diff --git a/README.md b/README.md index bbac1bf..e97a988 100644 --- a/README.md +++ b/README.md @@ -294,6 +294,8 @@ is tested on a throwaway virtual machine rather than on anybody's desk: e2e/vm/run.sh e2e/cases/status-export.sh intent against reality, and the file OLD_REF= e2e/vm/run.sh e2e/cases/upgrade-volumes.sh an install from before #5, upgraded + DISTRO=fedora e2e/vm/run.sh e2e/cases/selinux.sh + the SELinux-sensitive paths, with it enforcing e2e/vm/down.sh remove it `SERVER_BINARY=path/to/inbuxa e2e/vm/run.sh …` installs a server build @@ -308,6 +310,15 @@ is free to break it and a failure is the installer's rather than the last run's leftovers. `e2e/vm/up.sh` needs qemu, KVM and xorriso; nothing needs root on your machine. +On Fedora and Rocky the machine runs SELinux enforcing, and the installer has +to live within it -- a container could not once read a directory mounted into +it. So on those two, `run.sh` reads the audit log after *any* case and fails +the run if the installer provoked an SELinux denial, even one the case +survived; `SELINUX_IGNORE=1` downgrades that to a warning. `e2e/cases/selinux.sh` +drives the paths that have been caught before (the backup archives the +server's volumes through a container) and checks them directly; on a +distribution without SELinux enforcing it skips and passes. + ## License AGPL-3.0-or-later. Some of this began as [ihasmail-oneshot], which is ours diff --git a/e2e/cases/selinux.sh b/e2e/cases/selinux.sh new file mode 100644 index 0000000..ff237a0 --- /dev/null +++ b/e2e/cases/selinux.sh @@ -0,0 +1,107 @@ +#!/bin/bash +# SPDX-FileCopyrightText: 2026 Coffey Labs LLC +# SPDX-License-Identifier: AGPL-3.0-or-later +# +# The SELinux tier: drive the paths SELinux has stopped the installer on +# before, with SELinux enforcing, and prove it provoked no denial. +# +# The path that bit, and the one this leans on, is the server backup: the +# installer archives the server's two volumes by mounting them, read-only, +# into a throwaway container running tar (internal/apply/upgrade.go). On +# Fedora and Rocky that is refused unless the volumes carry a label the +# container may read -- which is exactly the kind of thing a container-naive +# change breaks without noticing on Debian. So: install in containers, take a +# backup, converge, and after each step read the audit log for AVC denials in +# the installer's footprint, and confirm SELinux is still enforcing. +# +# Only meaningful where SELinux enforces, so on Debian/Ubuntu/Arch it skips and +# passes. Run it on the family it is for: +# DISTRO=fedora e2e/vm/run.sh e2e/cases/selinux.sh +# DISTRO=rocky9 e2e/vm/run.sh e2e/cases/selinux.sh +set -uo pipefail +export PATH="/usr/sbin:/sbin:$PATH" # getenforce and ausearch live here + +pass=0; fail=0 +ok() { echo " ok $*"; pass=$((pass+1)); } +bad() { echo " FAIL $*"; fail=$((fail+1)); } + +DIR=/var/lib/inbuxa + +if ! command -v getenforce >/dev/null 2>&1 || [ "$(getenforce)" != Enforcing ]; then + echo "SKIPPED: SELinux is not enforcing here ($(. /etc/os-release && echo "$PRETTY_NAME"))." + echo "This tier runs on Fedora or Rocky, e.g. DISTRO=fedora e2e/vm/run.sh e2e/cases/selinux.sh" + exit 0 +fi +echo "==> SELinux is enforcing ($(. /etc/os-release && echo "$ID $VERSION_ID"))" +ok "getenforce reports Enforcing" +command -v ausearch >/dev/null 2>&1 \ + && ok "ausearch is present to read the audit log" \ + || bad "ausearch is missing (install the audit package); the denial scan cannot run" + +# scan LABEL SINCE: fail if the installer provoked an AVC denial in its own +# footprint (its containers, or /var/lib/inbuxa) after the checkpoint. +scan() { + local d + d="$(ausearch -i -m avc,user_avc -ts "$2" 2>/dev/null \ + | grep -iE 'inbuxa|container_t|container_file_t|svirt|conmon' || true)" + if [ -z "$d" ]; then + ok "$1: no SELinux denials" + else + bad "$1: SELinux denied something" + echo "$d" | sed 's/^/ /' + fi +} +enforcing() { [ "$(getenforce)" = Enforcing ]; } + +# Whichever runtime the installer chose (podman on this family). +rt() { command -v docker >/dev/null 2>&1 && echo docker || echo podman; } +compose() { + if [ "$(rt)" = docker ]; then + docker compose -f "$DIR/compose.yaml" "$@" + else + DOCKER_HOST=unix:///run/podman/podman.sock \ + /usr/local/lib/docker/cli-plugins/docker-compose -f "$DIR/compose.yaml" "$@" + fi +} + +echo +echo "==> installing the suite in containers" +C=$(date +%T) +OUT="$(/tmp/inbuxa install --local --domain example.test --install-deps --yes 2>&1)"; rc=$? +[ $rc -eq 0 ] && ok "installed (exit 0)" || { bad "install exit $rc"; echo "$OUT" | tail -20 | sed 's/^/ /'; } +scan "install" "$C" +enforcing && ok "still enforcing after install" || bad "no longer enforcing after install" + +STATE="$(compose ps --format '{{.Service}} {{.State}}' 2>/dev/null)" +for s in server console webmail; do + grep -q "^$s running" <<<"$STATE" && ok "$s is running" || { bad "$s is not running"; echo "$STATE" | sed 's/^/ /'; } +done + +echo +echo "==> taking a backup (archives the server's volumes through a container)" +C=$(date +%T) +OUT="$(/tmp/inbuxa backup --yes 2>&1)"; rc=$? +[ $rc -eq 0 ] && ok "backup (exit 0)" || { bad "backup exit $rc"; echo "$OUT" | tail -20 | sed 's/^/ /'; } +scan "backup (archiveVolumes)" "$C" +ls "$DIR"/backups/server-*.tar.gz >/dev/null 2>&1 && ok "an archive was written: $(ls "$DIR"/backups)" || bad "no archive in $DIR/backups" +# Whole listing first: grep -q stops at the first match, and under pipefail the +# cut-off tar would fail the check on its own. +LIST="$(tar -tzf "$DIR"/backups/server-*.tar.gz 2>/dev/null)" +grep -q "config.json" <<<"$LIST" && ok "with the server's configuration in it" || bad "the archive has no config.json" +# The backup stops the server and must start it again. +grep -q "^server running" <<<"$(compose ps --format '{{.Service}} {{.State}}' 2>/dev/null)" \ + && ok "the server is running again after the backup" || bad "the server did not come back after the backup" + +echo +echo "==> running install again converges, under SELinux, without a denial" +C=$(date +%T) +/tmp/inbuxa install --local --domain example.test --yes >/dev/null 2>&1 && ok "converged (exit 0)" || bad "the second install failed" +scan "converge" "$C" + +echo +echo "==> SELinux is still enforcing at the end" +enforcing && ok "getenforce still reports Enforcing" || bad "SELinux is no longer enforcing" + +echo +echo "==> $pass passed, $fail failed" +[ "$fail" -eq 0 ] diff --git a/e2e/vm/lib.sh b/e2e/vm/lib.sh index 74cedf4..a64a4de 100644 --- a/e2e/vm/lib.sh +++ b/e2e/vm/lib.sh @@ -93,3 +93,33 @@ vm_scp() { scp -o StrictHostKeyChecking=no -o UserKnownHostsFile=/dev/null \ -o LogLevel=ERROR -P "$SSH_PORT" "$@" } + +# ---- the SELinux tier ---- +# +# On the Red Hat family the installer runs under SELinux enforcing, and SELinux +# has stopped it before (a container could not read a directory mounted into +# it; see internal/apply/upgrade.go). These let run.sh turn any case on Fedora +# or Rocky into a check that the installer provoked no SELinux denial. They run +# in the guest, where the audit log is, and report nothing anywhere SELinux is +# not enforcing, so they are harmless on Debian, Ubuntu and Arch. + +# vm_selinux_enforcing is true only where the guest has SELinux in enforcing +# mode. getenforce is in /sbin, which a non-login shell may not have on PATH, +# so this goes through sudo, whose secure_path does. +vm_selinux_enforcing() { + [ "$(vm_ssh 'sudo getenforce 2>/dev/null || echo Disabled')" = Enforcing ] +} + +# vm_audit_checkpoint prints a time to hand back to vm_selinux_denials, so a +# scan sees only what happened after it -- not the machine's boot. +vm_audit_checkpoint() { + vm_ssh 'date +%T' +} + +# vm_selinux_denials prints the guest's AVC denials since the checkpoint that +# touch the installer's footprint -- its containers and /var/lib/inbuxa -- one +# per line, or nothing. Reading the audit log needs root. +vm_selinux_denials() { + vm_ssh "sudo ausearch -i -m avc,user_avc -ts '$1' 2>/dev/null" \ + | grep -iE 'inbuxa|container_t|container_file_t|svirt|conmon' || true +} diff --git a/e2e/vm/run.sh b/e2e/vm/run.sh index c3048d8..4253255 100755 --- a/e2e/vm/run.sh +++ b/e2e/vm/run.sh @@ -61,4 +61,29 @@ vm_scp "$ROOT/$CASE" lab@127.0.0.1:/tmp/case.sh >/dev/null vm_ssh 'chmod +x /tmp/inbuxa /tmp/case.sh; [ ! -f /tmp/inbuxa-old ] || chmod +x /tmp/inbuxa-old' echo "==> running $(basename "$CASE")" -vm_ssh 'sudo -E bash /tmp/case.sh' +# On Fedora/Rocky the case runs under SELinux enforcing; note the time first so +# the scan afterwards sees only what the case provoked, not the boot. +CKPT="" +if vm_selinux_enforcing; then CKPT="$(vm_audit_checkpoint)"; fi +rc=0 +vm_ssh 'sudo -E bash /tmp/case.sh' || rc=$? + +# The SELinux tier: a case that passed its own checks still fails the run if +# the installer made SELinux deny something -- a denial it happened to survive +# today is a latent break. SELINUX_IGNORE=1 downgrades that to a warning. +if [ -n "$CKPT" ]; then + echo "==> SELinux: denials since the case began" + DEN="$(vm_selinux_denials "$CKPT")" + if [ -n "$DEN" ]; then + echo "$DEN" | sed 's/^/ /' + if [ "${SELINUX_IGNORE:-}" = 1 ]; then + echo " SELINUX_IGNORE=1: not failing the run on these." + else + echo " the installer must not provoke SELinux denials (SELINUX_IGNORE=1 to override)" >&2 + rc=1 + fi + else + echo " none -- SELinux stayed out of the installer's way" + fi +fi +exit "$rc"