dockerfile: run buildkitd within a cgroup namespace for cgroup v2

Introduce a new entrypoint script for the Linux image that, if cgroup v2
is in use, creates a new cgroup and mount namespace for buildkitd within
a new entrypoint using `unshare` and remounts `/sys/fs/cgroup` to
restrict its view of the unified cgroup hierarchy. This will ensure its
`init` cgroup and all OCI worker managed cgroups are kept beneath the
root cgroup of the initial entrypoint process.

When buildkitd is run in a managed environment like Kubernetes without
its own cgroup namespace (the default behavior of privileged pods in
Kubernetes where cgroup v2 is in use; see [cgroup v2 KEP][kep]), the OCI
worker will spawn processes in cgroups that are outside of the cgroup
hierarchy that was created for the buildkitd container, leading to
incorrect resource accounting and enforcement which in turn can cause
OOM errors and CPU contention on the node.

Example behavior without this change:

```console
root@k8s-node:/# cat /proc/$(pgrep -n buildkitd)/cgroup
0::/init
root@k8s-node:/# cat /proc/$(pgrep -n some-build-process)/cgroup
0::/buildkit/{runc-container-id}
```

Example behavior with this change:

```console
root@k8s-node:/# cat /proc/$(pgrep -n buildkitd)/cgroup
0::/kubepods/burstable/pod{pod-id}/{container-id}/init
root@k8s-node:/# cat /proc/$(pgrep -n some-build-process)/cgroup
0::/kubepods/burstable/pod{pod-id}/{container-id}/buildkit/{runc-container-id}
```

Note this was developed as an alternative approach to #6343

[kep]: https://github.com/kubernetes/enhancements/tree/6d3210f7dd5d547c8f7f6a33af6a09eb45193cd7/keps/sig-node/2254-cgroup-v2#cgroup-namespace

Signed-off-by: Dan Duvall <[email protected]>

Author: marxarelli

Dockerfile

@@ -205,7 +205,7 @@ FROM scratch AS release
 COPY --link --from=releaser /out/ /
 
 FROM alpine:${ALPINE_VERSION} AS buildkit-export-alpine
-RUN apk add --no-cache fuse3 git openssh openssl pigz xz iptables ip6tables \
+RUN apk add --no-cache fuse3 git openssh openssl pigz xz iptables ip6tables util-linux-misc \
   && ln -s fusermount3 /usr/bin/fusermount
 COPY --link examples/buildctl-daemonless/buildctl-daemonless.sh /usr/bin/
 VOLUME /var/lib/buildkit
@@ -380,8 +380,36 @@ EOT
 
 FROM buildkit-export AS buildkit-linux
 COPY --link --from=binaries / /usr/bin/
+COPY --link --chmod=755 <<EOF /usr/bin/buildkitd-entrypoint
+#!/bin/sh
+#
+# For cgroup v2, ensure buildkitd has a namespaced view of /sys/fs/cgroup by
+# running in a new cgroup and mount namespace and remounting /sys/fs/cgroup.
+# Assume we are already in own our cgroup ns if the current cgroup path is "/".
+
+set -e
+
+if [ -e /sys/fs/cgroup/cgroup.controllers ]; then
+  if [ "\$(cut -d: -f3 /proc/self/cgroup)" != "/" ]; then
+    echo creating cgroup namespace
+    exec /usr/bin/unshare --cgroup --mount /usr/bin/with-cgroupfs-remount buildkitd "\$@"
+  fi
+fi
+
+exec /usr/bin/buildkitd "\$@"
+EOF
+COPY --link --chmod=755 <<EOF /usr/bin/with-cgroupfs-remount
+#!/bin/sh
+set -e
+
+options="\$(awk '\$2 == "/sys/fs/cgroup" { print \$4 }' /proc/self/mounts)"
+umount /sys/fs/cgroup
+mount -t cgroup2 -o "\$options" cgroup2 /sys/fs/cgroup
+
+exec "\$@"
+EOF
 ENV BUILDKIT_SETUP_CGROUPV2_ROOT=1
-ENTRYPOINT ["buildkitd"]
+ENTRYPOINT ["/usr/bin/buildkitd-entrypoint"]
 
 FROM buildkit-linux AS buildkit-linux-debug
 COPY --link --from=dlv /out/dlv /usr/bin/dlv