add these unconditionally - none of them are automatically + // // part of the container, as some spec mounts are. + // namedVolumes := ctr.NamedVolumes() + // for _, namedVol := range namedVolumes { + // if _, exists := userVolumes[namedVol.Dest]; exists { + // userVolumes[namedVol.Dest] = true + // } + // + // if len(options) != 0 { + // namedVol.Options = options + // } + // + // if _, ok := finalMounts[namedVol.Dest]; ok { + // logrus.Debugf("Overriding named volume mount to %s with new named volume from container %s", namedVol.Dest, ctr.ID()) + // } + // finalNamedVolumes[namedVol.Dest] = namedVol + // } + // + // // Check if we missed any volumes + // for volDest, found := range userVolumes { + // if !found { + // logrus.Warnf("Unable to match volume %s from container %s for volumes-from", volDest, ctr.ID()) + // } + // } + //} + // + //return finalMounts, finalNamedVolumes, nil + return nil, nil, nil +} + +// getMounts takes user-provided input from the --mount flag and creates OCI +// spec mounts and Libpod named volumes. +// podman run --mount type=bind,src=/etc/resolv.conf,target=/etc/resolv.conf ... +// podman run --mount type=tmpfs,target=/dev/shm ... +// podman run --mount type=volume,source=test-volume, ... +func getMounts(mounts []string) (map[string]spec.Mount, map[string]*libpod.ContainerNamedVolume, error) { //nolint + finalMounts := make(map[string]spec.Mount) + finalNamedVolumes := make(map[string]*libpod.ContainerNamedVolume) + + errInvalidSyntax := errors.Errorf("incorrect mount format: should be --mount type=<bind|tmpfs|volume>,[src=<host-dir|volume-name>,]target=<ctr-dir>[,options]") + + // TODO(vrothberg): the manual parsing can be replaced with a regular expression + // to allow a more robust parsing of the mount format and to give + // precise errors regarding supported format versus supported options. + for _, mount := range mounts { + arr := strings.SplitN(mount, ",", 2) + if len(arr) < 2 { + return nil, nil, errors.Wrapf(errInvalidSyntax, "%q", mount) + } + kv := strings.Split(arr[0], "=") + // TODO: type is not explicitly required in Docker. + // If not specified, it defaults to "volume". + if len(kv) != 2 || kv[0] != "type" { + return nil, nil, errors.Wrapf(errInvalidSyntax, "%q", mount) + } + + tokens := strings.Split(arr[1], ",") + switch kv[1] { + case TypeBind: + mount, err := getBindMount(tokens) + if err != nil { + return nil, nil, err + } + if _, ok := finalMounts[mount.Destination]; ok { + return nil, nil, errors.Wrapf(errDuplicateDest, mount.Destination) + } + finalMounts[mount.Destination] = mount + case TypeTmpfs: + mount, err := getTmpfsMount(tokens) + if err != nil { + return nil, nil, err + } + if _, ok := finalMounts[mount.Destination]; ok { + return nil, nil, errors.Wrapf(errDuplicateDest, mount.Destination) + } + finalMounts[mount.Destination] = mount + case "volume": + volume, err := getNamedVolume(tokens) + if err != nil { + return nil, nil, err + } + if _, ok := finalNamedVolumes[volume.Dest]; ok { + return nil, nil, errors.Wrapf(errDuplicateDest, volume.Dest) + } + finalNamedVolumes[volume.Dest] = volume + default: + return nil, nil, errors.Errorf("invalid filesystem type %q", kv[1]) + } + } + + return finalMounts, finalNamedVolumes, nil +} + +// Parse a single bind mount entry from the --mount flag. +func getBindMount(args []string) (spec.Mount, error) { //nolint + newMount := spec.Mount{ + Type: TypeBind, + } + + var setSource, setDest, setRORW, setSuid, setDev, setExec, setRelabel bool + + for _, val := range args { + kv := strings.Split(val, "=") + switch kv[0] { + case "bind-nonrecursive": + newMount.Options = append(newMount.Options, "bind") + case "ro", "rw": + if setRORW { + return newMount, errors.Wrapf(optionArgError, "cannot pass 'ro' or 'rw' options more than once") + } + setRORW = true + // Can be formatted as one of: + // ro + // ro=[true|false] + // rw + // rw=[true|false] + switch len(kv) { + case 1: + newMount.Options = append(newMount.Options, kv[0]) + case 2: + switch strings.ToLower(kv[1]) { + case "true": + newMount.Options = append(newMount.Options, kv[0]) + case "false": + // Set the opposite only for rw + // ro's opposite is the default + if kv[0] == "rw" { + newMount.Options = append(newMount.Options, "ro") + } + default: + return newMount, errors.Wrapf(optionArgError, "%s must be set to true or false, instead received %q", kv[0], kv[1]) + } + default: + return newMount, errors.Wrapf(optionArgError, "badly formatted option %q", val) + } + case "nosuid", "suid": + if setSuid { + return newMount, errors.Wrapf(optionArgError, "cannot pass 'nosuid' and 'suid' options more than once") + } + setSuid = true + newMount.Options = append(newMount.Options, kv[0]) + case "nodev", "dev": + if setDev { + return newMount, errors.Wrapf(optionArgError, "cannot pass 'nodev' and 'dev' options more than once") + } + setDev = true + newMount.Options = append(newMount.Options, kv[0]) + case "noexec", "exec": + if setExec { + return newMount, errors.Wrapf(optionArgError, "cannot pass 'noexec' and 'exec' options more than once") + } + setExec = true + newMount.Options = append(newMount.Options, kv[0]) + case "shared", "rshared", "private", "rprivate", "slave", "rslave", "Z", "z": + newMount.Options = append(newMount.Options, kv[0]) + case "bind-propagation": + if len(kv) == 1 { + return newMount, errors.Wrapf(optionArgError, kv[0]) + } + newMount.Options = append(newMount.Options, kv[1]) + case "src", "source": + if len(kv) == 1 { + return newMount, errors.Wrapf(optionArgError, kv[0]) + } + if err := parse.ValidateVolumeHostDir(kv[1]); err != nil { + return newMount, err + } + newMount.Source = kv[1] + setSource = true + case "target", "dst", "destination": + if len(kv) == 1 { + return newMount, errors.Wrapf(optionArgError, kv[0]) + } + if err := parse.ValidateVolumeCtrDir(kv[1]); err != nil { + return newMount, err + } + newMount.Destination = filepath.Clean(kv[1]) + setDest = true + case "relabel": + if setRelabel { + return newMount, errors.Wrapf(optionArgError, "cannot pass 'relabel' option more than once") + } + setRelabel = true + if len(kv) != 2 { + return newMount, errors.Wrapf(util.ErrBadMntOption, "%s mount option must be 'private' or 'shared'", kv[0]) + } + switch kv[1] { + case "private": + newMount.Options = append(newMount.Options, "z") + case "shared": + newMount.Options = append(newMount.Options, "Z") + default: + return newMount, errors.Wrapf(util.ErrBadMntOption, "%s mount option must be 'private' or 'shared'", kv[0]) + } + default: + return newMount, errors.Wrapf(util.ErrBadMntOption, kv[0]) + } + } + + if !setDest { + return newMount, noDestError + } + + if !setSource { + newMount.Source = newMount.Destination + } + + options, err := parse.ValidateVolumeOpts(newMount.Options) + if err != nil { + return newMount, err + } + newMount.Options = options + return newMount, nil +} + +// Parse a single tmpfs mount entry from the --mount flag +func getTmpfsMount(args []string) (spec.Mount, error) { //nolint + newMount := spec.Mount{ + Type: TypeTmpfs, + Source: TypeTmpfs, + } + + var setDest, setRORW, setSuid, setDev, setExec, setTmpcopyup bool + + for _, val := range args { + kv := strings.Split(val, "=") + switch kv[0] { + case "tmpcopyup", "notmpcopyup": + if setTmpcopyup { + return newMount, errors.Wrapf(optionArgError, "cannot pass 'tmpcopyup' and 'notmpcopyup' options more than once") + } + setTmpcopyup = true + newMount.Options = append(newMount.Options, kv[0]) + case "ro", "rw": + if setRORW { + return newMount, errors.Wrapf(optionArgError, "cannot pass 'ro' and 'rw' options more than once") + } + setRORW = true + newMount.Options = append(newMount.Options, kv[0]) + case "nosuid", "suid": + if setSuid { + return newMount, errors.Wrapf(optionArgError, "cannot pass 'nosuid' and 'suid' options more than once") + } + setSuid = true + newMount.Options = append(newMount.Options, kv[0]) + case "nodev", "dev": + if setDev { + return newMount, errors.Wrapf(optionArgError, "cannot pass 'nodev' and 'dev' options more than once") + } + setDev = true + newMount.Options = append(newMount.Options, kv[0]) + case "noexec", "exec": + if setExec { + return newMount, errors.Wrapf(optionArgError, "cannot pass 'noexec' and 'exec' options more than once") + } + setExec = true + newMount.Options = append(newMount.Options, kv[0]) + case "tmpfs-mode": + if len(kv) == 1 { + return newMount, errors.Wrapf(optionArgError, kv[0]) + } + newMount.Options = append(newMount.Options, fmt.Sprintf("mode=%s", kv[1])) + case "tmpfs-size": + if len(kv) == 1 { + return newMount, errors.Wrapf(optionArgError, kv[0]) + } + newMount.Options = append(newMount.Options, fmt.Sprintf("size=%s", kv[1])) + case "src", "source": + return newMount, errors.Errorf("source is not supported with tmpfs mounts") + case "target", "dst", "destination": + if len(kv) == 1 { + return newMount, errors.Wrapf(optionArgError, kv[0]) + } + if err := parse.ValidateVolumeCtrDir(kv[1]); err != nil { + return newMount, err + } + newMount.Destination = filepath.Clean(kv[1]) + setDest = true + default: + return newMount, errors.Wrapf(util.ErrBadMntOption, kv[0]) + } + } + + if !setDest { + return newMount, noDestError + } + + return newMount, nil +} + +// Parse a single volume mount entry from the --mount flag. +// Note that the volume-label option for named volumes is currently NOT supported. +// TODO: add support for --volume-label +func getNamedVolume(args []string) (*libpod.ContainerNamedVolume, error) { //nolint + newVolume := new(libpod.ContainerNamedVolume) + + var setSource, setDest, setRORW, setSuid, setDev, setExec bool + + for _, val := range args { + kv := strings.Split(val, "=") + switch kv[0] { + case "ro", "rw": + if setRORW { + return nil, errors.Wrapf(optionArgError, "cannot pass 'ro' and 'rw' options more than once") + } + setRORW = true + newVolume.Options = append(newVolume.Options, kv[0]) + case "nosuid", "suid": + if setSuid { + return nil, errors.Wrapf(optionArgError, "cannot pass 'nosuid' and 'suid' options more than once") + } + setSuid = true + newVolume.Options = append(newVolume.Options, kv[0]) + case "nodev", "dev": + if setDev { + return nil, errors.Wrapf(optionArgError, "cannot pass 'nodev' and 'dev' options more than once") + } + setDev = true + newVolume.Options = append(newVolume.Options, kv[0]) + case "noexec", "exec": + if setExec { + return nil, errors.Wrapf(optionArgError, "cannot pass 'noexec' and 'exec' options more than once") + } + setExec = true + newVolume.Options = append(newVolume.Options, kv[0]) + case "volume-label": + return nil, errors.Errorf("the --volume-label option is not presently implemented") + case "src", "source": + if len(kv) == 1 { + return nil, errors.Wrapf(optionArgError, kv[0]) + } + newVolume.Name = kv[1] + setSource = true + case "target", "dst", "destination": + if len(kv) == 1 { + return nil, errors.Wrapf(optionArgError, kv[0]) + } + if err := parse.ValidateVolumeCtrDir(kv[1]); err != nil { + return nil, err + } + newVolume.Dest = filepath.Clean(kv[1]) + setDest = true + default: + return nil, errors.Wrapf(util.ErrBadMntOption, kv[0]) + } + } + + if !setSource { + return nil, errors.Errorf("must set source volume") + } + if !setDest { + return nil, noDestError + } + + return newVolume, nil +} + +func getVolumeMounts(vols []string) (map[string]spec.Mount, map[string]*libpod.ContainerNamedVolume, error) { //nolint + mounts := make(map[string]spec.Mount) + volumes := make(map[string]*libpod.ContainerNamedVolume) + + volumeFormatErr := errors.Errorf("incorrect volume format, should be [host-dir:]ctr-dir[:option]") + + for _, vol := range vols { + var ( + options []string + src string + dest string + err error + ) + + splitVol := strings.Split(vol, ":") + if len(splitVol) > 3 { + return nil, nil, errors.Wrapf(volumeFormatErr, vol) + } + + src = splitVol[0] + if len(splitVol) == 1 { + // This is an anonymous named volume. Only thing given + // is destination. + // Name/source will be blank, and populated by libpod. + src = "" + dest = splitVol[0] + } else if len(splitVol) > 1 { + dest = splitVol[1] + } + if len(splitVol) > 2 { + if options, err = parse.ValidateVolumeOpts(strings.Split(splitVol[2], ",")); err != nil { + return nil, nil, err + } + } + + // Do not check source dir for anonymous volumes + if len(splitVol) > 1 { + if err := parse.ValidateVolumeHostDir(src); err != nil { + return nil, nil, err + } + } + if err := parse.ValidateVolumeCtrDir(dest); err != nil { + return nil, nil, err + } + + cleanDest := filepath.Clean(dest) + + if strings.HasPrefix(src, "/") || strings.HasPrefix(src, ".") { + // This is not a named volume + newMount := spec.Mount{ + Destination: cleanDest, + Type: string(TypeBind), + Source: src, + Options: options, + } + if _, ok := mounts[newMount.Destination]; ok { + return nil, nil, errors.Wrapf(errDuplicateDest, newMount.Destination) + } + mounts[newMount.Destination] = newMount + } else { + // This is a named volume + newNamedVol := new(libpod.ContainerNamedVolume) + newNamedVol.Name = src + newNamedVol.Dest = cleanDest + newNamedVol.Options = options + + if _, ok := volumes[newNamedVol.Dest]; ok { + return nil, nil, errors.Wrapf(errDuplicateDest, newNamedVol.Dest) + } + volumes[newNamedVol.Dest] = newNamedVol + } + + logrus.Debugf("User mount %s:%s options %v", src, dest, options) + } + + return mounts, volumes, nil +} + +// Get mounts for container's image volumes +// TODO deferred baude +func getImageVolumes() (map[string]spec.Mount, map[string]*libpod.ContainerNamedVolume, error) { //nolint + //mounts := make(map[string]spec.Mount) + //volumes := make(map[string]*define.ContainerNamedVolume) + // + //if config.ImageVolumeType == "ignore" { + // return mounts, volumes, nil + //} + // + //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{ + // Destination: cleanDest, + // Source: TypeTmpfs, + // Type: TypeTmpfs, + // Options: []string{"rprivate", "rw", "nodev", "exec"}, + // } + // mounts[cleanDest] = mount + // } else { + // // Anonymous volumes have no name. + // namedVolume := new(define.ContainerNamedVolume) + // namedVolume.Options = []string{"rprivate", "rw", "nodev", "exec"} + // namedVolume.Dest = cleanDest + // volumes[cleanDest] = namedVolume + // } + //} + // + //return mounts, volumes, nil + return nil, nil, nil +} + +// GetTmpfsMounts creates spec.Mount structs for user-requested tmpfs mounts +func getTmpfsMounts(mounts []string) (map[string]spec.Mount, error) { //nolint + m := make(map[string]spec.Mount) + for _, i := range mounts { + // Default options if nothing passed + var options []string + spliti := strings.Split(i, ":") + destPath := spliti[0] + if err := parse.ValidateVolumeCtrDir(spliti[0]); err != nil { + return nil, err + } + if len(spliti) > 1 { + options = strings.Split(spliti[1], ",") + } + + if _, ok := m[destPath]; ok { + return nil, errors.Wrapf(errDuplicateDest, destPath) + } + + mount := spec.Mount{ + Destination: filepath.Clean(destPath), + Type: string(TypeTmpfs), + Options: options, + Source: string(TypeTmpfs), + } + m[destPath] = mount + } + return m, nil +} + +// AddContainerInitBinary adds the init binary specified by path iff the +// container will run in a private PID namespace that is not shared with the +// host or another pre-existing container, where an init-like process is +// already running. +// +// Note that AddContainerInitBinary prepends "/dev/init" "--" to the command +// to execute the bind-mounted binary as PID 1. +// TODO this needs to be worked on to work in new env +func addContainerInitBinary(path string) (spec.Mount, error) { //nolint + mount := spec.Mount{ + Destination: "/dev/init", + Type: TypeBind, + Source: path, + Options: []string{TypeBind, "ro"}, + } + + //if path == "" { + // return mount, fmt.Errorf("please specify a path to the container-init binary") + //} + //if !config.Pid.PidMode.IsPrivate() { + // return mount, fmt.Errorf("cannot add init binary as PID 1 (PID namespace isn't private)") + //} + //if config.Systemd { + // return mount, fmt.Errorf("cannot use container-init binary with systemd") + //} + //if _, err := os.Stat(path); os.IsNotExist(err) { + // return mount, errors.Wrap(err, "container-init binary not found on the host") + //} + //config.Command = append([]string{"/dev/init", "--"}, config.Command...) + return mount, nil +} + +// Supersede existing mounts in the spec with new, user-specified mounts. +// 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 { + if len(mounts) > 0 { + // If we have overlappings mounts, remove them from the spec in favor of + // the user-added volume mounts + destinations := make(map[string]bool) + for _, mount := range mounts { + destinations[path.Clean(mount.Destination)] = true + } + // Copy all mounts from spec to defaultMounts, except for + // - mounts overridden by a user supplied mount; + // - all mounts under /dev if a user supplied /dev is present; + mountDev := destinations["/dev"] + for _, mount := range configMount { + if _, ok := destinations[path.Clean(mount.Destination)]; !ok { + if mountDev && strings.HasPrefix(mount.Destination, "/dev/") { + // filter out everything under /dev if /dev is user-mounted + continue + } + + logrus.Debugf("Adding mount %s", mount.Destination) + mounts = append(mounts, mount) + } + } + return mounts + } + return configMount +} + +func InitFSMounts(mounts []spec.Mount) error { + for i, m := range mounts { + switch { + case m.Type == TypeBind: + opts, err := util.ProcessOptions(m.Options, false, m.Source) + if err != nil { + return err + } + mounts[i].Options = opts + case m.Type == TypeTmpfs && filepath.Clean(m.Destination) != "/dev": + opts, err := util.ProcessOptions(m.Options, true, "") + if err != nil { + return err + } + mounts[i].Options = opts + } + } + return nil +} diff --git a/pkg/specgen/namespaces.go b/pkg/specgen/namespaces.go index fa2dee77d..2e7f80fe8 100644 --- a/pkg/specgen/namespaces.go +++ b/pkg/specgen/namespaces.go @@ -1,21 +1,15 @@ 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 ( + // Default indicates the spec generator should determine + // a sane default + Default NamespaceMode = "default" // Host means the the namespace is derived from // the host Host NamespaceMode = "host" @@ -34,9 +28,9 @@ const ( // Bridge indicates that a CNI network stack // should be used Bridge NamespaceMode = "bridge" - // Slirp indicates that a slirp4ns network stack should + // Slirp indicates that a slirp4netns network stack should // be used - Slirp NamespaceMode = "slirp4ns" + Slirp NamespaceMode = "slirp4netns" ) // Namespace describes the namespace @@ -83,7 +77,7 @@ func validateNetNS(n *Namespace) error { return nil } -// validate perform simple validation on the namespace to make sure it is not +// 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 { @@ -102,373 +96,3 @@ func (n *Namespace) validate() error { } 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/pod_validate.go b/pkg/specgen/pod_validate.go index 50309f096..9e9659fa9 100644 --- a/pkg/specgen/pod_validate.go +++ b/pkg/specgen/pod_validate.go @@ -15,7 +15,8 @@ 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 { +// Validate verifies the input is valid +func (p *PodSpecGenerator) Validate() error { // PodBasicConfig if p.NoInfra { if len(p.InfraCommand) > 0 { @@ -25,7 +26,7 @@ func (p *PodSpecGenerator) validate() error { return exclusivePodOptions("NoInfra", "InfraImage") } if len(p.SharedNamespaces) > 0 { - return exclusivePodOptions("NoInfo", "SharedNamespaces") + return exclusivePodOptions("NoInfra", "SharedNamespaces") } } diff --git a/pkg/specgen/specgen.go b/pkg/specgen/specgen.go index 89c76c273..1a05733f9 100644 --- a/pkg/specgen/specgen.go +++ b/pkg/specgen/specgen.go @@ -5,7 +5,6 @@ import ( "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" @@ -173,7 +172,7 @@ type ContainerStorageConfig struct { // These will supersede Image Volumes and VolumesFrom volumes where // there are conflicts. // Optional. - Volumes []*libpod.ContainerNamedVolume `json:"volumes,omitempty"` + Volumes []*Volumes `json:"volumes,omitempty"` // Devices are devices that will be added to the container. // Optional. Devices []spec.LinuxDevice `json:"devices,omitempty"` @@ -229,14 +228,6 @@ type ContainerSecurityConfig struct { // 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. @@ -371,6 +362,16 @@ type ContainerResourceConfig struct { // processes to kill for the container's process. // Optional. OOMScoreAdj *int `json:"oom_score_adj,omitempty"` + // Weight per cgroup per device, can override BlkioWeight + WeightDevice map[string]spec.LinuxWeightDevice `json:"weightDevice,omitempty"` + // IO read rate limit per cgroup per device, bytes per second + ThrottleReadBpsDevice map[string]spec.LinuxThrottleDevice `json:"throttleReadBpsDevice,omitempty"` + // IO write rate limit per cgroup per device, bytes per second + ThrottleWriteBpsDevice map[string]spec.LinuxThrottleDevice `json:"throttleWriteBpsDevice,omitempty"` + // IO read rate limit per cgroup per device, IO per second + ThrottleReadIOPSDevice map[string]spec.LinuxThrottleDevice `json:"throttleReadIOPSDevice,omitempty"` + // IO write rate limit per cgroup per device, IO per second + ThrottleWriteIOPSDevice map[string]spec.LinuxThrottleDevice `json:"throttleWriteIOPSDevice,omitempty"` } // ContainerHealthCheckConfig describes a container healthcheck with attributes @@ -392,6 +393,13 @@ type SpecGenerator struct { ContainerHealthCheckConfig } +// Volumes is a temporary struct to hold input from the User +type Volumes struct { + Name string + Dest string + Options []string +} + // NewSpecGenerator returns a SpecGenerator struct given one of two mandatory inputs func NewSpecGenerator(image string) *SpecGenerator { networkConfig := ContainerNetworkConfig{ |