prole/etc/init_knoe_users.sh

#!/usr/bin/env bash

set -euo pipefail

# init_knoe_users.sh
# Purpose:
# - Provision Kerberos principals and database accounts for knoe-system users
# - Creates: admin@PROLE.LOCAL (master password), guest@PROLE.LOCAL (read-only),
#   postgres service principal (keytab for GSS auth), developer group role
# - Cross-realm trust with myrddin.prole.org PROLE.ORG is activated via
#   PROLE_KDC_TRUST_REALM=PROLE.ORG in init_kdc.sh / init_knoe_auth.sh
# - Sets up service admin access: ArgoCD RBAC, Gitea, GitLab
#
# Prerequisites:
#   init_kdc.sh initialize  (with PROLE_KDC_REALM=PROLE.LOCAL,
#                             PROLE_KDC_TRUST_REALM=PROLE.ORG,
#                             KRB5_KDC=myrddin.prole.org)
#   init_cnpg_backup.sh     (CNPG cluster must exist)
#   init_argocd.sh          (ArgoCD must be running)
#
# Usage:
#   ./init_knoe_users.sh initialize   # create/update all user resources
#   ./init_knoe_users.sh status       # show current state
#   ./init_knoe_users.sh cleanup      # remove managed secrets/patches

SCRIPT_DIR=$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)

# shellcheck disable=SC1090
source "$SCRIPT_DIR/knoe_cfg.sh"

if [[ "${1:-}" == "--mode" || "${1:-}" == "-m" ]]; then
  knoe_set_mode "${2:-}"
  shift 2
elif [[ "${1:-}" == --mode=* || "${1:-}" == -m=* ]]; then
  knoe_set_mode "${1#*=}"
  shift
fi

knoe_ensure_kubeconfig >/dev/null 2>&1 || true
ensure_kube_context || exit 1

# Guard: we must never operate against the 'default' context — it means the
# desired context was not found and a placeholder was silently accepted.
_active_ctx=$(kubectl config current-context 2>/dev/null || true)
if [[ "${_active_ctx:-}" == "default" ]]; then
  printf '[ERROR] Active kubectl context is '\''default'\'' — the required context was not found.\n' >&2
  printf '        Add the correct kubeconfig to KUBECONFIG, e.g.:\n' >&2
  printf '          export KUBECONFIG=%s:~/.kube/config\n' "${KUBECONFIG:-/path/to/knoe-k3s.kubeconfig}" >&2
  printf '        Then retry.\n' >&2
  exit 1
fi
unset _active_ctx

ACTION="${1:-initialize}"

resolve_knoe_mode() {
  local raw_mode="${KNOE_MODE:-${DEPLOYMENT_MODE:-${CLUSTER_ENV:-k3s}}}"
  if command -v knoe_normalize_mode >/dev/null 2>&1; then
    knoe_normalize_mode "$raw_mode"
    return 0
  fi
  raw_mode=$(printf '%s' "$raw_mode" | tr 'A-Z' 'a-z')
  case "$raw_mode" in
    prod|production)
      printf 'k8s'
      ;;
    *)
      printf '%s' "$raw_mode"
      ;;
  esac
}

default_knoe_auth_deployment() {
  case "$(resolve_knoe_mode)" in
    k8s)
      printf 'authority-gcp-auth'
      ;;
    *)
      printf 'authority-knoe-auth'
      ;;
  esac
}

# ── Configurable variables ─────────────────────────────────────────────────
# PROLE_KDC_NAMESPACE is the canonical env set by init_kdc.sh; honour it here
# so both scripts look at the same namespace for knoe-kdc-secrets.
# Never fall through to "default" — use knoe-system as the last resort.
KNOE_USERS_NAMESPACE="${KNOE_USERS_NAMESPACE:-${SERVICE_NAMESPACE:-knoe-system}}"
KNOE_KDC_NAMESPACE="${KNOE_KDC_NAMESPACE:-${PROLE_KDC_NAMESPACE:-${KNOE_USERS_NAMESPACE}}}"
KNOE_DB_NAMESPACE="${KNOE_DB_NAMESPACE:-${DATABASE_NAMESPACE:-knoe-db}}"
KNOE_DB_CLUSTER="${KNOE_DB_CLUSTER:-knoe-db}"
KNOE_DEPLOYMENT_MODE="${KNOE_DEPLOYMENT_MODE:-$(resolve_knoe_mode)}"
KNOE_AUTH_DEPLOYMENT="${KNOE_AUTH_DEPLOYMENT:-${PROLE_KDC_NAME:-$(default_knoe_auth_deployment)}}"
KNOE_ADMIN_PRINCIPAL="${KNOE_ADMIN_PRINCIPAL:-admin}"
PROLE_KDC_REALM="${PROLE_KDC_REALM:-PROLE.LOCAL}"
GITEA_HOST="${GITEA_HOST:-git.prole.org}"
GITLAB_HOST="${GITLAB_HOST:-gitlab.prole.org}"
GITEA_NAMESPACE="${GITEA_NAMESPACE:-gitea}"
GITLAB_NAMESPACE="${GITLAB_NAMESPACE:-gitlab}"
ARGOCD_NAMESPACE="${ARGOCD_NAMESPACE:-argocd}"
# Provided externally or resolved from knoe-kdc-secrets
PROLE_KDC_MASTER_PASSWORD="${PROLE_KDC_MASTER_PASSWORD:-}"
KNOE_GUEST_PASSWORD="${KNOE_GUEST_PASSWORD:-}"
# UI-supplied credentials (set by knoe_users.py from the Database / Kerberos Auth screens)
PROLE_LOCAL_ADMIN_PASSWORD="${PROLE_LOCAL_ADMIN_PASSWORD:-}"
PROLE_ORG_ADMIN_PASSWORD="${PROLE_ORG_ADMIN_PASSWORD:-}"
# Optional: personal access tokens for service admin promotion
GITEA_ADMIN_TOKEN="${GITEA_ADMIN_TOKEN:-}"
GITLAB_ADMIN_TOKEN="${GITLAB_ADMIN_TOKEN:-}"

log()  { printf '[INFO]  %s\n' "$*"; }
err()  { printf '[ERROR] %s\n' "$*" >&2; }
warn() { printf '[WARN]  %s\n' "$*" >&2; }
die()  { err "$*"; exit 1; }

ensure_tools() {
  for t in kubectl curl python3; do
    command -v "$t" >/dev/null || die "Missing required tool: $t"
  done
}

b64_decode() {
  if base64 --decode </dev/null >/dev/null 2>&1; then
    base64 --decode
  elif base64 -d </dev/null >/dev/null 2>&1; then
    base64 -d
  else
    base64 -D
  fi
}

gen_password() {
  if command -v openssl >/dev/null 2>&1; then
    openssl rand -base64 18
    return 0
  fi
  python3 - <<'PY'
import secrets, string
alphabet = string.ascii_letters + string.digits
print(''.join(secrets.choice(alphabet) for _ in range(24)))
PY
}

# ── Secret helpers ─────────────────────────────────────────────────────────

# Read a base64-encoded key from a k8s Secret in KNOE_KDC_NAMESPACE.
# Mirrors get_secret_value() in init_kdc.sh (which reads from KDC_NAMESPACE).
get_secret_value() {
  local secret="$1" key="$2"
  kubectl -n "$KNOE_KDC_NAMESPACE" get secret "$secret" \
    -o "jsonpath={.data.${key}}" 2>/dev/null | b64_decode 2>/dev/null || true
}

# ── KDC helpers ────────────────────────────────────────────────────────────

get_kdc_pod() {
  # The kdc container lives inside the authority deployment pod selected by mode.
  local pod
  pod=$(kubectl -n "$KNOE_KDC_NAMESPACE" get pods \
    -l "app=${KNOE_AUTH_DEPLOYMENT}" \
    -o jsonpath='{.items[0].metadata.name}' 2>/dev/null || true)
  if [[ -n "$pod" ]]; then
    printf '%s' "$pod"
    return 0
  fi
  return 0
}

ensure_kdc_pod() {
  local kdc_pod init_kdc_script
  kdc_pod=$(get_kdc_pod)
  if [[ -n "$kdc_pod" ]]; then
    printf '%s' "$kdc_pod"
    return 0
  fi

  init_kdc_script="$SCRIPT_DIR/init_kdc.sh"
  [[ -f "$init_kdc_script" ]] || die "Required script not found: $init_kdc_script"

  log "No authority pod found in namespace ${KNOE_KDC_NAMESPACE}; bootstrapping ${KNOE_AUTH_DEPLOYMENT} via init_kdc.sh ..." >&2

  PROLE_KDC_NAMESPACE="$KNOE_KDC_NAMESPACE" \
  PROLE_KDC_MASTER_PASSWORD="$PROLE_KDC_MASTER_PASSWORD" \
  PROLE_KDC_NAME="$KNOE_AUTH_DEPLOYMENT" \
    bash "$init_kdc_script" initialize >&2

  kdc_pod=$(get_kdc_pod)
  [[ -n "$kdc_pod" ]] || die "No KDC pod found in namespace ${KNOE_KDC_NAMESPACE} (label app=${KNOE_AUTH_DEPLOYMENT})."
  printf '%s' "$kdc_pod"
}

kdc_principal_exists() {
  local pod="$1" principal="$2"
  kubectl -n "$KNOE_KDC_NAMESPACE" exec "$pod" -c kdc -- \
    kadmin.local -q "get_principal ${principal}" >/dev/null 2>&1
}

kdc_addprinc() {
  local pod="$1" principal="$2" password="$3"
  kubectl -n "$KNOE_KDC_NAMESPACE" exec "$pod" -c kdc -- \
    kadmin.local -q "addprinc -pw ${password} ${principal}"
}

kdc_addprinc_randkey() {
  local pod="$1" principal="$2"
  kubectl -n "$KNOE_KDC_NAMESPACE" exec "$pod" -c kdc -- \
    kadmin.local -q "addprinc -randkey ${principal}"
}

# ── CNPG helpers ───────────────────────────────────────────────────────────

get_cnpg_primary() {
  local pod=""
  # CNPG 1.20+: instanceRole label
  pod=$(kubectl -n "$KNOE_DB_NAMESPACE" get pods \
    -l "cnpg.io/cluster=${KNOE_DB_CLUSTER},cnpg.io/instanceRole=primary" \
    -o jsonpath='{.items[0].metadata.name}' 2>/dev/null || true)
  if [[ -z "$pod" ]]; then
    # Older label set
    pod=$(kubectl -n "$KNOE_DB_NAMESPACE" get pods \
      -l "cnpg.io/cluster=${KNOE_DB_CLUSTER},role=primary" \
      -o jsonpath='{.items[0].metadata.name}' 2>/dev/null || true)
  fi
  printf '%s' "$pod"
}

cnpg_psql() {
  local pod="$1"; shift
  kubectl -n "$KNOE_DB_NAMESPACE" exec "$pod" -- \
    psql -U postgres -d postgres "$@"
}

# ── Main: initialize ───────────────────────────────────────────────────────

initialize() {
  ensure_tools

  log "=== knoe-system user provisioning ==="
  log "Namespaces: kdc/authority=${KNOE_KDC_NAMESPACE}  db=${KNOE_DB_NAMESPACE}"
  log "Mode: ${KNOE_DEPLOYMENT_MODE}  authority deployment: ${KNOE_AUTH_DEPLOYMENT}"

  # 1. Resolve passwords — read from knoe-kdc-secrets (written by init_kdc.sh)
  if [[ -z "$PROLE_KDC_MASTER_PASSWORD" ]]; then
    PROLE_KDC_MASTER_PASSWORD=$(get_secret_value "knoe-kdc-secrets" "master_password")
  fi
  # Allow the UI-supplied PROLE_LOCAL_ADMIN_PASSWORD (DB master password field) to
  # override when the secret is absent or empty.
  if [[ -z "$PROLE_KDC_MASTER_PASSWORD" && -n "${PROLE_LOCAL_ADMIN_PASSWORD:-}" ]]; then
    PROLE_KDC_MASTER_PASSWORD="$PROLE_LOCAL_ADMIN_PASSWORD"
    log "Using PROLE_LOCAL_ADMIN_PASSWORD as PROLE_KDC_MASTER_PASSWORD"
  fi
  if [[ -z "$PROLE_KDC_MASTER_PASSWORD" ]]; then
    die "PROLE_KDC_MASTER_PASSWORD not found.
  The KDC master password must match the running KDC database and cannot be
  auto-generated.  Resolve one of:
    1. Run:   ./etc/init_kdc.sh initialize
       (creates/stores the secret in namespace '${KNOE_KDC_NAMESPACE}')
    2. Export: export PROLE_KDC_MASTER_PASSWORD=<password>
    3. Set the database master password on the Database screen and retry
    4. Check:  kubectl -n ${KNOE_KDC_NAMESPACE} get secret knoe-kdc-secrets -o yaml"
  fi

  if [[ -z "$KNOE_GUEST_PASSWORD" ]]; then
    KNOE_GUEST_PASSWORD=$(get_secret_value "knoe-kdc-secrets" "guest_password")
  fi
  if [[ -z "$KNOE_GUEST_PASSWORD" ]]; then
    KNOE_GUEST_PASSWORD=$(gen_password)
    log "Auto-generated guest password — storing in knoe-kdc-secrets ..."
    if kubectl -n "$KNOE_KDC_NAMESPACE" get secret knoe-kdc-secrets >/dev/null 2>&1; then
      local guest_b64
      guest_b64=$(printf '%s' "$KNOE_GUEST_PASSWORD" | base64 | tr -d '\n')
      kubectl -n "$KNOE_KDC_NAMESPACE" patch secret knoe-kdc-secrets \
        --type=merge \
        --patch "{\"data\":{\"guest_password\":\"${guest_b64}\"}}" >/dev/null
    else
      kubectl -n "$KNOE_KDC_NAMESPACE" create secret generic knoe-kdc-secrets \
        --from-literal=guest_password="$KNOE_GUEST_PASSWORD" \
        --dry-run=client -o yaml | kubectl apply -f - >/dev/null
    fi
  fi

  # 2. Locate (or bootstrap) KDC pod
  local kdc_pod
  kdc_pod=$(ensure_kdc_pod)
  log "KDC pod: $kdc_pod"

  # 2a. Pre-flight: verify the KDC database stash file is present and kadmin.local
  #     can access the database.  A missing stash means init_kdc.sh has not been
  #     run (or the pod restarted with ephemeral storage).
  log "Verifying KDC database accessibility ..."
  if ! kubectl -n "$KNOE_KDC_NAMESPACE" exec "$kdc_pod" -c kdc -- \
      sh -c 'kadmin.local -q listprincs' >/dev/null 2>&1; then
    die "KDC database not accessible in pod ${kdc_pod}.
  The KDC master-key stash file is missing or the database has not been initialised.
  Resolve by running:  ./etc/init_kdc.sh initialize
  Then click 'Provision Users & Auth' again."
  fi
  log "KDC database accessible"

  # 3. Create admin@PROLE.LOCAL (UI login with master password)
  local admin_princ="${KNOE_ADMIN_PRINCIPAL}@${PROLE_KDC_REALM}"
  if kdc_principal_exists "$kdc_pod" "$admin_princ"; then
    log "${admin_princ} already exists"
  else
    log "Creating ${admin_princ} ..."
    kdc_addprinc "$kdc_pod" "$admin_princ" "$PROLE_KDC_MASTER_PASSWORD"
  fi

  # 4. Create guest@PROLE.LOCAL (read-only)
  local guest_princ="guest@${PROLE_KDC_REALM}"
  if kdc_principal_exists "$kdc_pod" "$guest_princ"; then
    log "${guest_princ} already exists"
  else
    log "Creating ${guest_princ} ..."
    kdc_addprinc "$kdc_pod" "$guest_princ" "$KNOE_GUEST_PASSWORD"
  fi

  # 5. Ensure BOTH cross-realm trust principals exist in the MIT KDC
  #    Direction A: krbtgt/PROLE.ORG@PROLE.LOCAL  — cluster users → AD services
  #                 (created by init_kdc.sh PROLE_KDC_TRUST_REALM mechanism)
  #    Direction B: krbtgt/PROLE.LOCAL@PROLE.ORG  — AD users → cluster services ← THIS IS WHAT WE NEED
  local trust_princ_in="krbtgt/PROLE.ORG@${PROLE_KDC_REALM}"
  local trust_princ_out="krbtgt/${PROLE_KDC_REALM}@PROLE.ORG"
  local trust_shared_pw
  trust_shared_pw=$(kubectl -n "$KNOE_KDC_NAMESPACE" get secret knoe-kdc-secrets \
    -o jsonpath='{.data.trust_shared_password}' 2>/dev/null | b64_decode || true)
  if [[ -z "$trust_shared_pw" ]]; then
    warn "trust_shared_password not set in knoe-kdc-secrets; cross-realm trust principals will not be created"
    warn "Add trust_shared_password to knoe-kdc-secrets and re-run"
  else
    # Direction A (init_kdc.sh may already have created this)
    if kdc_principal_exists "$kdc_pod" "$trust_princ_in"; then
      log "${trust_princ_in} already exists"
    else
      log "Creating cross-realm principal ${trust_princ_in} ..."
      kdc_addprinc "$kdc_pod" "$trust_princ_in" "$trust_shared_pw"
    fi
    # Direction B — the principal that allows AD users to reach cluster services
    if kdc_principal_exists "$kdc_pod" "$trust_princ_out"; then
      log "${trust_princ_out} already exists"
    else
      log "Creating cross-realm principal ${trust_princ_out} ..."
      kdc_addprinc "$kdc_pod" "$trust_princ_out" "$trust_shared_pw"
    fi
  fi

  # 6. Create postgres service principal + export keytab
  local pg_spn="postgres/knoe-db-rw.${KNOE_DB_NAMESPACE}.svc.cluster.local@${PROLE_KDC_REALM}"
  if kdc_principal_exists "$kdc_pod" "$pg_spn"; then
    log "${pg_spn} already exists"
  else
    log "Creating service principal ${pg_spn} ..."
    kdc_addprinc_randkey "$kdc_pod" "$pg_spn"
  fi

  log "Exporting keytab for ${pg_spn} ..."
  kubectl -n "$KNOE_KDC_NAMESPACE" exec "$kdc_pod" -c kdc -- \
    kadmin.local -q "ktadd -k /tmp/pg.keytab -norandkey ${pg_spn}" >/dev/null

  log "Storing keytab in Secret knoe-db-pg-keytab ..."
  local keytab_b64
  keytab_b64=$(kubectl -n "$KNOE_KDC_NAMESPACE" exec "$kdc_pod" -c kdc -- \
    sh -c 'base64 /tmp/pg.keytab | tr -d "\\n"')
  # Write secret into the DB namespace where CNPG mounts it
  kubectl -n "$KNOE_DB_NAMESPACE" create secret generic knoe-db-pg-keytab \
    --from-literal=pg.keytab="$(printf '%s' "$keytab_b64" | b64_decode)" \
    --dry-run=client -o yaml | kubectl apply -f - >/dev/null
  log "knoe-db-pg-keytab stored in namespace ${KNOE_DB_NAMESPACE}"

  # 6. Patch CNPG cluster with managed.roles (idempotent merge)
  log "Patching CNPG cluster ${KNOE_DB_CLUSTER} managed.roles ..."
  kubectl -n "$KNOE_DB_NAMESPACE" patch cluster "$KNOE_DB_CLUSTER" \
    --type=merge --patch '{
      "spec": {
        "managed": {
          "roles": [
            {
              "name": "admin",
              "ensure": "present",
              "login": true,
              "superuser": true,
              "comment": "Kerberos admin principal — full cluster access"
            },
            {
              "name": "guest",
              "ensure": "present",
              "login": true,
              "superuser": false,
              "comment": "Kerberos guest principal — read-only access to demo schema"
            },
            {
              "name": "developer",
              "ensure": "present",
              "login": false,
              "superuser": false,
              "comment": "Developer group role — granted to knoe-system user accounts"
            }
          ]
        }
      }
    }'

  # 7. Locate CNPG primary and run SQL
  local primary
  primary=$(get_cnpg_primary)
  if [[ -z "$primary" ]]; then
    warn "CNPG primary pod not found — skipping SQL setup (run again once cluster is ready)"
  else
    log "CNPG primary pod: $primary"
    log "Setting up demo schema and role grants ..."
    cnpg_psql "$primary" <<'SQL'
-- demo schema for guest read-only access (evolves over time)
CREATE SCHEMA IF NOT EXISTS demo;

-- Ensure roles exist (CNPG managed.roles may not have reconciled yet)
DO $$
BEGIN
  IF NOT EXISTS (SELECT FROM pg_roles WHERE rolname = 'admin') THEN
    CREATE ROLE admin LOGIN SUPERUSER;
  ELSE
    ALTER ROLE admin SUPERUSER LOGIN;
  END IF;
END $$;

DO $$
BEGIN
  IF NOT EXISTS (SELECT FROM pg_roles WHERE rolname = 'guest') THEN
    CREATE ROLE guest LOGIN;
  END IF;
END $$;

DO $$
BEGIN
  IF NOT EXISTS (SELECT FROM pg_roles WHERE rolname = 'developer') THEN
    CREATE ROLE developer NOLOGIN;
  END IF;
END $$;

-- guest: read-only on demo schema
GRANT USAGE ON SCHEMA demo TO guest;
GRANT SELECT ON ALL TABLES IN SCHEMA demo TO guest;
ALTER DEFAULT PRIVILEGES IN SCHEMA demo GRANT SELECT ON TABLES TO guest;

-- developer: read/write access on public and demo schemas
GRANT USAGE ON SCHEMA public TO developer;
GRANT USAGE ON SCHEMA demo TO developer;
GRANT SELECT, INSERT, UPDATE, DELETE ON ALL TABLES IN SCHEMA public TO developer;
GRANT SELECT, INSERT, UPDATE, DELETE ON ALL TABLES IN SCHEMA demo TO developer;
ALTER DEFAULT PRIVILEGES IN SCHEMA public GRANT SELECT, INSERT, UPDATE, DELETE ON TABLES TO developer;
ALTER DEFAULT PRIVILEGES IN SCHEMA demo GRANT SELECT, INSERT, UPDATE, DELETE ON TABLES TO developer;
SQL
    log "SQL setup complete"
  fi

  # 8. Create knoe.user schema tables and bootstrap users
  if [[ -n "$primary" ]]; then
    create_knoe_schema "$primary"
    provision_user "$primary" "chrisfu" "chrisfu@prole.org" "Chris Fu" "admin"
    provision_user "$primary" "ron"     "ron@prole.org"     "Ron"      "developer"
  else
    warn "CNPG primary not found — skipping knoe.user schema and user provisioning"
  fi

  # 9. ArgoCD RBAC patch
  patch_argocd_rbac

  # 10. Gitea admin promotion
  promote_gitea_admin

  # 11. GitLab admin promotion
  promote_gitlab_admin

  log "=== Initialization complete ==="
  log ""
  log "Next steps:"
  log "  1. On myrddin.prole.org: add krbtgt/PROLE.LOCAL@PROLE.ORG trust principal"
  log "     (samba-tool domain trust or equivalent, using trust_shared_password from knoe-kdc-secrets)"
  log "  2. Confirm Grafana auth.proxy configured with: headers = Role:X-Knoe-Groups"
  log "  3. After admin logs in to Gitea/GitLab for the first time, re-run: $0 initialize"
}

# ── knoe.user schema + user provisioning ──────────────────────────────────

create_knoe_schema() {
  local primary="$1"
  log "Creating knoe.user schema tables (idempotent)..."
  cnpg_psql "$primary" <<'SQL'
CREATE TABLE IF NOT EXISTS knoe.user (
    id             SERIAL PRIMARY KEY,
    username       TEXT NOT NULL UNIQUE,
    realm          TEXT NOT NULL DEFAULT 'PROLE.LOCAL',
    email          TEXT,
    display_name   TEXT,
    tenant_realm   TEXT,
    is_realm_admin BOOLEAN DEFAULT false,
    created_at     TIMESTAMPTZ DEFAULT now(),
    updated_at     TIMESTAMPTZ DEFAULT now()
);
CREATE TABLE IF NOT EXISTS knoe.user_role (
    user_id    INT NOT NULL REFERENCES knoe.user(id) ON DELETE CASCADE,
    role       TEXT NOT NULL,
    granted_at TIMESTAMPTZ DEFAULT now(),
    PRIMARY KEY (user_id, role)
);
GRANT SELECT, INSERT, UPDATE ON knoe.user      TO knoe;
GRANT SELECT, INSERT, UPDATE ON knoe.user_role TO knoe;
GRANT USAGE, SELECT ON SEQUENCE knoe.user_id_seq TO knoe;
SQL
  log "knoe.user schema ready."
}

# provision_user <primary_pod> <username> <email> <display_name> <role>
# role: admin | developer | guest  (default: developer)
# Idempotent — safe to re-run.
provision_user() {
  local primary="$1" username="$2" email="${3:-}" display_name="${4:-$2}" role="${5:-developer}"

  log "Provisioning user: ${username} <${email}> role=${role}"

  # 1. Kerberos principal (randkey — user sets own password via kpasswd / kadmin)
  local user_princ="${username}@${PROLE_KDC_REALM}"
  local kdc_pod
  kdc_pod=$(get_kdc_pod)
  if [[ -n "$kdc_pod" ]]; then
    if kdc_principal_exists "$kdc_pod" "$user_princ"; then
      log "  Kerberos: ${user_princ} already exists"
    else
      log "  Kerberos: creating ${user_princ} ..."
      kdc_addprinc_randkey "$kdc_pod" "$user_princ"
    fi
  else
    warn "  No KDC pod — skipping Kerberos principal for ${username}"
  fi

  # 2. PostgreSQL role (GSS-API: username@PROLE.LOCAL → pg role 'username')
  cnpg_psql "$primary" -v username="$username" -v role="$role" <<SQL
DO \$\$
BEGIN
  IF NOT EXISTS (SELECT FROM pg_roles WHERE rolname = '${username}') THEN
    CREATE ROLE ${username} LOGIN;
  END IF;
END \$\$;
GRANT ${role} TO ${username};
SQL

  # 3. knoe.user record (upsert)
  cnpg_psql "$primary" <<SQL
INSERT INTO knoe.user (username, realm, email, display_name)
  VALUES ('${username}', '${PROLE_KDC_REALM}', '${email}', '${display_name}')
  ON CONFLICT (username) DO UPDATE SET
    email        = EXCLUDED.email,
    display_name = EXCLUDED.display_name,
    updated_at   = now();
INSERT INTO knoe.user_role (user_id, role)
  SELECT id, '${role}' FROM knoe.user WHERE username = '${username}'
  ON CONFLICT DO NOTHING;
SQL

  log "  User ${username} provisioned."
}

# ── ArgoCD RBAC ────────────────────────────────────────────────────────────

patch_argocd_rbac() {
  if ! kubectl -n "$ARGOCD_NAMESPACE" get configmap argocd-rbac-cm >/dev/null 2>&1; then
    warn "argocd-rbac-cm not found in namespace ${ARGOCD_NAMESPACE}; skipping ArgoCD RBAC patch"
    return 0
  fi

  local admin_line="g, ${KNOE_ADMIN_PRINCIPAL}, role:admin"
  local current_csv
  current_csv=$(kubectl -n "$ARGOCD_NAMESPACE" get configmap argocd-rbac-cm \
    -o jsonpath='{.data.policy\.csv}' 2>/dev/null || true)

  if printf '%s' "$current_csv" | grep -qF "$admin_line"; then
    log "ArgoCD RBAC already contains admin entry"
    return 0
  fi

  local new_csv="${current_csv}"$'\n'"${admin_line}"
  local escaped_csv
  escaped_csv=$(python3 -c 'import sys,json; print(json.dumps(sys.stdin.read()))' <<< "$new_csv")

  kubectl -n "$ARGOCD_NAMESPACE" patch configmap argocd-rbac-cm \
    --type=merge \
    --patch "{\"data\":{\"policy.csv\":${escaped_csv}}}" >/dev/null
  log "ArgoCD RBAC: added '${admin_line}'"
}

# ── Gitea admin promotion ───────────────────────────────────────────────────

promote_gitea_admin() {
  local token="$GITEA_ADMIN_TOKEN"
  if [[ -z "$token" ]]; then
    token=$(kubectl -n "$GITEA_NAMESPACE" get secret gitea-admin-token \
      -o jsonpath='{.data.token}' 2>/dev/null | b64_decode || true)
  fi
  if [[ -z "$token" ]]; then
    warn "No GITEA_ADMIN_TOKEN available — skipping Gitea admin promotion"
    warn "  Manually: curl -X PATCH https://${GITEA_HOST}/api/v1/admin/users/${KNOE_ADMIN_PRINCIPAL} \\"
    warn "    -H 'Authorization: token <token>' -H 'Content-Type: application/json' \\"
    warn "    -d '{\"admin\":true,\"login_name\":\"${KNOE_ADMIN_PRINCIPAL}\",\"source_id\":0}'"
    return 0
  fi

  local http_code
  http_code=$(curl -s -o /dev/null -w '%{http_code}' \
    -H "Authorization: token ${token}" \
    "https://${GITEA_HOST}/api/v1/users/${KNOE_ADMIN_PRINCIPAL}" 2>/dev/null || true)

  if [[ "$http_code" == "404" ]]; then
    warn "Gitea: user '${KNOE_ADMIN_PRINCIPAL}' not found (must log in first)"
    warn "  Re-run '$0 initialize' after first login"
    return 0
  fi

  http_code=$(curl -s -o /dev/null -w '%{http_code}' \
    -X PATCH \
    -H "Authorization: token ${token}" \
    -H "Content-Type: application/json" \
    -d "{\"admin\":true,\"login_name\":\"${KNOE_ADMIN_PRINCIPAL}\",\"source_id\":0}" \
    "https://${GITEA_HOST}/api/v1/admin/users/${KNOE_ADMIN_PRINCIPAL}" 2>/dev/null || true)

  if [[ "$http_code" =~ ^2 ]]; then
    log "Gitea: '${KNOE_ADMIN_PRINCIPAL}' promoted to admin"
  else
    warn "Gitea admin promotion returned HTTP ${http_code}"
  fi
}

# ── GitLab admin promotion ─────────────────────────────────────────────────

promote_gitlab_admin() {
  local token="$GITLAB_ADMIN_TOKEN"
  if [[ -z "$token" ]]; then
    warn "No GITLAB_ADMIN_TOKEN (personal access token) available — skipping GitLab admin promotion"
    warn "  Set GITLAB_ADMIN_TOKEN and re-run, or use:"
    warn "  kubectl exec -n ${GITLAB_NAMESPACE} <rails-pod> -- gitlab-rails runner \\"
    warn "    \"User.find_by_username('${KNOE_ADMIN_PRINCIPAL}')&.update(admin: true)\""
    return 0
  fi

  # Look up user ID by username
  local user_json user_id
  user_json=$(curl -s \
    -H "PRIVATE-TOKEN: ${token}" \
    "https://${GITLAB_HOST}/api/v4/users?username=${KNOE_ADMIN_PRINCIPAL}" 2>/dev/null || true)
  user_id=$(python3 -c \
    'import sys,json; u=json.load(sys.stdin); print(u[0]["id"] if u else "")' \
    <<< "$user_json" 2>/dev/null || true)

  if [[ -z "$user_id" ]]; then
    warn "GitLab: user '${KNOE_ADMIN_PRINCIPAL}' not found (must log in first)"
    warn "  Re-run '$0 initialize' after first login"
    return 0
  fi

  local http_code
  http_code=$(curl -s -o /dev/null -w '%{http_code}' \
    -X PUT \
    -H "PRIVATE-TOKEN: ${token}" \
    -H "Content-Type: application/json" \
    -d '{"admin":true}' \
    "https://${GITLAB_HOST}/api/v4/users/${user_id}" 2>/dev/null || true)

  if [[ "$http_code" =~ ^2 ]]; then
    log "GitLab: '${KNOE_ADMIN_PRINCIPAL}' promoted to admin"
  else
    warn "GitLab admin promotion returned HTTP ${http_code}"
  fi
}

# ── Status ─────────────────────────────────────────────────────────────────

show_status() {
  log "=== KDC Principals ==="
  local kdc_pod
  kdc_pod=$(get_kdc_pod)
  if [[ -n "$kdc_pod" ]]; then
    kubectl -n "$KNOE_KDC_NAMESPACE" exec "$kdc_pod" -c kdc -- \
      kadmin.local -q "list_principals" 2>/dev/null \
      | grep -E "^(admin|guest|developer|postgres/|krbtgt/)" || true
  else
    warn "No KDC pod found"
  fi

  log ""
  log "=== CNPG Managed Roles ==="
  kubectl -n "$KNOE_DB_NAMESPACE" get cluster "$KNOE_DB_CLUSTER" \
    -o jsonpath='{.spec.managed.roles}' 2>/dev/null \
    | python3 -m json.tool 2>/dev/null || warn "Could not read managed.roles"

  log ""
  log "=== PostgreSQL Roles ==="
  local primary
  primary=$(get_cnpg_primary)
  if [[ -n "$primary" ]]; then
    cnpg_psql "$primary" -c "\\du admin guest developer" 2>/dev/null || true
  else
    warn "No CNPG primary pod found"
  fi

  log ""
  log "=== Keytab Secret ==="
  if kubectl -n "$KNOE_DB_NAMESPACE" get secret knoe-db-pg-keytab >/dev/null 2>&1; then
    log "knoe-db-pg-keytab exists in namespace ${KNOE_DB_NAMESPACE}"
  else
    warn "knoe-db-pg-keytab NOT FOUND in namespace ${KNOE_DB_NAMESPACE}"
  fi

  log ""
  log "=== ArgoCD RBAC ==="
  kubectl -n "$ARGOCD_NAMESPACE" get configmap argocd-rbac-cm \
    -o jsonpath='{.data.policy\.csv}' 2>/dev/null || warn "argocd-rbac-cm not found"
}

# ── Cleanup ────────────────────────────────────────────────────────────────

cleanup() {
  warn "cleanup removes the pg keytab secret and clears managed.roles from CNPG."
  warn "KDC principals are NOT deleted (use kadmin.local to remove them manually)."

  kubectl -n "$KNOE_DB_NAMESPACE" delete secret knoe-db-pg-keytab \
    --ignore-not-found >/dev/null && log "Deleted knoe-db-pg-keytab"

  kubectl -n "$KNOE_DB_NAMESPACE" patch cluster "$KNOE_DB_CLUSTER" \
    --type=merge --patch '{"spec":{"managed":null}}' >/dev/null 2>&1 \
    && log "Cleared managed.roles from cluster ${KNOE_DB_CLUSTER}" || true
}

# ── Dispatch ───────────────────────────────────────────────────────────────

case "$ACTION" in
  initialize|init) initialize ;;
  status)          show_status ;;
  cleanup)         cleanup ;;
  *)
    err "Unknown action: ${ACTION}"
    err "Usage: $(basename "$0") [initialize|status|cleanup]"
    exit 1
    ;;
esac