fix(k8s): harden control-plane backup safety

k8s/backup-k3s-control-plane-config.sh

@@ -41,6 +41,7 @@ CURL_BIN="${CURL_BIN:-curl}"
 FLOCK_BIN="${FLOCK_BIN:-flock}"
 MKTEMP_BIN="${MKTEMP_BIN:-mktemp}"
 DATE_BIN="${DATE_BIN:-date}"
+REALPATH_BIN="${REALPATH_BIN:-realpath}"
 REMOTE_BASH_BIN="${REMOTE_BASH_BIN:-bash}"
 
 # Notifications follow the JSON formats used by automated-nfs-backup.sh.
@@ -55,6 +56,7 @@ declare -a ARCHIVE_ENTRIES=()
 declare -a SERVICE_FILE_LIST=()
 declare -a SSH_ARGS=()
 declare -a SCP_ARGS=()
+declare -a PROTECTED_PATHS=()
 readonly -a K3S_SERVER_ENTRIES=(cred etc manifests static tls token agent-token node-token)
 
 RUN_TMP_DIR=""
@@ -175,16 +177,83 @@ require_absolute_path() {
     fi
 }
 
+path_is_ancestor_or_same() {
+    local ancestor="${1:?ancestor path is required}"
+    local path="${2:?path is required}"
+
+    [[ "$ancestor" == "/" || "$path" == "$ancestor" || "$path" == "$ancestor"/* ]]
+}
+
+paths_overlap() {
+    local first="${1:?first path is required}"
+    local second="${2:?second path is required}"
+
+    path_is_ancestor_or_same "$first" "$second" || \
+        path_is_ancestor_or_same "$second" "$first"
+}
+
+canonicalize_protected_path() {
+    local path="${1:?protected path is required}"
+    local canonical_path
+    local canonical_parent
+    local basename
+
+    if [[ -e "$path" || -L "$path" ]]; then
+        canonical_path="$("$REALPATH_BIN" -e -- "$path" 2>/dev/null)" || return 1
+        printf '%s' "$canonical_path"
+        return 0
+    fi
+
+    basename="${path##*/}"
+    canonical_parent="${path%/*}"
+    if ! canonical_parent="$("$REALPATH_BIN" -e -- "$canonical_parent" 2>/dev/null)"; then
+        canonical_parent="$("$REALPATH_BIN" -m -- "$canonical_parent" 2>/dev/null)" || return 1
+    fi
+    printf '%s/%s' "${canonical_parent%/}" "$basename"
+}
+
+determine_protected_paths() {
+    local canonical_server_dir
+    local protected_path
+
+    if ! canonical_server_dir="$("$REALPATH_BIN" -e -- "$K3S_SERVER_DIR" 2>/dev/null)"; then
+        canonical_server_dir="$("$REALPATH_BIN" -m -- "$K3S_SERVER_DIR" 2>/dev/null)" || \
+            die "Configured source path is not safe to archive"
+    fi
+
+    PROTECTED_PATHS=()
+    if ! protected_path="$(canonicalize_protected_path "${canonical_server_dir%/}/db")"; then
+        die "Configured source path is not safe to archive"
+    fi
+    PROTECTED_PATHS+=("$protected_path")
+    if ! protected_path="$(canonicalize_protected_path "${canonical_server_dir%/}/kine.sock")"; then
+        die "Configured source path is not safe to archive"
+    fi
+    PROTECTED_PATHS+=("$protected_path")
+}
+
 add_archive_entry_if_present() {
     local absolute_path="${1:?source path is required}"
+    local canonical_path
     local relative_path
+    local protected_path
 
     if [[ ! -e "$absolute_path" && ! -L "$absolute_path" ]]; then
         return 0
     fi
 
-    relative_path="${absolute_path#/}"
-    if [[ -z "$relative_path" || "$relative_path" == *"/../"* || "$relative_path" == ../* ]]; then
+    if ! canonical_path="$("$REALPATH_BIN" -e -- "$absolute_path" 2>/dev/null)"; then
+        die "Configured source path is not safe to archive"
+    fi
+
+    for protected_path in "${PROTECTED_PATHS[@]}"; do
+        if paths_overlap "$canonical_path" "$protected_path"; then
+            die "Configured source path is not safe to archive"
+        fi
+    done
+
+    relative_path="${canonical_path#/}"
+    if [[ -z "$relative_path" ]]; then
         die "Configured source path is not safe to archive"
     fi
     ARCHIVE_ENTRIES+=("$relative_path")
@@ -211,6 +280,7 @@ validate_inputs() {
     require_cmd "$FLOCK_BIN"
     require_cmd "$MKTEMP_BIN"
     require_cmd "$DATE_BIN"
+    require_cmd "$REALPATH_BIN"
     if [[ -n "$NOTIFY_SUCCESS_URL" || -n "$NOTIFY_FAILURE_URL" ]]; then
         require_cmd "$CURL_BIN"
     fi
@@ -224,6 +294,7 @@ validate_inputs() {
     if [[ ! -d "$K3S_CONFIG_DIR" ]]; then
         die "K3S_CONFIG_DIR is not an available directory"
     fi
+    determine_protected_paths
     if [[ ! "$ARCHIVE_PREFIX" =~ ^[[:alnum:]][[:alnum:]._-]*$ ]]; then
         die "ARCHIVE_PREFIX may contain only letters, numbers, dots, underscores, and hyphens"
     fi
@@ -300,7 +371,7 @@ create_and_validate_archive() {
     log_info "Creating k3s control-plane configuration archive"
     (
         cd /
-        "$ZIP_BIN" -r "$LOCAL_ARCHIVE_PATH" "${ARCHIVE_ENTRIES[@]}" >/dev/null
+        "$ZIP_BIN" -r "$LOCAL_ARCHIVE_PATH" -- "${ARCHIVE_ENTRIES[@]}" >/dev/null
     )
     "$UNZIP_BIN" -t "$LOCAL_ARCHIVE_PATH" >/dev/null
 
@@ -363,22 +434,20 @@ remote_dir="$1"
 archive_prefix="$2"
 keep_count="$3"
 
-shopt -s nullglob
-candidates=("${remote_dir%/}/${archive_prefix}_"*.zip)
-shopt -u nullglob
+mapfile -d '' -t sorted_archives < <(
+    find -P "$remote_dir" -maxdepth 1 -type f \
+        -name "${archive_prefix}_*.zip" -printf '%T@\t%p\0' |
+        sort -z -t $'\t' -k1,1nr
+)
 
-matching_archives=()
-for candidate in "${candidates[@]}"; do
-    [[ -f "$candidate" ]] && matching_archives+=("$candidate")
-done
-
-if (( ${#matching_archives[@]} <= keep_count )); then
+if (( ${#sorted_archives[@]} <= keep_count )); then
     exit 0
 fi
 
-mapfile -t sorted_archives < <(ls -1t -- "${matching_archives[@]}")
 for ((index = keep_count; index < ${#sorted_archives[@]}; index++)); do
-    rm -f -- "${sorted_archives[$index]}"
+    archive_record="${sorted_archives[$index]}"
+    archive_path="${archive_record#*$'\t'}"
+    rm -f -- "$archive_path"
 done
 REMOTE_CLEANUP
 

k8s/test/test-backup-k3s-control-plane-config.sh

@@ -34,7 +34,7 @@ is_nonzero() { [[ "$1" -ne 0 ]]; }
 path_exists() { [[ -e "$1" ]]; }
 path_does_not_exist() { [[ ! -e "$1" ]]; }
 file_contains() { grep -Fq -- "$2" "$1"; }
-file_lacks() { ! grep -Fq -- "$2" "$1"; }
+file_lacks() { [[ ! -e "$1" ]] || ! grep -Fq -- "$2" "$1"; }
 file_line_count_is() { [[ "$(grep -Fc -- "$2" "$1")" -eq "$3" ]]; }
 
 matching_archive_count_is() {
@@ -74,6 +74,8 @@ set -Eeuo pipefail
 [[ "${1:-}" == "-r" ]]
 archive="${2:?archive path required}"
 shift 2
+[[ "${1:-}" == "--" ]]
+shift
 : > "$archive"
 for entry in "$@"; do
     source_path="/$entry"
@@ -154,18 +156,10 @@ fi
 
 if [[ "$command" == *" -s -- "* ]]; then
     value_count="${#quoted_values[@]}"
-    prefix="${quoted_values[value_count-2]}"
-    keep_count="${quoted_values[value_count-1]}"
-    shopt -s nullglob
-    archives=("${FAKE_REMOTE_ROOT}/${prefix}_"*.zip)
-    shopt -u nullglob
-    if (( ${#archives[@]} > keep_count )); then
-        mapfile -t sorted_archives < <(ls -1t -- "${archives[@]}")
-        for ((i = keep_count; i < ${#sorted_archives[@]}; i++)); do
-            rm -f -- "${sorted_archives[$i]}"
-        done
-    fi
-    cat >/dev/null
+    (
+        cd -- "$FAKE_REMOTE_ROOT"
+        bash -s -- "$FAKE_REMOTE_ROOT" "${quoted_values[value_count-2]}" "${quoted_values[value_count-1]}"
+    )
     exit 0
 fi
 
@@ -208,7 +202,7 @@ run_backup() {
     shift 4
     local -a overrides=("$@")
 
-    env "${overrides[@]}" \
+    env \
         K3S_CONFIG_DIR="${fixture_root}/etc/rancher/k3s" \
         K3S_SERVER_DIR="${fixture_root}/var/lib/rancher/k3s/server" \
         K3S_SERVICE_FILES="${fixture_root}/etc/systemd/system/k3s.service:${fixture_root}/etc/systemd/system/k3s.service.env:${fixture_root}/etc/default/k3s:${fixture_root}/etc/sysconfig/k3s" \
@@ -229,6 +223,7 @@ run_backup() {
         PATH="${TMPD}/bin:${PATH}" \
         TMPD="$TMPD" \
         FAKE_REMOTE_ROOT="$remote_root" \
+        "${overrides[@]}" \
         bash "$SCRIPT_UNDER_TEST" >"$log_file" 2>&1
 }
 
@@ -279,6 +274,11 @@ for day in 1 2 3 4 5; do
 done
 printf 'legacy backup\n' > "${success_remote}/k8s-backup_2026-07-01.7z"
 printf 'other zip\n' > "${success_remote}/other-prefix_2026-07-01.zip"
+newline_archive="${success_remote}/k3s-control-plane-config_2026-06-01_12-00-00.zip"$'\nunrelated.zip'
+touch -t "202606011200" "$newline_archive"
+: > "$newline_archive"
+touch -t "202606011200" "$newline_archive"
+: > "${success_remote}/unrelated.zip"
 : > "${TMPD}/curl_trace.log"
 : > "${TMPD}/unzip_trace.log"
 : > "${TMPD}/zip_invocations.log"
@@ -302,11 +302,42 @@ assert "archive excludes kine socket" file_lacks "${success_remote}/k3s-control-
 assert "retention leaves exactly four newest matching zip archives" matching_archive_count_is "$success_remote" 4
 assert "retention removes the two oldest matching zip archives" path_does_not_exist "${success_remote}/k3s-control-plane-config_2026-07-01_12-00-00.zip"
 assert "retention removes the next-oldest matching zip archive" path_does_not_exist "${success_remote}/k3s-control-plane-config_2026-07-02_12-00-00.zip"
+assert "retention removes matching old archives with newlines in their filenames" path_does_not_exist "$newline_archive"
+assert "retention keeps the standalone unrelated.zip sentinel" path_exists "${success_remote}/unrelated.zip"
 assert "retention keeps unrelated legacy .7z backups" path_exists "${success_remote}/k8s-backup_2026-07-01.7z"
 assert "retention keeps unrelated zip backups" path_exists "${success_remote}/other-prefix_2026-07-01.zip"
 assert "local temporary archive state is removed" directory_is_empty "$success_local_tmp"
 assert "successful backup sends a success notification" file_contains "${TMPD}/curl_trace.log" "http://example.invalid/success"
 
+# A configured source path ending in server/db/.. must be rejected before zip is
+# invoked, rather than recursively archiving server/db and its sibling socket.
+source_scope_dir="${TMPD}/source-scope"
+source_scope_fixture="${source_scope_dir}/fixture"
+source_scope_remote="${source_scope_dir}/remote"
+source_scope_local_tmp="${source_scope_dir}/local-tmp"
+source_scope_archive="${source_scope_remote}/k3s-control-plane-config_2026-07-21_12-00-00.zip"
+mkdir -p "$source_scope_remote" "$source_scope_local_tmp"
+create_source_fixture "$source_scope_fixture"
+: > "${TMPD}/zip_invocations.log"
+
+set +e
+run_backup \
+    "$source_scope_fixture" \
+    "$source_scope_remote" \
+    "$source_scope_local_tmp" \
+    "${source_scope_dir}/run.log" \
+    "K3S_CONFIG_DIR=${source_scope_fixture}/var/lib/rancher/k3s/server/db/.."
+source_scope_exit_code=$?
+set -e
+
+assert "unsafe source override exits non-zero" is_nonzero "$source_scope_exit_code"
+assert "unsafe source override reports the source-scope error" file_contains "${source_scope_dir}/run.log" "Configured source path is not safe to archive"
+assert "unsafe source override is rejected before zip runs" file_lacks "${TMPD}/zip_invocations.log" "server/db/.."
+assert "unsafe source override archive excludes server database content" file_lacks "$source_scope_archive" "state.db"
+assert "unsafe source override archive excludes kine socket" file_lacks "$source_scope_archive" "kine.sock"
+assert "unsafe source override does not publish an archive" directory_is_empty "$source_scope_remote"
+assert "unsafe source override removes local temporary state" directory_is_empty "$source_scope_local_tmp"
+
 # Upload failure: the partial remote temporary file is removed, no final archive
 # is published, local temporary data is removed, and one safe error notification
 # is attempted.