feat(verifier,as): gate the verifier and rvps for wasm32 by imlk0 · Pull Request #207 · openanolis/trustee

imlk0 · 2026-07-30T02:47:54Z

Make the verifier (notably TDX) and RVPS build and run under wasm32-unknown-unknown, so attestation-service can be embedded as a pure library/wasm module. The verifier's env/file reads, blocking per-GPU spawns, and Send-bound async traits all assume a native multi-threaded host; this PR makes each wasm-viable without forking the code path, using cfg_attr/cfg gates and injectable overrides so native behavior is unchanged.

Key changes worth reviewing

deps/verifier/src/tdx/mod.rs: new set_pccs_url(Option<String>) override (RwLock) takes precedence over the PCCS_URL env var / QCNL config-file resolution; those env + filesystem fallbacks are gated behind non-wasm targets, since wasm32 has no environment variable or filesystem access. Exported via verifier::tdx so an embedder can configure the PCCS endpoint directly.
deps/verifier/src/tdx/gpu/rim.rs: the forced wasm-skip of GPU RIM verification is replaced with injectable overrides — set_rim_service_url (wasm now fails closed instead of silently trying to fetch an unreachable default URL) and set_skip_gpu_verify (wasm now defaults to verify, not skip). The per-GPU tokio::spawn loop is swapped for futures::future::join_all, so a single code path works for both native (multi-threaded) and wasm (single-threaded) targets. This mirrors the TDX case: the NVIDIA RIM service URL is also normally resolved from an env var, which is unavailable on wasm, hence the same setter-based override pattern.
deps/verifier/src/tdx/verify/native.rs: reqwest request timeout is gated out for wasm (the wasm HTTP backend doesn't support this option).
deps/verifier + attestation-service traits: Verifier and RvpsApi (and their impls) use async_trait(?Send) on wasm32 via cfg_attr on the target triple, across the verifier/sample/sample_device/tdx and rvps modules.
Why: on wasm32, reqwest's HTTP backend delegates to the browser's fetch API through wasm-bindgen FFI bindings, which operate on JS types (Promise/JsValue) tied to the single-threaded JS runtime. JsValue is not Send, so any future that awaits an HTTP call transitively becomes !Send. Since async-trait normally boxes futures as Box<dyn Future + Send>, we need ?Send on wasm to drop that bound — the trait's async-ness itself doesn't change, only the auto-trait requirement.
getrandom/uuid: add the getrandom js feature for wasm32 in attestation-service and deps/verifier (uuid v4 generation and getrandom both need a JS/wasm-bindgen-backed RNG source in the browser, since there's no OS entropy source to call into). Mirrors attestation-service's existing native/wasm uuid split in deps/verifier: native uses ["v4"], wasm adds js on top.
web-time: switch std::time to web_time under wasm32 in verifier's sgx/tdx/tpm and rvps reference_value, plus the corresponding workspace dependency wiring.
Why: Rust's std SystemTime::now() / Instant::now() are unimplemented on wasm32-unknown-unknown and panic at runtime — there's no wall-clock/monotonic-clock syscall to bind to without going through JS. web-time is a drop-in replacement that reads time via the JS Date/Performance APIs instead.
(The challenge/mod.rs web-time hunk is split out into a follow-up PR that lands after the challenge module exists in the challenger PR.)

Open question for discussion

Both the TDX (PCCS_URL) and NVIDIA GPU RIM service URL originally come from environment variables (TDX additionally falls back to a QCNL config file on disk). Since wasm32 has no env/fs access, this PR adds a setter per config item (set_pccs_url, set_rim_service_url, set_skip_gpu_verify) as the wasm-compatible substitute.

This works, but doesn't scale cleanly if more verifiers need the same treatment — each one growing its own ad-hoc setter + RwLock. An alternative worth discussing: thread this configuration down from attestation-service's config (file/struct) into the verifier in a unified way, rather than having each verifier module manage its own override state. Happy to discuss the tradeoffs before this pattern spreads further.

Add a `set_pccs_url(Option<String>)` override (RwLock) that takes precedence over `PCCS_URL` env / QCNL-config-file resolution, and gate those env / filesystem fallbacks behind non-wasm targets since wasm has no env or filesystem. Export `set_pccs_url` via `verifier::tdx` so an embedder (binary or wasm host) can configure the PCCS endpoint directly. Signed-off-by: Kun Lai <laikun@linux.alibaba.com>

Replace the forced wasm skip of GPU RIM verification with injectable overrides mirroring `set_pccs_url` -- `set_rim_service_url` (wasm fails closed instead of fetching an unreachable default) and `set_skip_gpu_verify` (wasm defaults to verify, not skip). Swap the per-GPU `tokio::spawn` loop for `futures::future::join_all` so one code path works on both native and single-threaded wasm. Signed-off-by: Kun Lai <laikun@linux.alibaba.com>

reqwest's `.timeout()` is unavailable on wasm32-unknown-unknown, so split each reqwest client builder so the timeout call is applied only on non-wasm targets, across the three client constructions in gpu/rim.rs and the PCCS client in verify/native.rs. Signed-off-by: Kun Lai <laikun@linux.alibaba.com>

On wasm32-unknown-unknown futures default to !Send, so the `async_trait` expansions of the Verifier and RvpsApi traits (and their impls) must use `?Send`. Gate via `cfg_attr` on the wasm target triple across verifier, sample, sample_device, tdx, and rvps modules. Signed-off-by: Kun Lai <laikun@linux.alibaba.com>

The uuid v4 RNG and getrandom need a wasm backend (js-sys/wasm-bindgen) under wasm32-unknown-unknown. Add `getrandom` js in attestation-service and deps/verifier, and mirror attestation-service's target-specific uuid split in deps/verifier (native `["v4"]`, wasm adds `js`). Signed-off-by: Kun Lai <laikun@linux.alibaba.com>

Part 1 of splitting 284a091: the non-challenge web-time changes — switch std::time to web_time under wasm32 in verifier sgx/tdx/tpm and rvps reference_value, and add the web-time workspace/dep wiring. The challenge/mod.rs hunk is split into a follow-up commit so it can land in the challenger PR after the challenge module exists. Signed-off-by: Kun Lai <laikun@linux.alibaba.com>

ostest-bot · 2026-07-30T02:47:58Z

@imlk0 ，您好，您的请求已接收，请耐心等待结果。

ostest-bot · 2026-07-30T02:48:01Z

@imlk0 ，您好，未检测到有镜像需要构建，如需重新检测请评论 /start 。

ostest-bot · 2026-07-30T02:59:45Z

@imlk0 ，您好，您的请求已接收，请耐心等待结果。

ostest-bot · 2026-07-30T02:59:47Z

@imlk0 ，您好，未检测到有镜像需要构建，如需重新检测请评论 /start 。

CI's `cargo fmt --check` wanted the long `#[cfg_attr(not(all(...)), async_trait)]` one-liners broken across lines, and `cargo clippy` flagged `clone_on_copy` on the `Option<bool>` skip-gpu-verify override read. Apply rustfmt to the wasm cfg_attr lines and deref the RwLockReadGuard instead of cloning. Signed-off-by: Kun Lai <laikun@linux.alibaba.com>

ostest-bot · 2026-07-30T03:06:04Z

@imlk0 ，您好，您的请求已接收，请耐心等待结果。

ostest-bot · 2026-07-30T03:06:06Z

@imlk0 ，您好，未检测到有镜像需要构建，如需重新检测请评论 /start 。

jialez0 · 2026-07-30T03:34:55Z

-                        std::result::Result::Ok(gpu_evidence_claims),
-                    )) => {
-                        gpu_claims.insert(format!("nvidia_gpu.{}", index), gpu_evidence_claims);
+            // Evaluate each GPU concurrently without `tokio::spawn`: the RIM


这种方式，和原本的tokio::spawn有多少性能差距可以评估下吗？或者依然采用feature开关分流

jialez0

现在CI没有覆盖wsam模式下的编译和测试，最好添加一下

Restore `tokio::spawn` for the per-GPU evidence evaluation loop on native targets so the I/O-bound RIM fetches keep multi-core parallelism, and keep `futures::future::join_all` only on the single-threaded wasm target where `tokio::spawn` is unavailable. The per-GPU task body is unchanged; only the dispatch is differentiated via `cfg` on the wasm32 target triple. Addresses review feedback that switching to `join_all` uniformly lost multi-core parallelism on native. Signed-off-by: Kun Lai <laikun@linux.alibaba.com>

Add a `cargo check -p verifier --target wasm32-unknown-unknown --no-default-features` step to the CoCo-AS Check job. The verifier's `tdx` module is unconditionally compiled, so this exercises the wasm cfg gating. The attestation-service lib wasm check is intentionally excluded -- it does not compile until later PRs in this series. Addresses review feedback that CI did not cover wasm compilation. Signed-off-by: Kun Lai <laikun@linux.alibaba.com>

ostest-bot · 2026-07-30T07:34:27Z

@imlk0 ，您好，您的请求已接收，请耐心等待结果。

ostest-bot · 2026-07-30T07:34:30Z

@imlk0 ，您好，未检测到有镜像需要构建，如需重新检测请评论 /start 。

The inline `Wasm compile check (verifier)` step added to the `as-check` job failed on CI's pinned rustc 1.76: the wasm build pulls `js-sys`/`web-sys` 0.3.103 (already pinned in the lockfile on main), which require rustc >= 1.77. The wasm port inherently needs these (the getrandom `js` backend for RNG on wasm32). Move the check out of `as-check` so the native fmt/clippy/test steps on 1.76 run unblocked, into a dedicated `wasm-check` job that installs the `stable` toolchain (satisfies the js-sys MSRV), adds the wasm32 target, installs protoc, and runs `cargo check -p verifier --target wasm32-unknown-unknown --no-default-features`. Signed-off-by: Kun Lai <laikun@linux.alibaba.com>

ostest-bot · 2026-07-30T11:20:18Z

@imlk0 ，您好，您的请求已接收，请耐心等待结果。

ostest-bot · 2026-07-30T11:20:21Z

@imlk0 ，您好，未检测到有镜像需要构建，如需重新检测请评论 /start 。

jialez0 · 2026-07-31T02:22:29Z

几个小 comments：

WASM CI 没有覆盖TDX/GPU相关的verifier代码
建议固定一个明确的WASM模式rust版本
pccs_override_wins 当前没有#[test] 测试覆盖

pccs_override_wins was a bare fn with no #[test] attribute, so it never ran. Add #[test] and #[serial_test::serial] (serial_test is already a workspace dev-dep) since it mutates the shared global PCCS override RwLock; the serial attribute stops it racing other tests that touch the same global. The trailing assert_ne confirms the override is cleared so it cannot leak. Signed-off-by: Kun Lai <laikun@linux.alibaba.com>

Add a dedicated wasm-check job (runs on PR/push/create) that compiles the wasm-viable crates for wasm32-unknown-unknown. The main CI is pinned to 1.76 and never built wasm before; the pure-lib port adds wasm targets, so a dedicated job is needed. Pin the toolchain to a reproducible nightly (nightly-2025-07-07) and invoke each command with RUSTUP_TOOLCHAIN set to it. A separate job is required because js-sys/web-sys 0.3.103 (pre-existing on main's Cargo.lock, pulled by getrandom's js backend) need rustc >=1.77, while the native CI is 1.76. The check covers: - eventlog (default features) - reference-value-provider-service (--no-default-features; no 'fs' feature in wasm) - verifier (--no-default-features --features tdx-dcap-rust) so the full TDX + GPU verifier (PCCS override, collateral fetch, GPU RIM verify, the join_all per-GPU evaluation loop) compiles for wasm32; --no-default-features keeps the wasm-incompatible openssl/tss-esapi backends off. The attestation-service wasm check is left commented out: the only OPA policy engine still hard-depends on tokio::fs, which is unavailable on wasm32. It will be re-enabled once that dependency is lifted. Signed-off-by: Kun Lai <laikun@linux.alibaba.com>

ostest-bot · 2026-07-31T09:38:11Z

@imlk0 ，您好，您的请求已接收，请耐心等待结果。

ostest-bot · 2026-07-31T09:38:13Z

@imlk0 ，您好，未检测到有镜像需要构建，如需重新检测请评论 /start 。

imlk0 added 6 commits July 30, 2026 10:29

jialez0 reviewed Jul 30, 2026

View reviewed changes

imlk0 added 2 commits July 30, 2026 15:33

imlk0 added 2 commits July 31, 2026 11:42

Uh oh!

Conversation

imlk0 commented Jul 30, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Key changes worth reviewing

Open question for discussion

Uh oh!

ostest-bot commented Jul 30, 2026

Uh oh!

ostest-bot commented Jul 30, 2026

Uh oh!

ostest-bot commented Jul 30, 2026

Uh oh!

ostest-bot commented Jul 30, 2026

Uh oh!

ostest-bot commented Jul 30, 2026

Uh oh!

ostest-bot commented Jul 30, 2026

Uh oh!

jialez0 Jul 30, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

jialez0 left a comment

Choose a reason for hiding this comment

Uh oh!

ostest-bot commented Jul 30, 2026

Uh oh!

ostest-bot commented Jul 30, 2026

Uh oh!

ostest-bot commented Jul 30, 2026

Uh oh!

ostest-bot commented Jul 30, 2026

Uh oh!

jialez0 commented Jul 31, 2026

Uh oh!

ostest-bot commented Jul 31, 2026

Uh oh!

ostest-bot commented Jul 31, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

imlk0 commented Jul 30, 2026 •

edited

Loading

jialez0 Jul 30, 2026 •

edited

Loading