diff options
Diffstat (limited to 'libpod')
-rw-r--r-- | libpod/container.go | 3 | ||||
-rw-r--r-- | libpod/container_api.go | 38 | ||||
-rw-r--r-- | libpod/container_easyjson.go | 13 | ||||
-rw-r--r-- | libpod/container_internal.go | 4 | ||||
-rw-r--r-- | libpod/container_internal_linux.go | 6 | ||||
-rw-r--r-- | libpod/oci.go | 16 | ||||
-rw-r--r-- | libpod/oci_linux.go | 6 | ||||
-rw-r--r-- | libpod/oci_unsupported.go | 2 | ||||
-rw-r--r-- | libpod/options.go | 4 | ||||
-rw-r--r-- | libpod/runtime_pod_infra_linux.go | 6 | ||||
-rw-r--r-- | libpod/util.go | 44 |
11 files changed, 108 insertions, 34 deletions
diff --git a/libpod/container.go b/libpod/container.go index 16f61d021..a8a58f4d8 100644 --- a/libpod/container.go +++ b/libpod/container.go @@ -9,6 +9,7 @@ import ( "github.com/containernetworking/cni/pkg/types" cnitypes "github.com/containernetworking/cni/pkg/types/current" + "github.com/containers/libpod/pkg/namespaces" "github.com/containers/storage" "github.com/cri-o/ocicni/pkg/ocicni" spec "github.com/opencontainers/runtime-spec/specs-go" @@ -296,6 +297,8 @@ type ContainerConfig struct { HostAdd []string `json:"hostsAdd,omitempty"` // Network names (CNI) to add container to. Empty to use default network. Networks []string `json:"networks,omitempty"` + // Network mode specified for the default network. + NetMode namespaces.NetworkMode `json:"networkMode,omitempty"` // Image Config diff --git a/libpod/container_api.go b/libpod/container_api.go index df6b6e962..4789c0cd2 100644 --- a/libpod/container_api.go +++ b/libpod/container_api.go @@ -328,6 +328,11 @@ func (c *Container) Exec(tty, privileged bool, env, cmd []string, user string) e if err != nil { return errors.Wrapf(err, "error exec %s", c.ID()) } + chWait := make(chan error) + go func() { + chWait <- execCmd.Wait() + }() + defer close(chWait) pidFile := c.execPidPath(sessionID) // 60 second seems a reasonable time to wait @@ -336,18 +341,12 @@ func (c *Container) Exec(tty, privileged bool, env, cmd []string, user string) e const pidWaitTimeout = 60000 // Wait until the runtime makes the pidfile - // TODO: If runtime errors before the PID file is created, we have to - // wait for timeout here - if err := WaitForFile(pidFile, pidWaitTimeout*time.Millisecond); err != nil { - logrus.Debugf("Timed out waiting for pidfile from runtime for container %s exec", c.ID()) - - // Check if an error occurred in the process before we made a pidfile - // TODO: Wait() here is a poor choice - is there a way to see if - // a process has finished, instead of waiting for it to finish? - if err := execCmd.Wait(); err != nil { + exited, err := WaitForFile(pidFile, chWait, pidWaitTimeout*time.Millisecond) + if err != nil { + if exited { + // If the runtime exited, propagate the error we got from the process. return err } - return errors.Wrapf(err, "timed out waiting for runtime to create pidfile for exec session in container %s", c.ID()) } @@ -389,7 +388,10 @@ func (c *Container) Exec(tty, privileged bool, env, cmd []string, user string) e locked = false } - waitErr := execCmd.Wait() + var waitErr error + if !exited { + waitErr = <-chWait + } // Lock again if !c.batched { @@ -831,10 +833,16 @@ func (c *Container) Refresh(ctx context.Context) error { } // ContainerCheckpointOptions is a struct used to pass the parameters -// for checkpointing to corresponding functions +// for checkpointing (and restoring) to the corresponding functions type ContainerCheckpointOptions struct { - Keep bool + // Keep tells the API to not delete checkpoint artifacts + Keep bool + // KeepRunning tells the API to keep the container running + // after writing the checkpoint to disk KeepRunning bool + // TCPEstablished tells the API to checkpoint a container + // even if it contains established TCP connections + TCPEstablished bool } // Checkpoint checkpoints a container @@ -853,7 +861,7 @@ func (c *Container) Checkpoint(ctx context.Context, options ContainerCheckpointO } // Restore restores a container -func (c *Container) Restore(ctx context.Context, keep bool) (err error) { +func (c *Container) Restore(ctx context.Context, options ContainerCheckpointOptions) (err error) { logrus.Debugf("Trying to restore container %s", c) if !c.batched { c.lock.Lock() @@ -864,5 +872,5 @@ func (c *Container) Restore(ctx context.Context, keep bool) (err error) { } } - return c.restore(ctx, keep) + return c.restore(ctx, options) } diff --git a/libpod/container_easyjson.go b/libpod/container_easyjson.go index 041cc08ac..8bf5cb64f 100644 --- a/libpod/container_easyjson.go +++ b/libpod/container_easyjson.go @@ -8,6 +8,7 @@ import ( json "encoding/json" types "github.com/containernetworking/cni/pkg/types" current "github.com/containernetworking/cni/pkg/types/current" + namespaces "github.com/containers/libpod/pkg/namespaces" storage "github.com/containers/storage" idtools "github.com/containers/storage/pkg/idtools" ocicni "github.com/cri-o/ocicni/pkg/ocicni" @@ -1550,6 +1551,8 @@ func easyjson1dbef17bDecodeGithubComContainersLibpodLibpod2(in *jlexer.Lexer, ou } in.Delim(']') } + case "networkMode": + out.NetMode = namespaces.NetworkMode(in.String()) case "userVolumes": if in.IsNull() { in.Skip() @@ -2177,6 +2180,16 @@ func easyjson1dbef17bEncodeGithubComContainersLibpodLibpod2(out *jwriter.Writer, out.RawByte(']') } } + if in.NetMode != "" { + const prefix string = ",\"networkMode\":" + if first { + first = false + out.RawString(prefix[1:]) + } else { + out.RawString(prefix) + } + out.String(string(in.NetMode)) + } if len(in.UserVolumes) != 0 { const prefix string = ",\"userVolumes\":" if first { diff --git a/libpod/container_internal.go b/libpod/container_internal.go index 051e0aeb7..700773e7f 100644 --- a/libpod/container_internal.go +++ b/libpod/container_internal.go @@ -586,7 +586,7 @@ func (c *Container) completeNetworkSetup() error { if err := c.syncContainer(); err != nil { return err } - if rootless.IsRootless() { + if c.config.NetMode == "slirp4netns" { return c.runtime.setupRootlessNetNS(c) } return c.runtime.setupNetNS(c) @@ -606,7 +606,7 @@ func (c *Container) init(ctx context.Context) error { } // With the spec complete, do an OCI create - if err := c.runtime.ociRuntime.createContainer(c, c.config.CgroupParent, false); err != nil { + if err := c.runtime.ociRuntime.createContainer(c, c.config.CgroupParent, nil); err != nil { return err } diff --git a/libpod/container_internal_linux.go b/libpod/container_internal_linux.go index 677645e7d..6b21bc3d6 100644 --- a/libpod/container_internal_linux.go +++ b/libpod/container_internal_linux.go @@ -514,7 +514,7 @@ func (c *Container) checkpoint(ctx context.Context, options ContainerCheckpointO return c.save() } -func (c *Container) restore(ctx context.Context, keep bool) (err error) { +func (c *Container) restore(ctx context.Context, options ContainerCheckpointOptions) (err error) { if !criu.CheckForCriu() { return errors.Errorf("restoring a container requires at least CRIU %d", criu.MinCriuVersion) @@ -602,7 +602,7 @@ func (c *Container) restore(ctx context.Context, keep bool) (err error) { // Cleanup for a working restore. c.removeConmonFiles() - if err := c.runtime.ociRuntime.createContainer(c, c.config.CgroupParent, true); err != nil { + if err := c.runtime.ociRuntime.createContainer(c, c.config.CgroupParent, &options); err != nil { return err } @@ -610,7 +610,7 @@ func (c *Container) restore(ctx context.Context, keep bool) (err error) { c.state.State = ContainerStateRunning - if !keep { + if !options.Keep { // Delete all checkpoint related files. At this point, in theory, all files // should exist. Still ignoring errors for now as the container should be // restored and running. Not erroring out just because some cleanup operation diff --git a/libpod/oci.go b/libpod/oci.go index a7aec06e5..a8013aa47 100644 --- a/libpod/oci.go +++ b/libpod/oci.go @@ -227,7 +227,7 @@ func bindPorts(ports []ocicni.PortMapping) ([]*os.File, error) { return files, nil } -func (r *OCIRuntime) createOCIContainer(ctr *Container, cgroupParent string, restoreContainer bool) (err error) { +func (r *OCIRuntime) createOCIContainer(ctr *Container, cgroupParent string, restoreOptions *ContainerCheckpointOptions) (err error) { var stderrBuf bytes.Buffer runtimeDir, err := util.GetRootlessRuntimeDir() @@ -289,8 +289,11 @@ func (r *OCIRuntime) createOCIContainer(ctr *Container, cgroupParent string, res args = append(args, "--syslog") } - if restoreContainer { + if restoreOptions != nil { args = append(args, "--restore", ctr.CheckpointPath()) + if restoreOptions.TCPEstablished { + args = append(args, "--restore-arg", "--tcp-established") + } } logrus.WithFields(logrus.Fields{ @@ -316,6 +319,10 @@ func (r *OCIRuntime) createOCIContainer(ctr *Container, cgroupParent string, res cmd.Env = append(r.conmonEnv, fmt.Sprintf("_OCI_SYNCPIPE=%d", 3)) cmd.Env = append(cmd.Env, fmt.Sprintf("_OCI_STARTPIPE=%d", 4)) cmd.Env = append(cmd.Env, fmt.Sprintf("XDG_RUNTIME_DIR=%s", runtimeDir)) + cmd.Env = append(cmd.Env, fmt.Sprintf("_LIBPOD_USERNS_CONFIGURED=%s", os.Getenv("_LIBPOD_USERNS_CONFIGURED"))) + cmd.Env = append(cmd.Env, fmt.Sprintf("_LIBPOD_ROOTLESS_UID=%s", os.Getenv("_LIBPOD_ROOTLESS_UID"))) + cmd.Env = append(cmd.Env, fmt.Sprintf("HOME=%s", os.Getenv("HOME"))) + cmd.Env = append(cmd.Env, fmt.Sprintf("XDG_RUNTIME_DIR=%s", runtimeDir)) if r.reservePorts { ports, err := bindPorts(ctr.config.PortMappings) @@ -329,7 +336,7 @@ func (r *OCIRuntime) createOCIContainer(ctr *Container, cgroupParent string, res cmd.ExtraFiles = append(cmd.ExtraFiles, ports...) } - if rootless.IsRootless() { + if ctr.config.NetMode.IsSlirp4netns() { ctr.rootlessSlirpSyncR, ctr.rootlessSlirpSyncW, err = os.Pipe() if err != nil { return errors.Wrapf(err, "failed to create rootless network sync pipe") @@ -862,6 +869,9 @@ func (r *OCIRuntime) checkpointContainer(ctr *Container, options ContainerCheckp if options.KeepRunning { args = append(args, "--leave-running") } + if options.TCPEstablished { + args = append(args, "--tcp-established") + } args = append(args, ctr.ID()) return utils.ExecCmdWithStdStreams(os.Stdin, os.Stdout, os.Stderr, nil, r.path, args...) } diff --git a/libpod/oci_linux.go b/libpod/oci_linux.go index b159eae78..2737a641e 100644 --- a/libpod/oci_linux.go +++ b/libpod/oci_linux.go @@ -65,10 +65,10 @@ func newPipe() (parent *os.File, child *os.File, err error) { // CreateContainer creates a container in the OCI runtime // TODO terminal support for container // Presently just ignoring conmon opts related to it -func (r *OCIRuntime) createContainer(ctr *Container, cgroupParent string, restoreContainer bool) (err error) { +func (r *OCIRuntime) createContainer(ctr *Container, cgroupParent string, restoreOptions *ContainerCheckpointOptions) (err error) { if ctr.state.UserNSRoot == "" { // no need of an intermediate mount ns - return r.createOCIContainer(ctr, cgroupParent, restoreContainer) + return r.createOCIContainer(ctr, cgroupParent, restoreOptions) } var wg sync.WaitGroup wg.Add(1) @@ -106,7 +106,7 @@ func (r *OCIRuntime) createContainer(ctr *Container, cgroupParent string, restor if err != nil { return } - err = r.createOCIContainer(ctr, cgroupParent, restoreContainer) + err = r.createOCIContainer(ctr, cgroupParent, restoreOptions) }() wg.Wait() diff --git a/libpod/oci_unsupported.go b/libpod/oci_unsupported.go index b133eb402..8c084d1e2 100644 --- a/libpod/oci_unsupported.go +++ b/libpod/oci_unsupported.go @@ -15,7 +15,7 @@ func newPipe() (parent *os.File, child *os.File, err error) { return nil, nil, ErrNotImplemented } -func (r *OCIRuntime) createContainer(ctr *Container, cgroupParent string, restoreContainer bool) (err error) { +func (r *OCIRuntime) createContainer(ctr *Container, cgroupParent string, restoreOptions *ContainerCheckpointOptions) (err error) { return ErrNotImplemented } diff --git a/libpod/options.go b/libpod/options.go index 507847d65..7f4e3ac6b 100644 --- a/libpod/options.go +++ b/libpod/options.go @@ -7,6 +7,7 @@ import ( "regexp" "syscall" + "github.com/containers/libpod/pkg/namespaces" "github.com/containers/storage" "github.com/containers/storage/pkg/idtools" "github.com/cri-o/ocicni/pkg/ocicni" @@ -817,7 +818,7 @@ func WithDependencyCtrs(ctrs []*Container) CtrCreateOption { // namespace with a minimal configuration. // An optional array of port mappings can be provided. // Conflicts with WithNetNSFrom(). -func WithNetNS(portMappings []ocicni.PortMapping, postConfigureNetNS bool, networks []string) CtrCreateOption { +func WithNetNS(portMappings []ocicni.PortMapping, postConfigureNetNS bool, netmode string, networks []string) CtrCreateOption { return func(ctr *Container) error { if ctr.valid { return ErrCtrFinalized @@ -831,6 +832,7 @@ func WithNetNS(portMappings []ocicni.PortMapping, postConfigureNetNS bool, netwo ctr.config.CreateNetNS = true ctr.config.PortMappings = portMappings ctr.config.Networks = networks + ctr.config.NetMode = namespaces.NetworkMode(netmode) return nil } diff --git a/libpod/runtime_pod_infra_linux.go b/libpod/runtime_pod_infra_linux.go index 450a2fb32..8a5dbef56 100644 --- a/libpod/runtime_pod_infra_linux.go +++ b/libpod/runtime_pod_infra_linux.go @@ -50,7 +50,11 @@ func (r *Runtime) makeInfraContainer(ctx context.Context, p *Pod, imgName, imgID // Since user namespace sharing is not implemented, we only need to check if it's rootless networks := make([]string, 0) - options = append(options, WithNetNS(p.config.InfraContainer.PortBindings, isRootless, networks)) + netmode := "bridge" + if isRootless { + netmode = "slirp4netns" + } + options = append(options, WithNetNS(p.config.InfraContainer.PortBindings, isRootless, netmode, networks)) return r.newContainer(ctx, g.Config, options...) } diff --git a/libpod/util.go b/libpod/util.go index 7007b29cd..aa3494529 100644 --- a/libpod/util.go +++ b/libpod/util.go @@ -13,6 +13,7 @@ import ( "github.com/containers/image/signature" "github.com/containers/image/types" "github.com/containers/libpod/pkg/util" + "github.com/fsnotify/fsnotify" spec "github.com/opencontainers/runtime-spec/specs-go" "github.com/pkg/errors" ) @@ -90,31 +91,64 @@ func MountExists(specMounts []spec.Mount, dest string) bool { } // WaitForFile waits until a file has been created or the given timeout has occurred -func WaitForFile(path string, timeout time.Duration) error { +func WaitForFile(path string, chWait chan error, timeout time.Duration) (bool, error) { done := make(chan struct{}) chControl := make(chan struct{}) + + var inotifyEvents chan fsnotify.Event + var timer chan struct{} + watcher, err := fsnotify.NewWatcher() + if err == nil { + if err := watcher.Add(filepath.Dir(path)); err == nil { + inotifyEvents = watcher.Events + } + defer watcher.Close() + } + if inotifyEvents == nil { + // If for any reason we fail to create the inotify + // watcher, fallback to polling the file + timer = make(chan struct{}) + go func() { + select { + case <-chControl: + close(timer) + return + default: + time.Sleep(25 * time.Millisecond) + timer <- struct{}{} + } + }() + } + go func() { for { select { case <-chControl: return - default: + case <-timer: + _, err := os.Stat(path) + if err == nil { + close(done) + return + } + case <-inotifyEvents: _, err := os.Stat(path) if err == nil { close(done) return } - time.Sleep(25 * time.Millisecond) } } }() select { + case e := <-chWait: + return true, e case <-done: - return nil + return false, nil case <-time.After(timeout): close(chControl) - return errors.Wrapf(ErrInternal, "timed out waiting for file %s", path) + return false, errors.Wrapf(ErrInternal, "timed out waiting for file %s", path) } } |