Files
kubernetes/vendor/github.com/opencontainers/runc/libcontainer/cgroups/fs2/devices.go
Kir Kolyshkin f3cdfc488e vendor: bump runc to rc95
runc rc95 contains a fix for CVE-2021-30465.

runc rc94 provides fixes and improvements.

One notable change is cgroup manager's Set now accept Resources rather
than Cgroup (see https://github.com/opencontainers/runc/pull/2906).
Modify the code accordingly.

Also update runc dependencies (as hinted by hack/lint-depdendencies.sh):

        github.com/cilium/ebpf v0.5.0
        github.com/containerd/console v1.0.2
        github.com/coreos/go-systemd/v22 v22.3.1
        github.com/godbus/dbus/v5 v5.0.4
        github.com/moby/sys/mountinfo v0.4.1
        golang.org/x/sys v0.0.0-20210426230700-d19ff857e887
        github.com/google/go-cmp v0.5.4
        github.com/kr/pretty v0.2.1
        github.com/opencontainers/runtime-spec v1.0.3-0.20210326190908-1c3f411f0417

Signed-off-by: Kir Kolyshkin <kolyshkin@gmail.com>
2021-05-19 23:51:59 -07:00

91 lines
3.0 KiB
Go

// +build linux
package fs2
import (
"github.com/opencontainers/runc/libcontainer/cgroups/ebpf"
"github.com/opencontainers/runc/libcontainer/cgroups/ebpf/devicefilter"
"github.com/opencontainers/runc/libcontainer/configs"
"github.com/opencontainers/runc/libcontainer/devices"
"github.com/opencontainers/runc/libcontainer/userns"
"github.com/pkg/errors"
"golang.org/x/sys/unix"
)
func isRWM(perms devices.Permissions) bool {
var r, w, m bool
for _, perm := range perms {
switch perm {
case 'r':
r = true
case 'w':
w = true
case 'm':
m = true
}
}
return r && w && m
}
// This is similar to the logic applied in crun for handling errors from bpf(2)
// <https://github.com/containers/crun/blob/0.17/src/libcrun/cgroup.c#L2438-L2470>.
func canSkipEBPFError(r *configs.Resources) bool {
// If we're running in a user namespace we can ignore eBPF rules because we
// usually cannot use bpf(2), as well as rootless containers usually don't
// have the necessary privileges to mknod(2) device inodes or access
// host-level instances (though ideally we would be blocking device access
// for rootless containers anyway).
if userns.RunningInUserNS() {
return true
}
// We cannot ignore an eBPF load error if any rule if is a block rule or it
// doesn't permit all access modes.
//
// NOTE: This will sometimes trigger in cases where access modes are split
// between different rules but to handle this correctly would require
// using ".../libcontainer/cgroup/devices".Emulator.
for _, dev := range r.Devices {
if !dev.Allow || !isRWM(dev.Permissions) {
return false
}
}
return true
}
func setDevices(dirPath string, r *configs.Resources) error {
if r.SkipDevices {
return nil
}
// XXX: This is currently a white-list (but all callers pass a blacklist of
// devices). This is bad for a whole variety of reasons, but will need
// to be fixed with co-ordinated effort with downstreams.
insts, license, err := devicefilter.DeviceFilter(r.Devices)
if err != nil {
return err
}
dirFD, err := unix.Open(dirPath, unix.O_DIRECTORY|unix.O_RDONLY, 0600)
if err != nil {
return errors.Errorf("cannot get dir FD for %s", dirPath)
}
defer unix.Close(dirFD)
// XXX: This code is currently incorrect when it comes to updating an
// existing cgroup with new rules (new rulesets are just appended to
// the program list because this uses BPF_F_ALLOW_MULTI). If we didn't
// use BPF_F_ALLOW_MULTI we could actually atomically swap the
// programs.
//
// The real issue is that BPF_F_ALLOW_MULTI makes it hard to have a
// race-free blacklist because it acts as a whitelist by default, and
// having a deny-everything program cannot be overridden by other
// programs. You could temporarily insert a deny-everything program
// but that would result in spurrious failures during updates.
if _, err := ebpf.LoadAttachCgroupDeviceFilter(insts, license, dirFD); err != nil {
if !canSkipEBPFError(r) {
return err
}
}
return nil
}