summaryrefslogtreecommitdiff
path: root/libpod
diff options
context:
space:
mode:
Diffstat (limited to 'libpod')
-rw-r--r--libpod/container.go10
-rw-r--r--libpod/container_exec.go83
-rw-r--r--libpod/container_internal.go65
-rw-r--r--libpod/container_internal_linux.go89
-rw-r--r--libpod/container_internal_unsupported.go5
-rw-r--r--libpod/container_log.go24
-rw-r--r--libpod/define/containerstate.go10
-rw-r--r--libpod/events.go39
-rw-r--r--libpod/events/config.go2
-rw-r--r--libpod/events/events.go2
-rw-r--r--libpod/networking_linux.go42
-rw-r--r--libpod/oci.go7
-rw-r--r--libpod/oci_attach_linux.go13
-rw-r--r--libpod/oci_conmon_exec_linux.go19
-rw-r--r--libpod/oci_conmon_linux.go6
-rw-r--r--libpod/oci_missing.go5
-rw-r--r--libpod/options.go19
-rw-r--r--libpod/runtime.go7
-rw-r--r--libpod/stats.go10
-rw-r--r--libpod/volume_internal.go19
20 files changed, 345 insertions, 131 deletions
diff --git a/libpod/container.go b/libpod/container.go
index c6f0cd618..4b9bea5fc 100644
--- a/libpod/container.go
+++ b/libpod/container.go
@@ -957,6 +957,12 @@ func (c *Container) cGroupPath() (string, error) {
// is the libpod-specific one we're looking for.
//
// See #8397 on the need for the longest-path look up.
+ //
+ // And another workaround for containers running systemd as the payload.
+ // containers running systemd moves themselves into a child subgroup of
+ // the named systemd cgroup hierarchy. Ignore any named cgroups during
+ // the lookup.
+ // See #10602 for more details.
procPath := fmt.Sprintf("/proc/%d/cgroup", c.state.PID)
lines, err := ioutil.ReadFile(procPath)
if err != nil {
@@ -972,6 +978,10 @@ func (c *Container) cGroupPath() (string, error) {
logrus.Debugf("Error parsing cgroup: expected 3 fields but got %d: %s", len(fields), procPath)
continue
}
+ // Ignore named cgroups like name=systemd.
+ if bytes.Contains(fields[1], []byte("=")) {
+ continue
+ }
path := string(fields[2])
if len(path) > len(cgroupPath) {
cgroupPath = path
diff --git a/libpod/container_exec.go b/libpod/container_exec.go
index c359f1e5d..5d4bcb422 100644
--- a/libpod/container_exec.go
+++ b/libpod/container_exec.go
@@ -1,6 +1,7 @@
package libpod
import (
+ "context"
"io/ioutil"
"net/http"
"os"
@@ -276,9 +277,10 @@ func (c *Container) ExecStart(sessionID string) error {
}
// ExecStartAndAttach starts and attaches to an exec session in a container.
+// newSize resizes the tty to this size before the process is started, must be nil if the exec session has no tty
// TODO: Should we include detach keys in the signature to allow override?
// TODO: How do we handle AttachStdin/AttachStdout/AttachStderr?
-func (c *Container) ExecStartAndAttach(sessionID string, streams *define.AttachStreams) error {
+func (c *Container) ExecStartAndAttach(sessionID string, streams *define.AttachStreams, newSize *define.TerminalSize) error {
if !c.batched {
c.lock.Lock()
defer c.lock.Unlock()
@@ -309,7 +311,7 @@ func (c *Container) ExecStartAndAttach(sessionID string, streams *define.AttachS
return err
}
- pid, attachChan, err := c.ociRuntime.ExecContainer(c, session.ID(), opts, streams)
+ pid, attachChan, err := c.ociRuntime.ExecContainer(c, session.ID(), opts, streams, newSize)
if err != nil {
return err
}
@@ -372,7 +374,9 @@ func (c *Container) ExecStartAndAttach(sessionID string, streams *define.AttachS
}
// ExecHTTPStartAndAttach starts and performs an HTTP attach to an exec session.
-func (c *Container) ExecHTTPStartAndAttach(sessionID string, r *http.Request, w http.ResponseWriter, streams *HTTPAttachStreams, detachKeys *string, cancel <-chan bool, hijackDone chan<- bool) error {
+// newSize resizes the tty to this size before the process is started, must be nil if the exec session has no tty
+func (c *Container) ExecHTTPStartAndAttach(sessionID string, r *http.Request, w http.ResponseWriter,
+ streams *HTTPAttachStreams, detachKeys *string, cancel <-chan bool, hijackDone chan<- bool, newSize *define.TerminalSize) error {
// TODO: How do we combine streams with the default streams set in the exec session?
// Ensure that we don't leak a goroutine here
@@ -430,7 +434,7 @@ func (c *Container) ExecHTTPStartAndAttach(sessionID string, r *http.Request, w
close(holdConnOpen)
}()
- pid, attachChan, err := c.ociRuntime.ExecContainerHTTP(c, session.ID(), execOpts, r, w, streams, cancel, hijackDone, holdConnOpen)
+ pid, attachChan, err := c.ociRuntime.ExecContainerHTTP(c, session.ID(), execOpts, r, w, streams, cancel, hijackDone, holdConnOpen, newSize)
if err != nil {
session.State = define.ExecStateStopped
session.ExitCode = define.TranslateExecErrorToExitCode(define.ExecErrorCodeGeneric, err)
@@ -539,18 +543,7 @@ func (c *Container) ExecStop(sessionID string, timeout *uint) error {
var cleanupErr error
// Retrieve exit code and update status
- exitCode, err := c.readExecExitCode(session.ID())
- if err != nil {
- cleanupErr = err
- }
- session.ExitCode = exitCode
- session.PID = 0
- session.State = define.ExecStateStopped
-
- if err := c.save(); err != nil {
- if cleanupErr != nil {
- logrus.Errorf("Error stopping container %s exec session %s: %v", c.ID(), session.ID(), cleanupErr)
- }
+ if err := retrieveAndWriteExecExitCode(c, session.ID()); err != nil {
cleanupErr = err
}
@@ -592,15 +585,7 @@ func (c *Container) ExecCleanup(sessionID string) error {
return errors.Wrapf(define.ErrExecSessionStateInvalid, "cannot clean up container %s exec session %s as it is running", c.ID(), session.ID())
}
- exitCode, err := c.readExecExitCode(session.ID())
- if err != nil {
- return err
- }
- session.ExitCode = exitCode
- session.PID = 0
- session.State = define.ExecStateStopped
-
- if err := c.save(); err != nil {
+ if err := retrieveAndWriteExecExitCode(c, session.ID()); err != nil {
return err
}
}
@@ -637,9 +622,9 @@ func (c *Container) ExecRemove(sessionID string, force bool) error {
return err
}
if !running {
- session.State = define.ExecStateStopped
- // TODO: should we retrieve exit code here?
- // TODO: Might be worth saving state here.
+ if err := retrieveAndWriteExecExitCode(c, session.ID()); err != nil {
+ return err
+ }
}
}
@@ -653,6 +638,10 @@ func (c *Container) ExecRemove(sessionID string, force bool) error {
return err
}
+ if err := retrieveAndWriteExecExitCode(c, session.ID()); err != nil {
+ return err
+ }
+
if err := c.cleanupExecBundle(session.ID()); err != nil {
return err
}
@@ -733,7 +722,10 @@ func (c *Container) Exec(config *ExecConfig, streams *define.AttachStreams, resi
// API there.
// TODO: Refactor so this is closed here, before we remove the exec
// session.
+ var size *define.TerminalSize
if resize != nil {
+ s := <-resize
+ size = &s
go func() {
logrus.Debugf("Sending resize events to exec session %s", sessionID)
for resizeRequest := range resize {
@@ -751,16 +743,31 @@ func (c *Container) Exec(config *ExecConfig, streams *define.AttachStreams, resi
}()
}
- if err := c.ExecStartAndAttach(sessionID, streams); err != nil {
+ if err := c.ExecStartAndAttach(sessionID, streams, size); err != nil {
return -1, err
}
session, err := c.ExecSession(sessionID)
if err != nil {
+ if errors.Cause(err) == define.ErrNoSuchExecSession {
+ // TODO: If a proper Context is ever plumbed in here, we
+ // should use it.
+ // As things stand, though, it's not worth it - this
+ // should always terminate quickly since it's not
+ // streaming.
+ diedEvent, err := c.runtime.GetExecDiedEvent(context.Background(), c.ID(), sessionID)
+ if err != nil {
+ return -1, errors.Wrapf(err, "error retrieving exec session %s exit code", sessionID)
+ }
+ return diedEvent.ContainerExitCode, nil
+ }
return -1, err
}
exitCode := session.ExitCode
if err := c.ExecRemove(sessionID, false); err != nil {
+ if errors.Cause(err) == define.ErrNoSuchExecSession {
+ return exitCode, nil
+ }
return -1, err
}
@@ -927,6 +934,8 @@ func (c *Container) getActiveExecSessions() ([]string, error) {
session.PID = 0
session.State = define.ExecStateStopped
+ c.newExecDiedEvent(session.ID(), exitCode)
+
needSave = true
}
if err := c.cleanupExecBundle(id); err != nil {
@@ -1036,6 +1045,22 @@ func writeExecExitCode(c *Container, sessionID string, exitCode int) error {
return errors.Wrapf(err, "error syncing container %s state to remove exec session %s", c.ID(), sessionID)
}
+ return justWriteExecExitCode(c, sessionID, exitCode)
+}
+
+func retrieveAndWriteExecExitCode(c *Container, sessionID string) error {
+ exitCode, err := c.readExecExitCode(sessionID)
+ if err != nil {
+ return err
+ }
+
+ return justWriteExecExitCode(c, sessionID, exitCode)
+}
+
+func justWriteExecExitCode(c *Container, sessionID string, exitCode int) error {
+ // Write an event first
+ c.newExecDiedEvent(sessionID, exitCode)
+
session, ok := c.state.ExecSessions[sessionID]
if !ok {
// Exec session already removed.
diff --git a/libpod/container_internal.go b/libpod/container_internal.go
index f77825efd..545b78976 100644
--- a/libpod/container_internal.go
+++ b/libpod/container_internal.go
@@ -42,6 +42,7 @@ const (
// name of the directory holding the artifacts
artifactsDir = "artifacts"
execDirPermission = 0755
+ preCheckpointDir = "pre-checkpoint"
)
// rootFsSize gets the size of the container's root filesystem
@@ -141,7 +142,7 @@ func (c *Container) CheckpointPath() string {
// PreCheckpointPath returns the path to the directory containing the pre-checkpoint-images
func (c *Container) PreCheckPointPath() string {
- return filepath.Join(c.bundlePath(), "pre-checkpoint")
+ return filepath.Join(c.bundlePath(), preCheckpointDir)
}
// AttachSocketPath retrieves the path of the container's attach socket
@@ -427,7 +428,7 @@ func (c *Container) setupStorage(ctx context.Context) error {
},
LabelOpts: c.config.LabelOpts,
}
- if c.restoreFromCheckpoint {
+ if c.restoreFromCheckpoint && !c.config.Privileged {
// If restoring from a checkpoint, the root file-system
// needs to be mounted with the same SELinux labels as
// it was mounted previously.
@@ -1061,7 +1062,7 @@ func (c *Container) init(ctx context.Context, retainRetries bool) error {
}
for _, v := range c.config.NamedVolumes {
- if err := c.chownVolume(v.Name); err != nil {
+ if err := c.fixVolumePermissions(v); err != nil {
return err
}
}
@@ -1680,64 +1681,6 @@ func (c *Container) mountNamedVolume(v *ContainerNamedVolume, mountpoint string)
return vol, nil
}
-// Chown the specified volume if necessary.
-func (c *Container) chownVolume(volumeName string) error {
- vol, err := c.runtime.state.Volume(volumeName)
- if err != nil {
- return errors.Wrapf(err, "error retrieving named volume %s for container %s", volumeName, c.ID())
- }
-
- vol.lock.Lock()
- defer vol.lock.Unlock()
-
- // The volume may need a copy-up. Check the state.
- if err := vol.update(); err != nil {
- return err
- }
-
- // TODO: For now, I've disabled chowning volumes owned by non-Podman
- // drivers. This may be safe, but it's really going to be a case-by-case
- // thing, I think - safest to leave disabled now and re-enable later if
- // there is a demand.
- if vol.state.NeedsChown && !vol.UsesVolumeDriver() {
- vol.state.NeedsChown = false
-
- uid := int(c.config.Spec.Process.User.UID)
- gid := int(c.config.Spec.Process.User.GID)
-
- if c.config.IDMappings.UIDMap != nil {
- p := idtools.IDPair{
- UID: uid,
- GID: gid,
- }
- mappings := idtools.NewIDMappingsFromMaps(c.config.IDMappings.UIDMap, c.config.IDMappings.GIDMap)
- newPair, err := mappings.ToHost(p)
- if err != nil {
- return errors.Wrapf(err, "error mapping user %d:%d", uid, gid)
- }
- uid = newPair.UID
- gid = newPair.GID
- }
-
- vol.state.UIDChowned = uid
- vol.state.GIDChowned = gid
-
- if err := vol.save(); err != nil {
- return err
- }
-
- mountPoint, err := vol.MountPoint()
- if err != nil {
- return err
- }
-
- if err := os.Lchown(mountPoint, uid, gid); err != nil {
- return err
- }
- }
- return nil
-}
-
// cleanupStorage unmounts and cleans up the container's root filesystem
func (c *Container) cleanupStorage() error {
if !c.state.Mounted {
diff --git a/libpod/container_internal_linux.go b/libpod/container_internal_linux.go
index 94bf7855b..25db3cff0 100644
--- a/libpod/container_internal_linux.go
+++ b/libpod/container_internal_linux.go
@@ -909,14 +909,15 @@ func (c *Container) exportCheckpoint(options ContainerCheckpointOptions) error {
includeFiles := []string{
"artifacts",
"ctr.log",
- metadata.CheckpointDirectory,
metadata.ConfigDumpFile,
metadata.SpecDumpFile,
metadata.NetworkStatusFile,
}
if options.PreCheckPoint {
- includeFiles[0] = "pre-checkpoint"
+ includeFiles = append(includeFiles, preCheckpointDir)
+ } else {
+ includeFiles = append(includeFiles, metadata.CheckpointDirectory)
}
// Get root file-system changes included in the checkpoint archive
var addToTarFiles []string
@@ -1660,9 +1661,13 @@ func (c *Container) generateResolvConf() (string, error) {
// check if systemd-resolved is used, assume it is used when 127.0.0.53 is the only nameserver
if len(ns) == 1 && ns[0] == "127.0.0.53" {
// read the actual resolv.conf file for systemd-resolved
- contents, err = ioutil.ReadFile("/run/systemd/resolve/resolv.conf")
+ resolvedContents, err := ioutil.ReadFile("/run/systemd/resolve/resolv.conf")
if err != nil {
- return "", errors.Wrapf(err, "detected that systemd-resolved is in use, but could not locate real resolv.conf")
+ if !os.IsNotExist(err) {
+ return "", errors.Wrapf(err, "detected that systemd-resolved is in use, but could not locate real resolv.conf")
+ }
+ } else {
+ contents = resolvedContents
}
}
@@ -1815,7 +1820,7 @@ func (c *Container) getHosts() string {
if c.Hostname() != "" {
if c.config.NetMode.IsSlirp4netns() {
// When using slirp4netns, the interface gets a static IP
- slirp4netnsIP, err := GetSlirp4netnsGateway(c.slirp4netnsSubnet)
+ slirp4netnsIP, err := GetSlirp4netnsIP(c.slirp4netnsSubnet)
if err != nil {
logrus.Warn("failed to determine slirp4netnsIP: ", err.Error())
} else {
@@ -2425,3 +2430,77 @@ func (c *Container) createSecretMountDir() error {
return err
}
+
+// Fix ownership and permissions of the specified volume if necessary.
+func (c *Container) fixVolumePermissions(v *ContainerNamedVolume) error {
+ vol, err := c.runtime.state.Volume(v.Name)
+ if err != nil {
+ return errors.Wrapf(err, "error retrieving named volume %s for container %s", v.Name, c.ID())
+ }
+
+ vol.lock.Lock()
+ defer vol.lock.Unlock()
+
+ // The volume may need a copy-up. Check the state.
+ if err := vol.update(); err != nil {
+ return err
+ }
+
+ // TODO: For now, I've disabled chowning volumes owned by non-Podman
+ // drivers. This may be safe, but it's really going to be a case-by-case
+ // thing, I think - safest to leave disabled now and re-enable later if
+ // there is a demand.
+ if vol.state.NeedsChown && !vol.UsesVolumeDriver() {
+ vol.state.NeedsChown = false
+
+ uid := int(c.config.Spec.Process.User.UID)
+ gid := int(c.config.Spec.Process.User.GID)
+
+ if c.config.IDMappings.UIDMap != nil {
+ p := idtools.IDPair{
+ UID: uid,
+ GID: gid,
+ }
+ mappings := idtools.NewIDMappingsFromMaps(c.config.IDMappings.UIDMap, c.config.IDMappings.GIDMap)
+ newPair, err := mappings.ToHost(p)
+ if err != nil {
+ return errors.Wrapf(err, "error mapping user %d:%d", uid, gid)
+ }
+ uid = newPair.UID
+ gid = newPair.GID
+ }
+
+ vol.state.UIDChowned = uid
+ vol.state.GIDChowned = gid
+
+ if err := vol.save(); err != nil {
+ return err
+ }
+
+ mountPoint, err := vol.MountPoint()
+ if err != nil {
+ return err
+ }
+
+ if err := os.Lchown(mountPoint, uid, gid); err != nil {
+ return err
+ }
+
+ // Make sure the new volume matches the permissions of the target directory.
+ // https://github.com/containers/podman/issues/10188
+ st, err := os.Lstat(filepath.Join(c.state.Mountpoint, v.Dest))
+ if err == nil {
+ if err := os.Chmod(mountPoint, st.Mode()|0111); err != nil {
+ return err
+ }
+ stat := st.Sys().(*syscall.Stat_t)
+ atime := time.Unix(int64(stat.Atim.Sec), int64(stat.Atim.Nsec))
+ if err := os.Chtimes(mountPoint, atime, st.ModTime()); err != nil {
+ return err
+ }
+ } else if !os.IsNotExist(err) {
+ return err
+ }
+ }
+ return nil
+}
diff --git a/libpod/container_internal_unsupported.go b/libpod/container_internal_unsupported.go
index f979bcbde..125329ce5 100644
--- a/libpod/container_internal_unsupported.go
+++ b/libpod/container_internal_unsupported.go
@@ -57,3 +57,8 @@ func (c *Container) reloadNetwork() error {
func (c *Container) getUserOverrides() *lookup.Overrides {
return nil
}
+
+// Fix ownership and permissions of the specified volume if necessary.
+func (c *Container) fixVolumePermissions(v *ContainerNamedVolume) error {
+ return define.ErrNotImplemented
+}
diff --git a/libpod/container_log.go b/libpod/container_log.go
index a30e4f5cc..43b3f7736 100644
--- a/libpod/container_log.go
+++ b/libpod/container_log.go
@@ -4,10 +4,12 @@ import (
"context"
"fmt"
"os"
+ "time"
"github.com/containers/podman/v3/libpod/define"
"github.com/containers/podman/v3/libpod/events"
"github.com/containers/podman/v3/libpod/logs"
+ "github.com/hpcloud/tail/watch"
"github.com/pkg/errors"
"github.com/sirupsen/logrus"
)
@@ -74,7 +76,7 @@ func (c *Container) readFromLogFile(ctx context.Context, options *logs.LogOption
}
nll, err := logs.NewLogLine(line.Text)
if err != nil {
- logrus.Error(err)
+ logrus.Errorf("Error getting new log line: %v", err)
continue
}
if nll.Partial() {
@@ -93,17 +95,20 @@ func (c *Container) readFromLogFile(ctx context.Context, options *logs.LogOption
}()
// Check if container is still running or paused
if options.Follow {
+ // If the container isn't running or if we encountered an error
+ // getting its state, instruct the logger to read the file
+ // until EOF.
state, err := c.State()
if err != nil || state != define.ContainerStateRunning {
- // If the container isn't running or if we encountered
- // an error getting its state, instruct the logger to
- // read the file until EOF.
+ // Make sure to wait at least for the poll duration
+ // before stopping the file logger (see #10675).
+ time.Sleep(watch.POLL_DURATION)
tailError := t.StopAtEOF()
if tailError != nil && fmt.Sprintf("%v", tailError) != "tail: stop at eof" {
- logrus.Error(tailError)
+ logrus.Errorf("Error stopping logger: %v", tailError)
}
- if errors.Cause(err) != define.ErrNoSuchCtr {
- logrus.Error(err)
+ if err != nil && errors.Cause(err) != define.ErrNoSuchCtr {
+ logrus.Errorf("Error getting container state: %v", err)
}
return nil
}
@@ -124,9 +129,12 @@ func (c *Container) readFromLogFile(ctx context.Context, options *logs.LogOption
// Now wait for the died event and signal to finish
// reading the log until EOF.
<-eventChannel
+ // Make sure to wait at least for the poll duration
+ // before stopping the file logger (see #10675).
+ time.Sleep(watch.POLL_DURATION)
tailError := t.StopAtEOF()
if tailError != nil && fmt.Sprintf("%v", tailError) != "tail: stop at eof" {
- logrus.Error(tailError)
+ logrus.Errorf("Error stopping logger: %v", tailError)
}
}()
}
diff --git a/libpod/define/containerstate.go b/libpod/define/containerstate.go
index 5d2bc9099..fc272beaa 100644
--- a/libpod/define/containerstate.go
+++ b/libpod/define/containerstate.go
@@ -1,6 +1,10 @@
package define
-import "github.com/pkg/errors"
+import (
+ "time"
+
+ "github.com/pkg/errors"
+)
// ContainerStatus represents the current state of a container
type ContainerStatus int
@@ -120,12 +124,14 @@ func (s ContainerExecStatus) String() string {
// ContainerStats contains the statistics information for a running container
type ContainerStats struct {
+ AvgCPU float64
ContainerID string
Name string
PerCPU []uint64
CPU float64
CPUNano uint64
CPUSystemNano uint64
+ DataPoints int64
SystemNano uint64
MemUsage uint64
MemLimit uint64
@@ -135,4 +141,6 @@ type ContainerStats struct {
BlockInput uint64
BlockOutput uint64
PIDs uint64
+ UpTime time.Duration
+ Duration uint64
}
diff --git a/libpod/events.go b/libpod/events.go
index 839229674..22c51aeec 100644
--- a/libpod/events.go
+++ b/libpod/events.go
@@ -46,7 +46,22 @@ func (c *Container) newContainerExitedEvent(exitCode int32) {
e.Type = events.Container
e.ContainerExitCode = int(exitCode)
if err := c.runtime.eventer.Write(e); err != nil {
- logrus.Errorf("unable to write pod event: %q", err)
+ logrus.Errorf("unable to write container exited event: %q", err)
+ }
+}
+
+// newExecDiedEvent creates a new event for an exec session's death
+func (c *Container) newExecDiedEvent(sessionID string, exitCode int) {
+ e := events.NewEvent(events.ExecDied)
+ e.ID = c.ID()
+ e.Name = c.Name()
+ e.Image = c.config.RootfsImageName
+ e.Type = events.Container
+ e.ContainerExitCode = exitCode
+ e.Attributes = make(map[string]string)
+ e.Attributes["execID"] = sessionID
+ if err := c.runtime.eventer.Write(e); err != nil {
+ logrus.Errorf("unable to write exec died event: %q", err)
}
}
@@ -154,3 +169,25 @@ func (r *Runtime) GetLastContainerEvent(ctx context.Context, nameOrID string, co
// return the last element in the slice
return containerEvents[len(containerEvents)-1], nil
}
+
+// GetExecDiedEvent takes a container name or ID, exec session ID, and returns
+// that exec session's Died event (if it has already occurred).
+func (r *Runtime) GetExecDiedEvent(ctx context.Context, nameOrID, execSessionID string) (*events.Event, error) {
+ filters := []string{
+ fmt.Sprintf("container=%s", nameOrID),
+ "event=exec_died",
+ "type=container",
+ fmt.Sprintf("label=execID=%s", execSessionID),
+ }
+
+ containerEvents, err := r.GetEvents(ctx, filters)
+ if err != nil {
+ return nil, err
+ }
+ // There *should* only be one event maximum.
+ // But... just in case... let's not blow up if there's more than one.
+ if len(containerEvents) < 1 {
+ return nil, errors.Wrapf(events.ErrEventNotFound, "exec died event for session %s (container %s) not found", execSessionID, nameOrID)
+ }
+ return containerEvents[len(containerEvents)-1], nil
+}
diff --git a/libpod/events/config.go b/libpod/events/config.go
index 085fa9d52..d88d7b6e3 100644
--- a/libpod/events/config.go
+++ b/libpod/events/config.go
@@ -127,6 +127,8 @@ const (
Create Status = "create"
// Exec ...
Exec Status = "exec"
+ // ExecDied indicates that an exec session in a container died.
+ ExecDied Status = "exec_died"
// Exited indicates that a container's process died
Exited Status = "died"
// Export ...
diff --git a/libpod/events/events.go b/libpod/events/events.go
index 01ea6a386..e03215eff 100644
--- a/libpod/events/events.go
+++ b/libpod/events/events.go
@@ -149,6 +149,8 @@ func StringToStatus(name string) (Status, error) {
return Create, nil
case Exec.String():
return Exec, nil
+ case ExecDied.String():
+ return ExecDied, nil
case Exited.String():
return Exited, nil
case Export.String():
diff --git a/libpod/networking_linux.go b/libpod/networking_linux.go
index c928e02a6..1b775a4f3 100644
--- a/libpod/networking_linux.go
+++ b/libpod/networking_linux.go
@@ -46,6 +46,9 @@ const (
// rootlessCNINSName is the file name for the rootless network namespace bind mount
rootlessCNINSName = "rootless-cni-ns"
+
+ // persistentCNIDir is the directory where the CNI files are stored
+ persistentCNIDir = "/var/lib/cni"
)
// Get an OCICNI network config
@@ -150,14 +153,31 @@ func (r *RootlessCNI) Do(toRun func() error) error {
}
}
- // cni plugins need access to /var and /run
- runDir := filepath.Join(r.dir, "run")
- varDir := filepath.Join(r.dir, "var")
+ // cni plugins need access to /var/lib/cni and /run
+ varDir := ""
+ varTarget := persistentCNIDir
+ // we can only mount to a target dir which exists, check /var/lib/cni recursively
+ // while we could always use /var there are cases where a user might store the cni
+ // configs under /var/custom and this would break
+ for {
+ if _, err := os.Stat(varTarget); err == nil {
+ varDir = filepath.Join(r.dir, strings.TrimPrefix(varTarget, "/"))
+ break
+ }
+ varTarget = filepath.Base(varTarget)
+ if varTarget == "/" {
+ break
+ }
+ }
+ if varDir == "" {
+ return errors.New("failed to stat /var directory")
+ }
// make sure to mount var first
- err = unix.Mount(varDir, "/var", "none", unix.MS_BIND, "")
+ err = unix.Mount(varDir, varTarget, "none", unix.MS_BIND, "")
if err != nil {
- return errors.Wrap(err, "failed to mount /var for rootless cni")
+ return errors.Wrapf(err, "failed to mount %s for rootless cni", varTarget)
}
+ runDir := filepath.Join(r.dir, "run")
// recursive mount to keep the netns mount
err = unix.Mount(runDir, "/run", "none", unix.MS_BIND|unix.MS_REC, "")
if err != nil {
@@ -385,7 +405,7 @@ func (r *Runtime) GetRootlessCNINetNs(new bool) (*RootlessCNI, error) {
// create cni directories to store files
// they will be bind mounted to the correct location in a extra mount ns
- err = os.MkdirAll(filepath.Join(cniDir, "var"), 0700)
+ err = os.MkdirAll(filepath.Join(cniDir, strings.TrimPrefix(persistentCNIDir, "/")), 0700)
if err != nil {
return nil, errors.Wrap(err, "could not create rootless-cni var directory")
}
@@ -888,6 +908,10 @@ func (c *Container) getContainerNetworkInfo() (*define.InspectNetworkSettings, e
if err != nil {
return nil, err
}
+ // see https://github.com/containers/podman/issues/10090
+ // the container has to be locked for syncContainer()
+ netNsCtr.lock.Lock()
+ defer netNsCtr.lock.Unlock()
// Have to sync to ensure that state is populated
if err := netNsCtr.syncContainer(); err != nil {
return nil, err
@@ -1043,7 +1067,7 @@ func resultToBasicNetworkConfig(result *cnitypes.Result) (define.InspectBasicNet
// after itself on an unclean reboot. Return what we're pretty sure is the path
// to CNI's internal files (it's not really exposed to us).
func getCNINetworksDir() (string, error) {
- return "/var/lib/cni/networks", nil
+ return filepath.Join(persistentCNIDir, "networks"), nil
}
type logrusDebugWriter struct {
@@ -1090,7 +1114,7 @@ func (c *Container) NetworkDisconnect(nameOrID, netName string, force bool) erro
}
c.newNetworkEvent(events.NetworkDisconnect, netName)
- if c.state.State != define.ContainerStateRunning {
+ if !c.ensureState(define.ContainerStateRunning, define.ContainerStateCreated) {
return nil
}
@@ -1145,7 +1169,7 @@ func (c *Container) NetworkConnect(nameOrID, netName string, aliases []string) e
return err
}
c.newNetworkEvent(events.NetworkConnect, netName)
- if c.state.State != define.ContainerStateRunning {
+ if !c.ensureState(define.ContainerStateRunning, define.ContainerStateCreated) {
return nil
}
if c.state.NetNS == nil {
diff --git a/libpod/oci.go b/libpod/oci.go
index 1f2c7dd71..c92d9a077 100644
--- a/libpod/oci.go
+++ b/libpod/oci.go
@@ -72,13 +72,16 @@ type OCIRuntime interface {
// has completed, as one might expect. The attach session will remain
// running, in a goroutine that will return via the chan error in the
// return signature.
- ExecContainer(ctr *Container, sessionID string, options *ExecOptions, streams *define.AttachStreams) (int, chan error, error)
+ // newSize resizes the tty to this size before the process is started, must be nil if the exec session has no tty
+ ExecContainer(ctr *Container, sessionID string, options *ExecOptions, streams *define.AttachStreams, newSize *define.TerminalSize) (int, chan error, error)
// ExecContainerHTTP executes a command in a running container and
// attaches its standard streams to a provided hijacked HTTP session.
// Maintains the same invariants as ExecContainer (returns on session
// start, with a goroutine running in the background to handle attach).
// The HTTP attach itself maintains the same invariants as HTTPAttach.
- ExecContainerHTTP(ctr *Container, sessionID string, options *ExecOptions, r *http.Request, w http.ResponseWriter, streams *HTTPAttachStreams, cancel <-chan bool, hijackDone chan<- bool, holdConnOpen <-chan bool) (int, chan error, error)
+ // newSize resizes the tty to this size before the process is started, must be nil if the exec session has no tty
+ ExecContainerHTTP(ctr *Container, sessionID string, options *ExecOptions, r *http.Request, w http.ResponseWriter,
+ streams *HTTPAttachStreams, cancel <-chan bool, hijackDone chan<- bool, holdConnOpen <-chan bool, newSize *define.TerminalSize) (int, chan error, error)
// ExecContainerDetached executes a command in a running container, but
// does not attach to it. Returns the PID of the exec session and an
// error (if starting the exec session failed)
diff --git a/libpod/oci_attach_linux.go b/libpod/oci_attach_linux.go
index b5040de3e..de435b58a 100644
--- a/libpod/oci_attach_linux.go
+++ b/libpod/oci_attach_linux.go
@@ -94,17 +94,18 @@ func (c *Container) attach(streams *define.AttachStreams, keys string, resize <-
// this ensures attachToExec gets all of the output of the called process
// conmon will then send the exit code of the exec process, or an error in the exec session
// startFd must be the input side of the fd.
+// newSize resizes the tty to this size before the process is started, must be nil if the exec session has no tty
// conmon will wait to start the exec session until the parent process has setup the console socket.
// Once attachToExec successfully attaches to the console socket, the child conmon process responsible for calling runtime exec
// will read from the output side of start fd, thus learning to start the child process.
// Thus, the order goes as follow:
// 1. conmon parent process sets up its console socket. sends on attachFd
-// 2. attachToExec attaches to the console socket after reading on attachFd
+// 2. attachToExec attaches to the console socket after reading on attachFd and resizes the tty
// 3. child waits on startFd for attachToExec to attach to said console socket
// 4. attachToExec sends on startFd, signalling it has attached to the socket and child is ready to go
// 5. child receives on startFd, runs the runtime exec command
// attachToExec is responsible for closing startFd and attachFd
-func (c *Container) attachToExec(streams *define.AttachStreams, keys *string, sessionID string, startFd, attachFd *os.File) error {
+func (c *Container) attachToExec(streams *define.AttachStreams, keys *string, sessionID string, startFd, attachFd *os.File, newSize *define.TerminalSize) error {
if !streams.AttachOutput && !streams.AttachError && !streams.AttachInput {
return errors.Wrapf(define.ErrInvalidArg, "must provide at least one stream to attach to")
}
@@ -137,6 +138,14 @@ func (c *Container) attachToExec(streams *define.AttachStreams, keys *string, se
return err
}
+ // resize before we start the container process
+ if newSize != nil {
+ err = c.ociRuntime.ExecAttachResize(c, sessionID, *newSize)
+ if err != nil {
+ logrus.Warn("resize failed", err)
+ }
+ }
+
// 2: then attach
conn, err := openUnixSocket(sockPath)
if err != nil {
diff --git a/libpod/oci_conmon_exec_linux.go b/libpod/oci_conmon_exec_linux.go
index 76338b86c..09d3d1833 100644
--- a/libpod/oci_conmon_exec_linux.go
+++ b/libpod/oci_conmon_exec_linux.go
@@ -25,7 +25,7 @@ import (
)
// ExecContainer executes a command in a running container
-func (r *ConmonOCIRuntime) ExecContainer(c *Container, sessionID string, options *ExecOptions, streams *define.AttachStreams) (int, chan error, error) {
+func (r *ConmonOCIRuntime) ExecContainer(c *Container, sessionID string, options *ExecOptions, streams *define.AttachStreams, newSize *define.TerminalSize) (int, chan error, error) {
if options == nil {
return -1, nil, errors.Wrapf(define.ErrInvalidArg, "must provide an ExecOptions struct to ExecContainer")
}
@@ -68,7 +68,7 @@ func (r *ConmonOCIRuntime) ExecContainer(c *Container, sessionID string, options
attachChan := make(chan error)
go func() {
// attachToExec is responsible for closing pipes
- attachChan <- c.attachToExec(streams, options.DetachKeys, sessionID, pipes.startPipe, pipes.attachPipe)
+ attachChan <- c.attachToExec(streams, options.DetachKeys, sessionID, pipes.startPipe, pipes.attachPipe, newSize)
close(attachChan)
}()
@@ -83,7 +83,8 @@ func (r *ConmonOCIRuntime) ExecContainer(c *Container, sessionID string, options
// ExecContainerHTTP executes a new command in an existing container and
// forwards its standard streams over an attach
-func (r *ConmonOCIRuntime) ExecContainerHTTP(ctr *Container, sessionID string, options *ExecOptions, req *http.Request, w http.ResponseWriter, streams *HTTPAttachStreams, cancel <-chan bool, hijackDone chan<- bool, holdConnOpen <-chan bool) (int, chan error, error) {
+func (r *ConmonOCIRuntime) ExecContainerHTTP(ctr *Container, sessionID string, options *ExecOptions, req *http.Request, w http.ResponseWriter,
+ streams *HTTPAttachStreams, cancel <-chan bool, hijackDone chan<- bool, holdConnOpen <-chan bool, newSize *define.TerminalSize) (int, chan error, error) {
if streams != nil {
if !streams.Stdin && !streams.Stdout && !streams.Stderr {
return -1, nil, errors.Wrapf(define.ErrInvalidArg, "must provide at least one stream to attach to")
@@ -133,7 +134,7 @@ func (r *ConmonOCIRuntime) ExecContainerHTTP(ctr *Container, sessionID string, o
conmonPipeDataChan := make(chan conmonPipeData)
go func() {
// attachToExec is responsible for closing pipes
- attachChan <- attachExecHTTP(ctr, sessionID, req, w, streams, pipes, detachKeys, options.Terminal, cancel, hijackDone, holdConnOpen, execCmd, conmonPipeDataChan, ociLog)
+ attachChan <- attachExecHTTP(ctr, sessionID, req, w, streams, pipes, detachKeys, options.Terminal, cancel, hijackDone, holdConnOpen, execCmd, conmonPipeDataChan, ociLog, newSize)
close(attachChan)
}()
@@ -486,7 +487,7 @@ func (r *ConmonOCIRuntime) startExec(c *Container, sessionID string, options *Ex
}
// Attach to a container over HTTP
-func attachExecHTTP(c *Container, sessionID string, r *http.Request, w http.ResponseWriter, streams *HTTPAttachStreams, pipes *execPipes, detachKeys []byte, isTerminal bool, cancel <-chan bool, hijackDone chan<- bool, holdConnOpen <-chan bool, execCmd *exec.Cmd, conmonPipeDataChan chan<- conmonPipeData, ociLog string) (deferredErr error) {
+func attachExecHTTP(c *Container, sessionID string, r *http.Request, w http.ResponseWriter, streams *HTTPAttachStreams, pipes *execPipes, detachKeys []byte, isTerminal bool, cancel <-chan bool, hijackDone chan<- bool, holdConnOpen <-chan bool, execCmd *exec.Cmd, conmonPipeDataChan chan<- conmonPipeData, ociLog string, newSize *define.TerminalSize) (deferredErr error) {
// NOTE: As you may notice, the attach code is quite complex.
// Many things happen concurrently and yet are interdependent.
// If you ever change this function, make sure to write to the
@@ -524,6 +525,14 @@ func attachExecHTTP(c *Container, sessionID string, r *http.Request, w http.Resp
return err
}
+ // resize before we start the container process
+ if newSize != nil {
+ err = c.ociRuntime.ExecAttachResize(c, sessionID, *newSize)
+ if err != nil {
+ logrus.Warn("resize failed", err)
+ }
+ }
+
// 2: then attach
conn, err := openUnixSocket(sockPath)
if err != nil {
diff --git a/libpod/oci_conmon_linux.go b/libpod/oci_conmon_linux.go
index 3da49b85f..2914bd1a1 100644
--- a/libpod/oci_conmon_linux.go
+++ b/libpod/oci_conmon_linux.go
@@ -787,7 +787,11 @@ func (r *ConmonOCIRuntime) CheckpointContainer(ctr *Container, options Container
args = append(args, "--pre-dump")
}
if !options.PreCheckPoint && options.WithPrevious {
- args = append(args, "--parent-path", ctr.PreCheckPointPath())
+ args = append(
+ args,
+ "--parent-path",
+ filepath.Join("..", preCheckpointDir),
+ )
}
runtimeDir, err := util.GetRuntimeDir()
if err != nil {
diff --git a/libpod/oci_missing.go b/libpod/oci_missing.go
index 10526f368..fcf2ffca8 100644
--- a/libpod/oci_missing.go
+++ b/libpod/oci_missing.go
@@ -119,12 +119,13 @@ func (r *MissingRuntime) AttachResize(ctr *Container, newSize define.TerminalSiz
}
// ExecContainer is not available as the runtime is missing
-func (r *MissingRuntime) ExecContainer(ctr *Container, sessionID string, options *ExecOptions, streams *define.AttachStreams) (int, chan error, error) {
+func (r *MissingRuntime) ExecContainer(ctr *Container, sessionID string, options *ExecOptions, streams *define.AttachStreams, newSize *define.TerminalSize) (int, chan error, error) {
return -1, nil, r.printError()
}
// ExecContainerHTTP is not available as the runtime is missing
-func (r *MissingRuntime) ExecContainerHTTP(ctr *Container, sessionID string, options *ExecOptions, req *http.Request, w http.ResponseWriter, streams *HTTPAttachStreams, cancel <-chan bool, hijackDone chan<- bool, holdConnOpen <-chan bool) (int, chan error, error) {
+func (r *MissingRuntime) ExecContainerHTTP(ctr *Container, sessionID string, options *ExecOptions, req *http.Request, w http.ResponseWriter,
+ streams *HTTPAttachStreams, cancel <-chan bool, hijackDone chan<- bool, holdConnOpen <-chan bool, newSize *define.TerminalSize) (int, chan error, error) {
return -1, nil, r.printError()
}
diff --git a/libpod/options.go b/libpod/options.go
index f942d264b..f2468e41b 100644
--- a/libpod/options.go
+++ b/libpod/options.go
@@ -7,6 +7,7 @@ import (
"strings"
"syscall"
+ "github.com/containers/buildah/pkg/parse"
"github.com/containers/common/pkg/config"
"github.com/containers/common/pkg/secrets"
"github.com/containers/image/v5/manifest"
@@ -268,8 +269,11 @@ func WithRegistriesConf(path string) RuntimeOption {
return errors.Wrap(err, "error locating specified registries.conf")
}
if rt.imageContext == nil {
- rt.imageContext = &types.SystemContext{}
+ rt.imageContext = &types.SystemContext{
+ BigFilesTemporaryDir: parse.GetTempDir(),
+ }
}
+
rt.imageContext.SystemRegistriesConfPath = path
return nil
}
@@ -1641,6 +1645,19 @@ func WithVolumeGID(gid int) VolumeCreateOption {
}
}
+// WithVolumeNoChown prevents the volume from being chowned to the process uid at first use.
+func WithVolumeNoChown() VolumeCreateOption {
+ return func(volume *Volume) error {
+ if volume.valid {
+ return define.ErrVolumeFinalized
+ }
+
+ volume.state.NeedsChown = false
+
+ return nil
+ }
+}
+
// withSetAnon sets a bool notifying libpod that this volume is anonymous and
// should be removed when containers using it are removed and volumes are
// specified for removal.
diff --git a/libpod/runtime.go b/libpod/runtime.go
index d775b55e1..f53789e89 100644
--- a/libpod/runtime.go
+++ b/libpod/runtime.go
@@ -15,6 +15,7 @@ import (
"syscall"
"time"
+ "github.com/containers/buildah/pkg/parse"
"github.com/containers/common/libimage"
"github.com/containers/common/pkg/config"
"github.com/containers/common/pkg/defaultnet"
@@ -381,7 +382,9 @@ func makeRuntime(ctx context.Context, runtime *Runtime) (retErr error) {
// Set up containers/image
if runtime.imageContext == nil {
- runtime.imageContext = &types.SystemContext{}
+ runtime.imageContext = &types.SystemContext{
+ BigFilesTemporaryDir: parse.GetTempDir(),
+ }
}
runtime.imageContext.SignaturePolicyPath = runtime.config.Engine.SignaturePolicyPath
@@ -465,7 +468,7 @@ func makeRuntime(ctx context.Context, runtime *Runtime) (retErr error) {
}
// Set up the CNI net plugin
- netPlugin, err := ocicni.InitCNI(runtime.config.Network.DefaultNetwork, runtime.config.Network.NetworkConfigDir, runtime.config.Network.CNIPluginDirs...)
+ netPlugin, err := ocicni.InitCNINoInotify(runtime.config.Network.DefaultNetwork, runtime.config.Network.NetworkConfigDir, "", runtime.config.Network.CNIPluginDirs...)
if err != nil {
return errors.Wrapf(err, "error configuring CNI network plugin")
}
diff --git a/libpod/stats.go b/libpod/stats.go
index f4732b4fc..6f0360ef1 100644
--- a/libpod/stats.go
+++ b/libpod/stats.go
@@ -56,7 +56,11 @@ func (c *Container) GetContainerStats(previousStats *define.ContainerStats) (*de
previousCPU := previousStats.CPUNano
now := uint64(time.Now().UnixNano())
+ stats.Duration = cgroupStats.CPU.Usage.Total
+ stats.UpTime = time.Duration(stats.Duration)
stats.CPU = calculateCPUPercent(cgroupStats, previousCPU, now, previousStats.SystemNano)
+ stats.AvgCPU = calculateAvgCPU(stats.CPU, previousStats.AvgCPU, previousStats.DataPoints)
+ stats.DataPoints = previousStats.DataPoints + 1
stats.MemUsage = cgroupStats.Memory.Usage.Usage
stats.MemLimit = getMemLimit(cgroupStats.Memory.Usage.Limit)
stats.MemPerc = (float64(stats.MemUsage) / float64(stats.MemLimit)) * 100
@@ -127,3 +131,9 @@ func calculateBlockIO(stats *cgroups.Metrics) (read uint64, write uint64) {
}
return
}
+
+// calculateAvgCPU calculates the avg CPU percentage given the previous average and the number of data points.
+func calculateAvgCPU(statsCPU float64, prevAvg float64, prevData int64) float64 {
+ avgPer := ((prevAvg * float64(prevData)) + statsCPU) / (float64(prevData) + 1)
+ return avgPer
+}
diff --git a/libpod/volume_internal.go b/libpod/volume_internal.go
index 694cdd149..19008a253 100644
--- a/libpod/volume_internal.go
+++ b/libpod/volume_internal.go
@@ -39,8 +39,23 @@ func (v *Volume) needsMount() bool {
return true
}
- // Local driver with options needs mount
- return len(v.config.Options) > 0
+ // Commit 28138dafcc added the UID and GID options to this map
+ // However we should only mount when options other than uid and gid are set.
+ // see https://github.com/containers/podman/issues/10620
+ index := 0
+ if _, ok := v.config.Options["UID"]; ok {
+ index++
+ }
+ if _, ok := v.config.Options["GID"]; ok {
+ index++
+ }
+ // when uid or gid is set there is also the "o" option
+ // set so we have to ignore this one as well
+ if index > 0 {
+ index++
+ }
+ // Local driver with options other than uid,gid needs mount
+ return len(v.config.Options) > index
}
// update() updates the volume state from the DB.
generated by cgit v1.2.3-54-g00ecf (git 2.39.0) at 2025-01-13 11:53:56 +0000