diff options
Diffstat (limited to 'libpod')
-rw-r--r-- | libpod/container_internal_linux.go | 56 | ||||
-rw-r--r-- | libpod/networking_linux.go | 90 | ||||
-rw-r--r-- | libpod/runtime_pod_linux.go | 71 |
3 files changed, 181 insertions, 36 deletions
diff --git a/libpod/container_internal_linux.go b/libpod/container_internal_linux.go index cef9e2c04..1517a7df7 100644 --- a/libpod/container_internal_linux.go +++ b/libpod/container_internal_linux.go @@ -173,6 +173,57 @@ func (c *Container) prepare() error { return nil } +// isWorkDirSymlink returns true if resolved workdir is symlink or a chain of symlinks, +// and final resolved target is present either on volume, mount or inside of container +// otherwise it returns false. Following function is meant for internal use only and +// can change at any point of time. +func (c *Container) isWorkDirSymlink(resolvedPath string) bool { + // We cannot create workdir since explicit --workdir is + // set in config but workdir could also be a symlink. + // If its a symlink lets check if resolved link is present + // on the container or not. + + // If we can resolve symlink and resolved link is present on the container + // then return nil cause its a valid use-case. + + maxSymLinks := 0 + for { + // Linux only supports a chain of 40 links. + // Reference: https://github.com/torvalds/linux/blob/master/include/linux/namei.h#L13 + if maxSymLinks > 40 { + break + } + resolvedSymlink, err := os.Readlink(resolvedPath) + if err != nil { + // End sym-link resolution loop. + break + } + if resolvedSymlink != "" { + _, resolvedSymlinkWorkdir, err := c.resolvePath(c.state.Mountpoint, resolvedSymlink) + if isPathOnVolume(c, resolvedSymlinkWorkdir) || isPathOnBindMount(c, resolvedSymlinkWorkdir) { + // Resolved symlink exists on external volume or mount + return true + } + if err != nil { + // Could not resolve path so end sym-link resolution loop. + break + } + if resolvedSymlinkWorkdir != "" { + resolvedPath = resolvedSymlinkWorkdir + _, err := os.Stat(resolvedSymlinkWorkdir) + if err == nil { + // Symlink resolved successfully and resolved path exists on container, + // this is a valid use-case so return nil. + logrus.Debugf("Workdir is a symlink with target to %q and resolved symlink exists on container", resolvedSymlink) + return true + } + } + } + maxSymLinks++ + } + return false +} + // resolveWorkDir resolves the container's workdir and, depending on the // configuration, will create it, or error out if it does not exist. // Note that the container must be mounted before. @@ -205,6 +256,11 @@ func (c *Container) resolveWorkDir() error { // the path exists on the container. if err != nil { if os.IsNotExist(err) { + // If resolved Workdir path gets marked as a valid symlink, + // return nil cause this is valid use-case. + if c.isWorkDirSymlink(resolvedWorkdir) { + return nil + } return errors.Errorf("workdir %q does not exist on container %s", workdir, c.ID()) } // This might be a serious error (e.g., permission), so diff --git a/libpod/networking_linux.go b/libpod/networking_linux.go index 29b9941fe..7fd80927b 100644 --- a/libpod/networking_linux.go +++ b/libpod/networking_linux.go @@ -30,6 +30,7 @@ import ( "github.com/containers/podman/v4/pkg/util" "github.com/containers/podman/v4/utils" "github.com/containers/storage/pkg/lockfile" + spec "github.com/opencontainers/runtime-spec/specs-go" "github.com/opencontainers/selinux/go-selinux/label" "github.com/pkg/errors" "github.com/sirupsen/logrus" @@ -990,8 +991,20 @@ func (c *Container) getContainerNetworkInfo() (*define.InspectNetworkSettings, e return nil, err } - // We can't do more if the network is down. if c.state.NetNS == nil { + if networkNSPath := c.joinedNetworkNSPath(); networkNSPath != "" { + if result, err := c.inspectJoinedNetworkNS(networkNSPath); err == nil { + if basicConfig, err := resultToBasicNetworkConfig(result); err == nil { + // fallback to dummy configuration + settings.InspectBasicNetworkConfig = basicConfig + return settings, nil + } + } + // do not propagate error inspecting a joined network ns + logrus.Errorf("Error inspecting network namespace: %s of container %s: %v", networkNSPath, c.ID(), err) + } + // We can't do more if the network is down. + // We still want to make dummy configurations for each CNI net // the container joined. if len(networks) > 0 { @@ -1065,11 +1078,84 @@ func (c *Container) getContainerNetworkInfo() (*define.InspectNetworkSettings, e return settings, nil } +func (c *Container) joinedNetworkNSPath() string { + for _, namespace := range c.config.Spec.Linux.Namespaces { + if namespace.Type == spec.NetworkNamespace { + return namespace.Path + } + } + return "" +} + +func (c *Container) inspectJoinedNetworkNS(networkns string) (q types.StatusBlock, retErr error) { + var result types.StatusBlock + err := ns.WithNetNSPath(networkns, func(_ ns.NetNS) error { + ifaces, err := net.Interfaces() + if err != nil { + return err + } + routes, err := netlink.RouteList(nil, netlink.FAMILY_ALL) + if err != nil { + return err + } + var gateway net.IP + for _, route := range routes { + // default gateway + if route.Dst == nil { + gateway = route.Gw + } + } + result.Interfaces = make(map[string]types.NetInterface) + for _, iface := range ifaces { + if iface.Flags&net.FlagLoopback != 0 { + continue + } + addrs, err := iface.Addrs() + if err != nil { + continue + } + if len(addrs) == 0 { + continue + } + subnets := make([]types.NetAddress, 0, len(addrs)) + for _, address := range addrs { + if ipnet, ok := address.(*net.IPNet); ok { + if ipnet.IP.IsLinkLocalMulticast() || ipnet.IP.IsLinkLocalUnicast() { + continue + } + subnet := types.NetAddress{ + IPNet: types.IPNet{ + IPNet: *ipnet, + }, + } + if ipnet.Contains(gateway) { + subnet.Gateway = gateway + } + subnets = append(subnets, subnet) + } + } + result.Interfaces[iface.Name] = types.NetInterface{ + Subnets: subnets, + MacAddress: types.HardwareAddr(iface.HardwareAddr), + } + } + return nil + }) + return result, err +} + // resultToBasicNetworkConfig produces an InspectBasicNetworkConfig from a CNI // result func resultToBasicNetworkConfig(result types.StatusBlock) (define.InspectBasicNetworkConfig, error) { config := define.InspectBasicNetworkConfig{} - for _, netInt := range result.Interfaces { + interfaceNames := make([]string, len(result.Interfaces)) + for interfaceName := range result.Interfaces { + interfaceNames = append(interfaceNames, interfaceName) + } + // ensure consistent inspect results by sorting + sort.Strings(interfaceNames) + for _, interfaceName := range interfaceNames { + netInt := result.Interfaces[interfaceName] for _, netAddress := range netInt.Subnets { size, _ := netAddress.IPNet.Mask.Size() if netAddress.IPNet.IP.To4() != nil { diff --git a/libpod/runtime_pod_linux.go b/libpod/runtime_pod_linux.go index 155ad5c2d..230491c1a 100644 --- a/libpod/runtime_pod_linux.go +++ b/libpod/runtime_pod_linux.go @@ -1,3 +1,4 @@ +//go:build linux // +build linux package libpod @@ -59,50 +60,52 @@ func (r *Runtime) NewPod(ctx context.Context, p specgen.PodSpecGenerator, option pod.valid = true // Check Cgroup parent sanity, and set it if it was not set - switch r.config.Engine.CgroupManager { - case config.CgroupfsCgroupsManager: - canUseCgroup := !rootless.IsRootless() || isRootlessCgroupSet(pod.config.CgroupParent) - if canUseCgroup { + if r.config.Cgroups() != "disabled" { + switch r.config.Engine.CgroupManager { + case config.CgroupfsCgroupsManager: + canUseCgroup := !rootless.IsRootless() || isRootlessCgroupSet(pod.config.CgroupParent) + if canUseCgroup { + if pod.config.CgroupParent == "" { + pod.config.CgroupParent = CgroupfsDefaultCgroupParent + } else if strings.HasSuffix(path.Base(pod.config.CgroupParent), ".slice") { + return nil, errors.Wrapf(define.ErrInvalidArg, "systemd slice received as cgroup parent when using cgroupfs") + } + // If we are set to use pod cgroups, set the cgroup parent that + // all containers in the pod will share + // No need to create it with cgroupfs - the first container to + // launch should do it for us + if pod.config.UsePodCgroup { + pod.state.CgroupPath = filepath.Join(pod.config.CgroupParent, pod.ID()) + if p.InfraContainerSpec != nil { + p.InfraContainerSpec.CgroupParent = pod.state.CgroupPath + } + } + } + case config.SystemdCgroupsManager: if pod.config.CgroupParent == "" { - pod.config.CgroupParent = CgroupfsDefaultCgroupParent - } else if strings.HasSuffix(path.Base(pod.config.CgroupParent), ".slice") { - return nil, errors.Wrapf(define.ErrInvalidArg, "systemd slice received as cgroup parent when using cgroupfs") + if rootless.IsRootless() { + pod.config.CgroupParent = SystemdDefaultRootlessCgroupParent + } else { + pod.config.CgroupParent = SystemdDefaultCgroupParent + } + } else if len(pod.config.CgroupParent) < 6 || !strings.HasSuffix(path.Base(pod.config.CgroupParent), ".slice") { + return nil, errors.Wrapf(define.ErrInvalidArg, "did not receive systemd slice as cgroup parent when using systemd to manage cgroups") } // If we are set to use pod cgroups, set the cgroup parent that // all containers in the pod will share - // No need to create it with cgroupfs - the first container to - // launch should do it for us if pod.config.UsePodCgroup { - pod.state.CgroupPath = filepath.Join(pod.config.CgroupParent, pod.ID()) + cgroupPath, err := systemdSliceFromPath(pod.config.CgroupParent, fmt.Sprintf("libpod_pod_%s", pod.ID())) + if err != nil { + return nil, errors.Wrapf(err, "unable to create pod cgroup for pod %s", pod.ID()) + } + pod.state.CgroupPath = cgroupPath if p.InfraContainerSpec != nil { p.InfraContainerSpec.CgroupParent = pod.state.CgroupPath } } + default: + return nil, errors.Wrapf(define.ErrInvalidArg, "unsupported Cgroup manager: %s - cannot validate cgroup parent", r.config.Engine.CgroupManager) } - case config.SystemdCgroupsManager: - if pod.config.CgroupParent == "" { - if rootless.IsRootless() { - pod.config.CgroupParent = SystemdDefaultRootlessCgroupParent - } else { - pod.config.CgroupParent = SystemdDefaultCgroupParent - } - } else if len(pod.config.CgroupParent) < 6 || !strings.HasSuffix(path.Base(pod.config.CgroupParent), ".slice") { - return nil, errors.Wrapf(define.ErrInvalidArg, "did not receive systemd slice as cgroup parent when using systemd to manage cgroups") - } - // If we are set to use pod cgroups, set the cgroup parent that - // all containers in the pod will share - if pod.config.UsePodCgroup { - cgroupPath, err := systemdSliceFromPath(pod.config.CgroupParent, fmt.Sprintf("libpod_pod_%s", pod.ID())) - if err != nil { - return nil, errors.Wrapf(err, "unable to create pod cgroup for pod %s", pod.ID()) - } - pod.state.CgroupPath = cgroupPath - if p.InfraContainerSpec != nil { - p.InfraContainerSpec.CgroupParent = pod.state.CgroupPath - } - } - default: - return nil, errors.Wrapf(define.ErrInvalidArg, "unsupported Cgroup manager: %s - cannot validate cgroup parent", r.config.Engine.CgroupManager) } if pod.config.UsePodCgroup { |