Merge pull request #362 from rancher-sandbox/etc-environment

Write /etc/environment before the lingering session is started

hack/test-example.sh

@@ -92,6 +92,8 @@ function diagnose() {
 	set +x -e
 }
 
+export ftp_proxy=http://localhost:2121
+
 INFO "Starting \"$NAME\" from \"$FILE\""
 defer "limactl delete -f \"$NAME\""
 set -x
@@ -106,6 +108,17 @@ limactl shell "$NAME" uname -a
 limactl shell "$NAME" cat /etc/os-release
 set +x
 
+INFO "Testing proxy settings are imported"
+got=$(limactl shell "$NAME" env | grep FTP_PROXY)
+# Expected: FTP_PROXY is set in addition to ftp_proxy, localhost is replaced
+# by the gateway address, and the value is set immediately without a restart
+expected="FTP_PROXY=http://192.168.5.2:2121"
+INFO "FTP_PROXY: expected=${expected} got=${got}"
+if [ "$got" != "$expected" ]; then
+	ERROR "proxy environment variable not set to correct value"
+	exit 1
+fi
+
 INFO "Testing limactl copy command"
 tmpfile="$HOME/lima-hostname"
 rm -f "$tmpfile"
@@ -212,9 +225,19 @@ if [[ -n ${CHECKS["restart"]} ]]; then
 	limactl stop "$NAME"
 	sleep 3
 
+	export ftp_proxy=my.proxy:8021
 	INFO "Restarting \"$NAME\""
 	limactl start "$NAME"
 
+	INFO "Make sure proxy setting is updated"
+	got=$(limactl shell "$NAME" env | grep FTP_PROXY)
+	expected="FTP_PROXY=my.proxy:8021"
+	INFO "FTP_PROXY: expected=${expected} got=${got}"
+	if [ "$got" != "$expected" ]; then
+		ERROR "proxy environment variable not set to correct value"
+		exit 1
+	fi
+
 	# shellcheck disable=SC2016
 	if ! limactl shell "$NAME" sh -c 'test -f $HOME/sweet-home'; then
 		ERROR "Guest home directory does not persist across restarts"

pkg/cidata/cidata.TEMPLATE.d/boot.sh

@@ -19,9 +19,9 @@ done <"${LIMA_CIDATA_MNT}"/etc_environment
 
 CODE=0
 
-# Don't make any changes to /etc or /var/lib until the boot/* scripts have run
-# because they might move the directories to /mnt/data on first boot, so changes
-# made on restart would be lost.
+# Don't make any changes to /etc or /var/lib until boot/05-persistent-data-volume.sh
+# has run because it might move the directories to /mnt/data on first boot. In that
+# case changes made on restart would be lost.
 
 for f in "${LIMA_CIDATA_MNT}"/boot/*; do
 	INFO "Executing $f"
@@ -31,11 +31,6 @@ for f in "${LIMA_CIDATA_MNT}"/boot/*; do
 	fi
 done
 
-if [ -e /etc/environment ]; then
-	sed -i '/#LIMA-START/,/#LIMA-END/d' /etc/environment
-fi
-cat "${LIMA_CIDATA_MNT}/etc_environment" >>/etc/environment
-
 if [ -d "${LIMA_CIDATA_MNT}"/provision.system ]; then
 	for f in "${LIMA_CIDATA_MNT}"/provision.system/*; do
 		INFO "Executing $f"
@@ -64,5 +59,9 @@ if [ -d "${LIMA_CIDATA_MNT}"/provision.user ]; then
 	done
 fi
 
+# Signal that provisioning is done. The instance-id in the meta-data file changes on every boot,
+# so any copy from a previous boot cycle will have different content.
+cp "${LIMA_CIDATA_MNT}"/meta-data /run/lima-boot-done
+
 INFO "Exiting with code $CODE"
 exit "$CODE"

pkg/cidata/cidata.TEMPLATE.d/boot/07-etc-environment.sh

@@ -0,0 +1,25 @@
+#!/bin/sh
+set -eux
+
+# /etc/environment must be written after 05-persistent-data-volume.sh has run to
+# make sure the changes on a restart are applied to the persisted version.
+
+if [ -e /etc/environment ]; then
+	sed -i '/#LIMA-START/,/#LIMA-END/d' /etc/environment
+fi
+cat "${LIMA_CIDATA_MNT}/etc_environment" >>/etc/environment
+
+# It is possible that a requirements script has started an ssh session before
+# /etc/environment was updated, so we need to kill it to make sure it will
+# restart with the updated environment before "linger" is being enabled.
+
+if command -v loginctl >/dev/null 2>&1; then
+	loginctl terminate-user "${LIMA_CIDATA_USER}" || true
+fi
+
+# Make sure the guestagent socket from a previous boot is removed before we open the "lima-ssh-ready" gate.
+rm -f /run/lima-guest-agent.sock
+
+# Signal that provisioning is done. The instance-id in the meta-data file changes on every boot,
+# so any copy from a previous boot cycle will have different content.
+cp "${LIMA_CIDATA_MNT}"/meta-data /run/lima-ssh-ready

pkg/cidata/cidata.TEMPLATE.d/boot/30-install-packages.sh

@@ -112,7 +112,7 @@ if [ -n "${LIMA_CIDATA_TCP_DNS_LOCAL_PORT}" ] && [ "${LIMA_CIDATA_TCP_DNS_LOCAL_
 fi
 if [ "${SETUP_DNS}" = 1 ]; then
 	# Try to setup iptables rule again, in case we just installed iptables
-	"${LIMA_CIDATA_MNT}/boot/07-host-dns-setup.sh"
+	"${LIMA_CIDATA_MNT}/boot/09-host-dns-setup.sh"
 fi
 
 # update_fuse_conf has to be called after installing all the packages,

pkg/hostagent/hostagent.go

@@ -388,6 +388,9 @@ func (a *HostAgent) startHostAgentRoutines(ctx context.Context) error {
 	if err := a.waitForRequirements(ctx, "optional", a.optionalRequirements()); err != nil {
 		mErr = multierror.Append(mErr, err)
 	}
+	if err := a.waitForRequirements(ctx, "final", a.finalRequirements()); err != nil {
+		mErr = multierror.Append(mErr, err)
+	}
 	return mErr
 }
 

pkg/hostagent/requirements.go

@@ -60,16 +60,33 @@ type requirement struct {
 
 func (a *HostAgent) essentialRequirements() []requirement {
 	req := make([]requirement, 0)
-	req = append(req, requirement{
-		description: "ssh",
-		script: `#!/bin/bash
+	req = append(req,
+		requirement{
+			description: "ssh",
+			script: `#!/bin/bash
 true
 `,
-		debugHint: `Failed to SSH into the guest.
+			debugHint: `Failed to SSH into the guest.
 Make sure that the YAML field "ssh.localPort" is not used by other processes on the host.
 If any private key under ~/.ssh is protected with a passphrase, you need to have ssh-agent to be running.
 `,
-	})
+		},
+		requirement{
+			description: "user session is ready for ssh",
+			script: `#!/bin/bash
+set -eux -o pipefail
+if ! timeout 30s bash -c "until sudo diff -q /run/lima-ssh-ready /mnt/lima-cidata/meta-data 2>/dev/null; do sleep 3; done"; then
+	echo >&2 "not ready to start persistent ssh session"
+	exit 1
+fi
+`,
+			debugHint: `The boot sequence will terminate any existing user session after updating
+/etc/environment to make sure the session includes the new values.
+Terminating the session will break the persistent SSH tunnel, so
+it must not be created until the session reset is done.
+`,
+		})
+
 	if len(a.y.Mounts) > 0 {
 		req = append(req, requirement{
 			description: "sshfs binary to be installed",
@@ -164,3 +181,23 @@ Also see "/var/log/cloud-init-output.log" in the guest.
 	}
 	return req
 }
+
+func (a *HostAgent) finalRequirements() []requirement {
+	req := make([]requirement, 0)
+	req = append(req,
+		requirement{
+			description: "boot scripts must have finished",
+			script: `#!/bin/bash
+set -eux -o pipefail
+if ! timeout 30s bash -c "until sudo diff -q /run/lima-boot-done /mnt/lima-cidata/meta-data 2>/dev/null; do sleep 3; done"; then
+	echo >&2 "boot scripts have not finished"
+	exit 1
+fi
+`,
+			debugHint: `All boot scripts, provisioning scripts, and readiness probes must
+finish before the instance is considered "ready".
+Check "/var/log/cloud-init-output.log" in the guest to see where the process is blocked!
+`,
+		})
+	return req
+}