fix(interceptors): harden the §9.4 slice — seal, fail-closed, re-entrancy, shared coverage

An end-to-end audit of the (unmerged) interceptor slice surfaced seven ways a
KV allow/deny guard could be silently defeated or misbehave. Close all of them:

1. Shared-collection bypass — `kv::shared_collection(...).set/delete` skipped
   the hook entirely, so any `(kv, set/delete)` guard was circumvented by
   choosing the shared handle. `GroupKvHandle` now runs the same before-op hook.

2. Seal to the declaring owner — the marker resolved nearest-owner-wins but its
   script name was then re-resolved on the CALLER's chain, so a descendant app
   could shadow a group's mandatory guard with a same-named local script.
   `resolve_before` now returns the script sealed to the owner that declared the
   marker (one LEFT JOIN in manager-core), so the executor never re-resolves by
   name. A descendant can only override with its OWN explicit marker.

3. Re-entrancy — an "allow + audit-log" guard that itself wrote KV re-triggered
   itself to the depth cap and then DENIED the original write (plus ~8x
   resolve/compile/execute amplification). A thread-local guard makes a write
   performed by an interceptor bypass interception.

4. Fail-closed verdict — only `#{ allowed: false }` denied; a bare bool, a
   typo'd key, a non-bool, or a bare unit all ALLOWED. Now allow ONLY on an
   explicit `#{ allowed: true }`; every other shape denies.

5/6. Fail-closed edges — a dangling (missing/disabled) interceptor script and a
   missing engine back-reference now DENY instead of allowing.

7. AppInvoke coupling — resolution no longer routes through `invoke.resolve`, so
   installing a guard no longer denies writes to principals who hold KV-write
   but not AppInvoke.

The seam (`InterceptorService::resolve_before`) now returns an
`InterceptorResolution { None | Dangling | Run(ResolvedScript) }`; the executor
runs the sealed script straight through the shared `run_resolved_blocking` core
and drops its `InvokeService` dependency. executor-core stays Postgres-free.
No migration change (0073 unchanged).

Pinned by three new journeys in tests/interceptors.rs: the seal (a same-named
app script does NOT shadow the group's guard), shared-collection coverage, and
the fail-closed verdict. Full journey suite 157/157.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
This commit is contained in:
MechaCat02
2026-07-13 21:20:20 +02:00
parent 2fc9476f9e
commit d08df88df5
7 changed files with 504 additions and 102 deletions

View File

@@ -7,8 +7,10 @@
//! functions over `&PgPool` / `&mut Transaction`, keyed by [`ScriptOwner`]),
//! plus the runtime [`resolve_before`] chain walk (nearest-owner-wins).
use chrono::{DateTime, Utc};
use picloud_shared::{AppId, ScriptOwner};
use sqlx::{PgPool, Postgres, Transaction};
use uuid::Uuid;
use crate::config_resolver::CHAIN_LEVELS_CTE;
@@ -84,29 +86,95 @@ pub async fn list_on_app_chain(
.collect())
}
/// Resolve the interceptor script guarding `(service, op)` for a calling app —
/// the nearest declaration on the app's chain (app, then nearest ancestor
/// group). `None` when un-hooked. **This chain walk is the resolution boundary**
/// — a sibling-subtree app never sees another subtree's interceptor.
/// The nearest interceptor marker on an app's chain, with its script resolved
/// **at the declaring owner** (the seal). `script` is `None` when the marker
/// names a script that is missing or disabled at that owner — a dangling hook
/// the caller must fail closed on.
#[derive(Debug, Clone)]
pub struct SealedInterceptor {
pub marker_app: Option<Uuid>,
pub marker_group: Option<Uuid>,
pub script_name: String,
/// `(script_id, source, updated_at)` of the sealed script, or `None` if it
/// is missing/disabled at the declaring owner.
pub script: Option<(Uuid, String, DateTime<Utc>)>,
}
impl SealedInterceptor {
/// The owner that DECLARED the marker (and therefore owns the resolved
/// script) — the lexical origin for the interceptor's own `import`s (§5.5).
#[must_use]
pub fn sealing_owner(&self) -> ScriptOwner {
if let Some(g) = self.marker_group {
ScriptOwner::Group(g.into())
} else {
ScriptOwner::App(
self.marker_app
.expect("marker XOR: app when not group")
.into(),
)
}
}
}
/// Resolve the interceptor guarding `(service, op)` for a calling app — the
/// nearest MARKER on the app's chain (app, then nearest ancestor group), with
/// its script **sealed to the owner that declared the marker** (a descendant
/// app cannot shadow it with a same-named local script). `None` when un-hooked.
/// **This chain walk is the resolution boundary** — a sibling-subtree app never
/// sees another subtree's interceptor. The script join is `LEFT` so a marker
/// whose script is missing/disabled still returns a row (with `script: None`)
/// rather than silently falling through to a farther, weaker marker.
pub async fn resolve_before(
pool: &PgPool,
app_id: AppId,
service: &str,
op: &str,
) -> Result<Option<String>, sqlx::Error> {
let row: Option<(String,)> = sqlx::query_as(&format!(
"{CHAIN_LEVELS_CTE} \
SELECT i.interceptor_script FROM chain c \
JOIN interceptors i ON (i.app_id = c.app_owner OR i.group_id = c.group_owner) \
WHERE i.service = $2 AND i.op = $3 \
ORDER BY c.depth ASC LIMIT 1",
) -> Result<Option<SealedInterceptor>, sqlx::Error> {
#[allow(clippy::type_complexity)]
let row: Option<(
Option<Uuid>,
Option<Uuid>,
String,
Option<Uuid>,
Option<String>,
Option<DateTime<Utc>>,
)> = sqlx::query_as(&format!(
"{CHAIN_LEVELS_CTE}, \
nearest AS ( \
SELECT i.interceptor_script AS script_name, \
i.app_id AS marker_app, i.group_id AS marker_group, c.depth AS depth \
FROM chain c \
JOIN interceptors i ON (i.app_id = c.app_owner OR i.group_id = c.group_owner) \
WHERE i.service = $2 AND i.op = $3 \
ORDER BY c.depth ASC LIMIT 1 \
) \
SELECT n.marker_app, n.marker_group, n.script_name, \
s.id, s.source, s.updated_at \
FROM nearest n \
LEFT JOIN scripts s ON ( \
(n.marker_app IS NOT NULL AND s.app_id = n.marker_app \
AND s.name = n.script_name AND s.enabled) \
OR (n.marker_group IS NOT NULL AND s.group_id = n.marker_group \
AND s.name = n.script_name AND s.enabled) \
)",
))
.bind(app_id.into_inner())
.bind(service)
.bind(op)
.fetch_optional(pool)
.await?;
Ok(row.map(|(s,)| s))
Ok(row.map(
|(marker_app, marker_group, script_name, sid, src, upd)| SealedInterceptor {
marker_app,
marker_group,
script_name,
script: match (sid, src, upd) {
(Some(id), Some(source), Some(updated_at)) => Some((id, source, updated_at)),
_ => None,
},
},
))
}
/// Upsert a marker at `owner` in the apply transaction. A re-apply that changes

View File

@@ -1,12 +1,14 @@
//! `InterceptorServiceImpl` — the Postgres-backed §9.4 interceptor resolver
//! injected into `Services`. Resolve-only: it maps `(cx.app_id, service, op)`
//! to the nearest interceptor script name on the calling app's chain (via
//! [`crate::interceptor_repo::resolve_before`]). Running that script is the
//! executor's job (the `invoke()` re-entry path), which keeps `executor-core`
//! Postgres-free.
//! injected into `Services`. It maps `(cx.app_id, service, op)` to the nearest
//! interceptor marker on the calling app's chain and materializes its script
//! **sealed to the declaring owner** (via [`crate::interceptor_repo`]). Running
//! that script is the executor's job (the `invoke()` re-entry core), which keeps
//! `executor-core` Postgres-free.
use async_trait::async_trait;
use picloud_shared::{InterceptorService, SdkCallCx};
use picloud_shared::{
InterceptorResolution, InterceptorService, ResolvedScript, ScriptId, SdkCallCx,
};
use sqlx::PgPool;
pub struct InterceptorServiceImpl {
@@ -27,11 +29,31 @@ impl InterceptorService for InterceptorServiceImpl {
cx: &SdkCallCx,
service: &str,
op: &str,
) -> Result<Option<String>, String> {
) -> Result<InterceptorResolution, String> {
// `app_id` derives from `cx` (never a script arg) — the isolation
// boundary; the chain walk then scopes resolution to this app's subtree.
crate::interceptor_repo::resolve_before(&self.pool, cx.app_id, service, op)
let sealed = crate::interceptor_repo::resolve_before(&self.pool, cx.app_id, service, op)
.await
.map_err(|e| e.to_string())
.map_err(|e| e.to_string())?;
let Some(sealed) = sealed else {
return Ok(InterceptorResolution::None);
};
// A marker exists but its sealed script is missing/disabled → dangling,
// the executor fails closed (denies) rather than silently allowing.
let Some((sid, source, updated_at)) = sealed.script.clone() else {
return Ok(InterceptorResolution::Dangling(sealed.script_name));
};
// The execution boundary is always the CALLER's app; the `owner` is the
// sealing (declaring) owner, the lexical origin for the interceptor's
// own `import`s (§5.5) — a group marker's script runs its imports from
// the group.
Ok(InterceptorResolution::Run(Box::new(ResolvedScript {
script_id: ScriptId::from(sid),
app_id: cx.app_id,
owner: Some(sealed.sealing_owner()),
source,
updated_at,
name: sealed.script_name,
})))
}
}