Support OEM systemd-sysext images and Flatcar extensions

The OEM software and future official Flatcar extensions will be shipped
as systemd-sysext images, coupled to the Flatcar version and A/B
updated. Instead of adding support for this in the update-engine C++
code the plan was to use a new helper binary in the post-inst action
for downloading of the payload. Since this is not ready we use curl and
a small script to decode the payload. The A/B update mechanism includes
a migration path for old instances that first need a fallback download,
because the old update-engine client doesn't pass the XML dump, then
they need to go through another update cycle to have systemd-sysext
images for both partitions, and then the migration can take place in
the initrd where the old OEM contents get cleaned up.

decode_payload

@@ -4,12 +4,15 @@ set -euo pipefail
 CHECKINPUT=0
 SCRIPTFOLDER="$(dirname "$(readlink -f "$0")")"
 
+PROTOPATH="${PROTOPATH-"${SCRIPTFOLDER}"/src/update_engine}"
+
 if [ $# -lt 3 ] || [ "$1" = "-h" ] || [ "$1" = "--help" ]; then
   echo "Usage: $0 PUBKEY PAYLOAD OUTPUT [KERNELOUTPUT]"
   echo "Decodes a payload, writing the decoded payload to stdout and payload information to stderr"
   echo "Only one optional kernel payload is supported (the output file will be zero-sized if no payload was found)."
   echo "Only a signle signature is supported as only one pubkey is accepted and in the multi-signature case only the second signature is looked at."
   echo "Pass CHECKINPUT=1 to process the inline checksums in addition to the final checksum."
+  echo "Pass PROTOPATH=folder/ to specify the location of the protobuf file."
   exit 1
 fi
 
@@ -23,7 +26,7 @@ KERNEL="${4-}"
 MLEN=$(dd status=none bs=1 skip=12 count=8 if="${FILE}" | od --endian=big -An -vtu8 -w1024 | tr -d ' ')
 
 # The manifest starts at offset 20 (with tail we do a +1 compared to dd) and we feed it into protoc for decoding (we assume that the text output format is stable)
-DESC=$(protoc --decode=chromeos_update_engine.DeltaArchiveManifest --proto_path "${SCRIPTFOLDER}"/src/update_engine "${SCRIPTFOLDER}"/src/update_engine/update_metadata.proto < <({ tail -c +21 "${FILE}" || true ; } |head -c "${MLEN}"))
+DESC=$(protoc --decode=chromeos_update_engine.DeltaArchiveManifest --proto_path "${PROTOPATH}" "${PROTOPATH}"/update_metadata.proto < <({ tail -c +21 "${FILE}" || true ; } |head -c "${MLEN}"))
 
 echo "${DESC}" >&2
 
@@ -117,7 +120,7 @@ fi
 
 # The signature protobuf message is at the signature offset
 # Decoding the "data" field caues some troubles and needs a workaround below for the dev key
-SIGDESC=$(protoc --decode=chromeos_update_engine.Signatures --proto_path "${SCRIPTFOLDER}"/src/update_engine "${SCRIPTFOLDER}"/src/update_engine/update_metadata.proto < <({ tail -c +$((21 + MLEN + SIGOFFSET)) "${FILE}" || true ; } |head -c "${SIGSIZE}"))
+SIGDESC=$(protoc --decode=chromeos_update_engine.Signatures --proto_path "${PROTOPATH}" "${PROTOPATH}"/update_metadata.proto < <({ tail -c +$((21 + MLEN + SIGOFFSET)) "${FILE}" || true ; } |head -c "${SIGSIZE}"))
 echo "${SIGDESC}" >&2
 VERSION=2 # Init for the single-signature case, in the many-signature case the first signature ("version 1" but better would be "number 1") is,
 # at least for Flatcar production, a dummy and parsing it overwrites this variable with 1 and it will be ignored,

flatcar-postinst

@@ -35,6 +35,7 @@ fi
 # shellcheck source=/dev/null
 source "${INSTALL_MNT}/lib/os-release"
 NEXT_VERSION_ID=${VERSION_ID}
+NEXT_VERSION="${VERSION}"
 
 # shellcheck source=/dev/null
 source /usr/lib/os-release
@@ -43,6 +44,167 @@ tee_journal() {
     tee >(systemd-cat -t coreos-postinst)
 }
 
+OEMID=$({ grep -m 1 -o "^ID=.*" "${OEM_MNT}"/oem-release || true ; } | cut -d = -f 2)
+
+# Declare which OEM sysext images are expected to exist, and the files to delete in the initramfs migration action (Use /oem instead of /usr/share/oem/ to avoid symlinks from the initrd.)
+# Note: Because associative arrays can't hold lists, we store the files newline-separated (not simply space separated, it must be only one file per line!)
+NVIDIA_FILES="/etc/systemd/system/nvidia.service
+/oem/bin/setup-nvidia
+/oem/bin/install-nvidia
+/oem/units/nvidia.service"
+declare -A OEM_SYSEXTS
+OEM_SYSEXTS[qemu]=""
+OEM_SYSEXTS[azure]="${NVIDIA_FILES}
+/etc/systemd/system/oem-cloudinit.service
+/etc/systemd/system/multi-user.target.wants/oem-cloudinit.service
+/etc/systemd/system/waagent.service
+/etc/systemd/system/multi-user.target.wants/waagent.service
+/oem/waagent.conf
+/oem/python/
+/oem/bin/
+/oem/units/
+/oem/base/
+"
+# TODO: For new OEM sysext images, declare an entry like OEM_SYSEXTS[ami]="${NVIDIA_FILES}..." or even just OEM_SYSEXTS[qemu]="" to enable fallback downloads for old clients
+
+# Must not be used as "if sysext_download; then" or "sysext_download ||" because that makes set -e a no-op, and also must not use "( sysext_download )" because we want to set the global SUCCESS variable.
+sysext_download() {
+    local name="$1" # Payload name
+    local target="$2" # Path to write the payload to, writing does not need to be atomic because the caller later does an atomic move
+    local from="${3-}" # Either path to XML dump or the constant "release-server"
+    local base=""
+    local entries=""
+    local hash=""
+    local size=""
+    local url=""
+    local ret
+    SUCCESS=false
+    set +e
+    (
+    set -e
+    # TODO: Replace the below with invoking an ue-rs helper binary for downloading the payload "name", either from the XML data or the release server ("from"), and write unpacked, verified file to "target"
+    if [ "${from}" = "release-server" ]; then
+      url="https://update.release.flatcar-linux.net/${FLATCAR_BOARD}/${NEXT_VERSION}/${name}"
+    else
+      base=$(grep -m 1 -o 'codebase="[^"]*"' "${from}" | cut -d '"' -f 2)
+      entries=$(grep -m 1 -o "<package name=\"${name}\"[^>]*" "${from}")
+      url="${base}/${name}"
+      size=$(echo "${entries}" | grep -o 'size="[0-9]*' | cut -d '"' -f 2)
+      hash=$(echo "${entries}" | grep -o -P 'hash="[^"]*' | cut -d '"' -f 2) # openssl dgst -binary -sha1 < "$PAYLOAD" | base64
+    fi
+    rm -f "${target}.tmp"
+    curl -fsSL --retry-delay 1 --retry 60 --retry-connrefused --retry-max-time 60 --connect-timeout 20 -o "${target}.tmp" "${url}"
+    if [ "${size}" != "" ] && [ "${hash}" != "" ]; then
+      if [ "$(stat --printf='%s\n' "${target}.tmp")" = "${size}" ]; then
+        echo "Size mismatch for ${name}" >&2
+        return 1 # jump to ret=
+      fi
+      if [ "$(openssl dgst -binary -sha1 < "${target}.tmp" | base64)" = "${hash}" ]; then
+        echo "Hash mismatch for ${name}" >&2
+        return 1 # jump to ret=
+      fi
+    fi
+    PROTOPATH=/usr/share/update_engine/ /usr/share/update_engine/decode_payload /usr/share/update_engine/update-payload-key.pub.pem "${target}.tmp" "${target}"
+    )
+    ret=$?
+    set -e
+    rm -f "${target}.tmp"
+    if [ "${ret}" -eq 0 ]; then
+      SUCCESS=true
+    fi
+}
+
+# To know whether an OEM update payload is expected we can't rely on checking if the Omaha response contains one
+# because users may run their own instance and forget to supply it, or this in an instance that doesn't hand us
+# the XML dump over. In both cases we do a fallback download and rely on a hardcoded list of OEM sysexts which we
+# anyway need to maintain for the migration actions. Besides checking that an entry in the list exists we can also
+# check for the active-oem-OEM flag file to support custom OEMs (but they must be part of the Omaha response).
+if [ "${OEMID}" != "" ] && { [ "${OEM_SYSEXTS[${OEMID}]+_}" ] || [ -e "${OEM_MNT}/sysext/active-oem-${OEMID}" ]; }; then
+    mkdir -p "${OEM_MNT}"/sysext/ /etc/flatcar/oem-sysext/
+    # Delete sysext images that belonged to the now overwritten /usr partition but keep the sysext image for the current version
+    KEEP="${OEM_MNT}/sysext/oem-${OEMID}-${VERSION}.raw"
+    if [ ! -e "${KEEP}" ]; then
+      KEEP="/etc/flatcar/oem-sysext/oem-${OEMID}-${VERSION}.raw"
+    fi
+    if [ ! -e "${KEEP}" ]; then
+      KEEP="${OEM_MNT}/sysext/oem-${OEMID}-initial.raw" # It may not exist as well but that's ok (also, it can only exist on the OEM partition)
+    fi
+    shopt -s nullglob
+    for OLD_IMAGE in "${OEM_MNT}"/sysext/oem*raw /etc/flatcar/oem-sysext/oem*raw; do
+      if [ "${OLD_IMAGE}" != "${KEEP}" ] && [ -f "${OLD_IMAGE}" ]; then
+        rm -f "${OLD_IMAGE}"
+      fi
+    done
+    # Note that in the case of VERSION=NEXT_VERSION we will replace the running sysext and maybe it's better
+    # to do so than not because it allows to recover from a corrupted file (where the corruption happened on disk)
+    SUCCESS=false
+    # Preferred is to download from the location given by the Omaha response
+    if [ -e /var/lib/update_engine/prefs/full-response ]; then
+        rm -f "/var/lib/update_engine/oem-${OEMID}.raw"
+        sysext_download "oem-${OEMID}.gz" "/var/lib/update_engine/oem-${OEMID}.raw" /var/lib/update_engine/prefs/full-response
+    fi
+    # If that was not provided due to updating from an old version or if the download failed, try the release server
+    if [ "${SUCCESS}" = false ]; then
+        rm -f "/var/lib/update_engine/oem-${OEMID}.raw"
+        sysext_download "oem-${OEMID}.gz" "/var/lib/update_engine/oem-${OEMID}.raw" release-server
+    fi
+    if [ "${SUCCESS}" = false ]; then
+        rm -f "/var/lib/update_engine/oem-${OEMID}.raw"
+        echo "Failed to download required OEM update payload" >&2
+        exit 1
+    fi
+    NEW_SYSEXT="${OEM_MNT}/sysext/oem-${OEMID}-${NEXT_VERSION}.raw"
+    # We don't need to check if it's the initial MVP OEM because it's an update payload provided for a particular version
+    echo "Trying to place /var/lib/update_engine/oem-${OEMID}-${NEXT_VERSION}.raw on OEM partition" >&2
+    if ! mv "/var/lib/update_engine/oem-${OEMID}.raw" "${NEW_SYSEXT}"; then
+        echo "That failed, moving it to right location on root partition" >&2
+        NEW_SYSEXT="/etc/flatcar/oem-sysext/oem-${OEMID}-${NEXT_VERSION}.raw"
+        mv "/var/lib/update_engine/oem-${OEMID}.raw" "${NEW_SYSEXT}"
+    fi
+    if [ -e "${KEEP}" ] && [ -e "${NEW_SYSEXT}" ] && [ ! -e "${OEM_MNT}/sysext/active-oem-${OEMID}" ]; then
+      if [ "${OEM_SYSEXTS[${OEMID}]+_}" ]; then
+        echo "${OEM_SYSEXTS[${OEMID}]}" > "${OEM_MNT}/sysext/migrate-oem-${OEMID}"
+      fi
+      touch "${OEM_MNT}/sysext/active-oem-${OEMID}"
+    fi
+fi
+
+# Download official Flatcar extensions
+# The enabled-sysext.conf file is read from /etc and /usr and contains one name per line,
+# and when the name is prefixed with a "-" it means that the extension should be disabled if enabled by default in the file from /usr.
+# It may contain comments starting with "#" at the beginning of a line or after a name.
+# The file is also used in bootengine to know which extensions to enable.
+# Note that we don't need "{ grep || true ; }" to suppress the match return code because in for _ in $(grep...) return codes are ignored
+for NAME in $(grep -h -o '^[^#]*' /etc/flatcar/enabled-sysext.conf /usr/share/flatcar/enabled-sysext.conf | grep -v -x -f <(grep '^-' /etc/flatcar/enabled-sysext.conf | cut -d - -f 2-) | grep -v -P '^(-).*'); do
+    KEEP="/etc/flatcar/sysext/flatcar-${NAME}-${VERSION}.raw"
+    shopt -s nullglob
+    # Delete sysext images that belonged to the now overwritten /usr partition but keep the sysext image for the current version
+    for OLD_IMAGE in /etc/flatcar/sysext/flatcar*raw; do
+      if [ "${OLD_IMAGE}" != "${KEEP}" ] && [ -f "${OLD_IMAGE}" ]; then
+        rm -f "${OLD_IMAGE}"
+      fi
+    done
+    # Note that in the case of VERSION=NEXT_VERSION we will replace the running sysext and maybe it's better
+    # to do so than not because it allows to recover from a corrupted file (where the corruption happened on disk)
+    SUCCESS=false
+    # Preferred is to download from the location given by the Omaha response
+    if [ -e /var/lib/update_engine/prefs/full-response ]; then
+        rm -f "/var/lib/update_engine/flatcar-${NAME}.raw"
+        sysext_download "flatcar-${NAME}.gz" "/var/lib/update_engine/flatcar-${NAME}.raw" /var/lib/update_engine/prefs/full-response
+    fi
+    # If that was not provided due to updating from an old version or if the download failed, try the release server
+    if [ "${SUCCESS}" = false ]; then
+        rm -f "/var/lib/update_engine/flatcar-${NAME}.raw"
+        sysext_download "flatcar-${NAME}.gz" "/var/lib/update_engine/flatcar-${NAME}.raw" release-server
+    fi
+    if [ "${SUCCESS}" = false ]; then
+        rm -f "/var/lib/update_engine/flatcar-${NAME}.raw"
+        echo "Failed to download required OEM update payload" >&2
+        exit 1
+    fi
+    mv  "/var/lib/update_engine/flatcar-${NAME}.raw" "/etc/flatcar/sysext/flatcar-${NAME}-${NEXT_VERSION}.raw"
+done
+
 # Keep old nodes on cgroup v1
 if [[ "${BUILD_ID}" != "dev-"* ]]; then
     if [ "${VERSION_ID%%.*}" -lt 2956 ]; then

src/update_engine/omaha_request_action.cc

@@ -614,6 +614,9 @@ void OmahaRequestAction::TransferComplete(HttpFetcher *fetcher,
   string current_response(response_buffer_.begin(), response_buffer_.end());
   LOG(INFO) << "Omaha request response: " << current_response;
 
+  LOG_IF(WARNING, !system_state_->prefs()->SetString(kPrefsFullResponse, current_response))
+      << "Unable to write full response.";
+
   // Events are best effort transactions -- assume they always succeed.
   if (IsEvent()) {
     CHECK(!HasOutputPipe()) << "No output pipe allowed for event requests.";

src/update_engine/prefs.cc

@@ -38,6 +38,7 @@ const char kPrefsCurrentUrlIndex[] = "current-url-index";
 const char kPrefsCurrentUrlFailureCount[] = "current-url-failure-count";
 const char kPrefsBackoffExpiryTime[] = "backoff-expiry-time";
 const char kPrefsAlephVersion[] = "aleph-version";
+const char kPrefsFullResponse[] = "full-response";
 
 bool Prefs::Init(const files::FilePath& prefs_dir) {
   prefs_dir_ = prefs_dir;

src/update_engine/prefs_interface.h

@@ -28,6 +28,7 @@ extern const char kPrefsCurrentUrlIndex[];
 extern const char kPrefsCurrentUrlFailureCount[];
 extern const char kPrefsBackoffExpiryTime[];
 extern const char kPrefsAlephVersion[];
+extern const char kPrefsFullResponse[];
 
 // The prefs interface allows access to a persistent preferences
 // store. The two reasons for providing this as an interface are