diff --git a/.github/workflows/e2e.yml b/.github/workflows/e2e.yml index 98cfe8321..ba1e87c31 100644 --- a/.github/workflows/e2e.yml +++ b/.github/workflows/e2e.yml @@ -434,19 +434,62 @@ jobs: # without an override ("bootupd is required for ostree-based # installs", run 43525356388), and its lifecycle reboots stay as # limited as before #1189. + # Two independent things drive how the installed disk boots: + # - COMPOSEFS: the image's install config (usr/lib/bootc/install/*.toml) + # sets composefs-backend, so bootc must install with --composefs-backend + # or the ostree backend aborts ("bootupd is required for ostree-based + # installs", #1215). Composefs deployments live under + # /sysroot/state/deploy/, not /ostree/deploy/default/deploy. + # - FIRMWARE_BOOT: the lifecycle suite reboots into staged deployments, + # which a direct -kernel boot can't do (#1023). It boots through OVMF + # and the image's own loader. Firmware boot needs the image to install + # its own loader — bootc does that through bootupd (ostree images), but + # composefs images install systemd-boot via --bootloader systemd and + # still need firmware boot, so detect composefs too. + COMPOSEFS=false FIRMWARE_BOOT=false if [[ "${SUITE_DIR}" == "lifecycle" ]] \ && sudo podman run --rm --quiet --entrypoint /bin/sh "${IMAGE}" -c 'command -v bootupctl' >/dev/null 2>&1; then FIRMWARE_BOOT=true + elif [[ "${SUITE_DIR}" == "lifecycle" ]] \ + && sudo podman run --rm --quiet --entrypoint /bin/sh "${IMAGE}" -c \ + 'grep -lq composefs-backend /usr/lib/bootc/install/*.toml 2>/dev/null' >/dev/null 2>&1; then + # Composefs images (e.g. dakota) ship no bootupd, so they fall through + # the bootupctl probe above. Detect them from the install config and + # still force firmware boot (#1215). + COMPOSEFS=true + FIRMWARE_BOOT=true elif [[ "${SUITE_DIR}" == "lifecycle" ]]; then - echo "::warning title=lifecycle without firmware boot::${IMAGE} has no bootupctl, so the lifecycle VM boots the first deployment's kernel directly and reboots cannot reach a staged deployment (#1023)" + echo "::warning title=lifecycle without firmware boot::${IMAGE} has no bootupctl and is not composefs-native, so the lifecycle VM boots the first deployment's kernel directly and reboots cannot reach a staged deployment (#1023)" fi - if [[ "${FIRMWARE_BOOT}" != "true" ]] \ - && sudo podman run --rm --quiet "${IMAGE}" bootc install to-disk --help 2>&1 | grep -q "\-\-bootloader"; then - BOOTLOADER_FLAG="--bootloader systemd" + # Composefs images install with --composefs-backend (the ostree backend + # fails on them, #1215) plus --bootloader systemd to install systemd-boot, + # since they ship no bootupd. Probe both flags: older bootc lacks one. + INSTALL_EXTRA="" + INSTALL_HELP="$(sudo podman run --rm --quiet "${IMAGE}" bootc install to-disk --help 2>&1)" + if [[ "${COMPOSEFS}" == "true" ]] \ + && printf '%s\n' "${INSTALL_HELP}" | grep -q -- "--composefs-backend" \ + && printf '%s\n' "${INSTALL_HELP}" | grep -q "\-\-bootloader"; then + INSTALL_EXTRA="--composefs-backend --bootloader systemd" + elif [[ "${FIRMWARE_BOOT}" != "true" ]] \ + && printf '%s\n' "${INSTALL_HELP}" | grep -q "\-\-bootloader"; then + INSTALL_EXTRA="--bootloader systemd" fi + BOOTLOADER_FLAG="${INSTALL_EXTRA}" + echo "COMPOSEFS=${COMPOSEFS}" >> "$GITHUB_ENV" echo "FIRMWARE_BOOT=${FIRMWARE_BOOT}" >> "$GITHUB_ENV" + # Backend-aware deployment prefix. Ostree images store deployments + # under /ostree/deploy/default/deploy/.. Composefs images + # store them under /sysroot/state/deploy/ (#1215). + if [[ "${COMPOSEFS}" == "true" ]]; then + DEPLOY_DIR="sysroot/state/deploy" + STATE_DIR="sysroot/state" + else + DEPLOY_DIR="ostree/deploy/default/deploy" + STATE_DIR="ostree/deploy/default" + fi + BOOTC_INSTALL_LOG="$(pwd)/bootc-install.log" BOOTC_INSTALL_ATTEMPTS=3 sudo mkdir -p /mnt/root @@ -480,7 +523,7 @@ jobs: # instead of ending the step on a bare "mount: wrong fs type". if sudo blkid -s TYPE -o value "${ROOT_PART}" >/dev/null 2>&1 \ && sudo mount "${ROOT_PART}" /mnt/root; then - DEPLOY=$(sudo find /mnt/root/ostree/deploy/default/deploy/ -mindepth 1 -maxdepth 1 -type d -printf '%f\n' 2>/dev/null | head -1) + DEPLOY=$(sudo find "/mnt/root/${DEPLOY_DIR}" -mindepth 1 -maxdepth 1 -type d -printf '%f\n' 2>/dev/null | head -1) if [[ -n "${DEPLOY}" ]]; then break fi @@ -516,8 +559,16 @@ jobs: fi ROOT_UUID=$(sudo blkid -s UUID -o value "${ROOT_PART}") echo "root=${ROOT_PART} uuid=${ROOT_UUID}" - D="/mnt/root/ostree/deploy/default/deploy/${DEPLOY}" - VAR="/mnt/root/ostree/deploy/default/var" + D="/mnt/root/${DEPLOY_DIR}/${DEPLOY}" + VAR="/mnt/root/${STATE_DIR}/var" + # Direct kernel boot prep. For firmware boot (OVMF) systemd-boot reads + # the deployment's kernel + BLS entry straight from the disk, so none of + # this ostree-only setup runs. Guard it so composefs deployments are never + # given a bogus /ostree/ tree and a missing initramfs.img can't abort the + # step under set -e. + if [[ "${FIRMWARE_BOOT}" == "true" ]]; then + : + else # el10/LTS images may have multiple entries in usr/lib/modules/ (e.g., a # modules-only dir for the old kernel + a full dir with vmlinuz for the new # one). Pick the version that actually contains a vmlinuz; fall back to the @@ -608,6 +659,7 @@ jobs: fi echo "ostree boot path: ${OSTREE_PATH}" + fi # Mask services not needed in CI to shorten first-boot time. # NetworkManager-wait-online can block for up to 120s — most # expensive to leave unmasked on a SLIRP QEMU network. @@ -636,7 +688,16 @@ jobs: if [[ "${FIRMWARE_BOOT}" == "true" ]]; then # shellcheck disable=SC2086 # split KERNEL_ARGS into one argument per line CI_KARGS=$(printf '%s\n' ${KERNEL_ARGS} | grep -vE '^(root=|rw$|ostree=)' | tr '\n' ' ') - ENTRIES=$(sudo find /mnt/root/boot/loader/entries -maxdepth 1 -name '*.conf' 2>/dev/null) + # systemd-boot reads BLS entries from the boot partition (xbootldr / + # p2). Some images also mirror them in the root's /boot (p3). Search + # both so composefs (p2) and ostree (p2 or p3) both get CI kargs. + mountpoint -q /mnt/boot || sudo mount "${LOOP}p2" /mnt/boot + ENTRIES="" + for ENTRY_ROOT in /mnt/root/boot/loader/entries /mnt/boot/loader/entries; do + [[ -d "${ENTRY_ROOT}" ]] || continue + ENTRIES+="$(sudo find "${ENTRY_ROOT}" -maxdepth 1 -name '*.conf' 2>/dev/null)" + done + ENTRIES=$(printf '%s\n' "${ENTRIES}" | grep -v '^$' | sort -u) if [[ -z "${ENTRIES}" ]]; then echo "::error title=no loader entries::bootc wrote no /boot/loader/entries/*.conf; nothing to boot through the image's own loader" exit 1 @@ -664,7 +725,7 @@ jobs: echo "SELinux: expected=${SELINUX_EXPECTED} policy=${SELINUX_POLICY}" echo "SELINUX_EXPECTED=${SELINUX_EXPECTED}" >> "$GITHUB_ENV" SELINUX_DIR="$(pwd)/results/selinux" - STATEROOT="/mnt/root/ostree/deploy/default" + STATEROOT="/mnt/root/${STATE_DIR}" mkdir -p "${SELINUX_DIR}" if [[ "${SELINUX_EXPECTED}" == enforcing ]]; then for DEP in $(sudo find "${STATEROOT}/deploy" -maxdepth 1 -mindepth 1 -type d 2>/dev/null); do @@ -681,7 +742,7 @@ jobs: # bootc may deploy with a different checksum than what `ls|head-1` # returns, so we iterate to ensure the booted deployment gets set up. BFT_UID=1001 - for DEP in $(sudo find /mnt/root/ostree/deploy/default/deploy \ + for DEP in $(sudo find "${STATEROOT}/deploy" \ -maxdepth 1 -mindepth 1 -type d 2>/dev/null); do # CI VMs are not users: opt out of the daily countme ping # (projectbluefin-countme.service honours this marker). @@ -835,7 +896,7 @@ jobs: sudo chmod 700 "${VAR}/home/bluefin-test/.ssh" sudo chmod 600 "${VAR}/home/bluefin-test/.ssh/authorized_keys" # Write key to AuthorizedKeysCommand target in every deployment - for DEP in $(sudo find /mnt/root/ostree/deploy/default/deploy \ + for DEP in $(sudo find "${STATEROOT}/deploy" \ -maxdepth 1 -mindepth 1 -type d 2>/dev/null); do sudo cp /tmp/vm_key.pub "${DEP}/etc/ssh/ci-authorized-keys" sudo chown root:root "${DEP}/etc/ssh/ci-authorized-keys" @@ -890,7 +951,7 @@ jobs: fi # Pre-boot diagnostics: confirm key files and sshd drop-in are correct. echo "=== ci-authorized-keys ===" - for DEP in $(sudo find /mnt/root/ostree/deploy/default/deploy \ + for DEP in $(sudo find "${STATEROOT}/deploy" \ -maxdepth 1 -mindepth 1 -type d 2>/dev/null); do echo "--- ${DEP}/etc/ssh/ci-authorized-keys ---" sudo cat "${DEP}/etc/ssh/ci-authorized-keys" || echo "(missing)" diff --git a/tests/unit/test_e2e_lifecycle_firmware_boot.py b/tests/unit/test_e2e_lifecycle_firmware_boot.py index 1682487f9..40eb9cdfa 100644 --- a/tests/unit/test_e2e_lifecycle_firmware_boot.py +++ b/tests/unit/test_e2e_lifecycle_firmware_boot.py @@ -26,7 +26,7 @@ def test_lifecycle_install_keeps_the_images_own_loader_and_needs_it(): lifecycle = install.index('if [[ "${SUITE_DIR}" == "lifecycle" ]]') probe = install.index("command -v bootupctl", lifecycle) firmware = install.index("FIRMWARE_BOOT=true", lifecycle) - override = install.index('BOOTLOADER_FLAG="--bootloader systemd"') + override = install.index('INSTALL_EXTRA="--bootloader systemd"') # Firmware boot needs bootupd in the image; the systemd-boot override # applies only when firmware boot is off (other suites, or an image # without bootupd such as dakota, run 43525356388). diff --git a/tests/unit/test_e2e_workflow.py b/tests/unit/test_e2e_workflow.py index 6ea35a27d..37b3f6542 100644 --- a/tests/unit/test_e2e_workflow.py +++ b/tests/unit/test_e2e_workflow.py @@ -39,6 +39,38 @@ def test_manual_runs_checkout_the_selected_fork_ref_without_changing_callers(): assert "test_repository: ${{ github.repository }}" in manual +def test_e2e_composefs_firmware_boot_branch_for_dakota(): + """Dakota is composefs-native with no bootupd, so its lifecycle reboots + must go through the OVMF firmware path, not direct kernel boot (#1215, + #1023). The install uses --composefs-backend and a detected composefs + backend forces FIRMWARE_BOOT=true.""" + workflow = _workflow("e2e.yml") + + # Composefs is detected from the image's install config, and a detected + # composefs backend forces the firmware-boot path (no bootupd to use). The + # composefs branch is a fallback after the bootupctl probe, and it sets + # FIRMWARE_BOOT=true. + assert "grep -lq composefs-backend /usr/lib/bootc/install/*.toml" in workflow + composefs_probe = workflow.index("grep -lq composefs-backend /usr/lib/bootc/install/*.toml") + assert workflow.index("FIRMWARE_BOOT=true", composefs_probe) > composefs_probe + + # Composefs images install with --composefs-backend (the ostree backend + # aborts without it, #1215) plus --bootloader systemd to install systemd-boot. + assert "--composefs-backend --bootloader systemd" in workflow + + # Deployment discovery is backend-aware: composefs deployments live under + # /sysroot/state/deploy/, not /ostree/deploy/default/deploy. + assert 'DEPLOY_DIR="sysroot/state/deploy"' in workflow + assert 'DEPLOY_DIR="ostree/deploy/default/deploy"' in workflow + assert 'D="/mnt/root/${DEPLOY_DIR}/${DEPLOY}"' in workflow + + # The ostree-only direct-boot prep (kernel extraction, boot.N symlink, + # ostree= arg) must not run for firmware boot, or it would write a bogus + # /ostree/ tree onto the composefs deployment. + assert "Direct kernel boot prep" in workflow + assert 'if [[ "${FIRMWARE_BOOT}" == "true" ]]; then' in workflow + + def test_e2e_ci_user_tmpfiles_has_no_recursive_home_relabel(): # A recursive `Z` on the test home walks the runner's Podman storage on # every later boot and trips systemd-tmpfiles SELinux denials (#1205).