gitea-actions/bao-secrets/action.yml

name: "Fetch OpenBao secrets"
description: "AppRole-login to OpenBao and export requested KV v2 secrets into the job env (masked). Phase 1, spikersoft-issues#545."

inputs:
  role-id:
    description: "AppRole role_id — the repo's BAO_ROLE_ID Actions secret."
    required: true
  secret-id:
    description: "AppRole secret_id — the repo's BAO_SECRET_ID Actions secret."
    required: true
  bao-addr:
    description: "OpenBao address."
    required: false
    default: "https://bao.spikersoft.com"
  secrets:
    description: |
      One mapping per line:  ENV_NAME  <kv-v2-path>  <field>
      The path is exactly what you'd pass to `bao kv get` (mount + logical path).
      Blank lines and `#` comments are ignored. Example:
        DOCKER_PASSWORD  secret/ci/shared/registry       password
        VIDEOS_S3_KEY    secret/ci/backend/minio/videos  secret_key
    required: true

runs:
  using: "composite"
  steps:
    - shell: bash
      env:
        BAO_ADDR: ${{ inputs.bao-addr }}
        BAO_ROLE_ID: ${{ inputs.role-id }}
        BAO_SECRET_ID: ${{ inputs.secret-id }}
        BAO_SECRETS: ${{ inputs.secrets }}
      run: |
        set -euo pipefail
        command -v curl >/dev/null || { echo "::error::bao-secrets needs 'curl' on the runner"; exit 1; }
        command -v jq   >/dev/null || { echo "::error::bao-secrets needs 'jq' on the runner"; exit 1; }

        TOKEN=""
        # HTTP with retry + backoff. When many CI jobs log in at once the vault
        # nodes can return an nginx 504 (login is a raft write) or a 429 from the
        # rate-limit quota — both are transient and non-JSON, which used to crash
        # jq and hard-fail the job. Retry those; surface real 4xx bodies as-is.
        # Prints the response BODY on stdout.
        bao_req() { # METHOD URL [DATA]
          local method="$1" url="$2" data="${3:-}" attempt=1 max=6 code tmp secs
          tmp="$(mktemp)"
          local -a args=(-s -o "$tmp" -w '%{http_code}' --max-time 15 -X "$method")
          [ -n "${TOKEN:-}" ] && args+=(-H "X-Vault-Token: ${TOKEN}")
          [ -n "$data" ] && args+=(--data "$data")
          args+=("$url")
          while :; do
            code=$(curl "${args[@]}" 2>/dev/null || echo 000)
            case "$code" in
              200|204) cat "$tmp"; rm -f "$tmp"; return 0 ;;
              000|429|500|502|503|504)
                if [ "$attempt" -ge "$max" ]; then
                  echo "::warning::bao $method failed after ${attempt} tries (last HTTP ${code})" >&2
                  cat "$tmp"; rm -f "$tmp"; return 1
                fi
                secs=$(( attempt * 4 + RANDOM % 4 ))   # ~4,8,12,16,20s + jitter
                echo "  bao ${method} -> HTTP ${code}; retry ${attempt}/${max} in ${secs}s (cluster busy/rate-limited)" >&2
                sleep "$secs"; attempt=$((attempt+1)) ;;
              *) cat "$tmp"; rm -f "$tmp"; return 1 ;;   # 4xx etc — surface to caller
            esac
          done
        }

        # 1) AppRole login -> short-lived token (batch token if the role is set so).
        login="$(bao_req POST "${BAO_ADDR}/v1/auth/approle/login" \
          "{\"role_id\":\"${BAO_ROLE_ID}\",\"secret_id\":\"${BAO_SECRET_ID}\"}")" || true
        TOKEN=$(printf '%s' "$login" | jq -r '.auth.client_token // empty' 2>/dev/null || true)
        if [ -z "$TOKEN" ]; then
          echo "::error::OpenBao AppRole login failed (after retries): $(printf '%s' "$login" | jq -rc '.errors // "non-JSON/timeout — cluster busy or rate-limited"' 2>/dev/null || echo 'non-JSON response')"
          exit 1
        fi
        echo "::add-mask::$TOKEN"

        # 2) fetch each requested secret into $GITHUB_ENV (masked, multiline-safe)
        while IFS= read -r line; do
          line="${line%%#*}"
          # shellcheck disable=SC2086
          set -- $line
          [ "$#" -eq 0 ] && continue
          if [ "$#" -ne 3 ]; then
            echo "::error::bad 'secrets' line (need: ENV_NAME <kv-v2-path> <field>): ${line}"
            exit 1
          fi
          env_name="$1"; path="$2"; field="$3"
          mount="${path%%/*}"; rest="${path#*/}"      # KV v2: <mount>/data/<rest>
          resp="$(bao_req GET "${BAO_ADDR}/v1/${mount}/data/${rest}")" || true
          val=$(printf '%s' "$resp" | jq -r --arg f "$field" '.data.data[$f] // empty' 2>/dev/null || true)
          if [ -z "$val" ]; then
            echo "::error::no value at '${path}' field '${field}' (wrong path/field, this role's policy denies it, or cluster busy): $(printf '%s' "$resp" | jq -rc '.errors // "empty/non-JSON"' 2>/dev/null || echo 'non-JSON')"
            exit 1
          fi
          # mask every line of the value (handles multiline secrets like keys)
          while IFS= read -r vline; do [ -n "$vline" ] && echo "::add-mask::$vline"; done <<< "$val"
          {
            echo "${env_name}<<__BAO_EOF__"
            printf '%s\n' "$val"
            echo "__BAO_EOF__"
          } >> "$GITHUB_ENV"
          echo "  ✓ ${env_name}  <-  ${path}#${field}"
        done <<< "$BAO_SECRETS"

        # 3) drop the token (batch tokens can't be revoked; ignore errors)
        curl -s --max-time 10 -H "X-Vault-Token: $TOKEN" \
          -X POST "${BAO_ADDR}/v1/auth/token/revoke-self" >/dev/null 2>&1 || true