workaround cgroup race in dind

.dockerignore

@@ -4,7 +4,7 @@
 # Allow files and directories.
 !cmd/
 !internal/
-!scripts/docker-entrypoint.sh
+!scripts/docker/
 !go.*
 
 # Ignore unnecessary files inside allowed directories.

Dockerfile

@@ -1,3 +1,5 @@
+ARG ALPINE_VERSION=3.20.3
+
 FROM golang:1.23.2-alpine AS uncloudd
 
 ARG TARGETOS
@@ -14,9 +16,9 @@ COPY . .
 RUN GOOS=${TARGETOS} GOARCH=${TARGETARCH} go build -o uncloudd cmd/uncloudd/main.go
 
 
-FROM alpine:latest AS corrosion-download
+FROM alpine:${ALPINE_VERSION} AS corrosion-download
 RUN wget -q -O /tmp/corrosion.tar.gz \
-      https://github.com/psviderski/corrosion/releases/latest/download/corrosion-aarch64-unknown-linux-gnu.tar.gz  \
+      https://github.com/psviderski/corrosion/releases/latest/download/corrosion-aarch64-unknown-linux-gnu.tar.gz \
     && tar -xzf /tmp/corrosion.tar.gz -C /tmp \
     && install /tmp/corrosion /usr/local/bin/corrosion \
     && rm /tmp/corrosion.tar.gz /tmp/corrosion
@@ -26,15 +28,23 @@ COPY --from=corrosion-download /usr/local/bin/corrosion /usr/local/bin/corrosion
 CMD ["corrosion", "agent"]
 
 
+FROM alpine:${ALPINE_VERSION} AS corrosion-image-tarball
+ARG CORROSION_IMAGE="ghcr.io/psviderski/corrosion:latest"
+
+RUN apk --no-cache add crane
+RUN crane pull "${CORROSION_IMAGE}" /corrosion.tar
+
+
 FROM docker:27.3.1-dind AS machine
 RUN apk --no-cache add \
     wireguard-tools
 # Create system group and user 'uncloud'.
 RUN addgroup -S uncloud && adduser -SHD -h /nonexistent -G uncloud -g "" uncloud
 
-COPY scripts/docker-entrypoint.sh /usr/local/bin/
+COPY --from=corrosion-image-tarball /corrosion.tar /images/corrosion.tar
+COPY scripts/docker/dind scripts/docker/entrypoint.sh /usr/local/bin/
 COPY --from=uncloudd /build/uncloudd /usr/local/bin/
 # TODO: socat to forward uncloud.sock unix socket?
 
-ENTRYPOINT ["docker-entrypoint.sh"]
+ENTRYPOINT ["entrypoint.sh"]
 CMD ["uncloudd"]

scripts/docker-entrypoint.sh

@@ -1,6 +0,0 @@
-#!/bin/sh
-set -eu
-
-dind dockerd &
-
-exec "$@"

scripts/docker/dind

@@ -0,0 +1,83 @@
+#!/bin/sh
+# This is a fork of https://github.com/moby/moby/blob/65cfcc28ab37cb75e1560e4b4738719c07c6618e/hack/dind
+# with a fix for cgroup v2 initialization races.
+
+set -e
+
+# DinD: a wrapper script which allows docker to be run inside a docker container.
+# Original version by Jerome Petazzoni <jerome@docker.com>
+# See the blog post: https://www.docker.com/blog/docker-can-now-run-within-docker/
+#
+# This script should be executed inside a docker container in privileged mode
+# ('docker run --privileged', introduced in docker 0.6).
+
+# Usage: dind CMD [ARG...]
+
+# apparmor sucks and Docker needs to know that it's in a container (c) @tianon
+#
+# Set the container env-var, so that AppArmor is enabled in the daemon and
+# containerd when running docker-in-docker.
+#
+# see: https://github.com/containerd/containerd/blob/787943dc1027a67f3b52631e084db0d4a6be2ccc/pkg/apparmor/apparmor_linux.go#L29-L45
+# see: https://github.com/moby/moby/commit/de191e86321f7d3136ff42ff75826b8107399497
+export container=docker
+
+# Allow AppArmor to work inside the container;
+#
+#     aa-status
+#     apparmor filesystem is not mounted.
+#     apparmor module is loaded.
+#
+#     mount -t securityfs none /sys/kernel/security
+#
+#     aa-status
+#     apparmor module is loaded.
+#     30 profiles are loaded.
+#     30 profiles are in enforce mode.
+#       /snap/snapd/18357/usr/lib/snapd/snap-confine
+#       ...
+#
+# Note: https://0xn3va.gitbook.io/cheat-sheets/container/escaping/sensitive-mounts#sys-kernel-security
+#
+#     ## /sys/kernel/security
+#
+#     In /sys/kernel/security mounted the securityfs interface, which allows
+#     configuration of Linux Security Modules. This allows configuration of
+#     AppArmor policies, and so access to this may allow a container to disable
+#     its MAC system.
+#
+# Given that we're running privileged already, this should not be an issue.
+if [ -d /sys/kernel/security ] && ! mountpoint -q /sys/kernel/security; then
+	mount -t securityfs none /sys/kernel/security || {
+		echo >&2 'Could not mount /sys/kernel/security.'
+		echo >&2 'AppArmor detection and --privileged mode might break.'
+	}
+fi
+
+# Mount /tmp (conditionally)
+if ! mountpoint -q /tmp; then
+	mount -t tmpfs none /tmp
+fi
+
+# cgroup v2: enable nesting
+if [ -f /sys/fs/cgroup/cgroup.controllers ]; then
+	# move the processes from the root group to the /init group,
+	# otherwise writing subtree_control fails with EBUSY.
+	# An error during moving non-existent process (i.e., "cat") is ignored.
+	mkdir -p /sys/fs/cgroup/init
+	xargs -rn1 < /sys/fs/cgroup/cgroup.procs > /sys/fs/cgroup/init/cgroup.procs || :
+	# enable controllers with retry on cgroup initialization races
+	timeout 5s sh -c "until sed -e 's/ / +/g' -e 's/^/+/' < /sys/fs/cgroup/cgroup.controllers \
+		> /sys/fs/cgroup/cgroup.subtree_control 2>/dev/null; do sleep 0.1; done"
+fi
+
+# Change mount propagation to shared to make the environment more similar to a
+# modern Linux system, e.g. with SystemD as PID 1.
+mount --make-rshared /
+
+if [ $# -gt 0 ]; then
+	exec "$@"
+fi
+
+echo >&2 'ERROR: No command specified.'
+echo >&2 'You probably want to run hack/make.sh, or maybe a shell?'

scripts/docker/entrypoint.sh

@@ -0,0 +1,12 @@
+#!/bin/sh
+set -eu
+
+dind dockerd &
+echo "Waiting for Docker in Docker to be ready..."
+timeout 5s sh -c "until docker info &> /dev/null; do sleep 0.1; done"
+echo "Docker in Docker is ready."
+
+echo "Loading corrosion image from /images/corrosion.tar..."
+docker load < /images/corrosion.tar
+
+exec "$@"