Files
shopdb-flask/shopdb/utils/apitoken_auth.py
cproudlock 784964c907 Bring the wilm branch into main: endpoints plugin, agent, policies, CSV import
Ports the corporate-GitHub wilm branch onto main, keeping the two branches
separate. Thirteen upstream commits arrive with their authorship intact; the
review fixes and the portability work below are on top.

What arrives:

  endpoints plugin  - a ShopDB agent reporting endpoint state, telemetry,
                      services and interfaces, with a command channel the agent
                      long-polls. Ingest stays on the generic collector route
                      (ADR-006); the agent key is a UI-managed single-scope
                      token whose revocation is an instant kill switch.
  computers policies - Configuration and Business Policies (RMM-style), scope
                      trees validated at write time, actions checked against
                      the endpoints command allowlist. Feature-gated: every
                      route 404s while the setting is off, so enabling it can
                      never half-expose the surface.
  asset CSV import  - templates, a collector script carrying the site path, and
                      a validate-then-commit import. A plugin can own a virtual
                      column plus a post-write row hook, which is how a PC's
                      machine number becomes a controls link without a
                      hand-built relationships sheet.
  agent/            - the Windows agent and its installer.

Contract 0.21.0 -> 0.24.0, additive throughout: register_import_hook (0.22),
mint/list/revoke_service_token (0.23), collector_authorized (0.24). The stability
table had no rows for any of them and now does.

MySQL 5.6 portability. The branch stored three policy columns as sa.JSON, the
only JSON columns in the product, and 5.6 has no JSON type: the migration failed
with a bare syntax error near the word JSON. They are TEXT now, through a
JSONText type decorator, so all seventeen call sites still read and write dicts.
The convention elsewhere is already TEXT (guides.blocksjson,
geenforce.manifestjson).

Review fixes carried on top:

  - six policy GETs were jwt_required(optional=True), which authenticates a
    caller without checking anything; they require computers.view now, and six
    redundant jwt_required decorators are gone
  - the agent installer took a default BaseUrl and CollectorKey, so a mistyped
    run silently pointed an endpoint at the wrong instance; both are mandatory
    and a non-https BaseUrl throws unless -AllowInsecureHttp is passed
  - a committing CSV import wrote no audit row

Documentation: the 34 new routes are described in docs/api-inventory.json, which
is hand-maintained on purpose, and the generated spec now covers 331 paths and
477 operations across 22 surfaces. DEVELOPMENT-SETUP's 1071 troubleshooting row
said "use MySQL 8 for dev" and now gives the actual fix, including the trap that
5.6 downgrades ROW_FORMAT=DYNAMIC to Compact with a warning instead of an error.

Full suite 2123 passed, vitest 260 passed, all seventeen plugin chains migrate
clean on MySQL 5.6 and the naming, docs, parity and read-surface gates are green.
2026-08-28 16:17:17 -04:00

241 lines
9.3 KiB
Python

"""Personal API token (PAT) authentication shim.
A request may send `Authorization: Bearer shopdb_pat_...`. This is recognized
BEFORE any JWT decode: a before_request hook validates the PAT (hash lookup,
active, not expired, active owner) and, on success, mints a short internal
request-scoped JWT for the token's user and swaps it into the request's
Authorization header.
Why mint a JWT instead of only stashing the user on g: every write route in
this app stacks a mandatory @jwt_required() ABOVE @require_permission /
@require_role. That mandatory decorator decodes the Authorization header
itself, so the ONLY way a PAT reaches the whole existing auth+authz stack
(jwt_required, require_permission, require_role, import_mode, current_user,
get_jwt_identity) unchanged is to present a genuine JWT downstream. The minted
token lives only in this request's environ and is never returned to the client.
Result: a PAT authenticates any route a login JWT would, acting as its owner,
with zero changes to the authz decorators or import-mode helpers.
"""
from datetime import datetime, timezone
from flask import g, request
from flask_jwt_extended import create_access_token
from shopdb.extensions import db
from shopdb.core.models.apitoken import ApiToken, TOKEN_SECRET_PREFIX
from shopdb.utils.responses import error_response, ErrorCodes
# Only rewrite lastusedat when it is older than this, to avoid a DB write on
# every single request a busy integration makes.
_LASTUSED_THROTTLE_SECONDS = 60
def _utcnow():
return datetime.now(timezone.utc).replace(tzinfo=None)
def _extract_pat_secret():
"""Return the PAT secret from the Authorization header, or None."""
header = request.headers.get('Authorization', '')
parts = header.split()
if len(parts) == 2 and parts[0] == 'Bearer' \
and parts[1].startswith(TOKEN_SECRET_PREFIX):
return parts[1]
return None
def resolve_api_token(secret):
"""Validate a PAT secret. Return (token, user) or None.
Shared validator: hash lookup, active token, unexpired, active owner. The
before_request shim and the collector API (which does not decode JWT) both
call this so the checks live in one place.
"""
from shopdb.core.models import User
token = ApiToken.query.filter_by(
tokenhash=ApiToken.hash_secret(secret), isactive=True).first()
if token is None or token.is_expired:
return None
user = db.session.get(User, token.userid)
if user is None or not user.isactive:
return None
return token, user
def touch_apitoken_lastused(token):
"""Throttled lastusedat write. Independent commit; nothing else is pending
this early in the request, so it cannot clobber route work."""
now = _utcnow()
if token.lastusedat is None \
or (now - token.lastusedat).total_seconds() > _LASTUSED_THROTTLE_SECONDS:
token.lastusedat = now
db.session.commit()
def authorized_service_token(scope):
"""Return the ApiToken authorizing this request for `scope`, or None.
Same checks as service_token_authorized (managed token scoped for `scope`,
active owner holding the permission), but hands back the token itself so a
caller can read its resource binding (token.resourcescopelist) without
reaching into core token internals. Touches lastusedat on success.
"""
from shopdb.core.models import User
api_key = request.headers.get('X-API-Key')
token = None
if api_key and api_key.startswith(TOKEN_SECRET_PREFIX):
resolved = resolve_api_token(api_key)
token = resolved[0] if resolved else None
else:
tokenid = getattr(g, 'apitokenid', None)
if tokenid is not None:
token = db.session.get(ApiToken, tokenid)
if token is None:
return None
scopelist = token.scopelist
if not scopelist or scope not in scopelist:
return None
user = db.session.get(User, token.userid)
if user is None or not user.isactive or not user.haspermission(scope):
return None
touch_apitoken_lastused(token)
return token
def mint_service_token(scope, name, owner_userid, revoke_existing=True):
"""Mint a managed service token scoped to exactly ONE permission.
The contract-surface way for a plugin to provision an unattended service
credential (an endpoint agent key, a collector key) without importing core
token internals. The returned secret is the ONLY time the full secret is
available - callers reveal it once and store only the prefix afterwards.
scope the single permission the token authorizes (e.g.
collector.ingest.endpoints). The owner must hold it.
name human label shown in the token list UI.
owner_userid the acting admin who owns (and can be audited for) the key.
revoke_existing when True (default), every currently-active token scoped to
EXACTLY this one permission is deactivated first, so
regenerating is a hard cutover: the old agent key stops
working the instant the new one is minted.
Returns (token, secret). Commits.
"""
from shopdb.core.models import User
owner = db.session.get(User, owner_userid)
if owner is None or not owner.isactive or not owner.haspermission(scope):
raise PermissionError(
'Token owner must be an active user holding ' + scope)
if revoke_existing:
for existing in list_service_tokens(scope):
existing.isactive = False
secret = ApiToken.generate_secret()
token = ApiToken(
userid=owner_userid,
name=name,
tokenprefix=ApiToken.prefix_of(secret),
tokenhash=ApiToken.hash_secret(secret),
)
token.scopelist = [scope]
db.session.add(token)
db.session.commit()
return token, secret
def list_service_tokens(scope, active_only=True):
"""Return managed tokens scoped to EXACTLY the single permission `scope`.
Used by a plugin settings page to render the current agent-key state
(prefix, created, last-used) without exposing token internals. Filters in
Python because scopes are stored as a JSON blob, not a queryable column.
"""
query = ApiToken.query
if active_only:
query = query.filter_by(isactive=True)
matches = []
for token in query.order_by(ApiToken.createddate.desc()).all():
scopelist = token.scopelist
if scopelist and len(scopelist) == 1 and scopelist[0] == scope:
matches.append(token)
return matches
def revoke_service_tokens(scope):
"""Deactivate every active managed token scoped to EXACTLY `scope`.
The kill switch: revoking the endpoint agent key stops all endpoint-agent
traffic authorized by a managed token. Returns the count revoked. Commits.
"""
revoked = 0
for token in list_service_tokens(scope):
token.isactive = False
revoked += 1
if revoked:
db.session.commit()
return revoked
def service_token_authorized(scope):
"""True when the current request carries a managed token scoped for `scope`
whose owner is active and holds that permission. Accepts X-API-Key or a
Bearer PAT (the before_request shim resolves Bearer into g.apitokenid).
Touches lastusedat on success.
The single contract-surface entry point for unattended SERVICE tokens
(collector.ingest, geenforce.fetch, ...), so plugins authorize a service
token without reaching into core token internals. Returns False on any
miss; the caller returns its own 401.
"""
return authorized_service_token(scope) is not None
def install_apitoken_auth(app):
"""Register the before_request PAT shim on the app."""
@app.before_request
def _apitoken_before_request():
secret = _extract_pat_secret()
if secret is None:
return
resolved = resolve_api_token(secret)
if resolved is None:
# The caller clearly meant to use a PAT (shopdb_pat_ prefix) but it
# is unknown, revoked, or expired. Reject with a clear 401 instead
# of letting the JWT decoder emit a confusing 422 on the non-JWT.
return error_response(
ErrorCodes.UNAUTHORIZED,
'Invalid, revoked, or expired API token',
http_code=401)
token, user = resolved
# Read claim inputs before the (possible) commit expires the instance.
claims = {
'username': user.username,
'roles': [role.rolename for role in user.roles],
}
# A scoped token carries a patscopes claim; authz reads it to grant ONLY
# the listed permissions and to deny role gates + import mode. Unscoped
# tokens carry no such claim and mint exactly as a login JWT would.
scopelist = token.scopelist
if scopelist is not None:
claims['patscopes'] = scopelist
# Expose the token/user for audit and introspection if a handler wants it.
g.apitokenid = token.tokenid
g.apitokenuser = user
touch_apitoken_lastused(token)
# Mint a request-scoped JWT for the owner and swap it into the header
# so the whole downstream auth stack authenticates as that user.
access_token = create_access_token(
identity=str(user.userid), additional_claims=claims)
request.environ['HTTP_AUTHORIZATION'] = f'Bearer {access_token}'