diff options
author | Daniel J Walsh <dwalsh@redhat.com> | 2018-12-06 03:20:16 -0500 |
---|---|---|
committer | Daniel J Walsh <dwalsh@redhat.com> | 2018-12-06 03:20:16 -0500 |
commit | 40678b119c0186bf7c56dc09d131e697c55cb9fd (patch) | |
tree | dfb1cc13b0bdc1abad1ada51970bf0ac03d46810 /libpod | |
parent | dcb68d1aa24b2dd00d793f5a66f96cc5403b43c3 (diff) | |
parent | 75b19ca8abe1957f3c48035767960a6b20c10519 (diff) | |
download | podman-40678b119c0186bf7c56dc09d131e697c55cb9fd.tar.gz podman-40678b119c0186bf7c56dc09d131e697c55cb9fd.tar.bz2 podman-40678b119c0186bf7c56dc09d131e697c55cb9fd.zip |
Merge branch 'master' of github.com:containers/libpod into vendor
Diffstat (limited to 'libpod')
32 files changed, 1598 insertions, 591 deletions
diff --git a/libpod/boltdb_state.go b/libpod/boltdb_state.go index 42f029379..cb661d4e9 100644 --- a/libpod/boltdb_state.go +++ b/libpod/boltdb_state.go @@ -3,7 +3,6 @@ package libpod import ( "bytes" "encoding/json" - "os" "strings" "sync" @@ -19,7 +18,6 @@ type BoltState struct { dbLock sync.Mutex namespace string namespaceBytes []byte - lockDir string runtime *Runtime } @@ -52,25 +50,15 @@ type BoltState struct { // containers/storage do not occur. // NewBoltState creates a new bolt-backed state database -func NewBoltState(path, lockDir string, runtime *Runtime) (State, error) { +func NewBoltState(path string, runtime *Runtime) (State, error) { state := new(BoltState) state.dbPath = path - state.lockDir = lockDir state.runtime = runtime state.namespace = "" state.namespaceBytes = nil logrus.Debugf("Initializing boltdb state at %s", path) - // Make the directory that will hold container lockfiles - if err := os.MkdirAll(lockDir, 0750); err != nil { - // The directory is allowed to exist - if !os.IsExist(err) { - return nil, errors.Wrapf(err, "error creating lockfiles dir %s", lockDir) - } - } - state.lockDir = lockDir - db, err := bolt.Open(path, 0600, nil) if err != nil { return nil, errors.Wrapf(err, "error opening database %s", path) @@ -115,11 +103,6 @@ func NewBoltState(path, lockDir string, runtime *Runtime) (State, error) { return nil, errors.Wrapf(err, "error creating initial database layout") } - // Check runtime configuration - if err := checkRuntimeConfig(db, runtime); err != nil { - return nil, err - } - state.valid = true return state, nil @@ -240,6 +223,72 @@ func (s *BoltState) Refresh() error { return err } +// GetDBConfig retrieves runtime configuration fields that were created when +// the database was first initialized +func (s *BoltState) GetDBConfig() (*DBConfig, error) { + if !s.valid { + return nil, ErrDBClosed + } + + cfg := new(DBConfig) + + db, err := s.getDBCon() + if err != nil { + return nil, err + } + defer s.closeDBCon(db) + + err = db.View(func(tx *bolt.Tx) error { + configBucket, err := getRuntimeConfigBucket(tx) + if err != nil { + return nil + } + + // Some of these may be nil + // When we convert to string, Go will coerce them to "" + // That's probably fine - we could raise an error if the key is + // missing, but just not including it is also OK. + libpodRoot := configBucket.Get(staticDirKey) + libpodTmp := configBucket.Get(tmpDirKey) + storageRoot := configBucket.Get(graphRootKey) + storageTmp := configBucket.Get(runRootKey) + graphDriver := configBucket.Get(graphDriverKey) + + cfg.LibpodRoot = string(libpodRoot) + cfg.LibpodTmp = string(libpodTmp) + cfg.StorageRoot = string(storageRoot) + cfg.StorageTmp = string(storageTmp) + cfg.GraphDriver = string(graphDriver) + + return nil + }) + if err != nil { + return nil, err + } + + return cfg, nil +} + +// ValidateDBConfig validates paths in the given runtime against the database +func (s *BoltState) ValidateDBConfig(runtime *Runtime) error { + if !s.valid { + return ErrDBClosed + } + + db, err := s.getDBCon() + if err != nil { + return err + } + defer s.closeDBCon(db) + + // Check runtime configuration + if err := checkRuntimeConfig(db, runtime); err != nil { + return err + } + + return nil +} + // SetNamespace sets the namespace that will be used for container and pod // retrieval func (s *BoltState) SetNamespace(ns string) error { diff --git a/libpod/boltdb_state_internal.go b/libpod/boltdb_state_internal.go index cc7d106cc..3f00657ea 100644 --- a/libpod/boltdb_state_internal.go +++ b/libpod/boltdb_state_internal.go @@ -30,6 +30,13 @@ const ( containersName = "containers" podIDName = "pod-id" namespaceName = "namespace" + + staticDirName = "static-dir" + tmpDirName = "tmp-dir" + runRootName = "run-root" + graphRootName = "graph-root" + graphDriverName = "graph-driver-name" + osName = "os" ) var ( @@ -49,21 +56,19 @@ var ( containersBkt = []byte(containersName) podIDKey = []byte(podIDName) namespaceKey = []byte(namespaceName) + + staticDirKey = []byte(staticDirName) + tmpDirKey = []byte(tmpDirName) + runRootKey = []byte(runRootName) + graphRootKey = []byte(graphRootName) + graphDriverKey = []byte(graphDriverName) + osKey = []byte(osName) ) // Check if the configuration of the database is compatible with the // configuration of the runtime opening it // If there is no runtime configuration loaded, load our own func checkRuntimeConfig(db *bolt.DB, rt *Runtime) error { - var ( - staticDir = []byte("static-dir") - tmpDir = []byte("tmp-dir") - runRoot = []byte("run-root") - graphRoot = []byte("graph-root") - graphDriverName = []byte("graph-driver-name") - osKey = []byte("os") - ) - err := db.Update(func(tx *bolt.Tx) error { configBkt, err := getRuntimeConfigBucket(tx) if err != nil { @@ -74,31 +79,31 @@ func checkRuntimeConfig(db *bolt.DB, rt *Runtime) error { return err } - if err := validateDBAgainstConfig(configBkt, "static dir", - rt.config.StaticDir, staticDir, ""); err != nil { + if err := validateDBAgainstConfig(configBkt, "libpod root directory (staticdir)", + rt.config.StaticDir, staticDirKey, ""); err != nil { return err } - if err := validateDBAgainstConfig(configBkt, "tmp dir", - rt.config.TmpDir, tmpDir, ""); err != nil { + if err := validateDBAgainstConfig(configBkt, "libpod temporary files directory (tmpdir)", + rt.config.TmpDir, tmpDirKey, ""); err != nil { return err } - if err := validateDBAgainstConfig(configBkt, "run root", - rt.config.StorageConfig.RunRoot, runRoot, + if err := validateDBAgainstConfig(configBkt, "storage temporary directory (runroot)", + rt.config.StorageConfig.RunRoot, runRootKey, storage.DefaultStoreOptions.RunRoot); err != nil { return err } - if err := validateDBAgainstConfig(configBkt, "graph root", - rt.config.StorageConfig.GraphRoot, graphRoot, + if err := validateDBAgainstConfig(configBkt, "storage graph root directory (graphroot)", + rt.config.StorageConfig.GraphRoot, graphRootKey, storage.DefaultStoreOptions.GraphRoot); err != nil { return err } - return validateDBAgainstConfig(configBkt, "graph driver name", + return validateDBAgainstConfig(configBkt, "storage graph driver", rt.config.StorageConfig.GraphDriverName, - graphDriverName, + graphDriverKey, storage.DefaultStoreOptions.GraphDriverName) }) @@ -261,7 +266,7 @@ func (s *BoltState) getContainerFromDB(id []byte, ctr *Container, ctrsBkt *bolt. } // Get the lock - lockPath := filepath.Join(s.lockDir, string(id)) + lockPath := filepath.Join(s.runtime.lockDir, string(id)) lock, err := storage.GetLockfile(lockPath) if err != nil { return errors.Wrapf(err, "error retrieving lockfile for container %s", string(id)) @@ -297,7 +302,7 @@ func (s *BoltState) getPodFromDB(id []byte, pod *Pod, podBkt *bolt.Bucket) error } // Get the lock - lockPath := filepath.Join(s.lockDir, string(id)) + lockPath := filepath.Join(s.runtime.lockDir, string(id)) lock, err := storage.GetLockfile(lockPath) if err != nil { return errors.Wrapf(err, "error retrieving lockfile for pod %s", string(id)) diff --git a/libpod/container.go b/libpod/container.go index 7bb5b2687..b6155a809 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 @@ -826,7 +829,7 @@ func (c *Container) IPs() ([]net.IPNet, error) { } if !c.config.CreateNetNS { - return nil, errors.Wrapf(ErrInvalidArg, "container %s network namespace is not managed by libpod") + return nil, errors.Wrapf(ErrInvalidArg, "container %s network namespace is not managed by libpod", c.ID()) } ips := make([]net.IPNet, 0) @@ -854,7 +857,7 @@ func (c *Container) Routes() ([]types.Route, error) { } if !c.config.CreateNetNS { - return nil, errors.Wrapf(ErrInvalidArg, "container %s network namespace is not managed by libpod") + return nil, errors.Wrapf(ErrInvalidArg, "container %s network namespace is not managed by libpod", c.ID()) } routes := make([]types.Route, 0) @@ -996,3 +999,15 @@ func (c *Container) IsInfra() bool { func (c *Container) IsReadOnly() bool { return c.config.Spec.Root.Readonly } + +// NetworkDisabled returns whether the container is running with a disabled network +func (c *Container) NetworkDisabled() bool { + if !c.config.PostConfigureNetNS { + for _, ns := range c.config.Spec.Linux.Namespaces { + if ns.Type == spec.NetworkNamespace { + return ns.Path == "" + } + } + } + return false +} diff --git a/libpod/container_api.go b/libpod/container_api.go index e522038a3..bc92cae69 100644 --- a/libpod/container_api.go +++ b/libpod/container_api.go @@ -10,8 +10,8 @@ import ( "time" "github.com/containers/libpod/libpod/driver" - "github.com/containers/libpod/pkg/chrootuser" "github.com/containers/libpod/pkg/inspect" + "github.com/containers/libpod/pkg/lookup" "github.com/containers/storage/pkg/stringid" "github.com/docker/docker/daemon/caps" "github.com/pkg/errors" @@ -39,16 +39,13 @@ func (c *Container) Init(ctx context.Context) (err error) { notRunning, err := c.checkDependenciesRunning() if err != nil { - return errors.Wrapf(err, "error checking dependencies for container %s") + return errors.Wrapf(err, "error checking dependencies for container %s", c.ID()) } if len(notRunning) > 0 { depString := strings.Join(notRunning, ",") return errors.Wrapf(ErrCtrStateInvalid, "some dependencies of container %s are not started: %s", c.ID(), depString) } - if err := c.prepare(); err != nil { - return err - } defer func() { if err != nil { if err2 := c.cleanup(ctx); err2 != nil { @@ -57,6 +54,10 @@ func (c *Container) Init(ctx context.Context) (err error) { } }() + if err := c.prepare(); err != nil { + return err + } + if c.state.State == ContainerStateStopped { // Reinitialize the container return c.reinit(ctx) @@ -92,16 +93,13 @@ func (c *Container) Start(ctx context.Context) (err error) { notRunning, err := c.checkDependenciesRunning() if err != nil { - return errors.Wrapf(err, "error checking dependencies for container %s") + return errors.Wrapf(err, "error checking dependencies for container %s", c.ID()) } if len(notRunning) > 0 { depString := strings.Join(notRunning, ",") return errors.Wrapf(ErrCtrStateInvalid, "some dependencies of container %s are not started: %s", c.ID(), depString) } - if err := c.prepare(); err != nil { - return err - } defer func() { if err != nil { if err2 := c.cleanup(ctx); err2 != nil { @@ -110,6 +108,10 @@ func (c *Container) Start(ctx context.Context) (err error) { } }() + if err := c.prepare(); err != nil { + return err + } + if c.state.State == ContainerStateStopped { // Reinitialize the container if we need to if err := c.reinit(ctx); err != nil { @@ -157,16 +159,13 @@ func (c *Container) StartAndAttach(ctx context.Context, streams *AttachStreams, notRunning, err := c.checkDependenciesRunning() if err != nil { - return nil, errors.Wrapf(err, "error checking dependencies for container %s") + return nil, errors.Wrapf(err, "error checking dependencies for container %s", c.ID()) } if len(notRunning) > 0 { depString := strings.Join(notRunning, ",") return nil, errors.Wrapf(ErrCtrStateInvalid, "some dependencies of container %s are not started: %s", c.ID(), depString) } - if err := c.prepare(); err != nil { - return nil, err - } defer func() { if err != nil { if err2 := c.cleanup(ctx); err2 != nil { @@ -175,6 +174,10 @@ func (c *Container) StartAndAttach(ctx context.Context, streams *AttachStreams, } }() + if err := c.prepare(); err != nil { + return nil, err + } + if c.state.State == ContainerStateStopped { // Reinitialize the container if we need to if err := c.reinit(ctx); err != nil { @@ -292,13 +295,13 @@ func (c *Container) Exec(tty, privileged bool, env, cmd []string, user string) e // the host hostUser := "" if user != "" { - uid, gid, err := chrootuser.GetUser(c.state.Mountpoint, user) + execUser, err := lookup.GetUserGroupInfo(c.state.Mountpoint, user, nil) if err != nil { - return errors.Wrapf(err, "error getting user to launch exec session as") + return err } // runc expects user formatted as uid:gid - hostUser = fmt.Sprintf("%d:%d", uid, gid) + hostUser = fmt.Sprintf("%d:%d", execUser.Uid, execUser.Gid) } // Generate exec session ID @@ -325,25 +328,25 @@ 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) - // 1 second seems a reasonable time to wait - // See https://github.com/containers/libpod/issues/1495 - const pidWaitTimeout = 1000 + // 60 second seems a reasonable time to wait + // https://github.com/containers/libpod/issues/1495 + // https://github.com/containers/libpod/issues/1816 + 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()) } @@ -385,7 +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 { @@ -685,7 +691,7 @@ func (c *Container) Sync() error { (c.state.State != ContainerStateConfigured) { oldState := c.state.State // TODO: optionally replace this with a stat for the exit file - if err := c.runtime.ociRuntime.updateContainerStatus(c); err != nil { + if err := c.runtime.ociRuntime.updateContainerStatus(c, true); err != nil { return err } // Only save back to DB if state changed @@ -712,7 +718,7 @@ func (c *Container) RestartWithTimeout(ctx context.Context, timeout uint) (err e notRunning, err := c.checkDependenciesRunning() if err != nil { - return errors.Wrapf(err, "error checking dependencies for container %s") + return errors.Wrapf(err, "error checking dependencies for container %s", c.ID()) } if len(notRunning) > 0 { depString := strings.Join(notRunning, ",") @@ -797,7 +803,7 @@ func (c *Container) Refresh(ctx context.Context) error { return err } - logrus.Debugf("Successfully refresh container %s state") + logrus.Debugf("Successfully refresh container %s state", c.ID()) // Initialize the container if it was created in runc if wasCreated || wasRunning || wasPaused { @@ -826,9 +832,22 @@ func (c *Container) Refresh(ctx context.Context) error { return nil } +// ContainerCheckpointOptions is a struct used to pass the parameters +// for checkpointing (and restoring) to the corresponding functions +type ContainerCheckpointOptions struct { + // 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 -func (c *Container) Checkpoint(ctx context.Context, keep bool) error { - logrus.Debugf("Trying to checkpoint container %s", c) +func (c *Container) Checkpoint(ctx context.Context, options ContainerCheckpointOptions) error { + logrus.Debugf("Trying to checkpoint container %s", c.ID()) if !c.batched { c.lock.Lock() defer c.lock.Unlock() @@ -838,12 +857,12 @@ func (c *Container) Checkpoint(ctx context.Context, keep bool) error { } } - return c.checkpoint(ctx, keep) + return c.checkpoint(ctx, options) } // Restore restores a container -func (c *Container) Restore(ctx context.Context, keep bool) (err error) { - logrus.Debugf("Trying to restore container %s", c) +func (c *Container) Restore(ctx context.Context, options ContainerCheckpointOptions) (err error) { + logrus.Debugf("Trying to restore container %s", c.ID()) if !c.batched { c.lock.Lock() defer c.lock.Unlock() @@ -853,5 +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_attach.go b/libpod/container_attach.go index 3c4e0775d..f925c3897 100644 --- a/libpod/container_attach.go +++ b/libpod/container_attach.go @@ -16,6 +16,10 @@ import ( "k8s.io/client-go/tools/remotecommand" ) +//#include <sys/un.h> +// extern int unix_path_length(){struct sockaddr_un addr; return sizeof(addr.sun_path) - 1;} +import "C" + /* Sync with stdpipe_t in conmon.c */ const ( AttachPipeStdin = 1 @@ -81,11 +85,19 @@ func (c *Container) attachContainerSocket(resize <-chan remotecommand.TerminalSi logrus.Warnf("Failed to write to control file to resize terminal: %v", err) } }) - logrus.Debug("connecting to socket ", c.AttachSocketPath()) - conn, err := net.DialUnix("unixpacket", nil, &net.UnixAddr{Name: c.AttachSocketPath(), Net: "unixpacket"}) + socketPath := c.AttachSocketPath() + + maxUnixLength := int(C.unix_path_length()) + if maxUnixLength < len(socketPath) { + socketPath = socketPath[0:maxUnixLength] + } + + logrus.Debug("connecting to socket ", socketPath) + + conn, err := net.DialUnix("unixpacket", nil, &net.UnixAddr{Name: socketPath, Net: "unixpacket"}) if err != nil { - return errors.Wrapf(err, "failed to connect to container's attach socket: %v", c.AttachSocketPath()) + return errors.Wrapf(err, "failed to connect to container's attach socket: %v", socketPath) } defer conn.Close() 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_graph_test.go b/libpod/container_graph_test.go index bba3d7aad..25461f1f4 100644 --- a/libpod/container_graph_test.go +++ b/libpod/container_graph_test.go @@ -205,6 +205,7 @@ func TestBuildContainerGraphFourContainersNoEdges(t *testing.T) { ctr3, err := getTestCtrN("3", tmpDir) assert.NoError(t, err) ctr4, err := getTestCtrN("4", tmpDir) + assert.NoError(t, err) graph, err := buildContainerGraph([]*Container{ctr1, ctr2, ctr3, ctr4}) assert.NoError(t, err) @@ -241,6 +242,7 @@ func TestBuildContainerGraphFourContainersTwoInCycle(t *testing.T) { ctr3, err := getTestCtrN("3", tmpDir) assert.NoError(t, err) ctr4, err := getTestCtrN("4", tmpDir) + assert.NoError(t, err) ctr1.config.IPCNsCtr = ctr2.config.ID ctr2.config.UserNsCtr = ctr1.config.ID @@ -260,6 +262,7 @@ func TestBuildContainerGraphFourContainersAllInCycle(t *testing.T) { ctr3, err := getTestCtrN("3", tmpDir) assert.NoError(t, err) ctr4, err := getTestCtrN("4", tmpDir) + assert.NoError(t, err) ctr1.config.IPCNsCtr = ctr2.config.ID ctr2.config.UserNsCtr = ctr3.config.ID ctr3.config.NetNsCtr = ctr4.config.ID @@ -281,6 +284,7 @@ func TestBuildContainerGraphFourContainersNoneInCycle(t *testing.T) { ctr3, err := getTestCtrN("3", tmpDir) assert.NoError(t, err) ctr4, err := getTestCtrN("4", tmpDir) + assert.NoError(t, err) ctr1.config.IPCNsCtr = ctr2.config.ID ctr1.config.NetNsCtr = ctr3.config.ID ctr2.config.UserNsCtr = ctr3.config.ID diff --git a/libpod/container_internal.go b/libpod/container_internal.go index 2af216358..934ad7a22 100644 --- a/libpod/container_internal.go +++ b/libpod/container_internal.go @@ -12,14 +12,14 @@ import ( "strconv" "strings" "syscall" + "time" "github.com/containers/buildah/imagebuildah" - "github.com/containers/libpod/pkg/chrootuser" + "github.com/containers/libpod/pkg/ctime" "github.com/containers/libpod/pkg/hooks" "github.com/containers/libpod/pkg/hooks/exec" - "github.com/containers/libpod/pkg/resolvconf" + "github.com/containers/libpod/pkg/lookup" "github.com/containers/libpod/pkg/rootless" - "github.com/containers/libpod/pkg/secrets" "github.com/containers/storage" "github.com/containers/storage/pkg/archive" "github.com/containers/storage/pkg/chrootarchive" @@ -30,6 +30,7 @@ import ( "github.com/pkg/errors" "github.com/sirupsen/logrus" "golang.org/x/text/language" + kwait "k8s.io/apimachinery/pkg/util/wait" ) const ( @@ -146,6 +147,77 @@ func (c *Container) execPidPath(sessionID string) string { return filepath.Join(c.state.RunDir, "exec_pid_"+sessionID) } +// exitFilePath gets the path to the container's exit file +func (c *Container) exitFilePath() string { + return filepath.Join(c.runtime.ociRuntime.exitsDir, c.ID()) +} + +// Wait for the container's exit file to appear. +// When it does, update our state based on it. +func (c *Container) waitForExitFileAndSync() error { + exitFile := c.exitFilePath() + + err := kwait.ExponentialBackoff( + kwait.Backoff{ + Duration: 500 * time.Millisecond, + Factor: 1.2, + Steps: 6, + }, + func() (bool, error) { + _, err := os.Stat(exitFile) + if err != nil { + // wait longer + return false, nil + } + return true, nil + }) + if err != nil { + // Exit file did not appear + // Reset our state + c.state.ExitCode = -1 + c.state.FinishedTime = time.Now() + c.state.State = ContainerStateStopped + + if err2 := c.save(); err2 != nil { + logrus.Errorf("Error saving container %s state: %v", c.ID(), err2) + } + + return err + } + + if err := c.runtime.ociRuntime.updateContainerStatus(c, false); err != nil { + return err + } + + return c.save() +} + +// Handle the container exit file. +// The exit file is used to supply container exit time and exit code. +// This assumes the exit file already exists. +func (c *Container) handleExitFile(exitFile string, fi os.FileInfo) error { + c.state.FinishedTime = ctime.Created(fi) + statusCodeStr, err := ioutil.ReadFile(exitFile) + if err != nil { + return errors.Wrapf(err, "failed to read exit file for container %s", c.ID()) + } + statusCode, err := strconv.Atoi(string(statusCodeStr)) + if err != nil { + return errors.Wrapf(err, "error converting exit status code (%q) for container %s to int", + c.ID(), statusCodeStr) + } + c.state.ExitCode = int32(statusCode) + + oomFilePath := filepath.Join(c.bundlePath(), "oom") + if _, err = os.Stat(oomFilePath); err == nil { + c.state.OOMKilled = true + } + + c.state.Exited = true + + return nil +} + // Sync this container with on-disk state and runtime status // Should only be called with container lock held // This function should suffice to ensure a container's state is accurate and @@ -161,7 +233,7 @@ func (c *Container) syncContainer() error { (c.state.State != ContainerStateExited) { oldState := c.state.State // TODO: optionally replace this with a stat for the exit file - if err := c.runtime.ociRuntime.updateContainerStatus(c); err != nil { + if err := c.runtime.ociRuntime.updateContainerStatus(c, false); err != nil { return err } // Only save back to DB if state changed @@ -201,6 +273,27 @@ func (c *Container) setupStorage(ctx context.Context) error { }, LabelOpts: c.config.LabelOpts, } + if c.config.Privileged { + privOpt := func(opt string) bool { + for _, privopt := range []string{"nodev", "nosuid", "noexec"} { + if opt == privopt { + return true + } + } + return false + } + defOptions, err := storage.GetDefaultMountOptions() + if err != nil { + return errors.Wrapf(err, "error getting default mount options") + } + var newOptions []string + for _, opt := range defOptions { + if !privOpt(opt) { + newOptions = append(newOptions, opt) + } + } + options.MountOpts = newOptions + } if c.config.Rootfs == "" { options.IDMappingOptions = c.config.IDMappings @@ -508,13 +601,13 @@ func (c *Container) checkDependenciesRunningLocked(depCtrs map[string]*Container } func (c *Container) completeNetworkSetup() error { - if !c.config.PostConfigureNetNS { + if !c.config.PostConfigureNetNS || c.NetworkDisabled() { return nil } 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) @@ -522,10 +615,6 @@ func (c *Container) completeNetworkSetup() error { // Initialize a container, creating it in the runtime func (c *Container) init(ctx context.Context) error { - if err := c.makeBindMounts(); err != nil { - return err - } - // Generate the OCI spec spec, err := c.generateSpec(ctx) if err != nil { @@ -538,7 +627,7 @@ 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 } @@ -622,9 +711,6 @@ func (c *Container) initAndStart(ctx context.Context) (err error) { return errors.Wrapf(ErrCtrStateInvalid, "cannot start paused container %s", c.ID()) } - if err := c.prepare(); err != nil { - return err - } defer func() { if err != nil { if err2 := c.cleanup(ctx); err2 != nil { @@ -633,6 +719,10 @@ func (c *Container) initAndStart(ctx context.Context) (err error) { } }() + if err := c.prepare(); err != nil { + return err + } + // If we are ContainerStateStopped we need to remove from runtime // And reset to ContainerStateConfigured if c.state.State == ContainerStateStopped { @@ -672,13 +762,8 @@ func (c *Container) stop(timeout uint) error { return err } - // Sync the container's state to pick up return code - if err := c.runtime.ociRuntime.updateContainerStatus(c); err != nil { - return err - } - - // Container should clean itself up - return nil + // Wait until we have an exit file, and sync once we do + return c.waitForExitFileAndSync() } // Internal, non-locking function to pause a container @@ -718,9 +803,6 @@ func (c *Container) restartWithTimeout(ctx context.Context, timeout uint) (err e return err } } - if err := c.prepare(); err != nil { - return err - } defer func() { if err != nil { if err2 := c.cleanup(ctx); err2 != nil { @@ -728,6 +810,9 @@ func (c *Container) restartWithTimeout(ctx context.Context, timeout uint) (err e } } }() + if err := c.prepare(); err != nil { + return err + } if c.state.State == ContainerStateStopped { // Reinitialize the container if we need to @@ -757,28 +842,22 @@ func (c *Container) mountStorage() (string, error) { return c.state.Mountpoint, nil } - if !rootless.IsRootless() { - // TODO: generalize this mount code so it will mount every mount in ctr.config.Mounts - mounted, err := mount.Mounted(c.config.ShmDir) - if err != nil { - return "", errors.Wrapf(err, "unable to determine if %q is mounted", c.config.ShmDir) - } + mounted, err := mount.Mounted(c.config.ShmDir) + if err != nil { + return "", errors.Wrapf(err, "unable to determine if %q is mounted", c.config.ShmDir) + } + if !mounted { + shmOptions := fmt.Sprintf("mode=1777,size=%d", c.config.ShmSize) + if err := c.mountSHM(shmOptions); err != nil { + return "", err + } if err := os.Chown(c.config.ShmDir, c.RootUID(), c.RootGID()); err != nil { return "", errors.Wrapf(err, "failed to chown %s", c.config.ShmDir) } - - if !mounted { - shmOptions := fmt.Sprintf("mode=1777,size=%d", c.config.ShmSize) - if err := c.mountSHM(shmOptions); err != nil { - return "", err - } - if err := os.Chown(c.config.ShmDir, c.RootUID(), c.RootGID()); err != nil { - return "", errors.Wrapf(err, "failed to chown %s", c.config.ShmDir) - } - } } + // TODO: generalize this mount code so it will mount every mount in ctr.config.Mounts mountPoint := c.config.Rootfs if mountPoint == "" { mountPoint, err = c.mount() @@ -916,86 +995,6 @@ func (c *Container) postDeleteHooks(ctx context.Context) (err error) { return nil } -// Make standard bind mounts to include in the container -func (c *Container) makeBindMounts() error { - if err := os.Chown(c.state.RunDir, c.RootUID(), c.RootGID()); err != nil { - return errors.Wrapf(err, "cannot chown run directory %s", c.state.RunDir) - } - - if c.state.BindMounts == nil { - c.state.BindMounts = make(map[string]string) - } - - // SHM is always added when we mount the container - c.state.BindMounts["/dev/shm"] = c.config.ShmDir - - // Make /etc/resolv.conf - if _, ok := c.state.BindMounts["/etc/resolv.conf"]; ok { - // If it already exists, delete so we can recreate - delete(c.state.BindMounts, "/etc/resolv.conf") - } - newResolv, err := c.generateResolvConf() - if err != nil { - return errors.Wrapf(err, "error creating resolv.conf for container %s", c.ID()) - } - c.state.BindMounts["/etc/resolv.conf"] = newResolv - - newPasswd, err := c.generatePasswd() - if err != nil { - return errors.Wrapf(err, "error creating temporary passwd file for container %s", c.ID()) - } - if newPasswd != "" { - // Make /etc/passwd - if _, ok := c.state.BindMounts["/etc/passwd"]; ok { - // If it already exists, delete so we can recreate - delete(c.state.BindMounts, "/etc/passwd") - } - logrus.Debugf("adding entry to /etc/passwd for non existent default user") - c.state.BindMounts["/etc/passwd"] = newPasswd - } - // Make /etc/hosts - if _, ok := c.state.BindMounts["/etc/hosts"]; ok { - // If it already exists, delete so we can recreate - delete(c.state.BindMounts, "/etc/hosts") - } - newHosts, err := c.generateHosts() - if err != nil { - return errors.Wrapf(err, "error creating hosts file for container %s", c.ID()) - } - c.state.BindMounts["/etc/hosts"] = newHosts - - // Make /etc/hostname - // This should never change, so no need to recreate if it exists - if _, ok := c.state.BindMounts["/etc/hostname"]; !ok { - hostnamePath, err := c.writeStringToRundir("hostname", c.Hostname()) - if err != nil { - return errors.Wrapf(err, "error creating hostname file for container %s", c.ID()) - } - c.state.BindMounts["/etc/hostname"] = hostnamePath - } - - // Make .containerenv - // Empty file, so no need to recreate if it exists - if _, ok := c.state.BindMounts["/run/.containerenv"]; !ok { - // Empty string for now, but we may consider populating this later - containerenvPath, err := c.writeStringToRundir(".containerenv", "") - if err != nil { - return errors.Wrapf(err, "error creating containerenv file for container %s", c.ID()) - } - c.state.BindMounts["/run/.containerenv"] = containerenvPath - } - - // Add Secret Mounts - secretMounts := secrets.SecretMountsWithUIDGID(c.config.MountLabel, c.state.RunDir, c.runtime.config.DefaultMountsFile, c.state.DestinationRunDir, c.RootUID(), c.RootGID()) - for _, mount := range secretMounts { - if _, ok := c.state.BindMounts[mount.Destination]; !ok { - c.state.BindMounts[mount.Destination] = mount.Source - } - } - - return nil -} - // writeStringToRundir copies the provided file to the runtimedir func (c *Container) writeStringToRundir(destFile, output string) (string, error) { destFileName := filepath.Join(c.state.RunDir, destFile) @@ -1024,135 +1023,8 @@ func (c *Container) writeStringToRundir(destFile, output string) (string, error) return filepath.Join(c.state.DestinationRunDir, destFile), nil } -// generatePasswd generates a container specific passwd file, -// iff g.config.User is a number -func (c *Container) generatePasswd() (string, error) { - var ( - groupspec string - gid uint32 - ) - if c.config.User == "" { - return "", nil - } - spec := strings.SplitN(c.config.User, ":", 2) - userspec := spec[0] - if len(spec) > 1 { - groupspec = spec[1] - } - // If a non numeric User, then don't generate passwd - uid, err := strconv.ParseUint(userspec, 10, 32) - if err != nil { - return "", nil - } - // if UID exists inside of container rootfs /etc/passwd then - // don't generate passwd - if _, _, err := chrootuser.LookupUIDInContainer(c.state.Mountpoint, uid); err == nil { - return "", nil - } - if err == nil && groupspec != "" { - if !c.state.Mounted { - return "", errors.Wrapf(ErrCtrStateInvalid, "container %s must be mounted in order to translate group field for passwd record", c.ID()) - } - gid, err = chrootuser.GetGroup(c.state.Mountpoint, groupspec) - if err != nil { - return "", errors.Wrapf(err, "unable to get gid from %s formporary passwd file") - } - } - - originPasswdFile := filepath.Join(c.state.Mountpoint, "/etc/passwd") - orig, err := ioutil.ReadFile(originPasswdFile) - if err != nil { - return "", errors.Wrapf(err, "unable to read passwd file %s", originPasswdFile) - } - - pwd := fmt.Sprintf("%s%d:x:%d:%d:container user:%s:/bin/sh\n", orig, uid, uid, gid, c.WorkingDir()) - passwdFile, err := c.writeStringToRundir("passwd", pwd) - if err != nil { - return "", errors.Wrapf(err, "failed to create temporary passwd file") - } - if os.Chmod(passwdFile, 0644); err != nil { - return "", err - } - return passwdFile, nil -} - -// generateResolvConf generates a containers resolv.conf -func (c *Container) generateResolvConf() (string, error) { - // Determine the endpoint for resolv.conf in case it is a symlink - resolvPath, err := filepath.EvalSymlinks("/etc/resolv.conf") - if err != nil { - return "", err - } - - contents, err := ioutil.ReadFile(resolvPath) - if err != nil { - return "", errors.Wrapf(err, "unable to read %s", resolvPath) - } - - // Process the file to remove localhost nameservers - // TODO: set ipv6 enable bool more sanely - resolv, err := resolvconf.FilterResolvDNS(contents, true) - if err != nil { - return "", errors.Wrapf(err, "error parsing host resolv.conf") - } - - // Make a new resolv.conf - nameservers := resolvconf.GetNameservers(resolv.Content) - if len(c.config.DNSServer) > 0 { - // We store DNS servers as net.IP, so need to convert to string - nameservers = []string{} - for _, server := range c.config.DNSServer { - nameservers = append(nameservers, server.String()) - } - } - - search := resolvconf.GetSearchDomains(resolv.Content) - if len(c.config.DNSSearch) > 0 { - search = c.config.DNSSearch - } - - options := resolvconf.GetOptions(resolv.Content) - if len(c.config.DNSOption) > 0 { - options = c.config.DNSOption - } - - destPath := filepath.Join(c.state.RunDir, "resolv.conf") - - if err := os.Remove(destPath); err != nil && !os.IsNotExist(err) { - return "", errors.Wrapf(err, "error removing resolv.conf for container %s", c.ID()) - } - - // Build resolv.conf - if _, err = resolvconf.Build(destPath, nameservers, search, options); err != nil { - return "", errors.Wrapf(err, "error building resolv.conf for container %s") - } - - // Relabel resolv.conf for the container - if err := label.Relabel(destPath, c.config.MountLabel, false); err != nil { - return "", err - } - - return filepath.Join(c.state.DestinationRunDir, "resolv.conf"), nil -} - -// generateHosts creates a containers hosts file -func (c *Container) generateHosts() (string, error) { - orig, err := ioutil.ReadFile("/etc/hosts") - if err != nil { - return "", errors.Wrapf(err, "unable to read /etc/hosts") - } - hosts := string(orig) - if len(c.config.HostAdd) > 0 { - for _, host := range c.config.HostAdd { - // the host format has already been verified at this point - fields := strings.SplitN(host, ":", 2) - hosts += fmt.Sprintf("%s %s\n", fields[1], fields[0]) - } - } - return c.writeStringToRundir("hosts", hosts) -} - func (c *Container) addLocalVolumes(ctx context.Context, g *generate.Generator) error { + var uid, gid int mountPoint := c.state.Mountpoint if !c.state.Mounted { return errors.Wrapf(ErrInternal, "container is not mounted") @@ -1176,6 +1048,18 @@ func (c *Container) addLocalVolumes(ctx context.Context, g *generate.Generator) } } + if c.config.User != "" { + if !c.state.Mounted { + return errors.Wrapf(ErrCtrStateInvalid, "container %s must be mounted in order to translate User field", c.ID()) + } + execUser, err := lookup.GetUserGroupInfo(c.state.Mountpoint, c.config.User, nil) + if err != nil { + return err + } + uid = execUser.Uid + gid = execUser.Gid + } + for k := range imageData.ContainerConfig.Volumes { mount := spec.Mount{ Destination: k, @@ -1186,19 +1070,6 @@ func (c *Container) addLocalVolumes(ctx context.Context, g *generate.Generator) continue } volumePath := filepath.Join(c.config.StaticDir, "volumes", k) - var ( - uid uint32 - gid uint32 - ) - if c.config.User != "" { - if !c.state.Mounted { - return errors.Wrapf(ErrCtrStateInvalid, "container %s must be mounted in order to translate User field", c.ID()) - } - uid, gid, err = chrootuser.GetUser(c.state.Mountpoint, c.config.User) - if err != nil { - return err - } - } // Ensure the symlinks are resolved resolvedSymlink, err := imagebuildah.ResolveSymLink(mountPoint, k) @@ -1218,7 +1089,7 @@ func (c *Container) addLocalVolumes(ctx context.Context, g *generate.Generator) return errors.Wrapf(err, "error creating directory %q for volume %q in container %q", volumePath, k, c.ID()) } - if err = os.Chown(srcPath, int(uid), int(gid)); err != nil { + if err = os.Chown(srcPath, uid, gid); err != nil { return errors.Wrapf(err, "error chowning directory %q for volume %q in container %q", srcPath, k, c.ID()) } } @@ -1228,7 +1099,7 @@ func (c *Container) addLocalVolumes(ctx context.Context, g *generate.Generator) return errors.Wrapf(err, "error creating directory %q for volume %q in container %q", volumePath, k, c.ID()) } - if err = os.Chown(volumePath, int(uid), int(gid)); err != nil { + if err = os.Chown(volumePath, uid, gid); err != nil { return errors.Wrapf(err, "error chowning directory %q for volume %q in container %q", volumePath, k, c.ID()) } @@ -1297,10 +1168,6 @@ func (c *Container) saveSpec(spec *spec.Spec) error { } func (c *Container) setupOCIHooks(ctx context.Context, config *spec.Spec) (extensionStageHooks map[string][]spec.Hook, err error) { - if len(c.runtime.config.HooksDir) == 0 { - return nil, nil - } - var locale string var ok bool for _, envVar := range []string{ @@ -1328,25 +1195,39 @@ func (c *Container) setupOCIHooks(ctx context.Context, config *spec.Spec) (exten } } - allHooks := make(map[string][]spec.Hook) - for _, hDir := range c.runtime.config.HooksDir { - manager, err := hooks.New(ctx, []string{hDir}, []string{"poststop"}, lang) - if err != nil { - if c.runtime.config.HooksDirNotExistFatal || !os.IsNotExist(err) { - return nil, err - } - logrus.Warnf("failed to load hooks: {}", err) + if c.runtime.config.HooksDir == nil { + if rootless.IsRootless() { return nil, nil } - hooks, err := manager.Hooks(config, c.Spec().Annotations, len(c.config.UserVolumes) > 0) - if err != nil { - return nil, err - } - for i, hook := range hooks { - allHooks[i] = hook + allHooks := make(map[string][]spec.Hook) + for _, hDir := range []string{hooks.DefaultDir, hooks.OverrideDir} { + manager, err := hooks.New(ctx, []string{hDir}, []string{"poststop"}, lang) + if err != nil { + if os.IsNotExist(err) { + continue + } + return nil, err + } + hooks, err := manager.Hooks(config, c.Spec().Annotations, len(c.config.UserVolumes) > 0) + if err != nil { + return nil, err + } + if len(hooks) > 0 || config.Hooks != nil { + logrus.Warnf("implicit hook directories are deprecated; set --hooks-dir=%q explicitly to continue to load hooks from this directory", hDir) + } + for i, hook := range hooks { + allHooks[i] = hook + } } + return allHooks, nil } - return allHooks, nil + + manager, err := hooks.New(ctx, c.runtime.config.HooksDir, []string{"poststop"}, lang) + if err != nil { + return nil, err + } + + return manager.Hooks(config, c.Spec().Annotations, len(c.config.UserVolumes) > 0) } // mount mounts the container's root filesystem diff --git a/libpod/container_internal_linux.go b/libpod/container_internal_linux.go index 0a1784ba7..b540bbeb8 100644 --- a/libpod/container_internal_linux.go +++ b/libpod/container_internal_linux.go @@ -11,6 +11,7 @@ import ( "os" "path" "path/filepath" + "strconv" "strings" "sync" "syscall" @@ -19,11 +20,12 @@ import ( cnitypes "github.com/containernetworking/cni/pkg/types/current" "github.com/containernetworking/plugins/pkg/ns" crioAnnotations "github.com/containers/libpod/pkg/annotations" - "github.com/containers/libpod/pkg/chrootuser" "github.com/containers/libpod/pkg/criu" + "github.com/containers/libpod/pkg/lookup" + "github.com/containers/libpod/pkg/resolvconf" "github.com/containers/libpod/pkg/rootless" + "github.com/containers/libpod/pkg/secrets" "github.com/containers/storage/pkg/idtools" - "github.com/cyphar/filepath-securejoin" "github.com/opencontainers/runc/libcontainer/user" spec "github.com/opencontainers/runtime-spec/specs-go" "github.com/opencontainers/runtime-tools/generate" @@ -59,7 +61,7 @@ func (c *Container) prepare() (err error) { networkStatus []*cnitypes.Result createNetNSErr, mountStorageErr error mountPoint string - saveNetworkStatus bool + tmpStateLock sync.Mutex ) wg.Add(2) @@ -68,17 +70,55 @@ func (c *Container) prepare() (err error) { defer wg.Done() // Set up network namespace if not already set up if c.config.CreateNetNS && c.state.NetNS == nil && !c.config.PostConfigureNetNS { - saveNetworkStatus = true netNS, networkStatus, createNetNSErr = c.runtime.createNetNS(c) + + tmpStateLock.Lock() + defer tmpStateLock.Unlock() + + // Assign NetNS attributes to container + if createNetNSErr == nil { + c.state.NetNS = netNS + c.state.NetworkStatus = networkStatus + } } }() // Mount storage if not mounted go func() { defer wg.Done() mountPoint, mountStorageErr = c.mountStorage() + + if mountStorageErr != nil { + return + } + + tmpStateLock.Lock() + defer tmpStateLock.Unlock() + + // Finish up mountStorage + c.state.Mounted = true + c.state.Mountpoint = mountPoint + if c.state.UserNSRoot == "" { + c.state.RealMountpoint = c.state.Mountpoint + } else { + c.state.RealMountpoint = filepath.Join(c.state.UserNSRoot, "mountpoint") + } + + logrus.Debugf("Created root filesystem for container %s at %s", c.ID(), c.state.Mountpoint) + }() + + defer func() { + if err != nil { + if err2 := c.cleanupNetwork(); err2 != nil { + logrus.Errorf("Error cleaning up container %s network: %v", c.ID(), err2) + } + if err2 := c.cleanupStorage(); err2 != nil { + logrus.Errorf("Error cleaning up container %s storage: %v", c.ID(), err2) + } + } }() wg.Wait() + if createNetNSErr != nil { if mountStorageErr != nil { logrus.Error(createNetNSErr) @@ -90,28 +130,15 @@ func (c *Container) prepare() (err error) { return mountStorageErr } - // Assign NetNS attributes to container - if saveNetworkStatus { - c.state.NetNS = netNS - c.state.NetworkStatus = networkStatus - } - - // Finish up mountStorage - c.state.Mounted = true - c.state.Mountpoint = mountPoint - if c.state.UserNSRoot == "" { - c.state.RealMountpoint = c.state.Mountpoint - } else { - c.state.RealMountpoint = filepath.Join(c.state.UserNSRoot, "mountpoint") - } - - logrus.Debugf("Created root filesystem for container %s at %s", c.ID(), c.state.Mountpoint) // Save the container return c.save() } // cleanupNetwork unmounts and cleans up the container's network func (c *Container) cleanupNetwork() error { + if c.NetworkDisabled() { + return nil + } if c.state.NetNS == nil { logrus.Debugf("Network is already cleaned up, skipping...") return nil @@ -135,6 +162,10 @@ func (c *Container) cleanupNetwork() error { // Generate spec for a container // Accepts a map of the container's dependencies func (c *Container) generateSpec(ctx context.Context) (*spec.Spec, error) { + execUser, err := lookup.GetUserGroupInfo(c.state.Mountpoint, c.config.User, nil) + if err != nil { + return nil, err + } g := generate.NewFromSpec(c.config.Spec) // If network namespace was requested, add it now @@ -145,6 +176,11 @@ func (c *Container) generateSpec(ctx context.Context) (*spec.Spec, error) { g.AddOrReplaceLinuxNamespace(spec.NetworkNamespace, c.state.NetNS.Path()) } } + + if err := c.makeBindMounts(); err != nil { + return nil, err + } + // Check if the spec file mounts contain the label Relabel flags z or Z. // If they do, relabel the source directory and then remove the option. for _, m := range g.Mounts() { @@ -188,11 +224,8 @@ func (c *Container) generateSpec(ctx context.Context) (*spec.Spec, error) { } } - var err error - if !rootless.IsRootless() { - if c.state.ExtensionStageHooks, err = c.setupOCIHooks(ctx, g.Config); err != nil { - return nil, errors.Wrapf(err, "error setting up OCI Hooks") - } + if c.state.ExtensionStageHooks, err = c.setupOCIHooks(ctx, g.Config); err != nil { + return nil, errors.Wrapf(err, "error setting up OCI Hooks") } // Bind builtin image volumes @@ -203,28 +236,15 @@ func (c *Container) generateSpec(ctx context.Context) (*spec.Spec, error) { } if c.config.User != "" { - if !c.state.Mounted { - return nil, errors.Wrapf(ErrCtrStateInvalid, "container %s must be mounted in order to translate User field", c.ID()) - } - uid, gid, err := chrootuser.GetUser(c.state.Mountpoint, c.config.User) - if err != nil { - return nil, err - } // User and Group must go together - g.SetProcessUID(uid) - g.SetProcessGID(gid) + g.SetProcessUID(uint32(execUser.Uid)) + g.SetProcessGID(uint32(execUser.Gid)) } // Add addition groups if c.config.GroupAdd is not empty if len(c.config.Groups) > 0 { - if !c.state.Mounted { - return nil, errors.Wrapf(ErrCtrStateInvalid, "container %s must be mounted in order to add additional groups", c.ID()) - } - for _, group := range c.config.Groups { - gid, err := chrootuser.GetGroup(c.state.Mountpoint, group) - if err != nil { - return nil, err - } + gids, _ := lookup.GetContainerGroups(c.config.Groups, c.state.Mountpoint, nil) + for _, gid := range gids { g.AddProcessAdditionalGid(gid) } } @@ -237,26 +257,6 @@ func (c *Container) generateSpec(ctx context.Context) (*spec.Spec, error) { // Look up and add groups the user belongs to, if a group wasn't directly specified if !rootless.IsRootless() && !strings.Contains(c.config.User, ":") { - var groupDest, passwdDest string - defaultExecUser := user.ExecUser{ - Uid: 0, - Gid: 0, - Home: "/", - } - - // Make sure the /etc/group and /etc/passwd destinations are not a symlink to something naughty - if groupDest, err = securejoin.SecureJoin(c.state.Mountpoint, "/etc/group"); err != nil { - logrus.Debug(err) - return nil, err - } - if passwdDest, err = securejoin.SecureJoin(c.state.Mountpoint, "/etc/passwd"); err != nil { - logrus.Debug(err) - return nil, err - } - execUser, err := user.GetExecUserPath(c.config.User, &defaultExecUser, passwdDest, groupDest) - if err != nil { - return nil, err - } for _, gid := range execUser.Sgids { g.AddProcessAdditionalGid(uint32(gid)) } @@ -351,8 +351,34 @@ func (c *Container) generateSpec(ctx context.Context) (*spec.Spec, error) { // Mounts need to be sorted so paths will not cover other paths mounts := sortMounts(g.Mounts()) g.ClearMounts() + + // Determine property of RootPropagation based on volume properties. If + // a volume is shared, then keep root propagation shared. This should + // work for slave and private volumes too. + // + // For slave volumes, it can be either [r]shared/[r]slave. + // + // For private volumes any root propagation value should work. + rootPropagation := "" for _, m := range mounts { g.AddMount(m) + for _, opt := range m.Options { + switch opt { + case MountShared, MountRShared: + if rootPropagation != MountShared && rootPropagation != MountRShared { + rootPropagation = MountShared + } + case MountSlave, MountRSlave: + if rootPropagation != MountShared && rootPropagation != MountRShared && rootPropagation != MountSlave && rootPropagation != MountRSlave { + rootPropagation = MountRSlave + } + } + } + } + + if rootPropagation != "" { + logrus.Debugf("set root propagation to %q", rootPropagation) + g.SetLinuxRootPropagation(rootPropagation) } return g.Config, nil } @@ -386,19 +412,31 @@ func (c *Container) setupSystemd(mounts []spec.Mount, g generate.Generator) erro g.AddMount(tmpfsMnt) } - cgroupPath, err := c.CGroupPath() - if err != nil { - return err - } - sourcePath := filepath.Join("/sys/fs/cgroup/systemd", cgroupPath) + // rootless containers have no write access to /sys/fs/cgroup, so don't + // add any mount into the container. + if !rootless.IsRootless() { + cgroupPath, err := c.CGroupPath() + if err != nil { + return err + } + sourcePath := filepath.Join("/sys/fs/cgroup/systemd", cgroupPath) - systemdMnt := spec.Mount{ - Destination: "/sys/fs/cgroup/systemd", - Type: "bind", - Source: sourcePath, - Options: []string{"bind", "private"}, + systemdMnt := spec.Mount{ + Destination: "/sys/fs/cgroup/systemd", + Type: "bind", + Source: sourcePath, + Options: []string{"bind", "private"}, + } + g.AddMount(systemdMnt) + } else { + systemdMnt := spec.Mount{ + Destination: "/sys/fs/cgroup/systemd", + Type: "bind", + Source: "/sys/fs/cgroup/systemd", + Options: []string{"bind", "nodev", "noexec", "nosuid"}, + } + g.AddMount(systemdMnt) } - g.AddMount(systemdMnt) return nil } @@ -423,7 +461,7 @@ func (c *Container) addNamespaceContainer(g *generate.Generator, ns LinuxNS, ctr return nil } -func (c *Container) checkpoint(ctx context.Context, keep bool) (err error) { +func (c *Container) checkpoint(ctx context.Context, options ContainerCheckpointOptions) (err error) { if !criu.CheckForCriu() { return errors.Errorf("checkpointing a container requires at least CRIU %d", criu.MinCriuVersion) @@ -432,7 +470,7 @@ func (c *Container) checkpoint(ctx context.Context, keep bool) (err error) { if c.state.State != ContainerStateRunning { return errors.Wrapf(ErrCtrStateInvalid, "%q is not running, cannot checkpoint", c.state.State) } - if err := c.runtime.ociRuntime.checkpointContainer(c); err != nil { + if err := c.runtime.ociRuntime.checkpointContainer(c, options); err != nil { return err } @@ -449,14 +487,16 @@ func (c *Container) checkpoint(ctx context.Context, keep bool) (err error) { logrus.Debugf("Checkpointed container %s", c.ID()) - c.state.State = ContainerStateStopped + if !options.KeepRunning { + c.state.State = ContainerStateStopped - // Cleanup Storage and Network - if err := c.cleanup(ctx); err != nil { - return err + // Cleanup Storage and Network + if err := c.cleanup(ctx); err != nil { + return err + } } - if !keep { + if !options.Keep { // Remove log file os.Remove(filepath.Join(c.bundlePath(), "dump.log")) // Remove statistic file @@ -466,7 +506,7 @@ func (c *Container) checkpoint(ctx context.Context, keep bool) (err error) { 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) @@ -510,9 +550,6 @@ func (c *Container) restore(ctx context.Context, keep bool) (err error) { } } - if err := c.prepare(); err != nil { - return err - } defer func() { if err != nil { if err2 := c.cleanup(ctx); err2 != nil { @@ -521,6 +558,10 @@ func (c *Container) restore(ctx context.Context, keep bool) (err error) { } }() + if err := c.prepare(); err != nil { + return err + } + // TODO: use existing way to request static IPs, once it is merged in ocicni // https://github.com/cri-o/ocicni/pull/23/ @@ -553,7 +594,7 @@ 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 } @@ -561,7 +602,7 @@ 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 @@ -582,3 +623,225 @@ func (c *Container) restore(ctx context.Context, keep bool) (err error) { return c.save() } + +// Make standard bind mounts to include in the container +func (c *Container) makeBindMounts() error { + if err := os.Chown(c.state.RunDir, c.RootUID(), c.RootGID()); err != nil { + return errors.Wrapf(err, "cannot chown run directory %s", c.state.RunDir) + } + + if c.state.BindMounts == nil { + c.state.BindMounts = make(map[string]string) + } + + if !c.NetworkDisabled() { + // Make /etc/resolv.conf + if _, ok := c.state.BindMounts["/etc/resolv.conf"]; ok { + // If it already exists, delete so we can recreate + delete(c.state.BindMounts, "/etc/resolv.conf") + } + newResolv, err := c.generateResolvConf() + if err != nil { + return errors.Wrapf(err, "error creating resolv.conf for container %s", c.ID()) + } + c.state.BindMounts["/etc/resolv.conf"] = newResolv + + // Make /etc/hosts + if _, ok := c.state.BindMounts["/etc/hosts"]; ok { + // If it already exists, delete so we can recreate + delete(c.state.BindMounts, "/etc/hosts") + } + newHosts, err := c.generateHosts() + if err != nil { + return errors.Wrapf(err, "error creating hosts file for container %s", c.ID()) + } + c.state.BindMounts["/etc/hosts"] = newHosts + + } + + // SHM is always added when we mount the container + c.state.BindMounts["/dev/shm"] = c.config.ShmDir + + newPasswd, err := c.generatePasswd() + if err != nil { + return errors.Wrapf(err, "error creating temporary passwd file for container %s", c.ID()) + } + if newPasswd != "" { + // Make /etc/passwd + if _, ok := c.state.BindMounts["/etc/passwd"]; ok { + // If it already exists, delete so we can recreate + delete(c.state.BindMounts, "/etc/passwd") + } + logrus.Debugf("adding entry to /etc/passwd for non existent default user") + c.state.BindMounts["/etc/passwd"] = newPasswd + } + + // Make /etc/hostname + // This should never change, so no need to recreate if it exists + if _, ok := c.state.BindMounts["/etc/hostname"]; !ok { + hostnamePath, err := c.writeStringToRundir("hostname", c.Hostname()) + if err != nil { + return errors.Wrapf(err, "error creating hostname file for container %s", c.ID()) + } + c.state.BindMounts["/etc/hostname"] = hostnamePath + } + + // Make .containerenv + // Empty file, so no need to recreate if it exists + if _, ok := c.state.BindMounts["/run/.containerenv"]; !ok { + // Empty string for now, but we may consider populating this later + containerenvPath, err := c.writeStringToRundir(".containerenv", "") + if err != nil { + return errors.Wrapf(err, "error creating containerenv file for container %s", c.ID()) + } + c.state.BindMounts["/run/.containerenv"] = containerenvPath + } + + // Add Secret Mounts + secretMounts := secrets.SecretMountsWithUIDGID(c.config.MountLabel, c.state.RunDir, c.runtime.config.DefaultMountsFile, c.state.DestinationRunDir, c.RootUID(), c.RootGID()) + for _, mount := range secretMounts { + if _, ok := c.state.BindMounts[mount.Destination]; !ok { + c.state.BindMounts[mount.Destination] = mount.Source + } + } + + return nil +} + +// generateResolvConf generates a containers resolv.conf +func (c *Container) generateResolvConf() (string, error) { + // Determine the endpoint for resolv.conf in case it is a symlink + resolvPath, err := filepath.EvalSymlinks("/etc/resolv.conf") + if err != nil { + return "", err + } + + contents, err := ioutil.ReadFile(resolvPath) + if err != nil { + return "", errors.Wrapf(err, "unable to read %s", resolvPath) + } + + // Ensure that the container's /etc/resolv.conf is compatible with its + // network configuration. + // TODO: set ipv6 enable bool more sanely + resolv, err := resolvconf.FilterResolvDNS(contents, true, c.config.CreateNetNS) + if err != nil { + return "", errors.Wrapf(err, "error parsing host resolv.conf") + } + + // Make a new resolv.conf + nameservers := resolvconf.GetNameservers(resolv.Content) + if len(c.config.DNSServer) > 0 { + // We store DNS servers as net.IP, so need to convert to string + nameservers = []string{} + for _, server := range c.config.DNSServer { + nameservers = append(nameservers, server.String()) + } + } + + search := resolvconf.GetSearchDomains(resolv.Content) + if len(c.config.DNSSearch) > 0 { + search = c.config.DNSSearch + } + + options := resolvconf.GetOptions(resolv.Content) + if len(c.config.DNSOption) > 0 { + options = c.config.DNSOption + } + + destPath := filepath.Join(c.state.RunDir, "resolv.conf") + + if err := os.Remove(destPath); err != nil && !os.IsNotExist(err) { + return "", errors.Wrapf(err, "error removing resolv.conf for container %s", c.ID()) + } + + // Build resolv.conf + if _, err = resolvconf.Build(destPath, nameservers, search, options); err != nil { + return "", errors.Wrapf(err, "error building resolv.conf for container %s", c.ID()) + } + + // Relabel resolv.conf for the container + if err := label.Relabel(destPath, c.config.MountLabel, false); err != nil { + return "", err + } + + return filepath.Join(c.state.DestinationRunDir, "resolv.conf"), nil +} + +// generateHosts creates a containers hosts file +func (c *Container) generateHosts() (string, error) { + orig, err := ioutil.ReadFile("/etc/hosts") + if err != nil { + return "", errors.Wrapf(err, "unable to read /etc/hosts") + } + hosts := string(orig) + if len(c.config.HostAdd) > 0 { + for _, host := range c.config.HostAdd { + // the host format has already been verified at this point + fields := strings.SplitN(host, ":", 2) + hosts += fmt.Sprintf("%s %s\n", fields[1], fields[0]) + } + } + if len(c.state.NetworkStatus) > 0 && len(c.state.NetworkStatus[0].IPs) > 0 { + ipAddress := strings.Split(c.state.NetworkStatus[0].IPs[0].Address.String(), "/")[0] + hosts += fmt.Sprintf("%s\t%s\n", ipAddress, c.Hostname()) + } + return c.writeStringToRundir("hosts", hosts) +} + +// generatePasswd generates a container specific passwd file, +// iff g.config.User is a number +func (c *Container) generatePasswd() (string, error) { + var ( + groupspec string + gid int + ) + if c.config.User == "" { + return "", nil + } + spec := strings.SplitN(c.config.User, ":", 2) + userspec := spec[0] + if len(spec) > 1 { + groupspec = spec[1] + } + // If a non numeric User, then don't generate passwd + uid, err := strconv.ParseUint(userspec, 10, 32) + if err != nil { + return "", nil + } + // Lookup the user to see if it exists in the container image + _, err = lookup.GetUser(c.state.Mountpoint, userspec) + if err != nil && err != user.ErrNoPasswdEntries { + return "", err + } + if err == nil { + return "", nil + } + if groupspec != "" { + ugid, err := strconv.ParseUint(groupspec, 10, 32) + if err == nil { + gid = int(ugid) + } else { + group, err := lookup.GetGroup(c.state.Mountpoint, groupspec) + if err != nil { + return "", errors.Wrapf(err, "unable to get gid %s from group file", groupspec) + } + gid = group.Gid + } + } + originPasswdFile := filepath.Join(c.state.Mountpoint, "/etc/passwd") + orig, err := ioutil.ReadFile(originPasswdFile) + if err != nil && !os.IsNotExist(err) { + return "", errors.Wrapf(err, "unable to read passwd file %s", originPasswdFile) + } + + pwd := fmt.Sprintf("%s%d:x:%d:%d:container user:%s:/bin/sh\n", orig, uid, uid, gid, c.WorkingDir()) + passwdFile, err := c.writeStringToRundir("passwd", pwd) + if err != nil { + return "", errors.Wrapf(err, "failed to create temporary passwd file") + } + if os.Chmod(passwdFile, 0644); err != nil { + return "", err + } + return passwdFile, nil +} diff --git a/libpod/image/errors.go b/libpod/image/errors.go new file mode 100644 index 000000000..4088946cb --- /dev/null +++ b/libpod/image/errors.go @@ -0,0 +1,15 @@ +package image + +import ( + "errors" +) + +// Copied directly from libpod errors to avoid circular imports +var ( + // ErrNoSuchCtr indicates the requested container does not exist + ErrNoSuchCtr = errors.New("no such container") + // ErrNoSuchPod indicates the requested pod does not exist + ErrNoSuchPod = errors.New("no such pod") + // ErrNoSuchImage indicates the requested image does not exist + ErrNoSuchImage = errors.New("no such image") +) diff --git a/libpod/image/image.go b/libpod/image/image.go index 7e520d97e..434f9031e 100644 --- a/libpod/image/image.go +++ b/libpod/image/image.go @@ -252,7 +252,7 @@ func (i *Image) getLocalImage() (*storage.Image, error) { // The image has a registry name in it and we made sure we looked for it locally // with a tag. It cannot be local. if decomposedImage.hasRegistry { - return nil, errors.Errorf("%s", imageError) + return nil, errors.Wrapf(ErrNoSuchImage, imageError) } @@ -275,7 +275,7 @@ func (i *Image) getLocalImage() (*storage.Image, error) { return repoImage, nil } - return nil, errors.Wrapf(err, imageError) + return nil, errors.Wrapf(ErrNoSuchImage, err.Error()) } // ID returns the image ID as a string @@ -869,6 +869,7 @@ func (i *Image) Inspect(ctx context.Context) (*inspect.ImageData, error) { GraphDriver: driver, ManifestType: manifestType, User: ociv1Img.Config.User, + History: ociv1Img.History, } return data, nil } diff --git a/libpod/in_memory_state.go b/libpod/in_memory_state.go index 78e765ccd..77eba0cc6 100644 --- a/libpod/in_memory_state.go +++ b/libpod/in_memory_state.go @@ -73,6 +73,18 @@ func (s *InMemoryState) Refresh() error { return nil } +// GetDBConfig is not implemented for in-memory state. +// As we do not store a config, return an empty one. +func (s *InMemoryState) GetDBConfig() (*DBConfig, error) { + return &DBConfig{}, nil +} + +// ValidateDBConfig is not implemented for the in-memory state. +// Since we do nothing just return no error. +func (s *InMemoryState) ValidateDBConfig(runtime *Runtime) error { + return nil +} + // SetNamespace sets the namespace for container and pod retrieval. func (s *InMemoryState) SetNamespace(ns string) error { s.namespace = ns diff --git a/libpod/info.go b/libpod/info.go index 4cbf3f734..5d8d160c8 100644 --- a/libpod/info.go +++ b/libpod/info.go @@ -12,6 +12,7 @@ import ( "strings" "time" + "github.com/containers/libpod/pkg/rootless" "github.com/containers/libpod/utils" "github.com/containers/storage/pkg/system" "github.com/pkg/errors" @@ -30,6 +31,7 @@ func (r *Runtime) hostInfo() (map[string]interface{}, error) { info["os"] = runtime.GOOS info["arch"] = runtime.GOARCH info["cpus"] = runtime.NumCPU() + info["rootless"] = rootless.IsRootless() mi, err := system.ReadMemInfo() if err != nil { return nil, errors.Wrapf(err, "error reading memory info") diff --git a/libpod/kube.go b/libpod/kube.go new file mode 100644 index 000000000..1a5f80878 --- /dev/null +++ b/libpod/kube.go @@ -0,0 +1,270 @@ +package libpod + +import ( + "fmt" + "strings" + + "github.com/containers/libpod/pkg/lookup" + "github.com/containers/libpod/pkg/util" + "github.com/cri-o/ocicni/pkg/ocicni" + "github.com/opencontainers/runtime-spec/specs-go" + "github.com/pkg/errors" + "github.com/sirupsen/logrus" + "k8s.io/api/core/v1" + "k8s.io/apimachinery/pkg/api/resource" + v12 "k8s.io/apimachinery/pkg/apis/meta/v1" +) + +// InspectForKube takes a slice of libpod containers and generates +// one v1.Pod description that includes just a single container. +func (c *Container) InspectForKube() (*v1.Pod, error) { + // Generate the v1.Pod yaml description + return simplePodWithV1Container(c) +} + +// simplePodWithV1Container is a function used by inspect when kube yaml needs to be generated +// for a single container. we "insert" that container description in a pod. +func simplePodWithV1Container(ctr *Container) (*v1.Pod, error) { + var containers []v1.Container + result, err := containerToV1Container(ctr) + if err != nil { + return nil, err + } + containers = append(containers, result) + + tm := v12.TypeMeta{ + Kind: "Pod", + APIVersion: "v1", + } + + // Add a label called "app" with the containers name as a value + labels := make(map[string]string) + labels["app"] = removeUnderscores(ctr.Name()) + om := v12.ObjectMeta{ + // The name of the pod is container_name-libpod + Name: fmt.Sprintf("%s-libpod", removeUnderscores(ctr.Name())), + Labels: labels, + // CreationTimestamp seems to be required, so adding it; in doing so, the timestamp + // will reflect time this is run (not container create time) because the conversion + // of the container create time to v1 Time is probably not warranted nor worthwhile. + CreationTimestamp: v12.Now(), + } + ps := v1.PodSpec{ + Containers: containers, + } + p := v1.Pod{ + TypeMeta: tm, + ObjectMeta: om, + Spec: ps, + } + return &p, nil +} + +// containerToV1Container converts information we know about a libpod container +// to a V1.Container specification. +func containerToV1Container(c *Container) (v1.Container, error) { + kubeContainer := v1.Container{} + kubeSec, err := generateKubeSecurityContext(c) + if err != nil { + return kubeContainer, err + } + + if len(c.config.Spec.Linux.Devices) > 0 { + // TODO Enable when we can support devices and their names + devices, err := generateKubeVolumeDeviceFromLinuxDevice(c.Spec().Linux.Devices) + if err != nil { + return kubeContainer, err + } + kubeContainer.VolumeDevices = devices + return kubeContainer, errors.Wrapf(ErrNotImplemented, "linux devices") + } + + if len(c.config.UserVolumes) > 0 { + // TODO When we until we can resolve what the volume name should be, this is disabled + // Volume names need to be coordinated "globally" in the kube files. + volumes, err := libpodMountsToKubeVolumeMounts(c) + if err != nil { + return kubeContainer, err + } + kubeContainer.VolumeMounts = volumes + return kubeContainer, errors.Wrapf(ErrNotImplemented, "volume names") + } + + envVariables, err := libpodEnvVarsToKubeEnvVars(c.config.Spec.Process.Env) + if err != nil { + return kubeContainer, nil + } + + ports, err := ocicniPortMappingToContainerPort(c.PortMappings()) + if err != nil { + return kubeContainer, nil + } + + containerCommands := c.Command() + kubeContainer.Name = removeUnderscores(c.Name()) + + _, image := c.Image() + kubeContainer.Image = image + kubeContainer.Stdin = c.Stdin() + kubeContainer.Command = containerCommands + // TODO need to figure out how we handle command vs entry point. Kube appears to prefer entrypoint. + // right now we just take the container's command + //container.Args = args + kubeContainer.WorkingDir = c.WorkingDir() + kubeContainer.Ports = ports + // This should not be applicable + //container.EnvFromSource = + kubeContainer.Env = envVariables + // TODO enable resources when we can support naming conventions + //container.Resources + kubeContainer.SecurityContext = kubeSec + kubeContainer.StdinOnce = false + kubeContainer.TTY = c.config.Spec.Process.Terminal + + return kubeContainer, nil +} + +// ocicniPortMappingToContainerPort takes an ocicni portmapping and converts +// it to a v1.ContainerPort format for kube output +func ocicniPortMappingToContainerPort(portMappings []ocicni.PortMapping) ([]v1.ContainerPort, error) { + var containerPorts []v1.ContainerPort + for _, p := range portMappings { + var protocol v1.Protocol + switch strings.ToUpper(p.Protocol) { + case "TCP": + protocol = v1.ProtocolTCP + case "UDP": + protocol = v1.ProtocolUDP + default: + return containerPorts, errors.Errorf("unknown network protocol %s", p.Protocol) + } + cp := v1.ContainerPort{ + // Name will not be supported + HostPort: p.HostPort, + HostIP: p.HostIP, + ContainerPort: p.ContainerPort, + Protocol: protocol, + } + containerPorts = append(containerPorts, cp) + } + return containerPorts, nil +} + +// libpodEnvVarsToKubeEnvVars converts a key=value string slice to []v1.EnvVar +func libpodEnvVarsToKubeEnvVars(envs []string) ([]v1.EnvVar, error) { + var envVars []v1.EnvVar + for _, e := range envs { + splitE := strings.SplitN(e, "=", 2) + if len(splitE) != 2 { + return envVars, errors.Errorf("environment variable %s is malformed; should be key=value", e) + } + ev := v1.EnvVar{ + Name: splitE[0], + Value: splitE[1], + } + envVars = append(envVars, ev) + } + return envVars, nil +} + +// Is this worth it? +func libpodMaxAndMinToResourceList(c *Container) (v1.ResourceList, v1.ResourceList) { //nolint + // It does not appear we can properly calculate CPU resources from the information + // we know in libpod. Libpod knows CPUs by time, shares, etc. + + // We also only know about a memory limit; no memory minimum + maxResources := make(map[v1.ResourceName]resource.Quantity) + minResources := make(map[v1.ResourceName]resource.Quantity) + config := c.Config() + maxMem := config.Spec.Linux.Resources.Memory.Limit + + _ = maxMem + + return maxResources, minResources +} + +func generateKubeVolumeMount(hostSourcePath string, mounts []specs.Mount) (v1.VolumeMount, error) { + vm := v1.VolumeMount{} + for _, m := range mounts { + if m.Source == hostSourcePath { + // TODO Name is not provided and is required by Kube; therefore, this is disabled earlier + //vm.Name = + vm.MountPath = m.Source + vm.SubPath = m.Destination + if util.StringInSlice("ro", m.Options) { + vm.ReadOnly = true + } + return vm, nil + } + } + return vm, errors.New("unable to find mount source") +} + +// libpodMountsToKubeVolumeMounts converts the containers mounts to a struct kube understands +func libpodMountsToKubeVolumeMounts(c *Container) ([]v1.VolumeMount, error) { + // At this point, I dont think we can distinguish between the default + // volume mounts and user added ones. For now, we pass them all. + var vms []v1.VolumeMount + for _, hostSourcePath := range c.config.UserVolumes { + vm, err := generateKubeVolumeMount(hostSourcePath, c.config.Spec.Mounts) + if err != nil { + return vms, err + } + vms = append(vms, vm) + } + return vms, nil +} + +// generateKubeSecurityContext generates a securityContext based on the existing container +func generateKubeSecurityContext(c *Container) (*v1.SecurityContext, error) { + priv := c.Privileged() + ro := c.IsReadOnly() + allowPrivEscalation := !c.Spec().Process.NoNewPrivileges + + // TODO enable use of capabilities when we can figure out how to extract cap-add|remove + //caps := v1.Capabilities{ + // //Add: c.config.Spec.Process.Capabilities + //} + sc := v1.SecurityContext{ + // TODO enable use of capabilities when we can figure out how to extract cap-add|remove + //Capabilities: &caps, + Privileged: &priv, + // TODO How do we know if selinux were passed into podman + //SELinuxOptions: + // RunAsNonRoot is an optional parameter; our first implementations should be root only; however + // I'm leaving this as a bread-crumb for later + //RunAsNonRoot: &nonRoot, + ReadOnlyRootFilesystem: &ro, + AllowPrivilegeEscalation: &allowPrivEscalation, + } + + if c.User() != "" { + // It is *possible* that + logrus.Debugf("Looking in container for user: %s", c.User()) + u, err := lookup.GetUser(c.state.Mountpoint, c.User()) + if err != nil { + return nil, err + } + user := int64(u.Uid) + sc.RunAsUser = &user + } + return &sc, nil +} + +// generateKubeVolumeDeviceFromLinuxDevice takes a list of devices and makes a VolumeDevice struct for kube +func generateKubeVolumeDeviceFromLinuxDevice(devices []specs.LinuxDevice) ([]v1.VolumeDevice, error) { + var volumeDevices []v1.VolumeDevice + for _, d := range devices { + vd := v1.VolumeDevice{ + // TBD How are we going to sync up these names + //Name: + DevicePath: d.Path, + } + volumeDevices = append(volumeDevices, vd) + } + return volumeDevices, nil +} + +func removeUnderscores(s string) string { + return strings.Replace(s, "_", "", -1) +} diff --git a/libpod/mounts_linux.go b/libpod/mounts_linux.go new file mode 100644 index 000000000..e6aa09eac --- /dev/null +++ b/libpod/mounts_linux.go @@ -0,0 +1,18 @@ +// +build linux + +package libpod + +const ( + // MountPrivate represents the private mount option. + MountPrivate = "private" + // MountRPrivate represents the rprivate mount option. + MountRPrivate = "rprivate" + // MountShared represents the shared mount option. + MountShared = "shared" + // MountRShared represents the rshared mount option. + MountRShared = "rshared" + // MountSlave represents the slave mount option. + MountSlave = "slave" + // MountRSlave represents the rslave mount option. + MountRSlave = "rslave" +) diff --git a/libpod/networking_linux.go b/libpod/networking_linux.go index 0d9ec2809..43d0a61a4 100644 --- a/libpod/networking_linux.go +++ b/libpod/networking_linux.go @@ -12,6 +12,7 @@ import ( "strconv" "strings" "syscall" + "time" cnitypes "github.com/containernetworking/cni/pkg/types/current" "github.com/containernetworking/plugins/pkg/ns" @@ -63,20 +64,20 @@ func (r *Runtime) configureNetNS(ctr *Container, ctrNS ns.NetNS) ([]*cnitypes.Re } }() - networkStatus := make([]*cnitypes.Result, 1) + networkStatus := make([]*cnitypes.Result, 0) for idx, r := range results { logrus.Debugf("[%d] CNI result: %v", idx, r.String()) resultCurrent, err := cnitypes.GetResult(r) if err != nil { return nil, errors.Wrapf(err, "error parsing CNI plugin result %q: %v", r.String(), err) } - networkStatus = append(ctr.state.NetworkStatus, resultCurrent) + networkStatus = append(networkStatus, resultCurrent) } // Add firewall rules to ensure the container has network access. // Will not be necessary once CNI firewall plugin merges upstream. // https://github.com/containernetworking/plugins/pull/75 - for _, netStatus := range ctr.state.NetworkStatus { + for _, netStatus := range networkStatus { firewallConf := &firewall.FirewallNetConf{ PrevResult: netStatus, } @@ -89,13 +90,16 @@ func (r *Runtime) configureNetNS(ctr *Container, ctrNS ns.NetNS) ([]*cnitypes.Re } // Create and configure a new network namespace for a container -func (r *Runtime) createNetNS(ctr *Container) (ns.NetNS, []*cnitypes.Result, error) { +func (r *Runtime) createNetNS(ctr *Container) (n ns.NetNS, q []*cnitypes.Result, err error) { ctrNS, err := netns.NewNS() if err != nil { return nil, nil, errors.Wrapf(err, "error creating network namespace for container %s", ctr.ID()) } defer func() { if err != nil { + if err2 := netns.UnmountNS(ctrNS); err2 != nil { + logrus.Errorf("Error unmounting partially created network namespace for container %s: %v", ctr.ID(), err2) + } if err2 := ctrNS.Close(); err2 != nil { logrus.Errorf("Error closing partially created network namespace for container %s: %v", ctr.ID(), err2) } @@ -134,12 +138,33 @@ func (r *Runtime) setupRootlessNetNS(ctr *Container) (err error) { cmd.ExtraFiles = append(cmd.ExtraFiles, ctr.rootlessSlirpSyncR, syncW) if err := cmd.Start(); err != nil { - return errors.Wrapf(err, "failed to start process") + return errors.Wrapf(err, "failed to start slirp4netns process") } + defer cmd.Process.Release() b := make([]byte, 16) - if _, err := syncR.Read(b); err != nil { - return errors.Wrapf(err, "failed to read from sync pipe") + for { + if err := syncR.SetDeadline(time.Now().Add(1 * time.Second)); err != nil { + return errors.Wrapf(err, "error setting slirp4netns pipe timeout") + } + if _, err := syncR.Read(b); err == nil { + break + } else { + if os.IsTimeout(err) { + // Check if the process is still running. + var status syscall.WaitStatus + _, err := syscall.Wait4(cmd.Process.Pid, &status, syscall.WNOHANG, nil) + if err != nil { + return errors.Wrapf(err, "failed to read slirp4netns process status") + } + if status.Exited() || status.Signaled() { + return errors.New("slirp4netns failed") + } + + continue + } + return errors.Wrapf(err, "failed to read from slirp4netns sync pipe") + } } return nil } diff --git a/libpod/oci.go b/libpod/oci.go index ca8f967c4..3222f9403 100644 --- a/libpod/oci.go +++ b/libpod/oci.go @@ -11,12 +11,10 @@ import ( "os/exec" "path/filepath" "runtime" - "strconv" "strings" "syscall" "time" - "github.com/containers/libpod/pkg/ctime" "github.com/containers/libpod/pkg/rootless" "github.com/containers/libpod/pkg/util" "github.com/coreos/go-systemd/activation" @@ -229,7 +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() @@ -291,8 +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{ @@ -318,6 +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) @@ -331,7 +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") @@ -352,7 +357,8 @@ func (r *OCIRuntime) createOCIContainer(ctr *Container, cgroupParent string, res // Set the label of the conmon process to be level :s0 // This will allow the container processes to talk to fifo-files // passed into the container by conmon - plabel, err := selinux.CurrentLabel() + var plabel string + plabel, err = selinux.CurrentLabel() if err != nil { childPipe.Close() return errors.Wrapf(err, "Failed to get current SELinux label") @@ -362,7 +368,7 @@ func (r *OCIRuntime) createOCIContainer(ctr *Container, cgroupParent string, res runtime.LockOSThread() if c["level"] != "s0" && c["level"] != "" { c["level"] = "s0" - if err := label.SetProcessLabel(c.Get()); err != nil { + if err = label.SetProcessLabel(c.Get()); err != nil { runtime.UnlockOSThread() return err } @@ -443,6 +449,7 @@ func (r *OCIRuntime) createOCIContainer(ctr *Container, cgroupParent string, res } return errors.Wrapf(ErrInternal, "container create failed") } + ctr.state.PID = ss.si.Pid case <-time.After(ContainerCreateTimeout): return errors.Wrapf(ErrInternal, "container creation timeout") } @@ -451,17 +458,47 @@ func (r *OCIRuntime) createOCIContainer(ctr *Container, cgroupParent string, res // updateContainerStatus retrieves the current status of the container from the // runtime. It updates the container's state but does not save it. -func (r *OCIRuntime) updateContainerStatus(ctr *Container) error { - state := new(spec.State) +// If useRunc is false, we will not directly hit runc to see the container's +// status, but will instead only check for the existence of the conmon exit file +// and update state to stopped if it exists. +func (r *OCIRuntime) updateContainerStatus(ctr *Container, useRunc bool) error { + exitFile := ctr.exitFilePath() runtimeDir, err := util.GetRootlessRuntimeDir() if err != nil { return err } + // If not using runc, we don't need to do most of this. + if !useRunc { + // If the container's not running, nothing to do. + if ctr.state.State != ContainerStateRunning { + return nil + } + + // Check for the exit file conmon makes + info, err := os.Stat(exitFile) + if err != nil { + if os.IsNotExist(err) { + // Container is still running, no error + return nil + } + + return errors.Wrapf(err, "error running stat on container %s exit file", ctr.ID()) + } + + // Alright, it exists. Transition to Stopped state. + ctr.state.State = ContainerStateStopped + + // Read the exit file to get our stopped time and exit code. + return ctr.handleExitFile(exitFile, info) + } + // Store old state so we know if we were already stopped oldState := ctr.state.State + state := new(spec.State) + cmd := exec.Command(r.path, "state", ctr.ID()) cmd.Env = append(cmd.Env, fmt.Sprintf("XDG_RUNTIME_DIR=%s", runtimeDir)) outPipe, err := cmd.StdoutPipe() @@ -480,6 +517,8 @@ func (r *OCIRuntime) updateContainerStatus(ctr *Container) error { } if strings.Contains(string(out), "does not exist") { ctr.removeConmonFiles() + ctr.state.ExitCode = -1 + ctr.state.FinishedTime = time.Now() ctr.state.State = ContainerStateExited return nil } @@ -514,7 +553,6 @@ func (r *OCIRuntime) updateContainerStatus(ctr *Container) error { // Only grab exit status if we were not already stopped // If we were, it should already be in the database if ctr.state.State == ContainerStateStopped && oldState != ContainerStateStopped { - exitFile := filepath.Join(r.exitsDir, ctr.ID()) var fi os.FileInfo err = kwait.ExponentialBackoff( kwait.Backoff{ @@ -538,24 +576,7 @@ func (r *OCIRuntime) updateContainerStatus(ctr *Container) error { return nil } - ctr.state.FinishedTime = ctime.Created(fi) - statusCodeStr, err := ioutil.ReadFile(exitFile) - if err != nil { - return errors.Wrapf(err, "failed to read exit file for container %s", ctr.ID()) - } - statusCode, err := strconv.Atoi(string(statusCodeStr)) - if err != nil { - return errors.Wrapf(err, "error converting exit status code for container %s to int", - ctr.ID()) - } - ctr.state.ExitCode = int32(statusCode) - - oomFilePath := filepath.Join(ctr.bundlePath(), "oom") - if _, err = os.Stat(oomFilePath); err == nil { - ctr.state.OOMKilled = true - } - - ctr.state.Exited = true + return ctr.handleExitFile(exitFile, fi) } return nil @@ -570,6 +591,9 @@ func (r *OCIRuntime) startContainer(ctr *Container) error { return err } env := []string{fmt.Sprintf("XDG_RUNTIME_DIR=%s", runtimeDir)} + if notify, ok := os.LookupEnv("NOTIFY_SOCKET"); ok { + env = append(env, fmt.Sprintf("NOTIFY_SOCKET=%s", notify)) + } if err := utils.ExecCmdWithStdStreams(os.Stdin, os.Stdout, os.Stderr, env, r.path, "start", ctr.ID()); err != nil { return err } @@ -601,6 +625,8 @@ func (r *OCIRuntime) killContainer(ctr *Container, signal uint) error { // Does not set finished time for container, assumes you will run updateStatus // after to pull the exit code func (r *OCIRuntime) stopContainer(ctr *Container, timeout uint) error { + logrus.Debugf("Stopping container %s (PID %d)", ctr.ID(), ctr.state.PID) + // Ping the container to see if it's alive // If it's not, it's already stopped, return err := unix.Kill(ctr.state.PID, 0) @@ -670,8 +696,12 @@ func (r *OCIRuntime) stopContainer(ctr *Container, timeout uint) error { // deleteContainer deletes a container from the OCI runtime func (r *OCIRuntime) deleteContainer(ctr *Container) error { - _, err := utils.ExecCmd(r.path, "delete", "--force", ctr.ID()) - return err + runtimeDir, err := util.GetRootlessRuntimeDir() + if err != nil { + return err + } + env := []string{fmt.Sprintf("XDG_RUNTIME_DIR=%s", runtimeDir)} + return utils.ExecCmdWithStdStreams(os.Stdin, os.Stdout, os.Stderr, env, r.path, "delete", "--force", ctr.ID()) } // pauseContainer pauses the given container @@ -725,6 +755,8 @@ func (r *OCIRuntime) execContainer(c *Container, cmd, capAdd, env []string, tty if tty { args = append(args, "--tty") + } else { + args = append(args, "--tty=false") } if user != "" { @@ -828,13 +860,25 @@ func (r *OCIRuntime) execStopContainer(ctr *Container, timeout uint) error { } // checkpointContainer checkpoints the given container -func (r *OCIRuntime) checkpointContainer(ctr *Container) error { +func (r *OCIRuntime) checkpointContainer(ctr *Container, options ContainerCheckpointOptions) error { // imagePath is used by CRIU to store the actual checkpoint files imagePath := ctr.CheckpointPath() // workPath will be used to store dump.log and stats-dump workPath := ctr.bundlePath() logrus.Debugf("Writing checkpoint to %s", imagePath) logrus.Debugf("Writing checkpoint logs to %s", workPath) - return utils.ExecCmdWithStdStreams(os.Stdin, os.Stdout, os.Stderr, nil, r.path, "checkpoint", - "--image-path", imagePath, "--work-path", workPath, ctr.ID()) + args := []string{} + args = append(args, "checkpoint") + args = append(args, "--image-path") + args = append(args, imagePath) + args = append(args, "--work-path") + args = append(args, workPath) + 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 0447670b3..2737a641e 100644 --- a/libpod/oci_linux.go +++ b/libpod/oci_linux.go @@ -19,6 +19,8 @@ import ( "golang.org/x/sys/unix" ) +const unknownPackage = "Unknown" + func (r *OCIRuntime) moveConmonToCgroup(ctr *Container, cgroupParent string, cmd *exec.Cmd) error { if os.Geteuid() == 0 { if r.cgroupManager == SystemdCgroupsManager { @@ -63,10 +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) @@ -74,7 +76,8 @@ func (r *OCIRuntime) createContainer(ctr *Container, cgroupParent string, restor defer wg.Done() runtime.LockOSThread() - fd, err := os.Open(fmt.Sprintf("/proc/%d/task/%d/ns/mnt", os.Getpid(), unix.Gettid())) + var fd *os.File + fd, err = os.Open(fmt.Sprintf("/proc/%d/task/%d/ns/mnt", os.Getpid(), unix.Gettid())) if err != nil { return } @@ -103,7 +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() @@ -111,7 +114,7 @@ func (r *OCIRuntime) createContainer(ctr *Container, cgroupParent string, restor } func rpmVersion(path string) string { - output := "Unknown" + output := unknownPackage cmd := exec.Command("/usr/bin/rpm", "-q", "-f", path) if outp, err := cmd.Output(); err == nil { output = string(outp) @@ -120,7 +123,7 @@ func rpmVersion(path string) string { } func dpkgVersion(path string) string { - output := "Unknown" + output := unknownPackage cmd := exec.Command("/usr/bin/dpkg", "-S", path) if outp, err := cmd.Output(); err == nil { output = string(outp) @@ -129,14 +132,14 @@ func dpkgVersion(path string) string { } func (r *OCIRuntime) pathPackage() string { - if out := rpmVersion(r.path); out != "Unknown" { + if out := rpmVersion(r.path); out != unknownPackage { return out } return dpkgVersion(r.path) } func (r *OCIRuntime) conmonPackage() string { - if out := rpmVersion(r.conmonPath); out != "Unknown" { + if out := rpmVersion(r.conmonPath); out != unknownPackage { return out } return dpkgVersion(r.conmonPath) 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 8d044313b..3e43d73f0 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" @@ -28,19 +29,40 @@ func WithStorageConfig(config storage.StoreOptions) RuntimeOption { return ErrRuntimeFinalized } - rt.config.StorageConfig.RunRoot = config.RunRoot - rt.config.StorageConfig.GraphRoot = config.GraphRoot - rt.config.StorageConfig.GraphDriverName = config.GraphDriverName - rt.config.StaticDir = filepath.Join(config.GraphRoot, "libpod") + if config.RunRoot != "" { + rt.config.StorageConfig.RunRoot = config.RunRoot + rt.configuredFrom.storageRunRootSet = true + } - rt.config.StorageConfig.GraphDriverOptions = make([]string, len(config.GraphDriverOptions)) - copy(rt.config.StorageConfig.GraphDriverOptions, config.GraphDriverOptions) + if config.GraphRoot != "" { + rt.config.StorageConfig.GraphRoot = config.GraphRoot + rt.configuredFrom.storageGraphRootSet = true - rt.config.StorageConfig.UIDMap = make([]idtools.IDMap, len(config.UIDMap)) - copy(rt.config.StorageConfig.UIDMap, config.UIDMap) + // Also set libpod static dir, so we are a subdirectory + // of the c/storage store by default + rt.config.StaticDir = filepath.Join(config.GraphRoot, "libpod") + rt.configuredFrom.libpodStaticDirSet = true + } - rt.config.StorageConfig.GIDMap = make([]idtools.IDMap, len(config.GIDMap)) - copy(rt.config.StorageConfig.GIDMap, config.GIDMap) + if config.GraphDriverName != "" { + rt.config.StorageConfig.GraphDriverName = config.GraphDriverName + rt.configuredFrom.storageGraphDriverSet = true + } + + if config.GraphDriverOptions != nil { + rt.config.StorageConfig.GraphDriverOptions = make([]string, len(config.GraphDriverOptions)) + copy(rt.config.StorageConfig.GraphDriverOptions, config.GraphDriverOptions) + } + + if config.UIDMap != nil { + rt.config.StorageConfig.UIDMap = make([]idtools.IDMap, len(config.UIDMap)) + copy(rt.config.StorageConfig.UIDMap, config.UIDMap) + } + + if config.GIDMap != nil { + rt.config.StorageConfig.GIDMap = make([]idtools.IDMap, len(config.GIDMap)) + copy(rt.config.StorageConfig.GIDMap, config.GIDMap) + } return nil } @@ -173,26 +195,26 @@ func WithStaticDir(dir string) RuntimeOption { } rt.config.StaticDir = dir + rt.configuredFrom.libpodStaticDirSet = true return nil } } -// WithHooksDir sets the directory to look for OCI runtime hooks config. -// Note we are not saving this in database, since this is really just for used -// for testing. -func WithHooksDir(hooksDir string) RuntimeOption { +// WithHooksDir sets the directories to look for OCI runtime hook configuration. +func WithHooksDir(hooksDirs ...string) RuntimeOption { return func(rt *Runtime) error { if rt.valid { return ErrRuntimeFinalized } - if hooksDir == "" { - return errors.Wrap(ErrInvalidArg, "empty-string hook directories are not supported") + for _, hooksDir := range hooksDirs { + if hooksDir == "" { + return errors.Wrap(ErrInvalidArg, "empty-string hook directories are not supported") + } } - rt.config.HooksDir = []string{hooksDir} - rt.config.HooksDirNotExistFatal = true + rt.config.HooksDir = hooksDirs return nil } } @@ -225,6 +247,7 @@ func WithTmpDir(dir string) RuntimeOption { } rt.config.TmpDir = dir + rt.configuredFrom.libpodTmpDirSet = true return nil } @@ -817,7 +840,7 @@ 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 +854,7 @@ 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 } @@ -1295,3 +1319,14 @@ func WithInfraContainer() PodCreateOption { return nil } } + +// WithInfraContainerPorts tells the pod to add port bindings to the pause container +func WithInfraContainerPorts(bindings []ocicni.PortMapping) PodCreateOption { + return func(pod *Pod) error { + if pod.valid { + return ErrPodFinalized + } + pod.config.InfraContainer.PortBindings = bindings + return nil + } +} diff --git a/libpod/pod.go b/libpod/pod.go index 8ac976f6a..07f41f5c6 100644 --- a/libpod/pod.go +++ b/libpod/pod.go @@ -4,6 +4,7 @@ import ( "time" "github.com/containers/storage" + "github.com/cri-o/ocicni/pkg/ocicni" "github.com/pkg/errors" ) @@ -96,7 +97,8 @@ type PodContainerInfo struct { // InfraContainerConfig is the configuration for the pod's infra container type InfraContainerConfig struct { - HasInfraContainer bool `json:"makeInfraContainer"` + HasInfraContainer bool `json:"makeInfraContainer"` + PortBindings []ocicni.PortMapping `json:"infraPortBindings"` } // ID retrieves the pod's ID diff --git a/libpod/pod_easyjson.go b/libpod/pod_easyjson.go index 6c1c939f3..8ea9a5e72 100644 --- a/libpod/pod_easyjson.go +++ b/libpod/pod_easyjson.go @@ -6,6 +6,7 @@ package libpod import ( json "encoding/json" + ocicni "github.com/cri-o/ocicni/pkg/ocicni" easyjson "github.com/mailru/easyjson" jlexer "github.com/mailru/easyjson/jlexer" jwriter "github.com/mailru/easyjson/jwriter" @@ -721,6 +722,29 @@ func easyjsonBe091417DecodeGithubComContainersLibpodLibpod5(in *jlexer.Lexer, ou switch key { case "makeInfraContainer": out.HasInfraContainer = bool(in.Bool()) + case "infraPortBindings": + if in.IsNull() { + in.Skip() + out.PortBindings = nil + } else { + in.Delim('[') + if out.PortBindings == nil { + if !in.IsDelim(']') { + out.PortBindings = make([]ocicni.PortMapping, 0, 1) + } else { + out.PortBindings = []ocicni.PortMapping{} + } + } else { + out.PortBindings = (out.PortBindings)[:0] + } + for !in.IsDelim(']') { + var v6 ocicni.PortMapping + easyjsonBe091417DecodeGithubComContainersLibpodVendorGithubComCriOOcicniPkgOcicni(in, &v6) + out.PortBindings = append(out.PortBindings, v6) + in.WantComma() + } + in.Delim(']') + } default: in.SkipRecursive() } @@ -745,5 +769,109 @@ func easyjsonBe091417EncodeGithubComContainersLibpodLibpod5(out *jwriter.Writer, } out.Bool(bool(in.HasInfraContainer)) } + { + const prefix string = ",\"infraPortBindings\":" + if first { + first = false + out.RawString(prefix[1:]) + } else { + out.RawString(prefix) + } + if in.PortBindings == nil && (out.Flags&jwriter.NilSliceAsEmpty) == 0 { + out.RawString("null") + } else { + out.RawByte('[') + for v7, v8 := range in.PortBindings { + if v7 > 0 { + out.RawByte(',') + } + easyjsonBe091417EncodeGithubComContainersLibpodVendorGithubComCriOOcicniPkgOcicni(out, v8) + } + out.RawByte(']') + } + } + out.RawByte('}') +} +func easyjsonBe091417DecodeGithubComContainersLibpodVendorGithubComCriOOcicniPkgOcicni(in *jlexer.Lexer, out *ocicni.PortMapping) { + isTopLevel := in.IsStart() + if in.IsNull() { + if isTopLevel { + in.Consumed() + } + in.Skip() + return + } + in.Delim('{') + for !in.IsDelim('}') { + key := in.UnsafeString() + in.WantColon() + if in.IsNull() { + in.Skip() + in.WantComma() + continue + } + switch key { + case "hostPort": + out.HostPort = int32(in.Int32()) + case "containerPort": + out.ContainerPort = int32(in.Int32()) + case "protocol": + out.Protocol = string(in.String()) + case "hostIP": + out.HostIP = string(in.String()) + default: + in.SkipRecursive() + } + in.WantComma() + } + in.Delim('}') + if isTopLevel { + in.Consumed() + } +} +func easyjsonBe091417EncodeGithubComContainersLibpodVendorGithubComCriOOcicniPkgOcicni(out *jwriter.Writer, in ocicni.PortMapping) { + out.RawByte('{') + first := true + _ = first + { + const prefix string = ",\"hostPort\":" + if first { + first = false + out.RawString(prefix[1:]) + } else { + out.RawString(prefix) + } + out.Int32(int32(in.HostPort)) + } + { + const prefix string = ",\"containerPort\":" + if first { + first = false + out.RawString(prefix[1:]) + } else { + out.RawString(prefix) + } + out.Int32(int32(in.ContainerPort)) + } + { + const prefix string = ",\"protocol\":" + if first { + first = false + out.RawString(prefix[1:]) + } else { + out.RawString(prefix) + } + out.String(string(in.Protocol)) + } + { + const prefix string = ",\"hostIP\":" + if first { + first = false + out.RawString(prefix[1:]) + } else { + out.RawString(prefix) + } + out.String(string(in.HostIP)) + } out.RawByte('}') } diff --git a/libpod/pod_internal.go b/libpod/pod_internal.go index 46162c7ef..39a25c004 100644 --- a/libpod/pod_internal.go +++ b/libpod/pod_internal.go @@ -48,7 +48,7 @@ func (p *Pod) updatePod() error { // Save pod state to database func (p *Pod) save() error { if err := p.runtime.state.SavePod(p); err != nil { - return errors.Wrapf(err, "error saving pod %s state") + return errors.Wrapf(err, "error saving pod %s state", p.ID()) } return nil diff --git a/libpod/runtime.go b/libpod/runtime.go index 318cd0369..6b6a8cc2d 100644 --- a/libpod/runtime.go +++ b/libpod/runtime.go @@ -12,7 +12,6 @@ import ( "github.com/containers/image/types" "github.com/containers/libpod/libpod/image" "github.com/containers/libpod/pkg/firewall" - "github.com/containers/libpod/pkg/hooks" sysreg "github.com/containers/libpod/pkg/registries" "github.com/containers/libpod/pkg/rootless" "github.com/containers/libpod/pkg/util" @@ -84,6 +83,7 @@ type Runtime struct { lock sync.RWMutex imageRuntime *image.Runtime firewallBackend firewall.FirewallBackend + configuredFrom *runtimeConfiguredFrom } // RuntimeConfig contains configuration options used to set up the runtime @@ -141,11 +141,11 @@ type RuntimeConfig struct { // CNIDefaultNetwork is the network name of the default CNI network // to attach pods to CNIDefaultNetwork string `toml:"cni_default_network,omitempty"` - // HooksDir Path to the directory containing hooks configuration files + // HooksDir holds paths to the directories containing hooks + // configuration files. When the same filename is present in in + // multiple directories, the file in the directory listed last in + // this slice takes precedence. HooksDir []string `toml:"hooks_dir"` - // HooksDirNotExistFatal switches between fatal errors and non-fatal - // warnings if the configured HooksDir does not exist. - HooksDirNotExistFatal bool `toml:"hooks_dir_not_exist_fatal"` // DefaultMountsFile is the path to the default mounts file for testing // purposes only DefaultMountsFile string `toml:"-"` @@ -175,6 +175,20 @@ type RuntimeConfig struct { EnableLabeling bool `toml:"label"` } +// runtimeConfiguredFrom is a struct used during early runtime init to help +// assemble the full RuntimeConfig struct from defaults. +// It indicated whether several fields in the runtime configuration were set +// explicitly. +// If they were not, we may override them with information from the database, +// if it exists and differs from what is present in the system already. +type runtimeConfiguredFrom struct { + storageGraphDriverSet bool + storageGraphRootSet bool + storageRunRootSet bool + libpodStaticDirSet bool + libpodTmpDirSet bool +} + var ( defaultRuntimeConfig = RuntimeConfig{ // Leave this empty so containers/storage will use its defaults @@ -203,7 +217,6 @@ var ( "PATH=/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin", }, CgroupManager: SystemdCgroupsManager, - HooksDir: []string{hooks.DefaultDir, hooks.OverrideDir}, StaticDir: filepath.Join(storage.DefaultStoreOptions.GraphRoot, "libpod"), TmpDir: "", MaxLogSize: -1, @@ -253,6 +266,7 @@ func SetXdgRuntimeDir(val string) error { func NewRuntime(options ...RuntimeOption) (runtime *Runtime, err error) { runtime = new(Runtime) runtime.config = new(RuntimeConfig) + runtime.configuredFrom = new(runtimeConfiguredFrom) // Copy the default configuration tmpDir, err := getDefaultTmpDir() @@ -262,8 +276,19 @@ func NewRuntime(options ...RuntimeOption) (runtime *Runtime, err error) { deepcopier.Copy(defaultRuntimeConfig).To(runtime.config) runtime.config.TmpDir = tmpDir + if rootless.IsRootless() { + // If we're rootless, override the default storage config + storageConf, err := util.GetDefaultStoreOptions() + if err != nil { + return nil, errors.Wrapf(err, "error retrieving rootless storage config") + } + runtime.config.StorageConfig = storageConf + runtime.config.StaticDir = filepath.Join(storageConf.GraphRoot, "libpod") + } + configPath := ConfigPath foundConfig := true + rootlessConfigPath := "" if rootless.IsRootless() { home := os.Getenv("HOME") if runtime.config.SignaturePolicyPath == "" { @@ -272,7 +297,10 @@ func NewRuntime(options ...RuntimeOption) (runtime *Runtime, err error) { runtime.config.SignaturePolicyPath = newPath } } - configPath = filepath.Join(home, ".config/containers/libpod.conf") + + rootlessConfigPath = filepath.Join(home, ".config/containers/libpod.conf") + + configPath = rootlessConfigPath if _, err := os.Stat(configPath); err != nil { foundConfig = false } @@ -303,6 +331,25 @@ func NewRuntime(options ...RuntimeOption) (runtime *Runtime, err error) { if err != nil { return nil, errors.Wrapf(err, "error reading configuration file %s", configPath) } + + // This is ugly, but we need to decode twice. + // Once to check if libpod static and tmp dirs were explicitly + // set (not enough to check if they're not the default value, + // might have been explicitly configured to the default). + // A second time to actually get a usable config. + tmpConfig := new(RuntimeConfig) + if _, err := toml.Decode(string(contents), tmpConfig); err != nil { + return nil, errors.Wrapf(err, "error decoding configuration file %s", + configPath) + } + + if tmpConfig.StaticDir != "" { + runtime.configuredFrom.libpodStaticDirSet = true + } + if tmpConfig.TmpDir != "" { + runtime.configuredFrom.libpodTmpDirSet = true + } + if _, err := toml.Decode(string(contents), runtime.config); err != nil { return nil, errors.Wrapf(err, "error decoding configuration file %s", configPath) } @@ -317,6 +364,22 @@ func NewRuntime(options ...RuntimeOption) (runtime *Runtime, err error) { if err := makeRuntime(runtime); err != nil { return nil, err } + + if !foundConfig && rootlessConfigPath != "" { + os.MkdirAll(filepath.Dir(rootlessConfigPath), 0755) + file, err := os.OpenFile(rootlessConfigPath, os.O_RDWR|os.O_CREATE|os.O_EXCL, 0666) + if err != nil && !os.IsExist(err) { + return nil, errors.Wrapf(err, "cannot open file %s", rootlessConfigPath) + } + if err == nil { + defer file.Close() + enc := toml.NewEncoder(file) + if err := enc.Encode(runtime.config); err != nil { + os.Remove(rootlessConfigPath) + } + } + } + return runtime, nil } @@ -328,6 +391,7 @@ func NewRuntime(options ...RuntimeOption) (runtime *Runtime, err error) { func NewRuntimeFromConfig(configPath string, options ...RuntimeOption) (runtime *Runtime, err error) { runtime = new(Runtime) runtime.config = new(RuntimeConfig) + runtime.configuredFrom = new(runtimeConfiguredFrom) // Set two fields not in the TOML config runtime.config.StateType = defaultRuntimeConfig.StateType @@ -406,6 +470,77 @@ func makeRuntime(runtime *Runtime) (err error) { runtime.config.ConmonPath) } + // Make the static files directory if it does not exist + if err := os.MkdirAll(runtime.config.StaticDir, 0700); err != nil { + // The directory is allowed to exist + if !os.IsExist(err) { + return errors.Wrapf(err, "error creating runtime static files directory %s", + runtime.config.StaticDir) + } + } + + // Set up the state + switch runtime.config.StateType { + case InMemoryStateStore: + state, err := NewInMemoryState() + if err != nil { + return err + } + runtime.state = state + case SQLiteStateStore: + return errors.Wrapf(ErrInvalidArg, "SQLite state is currently disabled") + case BoltDBStateStore: + dbPath := filepath.Join(runtime.config.StaticDir, "bolt_state.db") + + state, err := NewBoltState(dbPath, runtime) + if err != nil { + return err + } + runtime.state = state + default: + return errors.Wrapf(ErrInvalidArg, "unrecognized state type passed") + } + + // Grab config from the database so we can reset some defaults + dbConfig, err := runtime.state.GetDBConfig() + if err != nil { + return errors.Wrapf(err, "error retrieving runtime configuration from database") + } + + // Reset defaults if they were not explicitly set + if !runtime.configuredFrom.storageGraphDriverSet && dbConfig.GraphDriver != "" { + runtime.config.StorageConfig.GraphDriverName = dbConfig.GraphDriver + } + if !runtime.configuredFrom.storageGraphRootSet && dbConfig.StorageRoot != "" { + runtime.config.StorageConfig.GraphRoot = dbConfig.StorageRoot + } + if !runtime.configuredFrom.storageRunRootSet && dbConfig.StorageTmp != "" { + runtime.config.StorageConfig.RunRoot = dbConfig.StorageTmp + } + if !runtime.configuredFrom.libpodStaticDirSet && dbConfig.LibpodRoot != "" { + runtime.config.StaticDir = dbConfig.LibpodRoot + } + if !runtime.configuredFrom.libpodTmpDirSet && dbConfig.LibpodTmp != "" { + runtime.config.TmpDir = dbConfig.LibpodTmp + } + + logrus.Debugf("Using graph driver %s", runtime.config.StorageConfig.GraphDriverName) + logrus.Debugf("Using graph root %s", runtime.config.StorageConfig.GraphRoot) + logrus.Debugf("Using run root %s", runtime.config.StorageConfig.RunRoot) + logrus.Debugf("Using static dir %s", runtime.config.StaticDir) + logrus.Debugf("Using tmp dir %s", runtime.config.TmpDir) + + // Validate our config against the database, now that we've set our + // final storage configuration + if err := runtime.state.ValidateDBConfig(runtime); err != nil { + return err + } + + if err := runtime.state.SetNamespace(runtime.config.Namespace); err != nil { + return errors.Wrapf(err, "error setting libpod namespace in state") + } + logrus.Debugf("Set libpod namespace to %q", runtime.config.Namespace) + // Set up containers/storage var store storage.Store if rootless.SkipStorageSetup() { @@ -473,15 +608,6 @@ func makeRuntime(runtime *Runtime) (err error) { } runtime.ociRuntime = ociRuntime - // Make the static files directory if it does not exist - if err := os.MkdirAll(runtime.config.StaticDir, 0755); err != nil { - // The directory is allowed to exist - if !os.IsExist(err) { - return errors.Wrapf(err, "error creating runtime static files directory %s", - runtime.config.StaticDir) - } - } - // Make a directory to hold container lockfiles lockDir := filepath.Join(runtime.config.TmpDir, "lock") if err := os.MkdirAll(lockDir, 0755); err != nil { @@ -503,11 +629,13 @@ func makeRuntime(runtime *Runtime) (err error) { } // Set up the CNI net plugin - netPlugin, err := ocicni.InitCNI(runtime.config.CNIDefaultNetwork, runtime.config.CNIConfigDir, runtime.config.CNIPluginDir...) - if err != nil { - return errors.Wrapf(err, "error configuring CNI network plugin") + if !rootless.IsRootless() { + netPlugin, err := ocicni.InitCNI(runtime.config.CNIDefaultNetwork, runtime.config.CNIConfigDir, runtime.config.CNIPluginDir...) + if err != nil { + return errors.Wrapf(err, "error configuring CNI network plugin") + } + runtime.netPlugin = netPlugin } - runtime.netPlugin = netPlugin // Set up a firewall backend backendType := "" @@ -520,33 +648,6 @@ func makeRuntime(runtime *Runtime) (err error) { } runtime.firewallBackend = fwBackend - // Set up the state - switch runtime.config.StateType { - case InMemoryStateStore: - state, err := NewInMemoryState() - if err != nil { - return err - } - runtime.state = state - case SQLiteStateStore: - return errors.Wrapf(ErrInvalidArg, "SQLite state is currently disabled") - case BoltDBStateStore: - dbPath := filepath.Join(runtime.config.StaticDir, "bolt_state.db") - - state, err := NewBoltState(dbPath, runtime.lockDir, runtime) - if err != nil { - return err - } - runtime.state = state - default: - return errors.Wrapf(ErrInvalidArg, "unrecognized state type passed") - } - - if err := runtime.state.SetNamespace(runtime.config.Namespace); err != nil { - return errors.Wrapf(err, "error setting libpod namespace in state") - } - logrus.Debugf("Set libpod namespace to %q", runtime.config.Namespace) - // We now need to see if the system has restarted // We check for the presence of a file in our tmp directory to verify this // This check must be locked to prevent races diff --git a/libpod/runtime_ctr.go b/libpod/runtime_ctr.go index b63726f29..09d0ec042 100644 --- a/libpod/runtime_ctr.go +++ b/libpod/runtime_ctr.go @@ -246,7 +246,19 @@ func (r *Runtime) removeContainer(ctx context.Context, c *Container, force bool) } if c.state.State == ContainerStatePaused { - return errors.Wrapf(ErrCtrStateInvalid, "container %s is paused, cannot remove until unpaused", c.ID()) + if !force { + return errors.Wrapf(ErrCtrStateInvalid, "container %s is paused, cannot remove until unpaused", c.ID()) + } + if err := c.runtime.ociRuntime.killContainer(c, 9); err != nil { + return err + } + if err := c.unpause(); err != nil { + return err + } + // Need to update container state to make sure we know it's stopped + if err := c.waitForExitFileAndSync(); err != nil { + return err + } } // Check that the container's in a good state to be removed @@ -256,7 +268,7 @@ func (r *Runtime) removeContainer(ctx context.Context, c *Container, force bool) } // Need to update container state to make sure we know it's stopped - if err := c.syncContainer(); err != nil { + if err := c.waitForExitFileAndSync(); err != nil { return err } } else if !(c.state.State == ContainerStateConfigured || diff --git a/libpod/runtime_pod_infra_linux.go b/libpod/runtime_pod_infra_linux.go index fea79e994..8a5dbef56 100644 --- a/libpod/runtime_pod_infra_linux.go +++ b/libpod/runtime_pod_infra_linux.go @@ -7,7 +7,6 @@ import ( "github.com/containers/libpod/libpod/image" "github.com/containers/libpod/pkg/rootless" - "github.com/cri-o/ocicni/pkg/ocicni" spec "github.com/opencontainers/runtime-spec/specs-go" "github.com/opencontainers/runtime-tools/generate" ) @@ -50,9 +49,12 @@ func (r *Runtime) makeInfraContainer(ctx context.Context, p *Pod, imgName, imgID options = append(options, withIsInfra()) // Since user namespace sharing is not implemented, we only need to check if it's rootless - portMappings := make([]ocicni.PortMapping, 0) networks := make([]string, 0) - options = append(options, WithNetNS(portMappings, 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/runtime_pod_linux.go b/libpod/runtime_pod_linux.go index eb3d471dd..3d6fad52f 100644 --- a/libpod/runtime_pod_linux.go +++ b/libpod/runtime_pod_linux.go @@ -265,15 +265,26 @@ func (r *Runtime) removePod(ctx context.Context, p *Pod, removeCtrs, force bool) } case CgroupfsCgroupsManager: // Delete the cgroupfs cgroup + // Make sure the conmon cgroup is deleted first + // Since the pod is almost gone, don't bother failing + // hard - instead, just log errors. v1CGroups := GetV1CGroups(getExcludedCGroups()) + conmonCgroupPath := filepath.Join(p.state.CgroupPath, "conmon") + conmonCgroup, err := cgroups.Load(v1CGroups, cgroups.StaticPath(conmonCgroupPath)) + if err != nil && err != cgroups.ErrCgroupDeleted { + return err + } + if err == nil { + if err := conmonCgroup.Delete(); err != nil { + logrus.Errorf("Error deleting pod %s conmon cgroup %s: %v", p.ID(), conmonCgroupPath, err) + } + } cgroup, err := cgroups.Load(v1CGroups, cgroups.StaticPath(p.state.CgroupPath)) if err != nil && err != cgroups.ErrCgroupDeleted { return err - } else if err == nil { + } + if err == nil { if err := cgroup.Delete(); err != nil { - // The pod is already almost gone. - // No point in hard-failing if we fail - // this bit of cleanup. logrus.Errorf("Error deleting pod %s cgroup %s: %v", p.ID(), p.state.CgroupPath, err) } } diff --git a/libpod/state.go b/libpod/state.go index 273e81318..06c2003d8 100644 --- a/libpod/state.go +++ b/libpod/state.go @@ -1,5 +1,15 @@ package libpod +// DBConfig is a set of Libpod runtime configuration settings that are saved +// in a State when it is first created, and can subsequently be retrieved. +type DBConfig struct { + LibpodRoot string + LibpodTmp string + StorageRoot string + StorageTmp string + GraphDriver string +} + // State is a storage backend for libpod's current state. // A State is only initialized once per instance of libpod. // As such, initialization methods for State implementations may safely assume @@ -21,6 +31,22 @@ type State interface { // Refresh clears container and pod states after a reboot Refresh() error + // GetDBConfig retrieves several paths configured within the database + // when it was created - namely, Libpod root and tmp dirs, c/storage + // root and tmp dirs, and c/storage graph driver. + // This is not implemented by the in-memory state, as it has no need to + // validate runtime configuration. + GetDBConfig() (*DBConfig, error) + + // ValidateDBConfig validates the config in the given Runtime struct + // against paths stored in the configured database. + // Libpod root and tmp dirs and c/storage root and tmp dirs and graph + // driver are validated. + // This is not implemented by the in-memory state, as it has no need to + // validate runtime configuration that may change over multiple runs of + // the program. + ValidateDBConfig(runtime *Runtime) error + // SetNamespace() sets the namespace for the store, and will determine // what containers are retrieved with container and pod retrieval calls. // A namespace of "", the empty string, acts as no namespace, and diff --git a/libpod/state_test.go b/libpod/state_test.go index 04572fb29..d93a371f3 100644 --- a/libpod/state_test.go +++ b/libpod/state_test.go @@ -45,11 +45,16 @@ func getEmptyBoltState() (s State, p string, p2 string, err error) { dbPath := filepath.Join(tmpDir, "db.sql") lockDir := filepath.Join(tmpDir, "locks") + if err := os.Mkdir(lockDir, 0755); err != nil { + return nil, "", "", err + } + runtime := new(Runtime) runtime.config = new(RuntimeConfig) runtime.config.StorageConfig = storage.StoreOptions{} + runtime.lockDir = lockDir - state, err := NewBoltState(dbPath, lockDir, runtime) + state, err := NewBoltState(dbPath, runtime) if err != nil { return nil, "", "", err } diff --git a/libpod/testdata/config.toml b/libpod/testdata/config.toml index e19d36017..1d78f2083 100644 --- a/libpod/testdata/config.toml +++ b/libpod/testdata/config.toml @@ -14,7 +14,7 @@ seccomp_profile = "/etc/crio/seccomp.json" apparmor_profile = "crio-default" cgroup_manager = "cgroupfs" - hooks_dir_path = "/usr/share/containers/oci/hooks.d" + hooks_dir = ["/usr/share/containers/oci/hooks.d"] pids_limit = 2048 container_exits_dir = "/var/run/podman/exits" [crio.image] 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) } } diff --git a/libpod/version.go b/libpod/version.go index 5e7cd83c9..966588ae9 100644 --- a/libpod/version.go +++ b/libpod/version.go @@ -11,10 +11,10 @@ import ( var ( // GitCommit is the commit that the binary is being built from. // It will be populated by the Makefile. - GitCommit string + gitCommit string // BuildInfo is the time at which the binary was built // It will be populated by the Makefile. - BuildInfo string + buildInfo string ) //Version is an output struct for varlink @@ -30,9 +30,9 @@ type Version struct { func GetVersion() (Version, error) { var err error var buildTime int64 - if BuildInfo != "" { + if buildInfo != "" { // Converts unix time from string to int64 - buildTime, err = strconv.ParseInt(BuildInfo, 10, 64) + buildTime, err = strconv.ParseInt(buildInfo, 10, 64) if err != nil { return Version{}, err @@ -41,7 +41,7 @@ func GetVersion() (Version, error) { return Version{ Version: podmanVersion.Version, GoVersion: runtime.Version(), - GitCommit: GitCommit, + GitCommit: gitCommit, Built: buildTime, OsArch: runtime.GOOS + "/" + runtime.GOARCH, }, nil |