systemd: fix rootful-in-userns regression

shouldUseRootlessCgroupManager() was always returning true when we are inside a userns.

If the systemd OwnerUID is 0, we have rootful systemd inside the userns, and we can just use the rootful cgroup driver.

Fix #2724

Signed-off-by: Akihiro Suda <akihiro.suda.cz@hco.ntt.co.jp>
This commit is contained in:
Akihiro Suda
2021-01-08 14:08:12 +09:00
parent c751ba3fd9
commit 230a46b72b
+19
View File
@@ -5,7 +5,9 @@ package main
import (
"os"
"github.com/opencontainers/runc/libcontainer/cgroups/systemd"
"github.com/opencontainers/runc/libcontainer/system"
"github.com/sirupsen/logrus"
"github.com/urfave/cli"
)
@@ -28,6 +30,23 @@ func shouldUseRootlessCgroupManager(context *cli.Context) (bool, error) {
return false, nil
}
// euid = 0, in a userns.
//
// [systemd driver]
// We can call DetectUID() to parse the OwnerUID value from `busctl --user --no-pager status` result.
// The value corresponds to sd_bus_creds_get_owner_uid(3).
// If the value is 0, we have rootful systemd inside userns, so we do not need the rootless cgroup manager.
//
// On error, we assume we are root. An error may happen during shelling out to `busctl` CLI,
// mostly when $DBUS_SESSION_BUS_ADDRESS is unset.
if context.GlobalBool("systemd-cgroup") {
ownerUID, err := systemd.DetectUID()
if err != nil {
logrus.WithError(err).Debug("failed to get the OwnerUID value, assuming the value to be 0")
ownerUID = 0
}
return ownerUID != 0, nil
}
// [cgroupfs driver]
// As we are unaware of cgroups path, we can't determine whether we have the full
// access to the cgroups path.
// Either way, we can safely decide to use the rootless cgroups manager.