T5 Attribution across hops¶
Description¶
What a node can tell about the path of a request, and how a request can be followed afterwards. Uses the saved calls of T1 and a refreshed token of T4, in the same session.
From the LS Login team, 2026-11-03: act and can_act aren't supported, and are under consideration for the new stack (C1). What's left in the token is sub, the user, and aud, the next service. At best, client_id names the last caller.
Runs against: both. On the mock, the comparison follows from our mock/oidc.json. The hop limit and traceparent are the testbed's own, and work the same on both.
Expectations¶
- Provenance: direct and through the orchestrator, the node sees the same claims:
identical_at_last_hop: true, unlessclient_iddiffers, andact_present: [false, false]. So whether the orchestrator was involved can't be read from the token. - Accounting: usage can't be attributed to a portal or workflow system from the token alone.
- Loops: a gateway can't find itself in the chain, so it can't refuse by identity. A hop limit of 0 gives 508 with
refused: hop limit. With 1, the gateway forwards, and the next service sees a limit of 0. - Tracing: all rows of a call have the same
traceparent, and the logs of the services it passed show it.
Hop limit and traceparent come from proTES #391. Neither is signed, so any hop can reset them. They stop accidental loops and help with tracing. They prove nothing.
Steps¶
Provenance and accounting¶
Hop limit¶
A new token, as the one from T1 may have expired, in which case the services answer 401 before they get to the hop limit.
tb> refresh r3 --as client --from r2
tb> exchange to-gw5 --as client --from r3 --audience @gateway
tb> call gateway --token to-gw5 --hop-limit 0
tb> call gateway --token to-gw5 --hop-limit 1
Tracing¶
tb> call gateway --token to-gw5 --traceparent 00-1bf7651916cd43dd8448eb211c80319c-c7ad6b7169203331-01
Then look for it in the services' logs:
Revoking consent, last¶
Optional, and only after every other step of T1-T5, in both sessions, including T4's binding and reuse: it ends the refresh tokens of both, as both log in account A with tx-test-client (Can users see and revoke what a service holds?). In the profile of account A, revoke the consent of the client. Then, in the main session, refresh from its newest refresh token, r3 from Hop limit, right away, and again 5 minutes later, so that a delayed revocation shows:
If rv1 worked and showed refresh_token_rotated: true, use --from rv1 for the second one. In the expiry session, refresh once, too, from its newest refresh token (exp-after, or exp2 if you logged in again):
Questions it answers¶
Is act coming?¶
The LS Login team says that it's under consideration (2026-11-03, C1). What the standard describes is nested act (RFC 8693, section 4.1), set by the identity provider.
- Yes: the chain comes from nested
actclaims, and a gateway finding itself in the chain refuses the request. FOCA can readactas optional from the start, as that changes nothing today. - No: provenance and accounting stay in logs, with
traceparent.
What to record¶
| Step | Record | Result |
|---|---|---|
| Provenance | identical_at_last_hop, act_present, and what differs |
|
| Accounting | client_id at the node on both paths |
|
| Hop limit | Status for 0 and 1 | |
| Tracing | Whether the logs of the gateway and the node both show this traceparent |
|
| Revoking consent | Whether rv1, rv2 and rv-exp were refused, with errors, and the time between revoking and each |
|
act |
The LS Login team's answer |