Conversation
Resolve wasm guest modules from image manifests and tighten wasi-stdio mode handling. Use stdio envelope framing in the wasm function-bridge runner and keep wasm runner env minimal. Stabilize interactive PTY behavior with newline normalization and TTY attach output-mode fixes. Refactor exec lifecycle cleanup helpers and add PTY line-ending regression tests.
Enable wasm-node to connect to the existing qemu-network backend via a net socket bridge in the WASI runner, including custom socket/poll shims for stdin + network interop. Allow wasm-node VFS mounts to proceed by materializing mount placeholders in-guest and seeding the MITM CA certificate so HTTPS mediation works without -k. tcp-forward channels remain capability-gated.
Restore explicit capability gating for wasm-node VFS mounts and remove placeholder mount-path materialization that made /workspace appear mounted while empty. Keep wasm-node behavior aligned with backend capabilities and update docs/tests to reflect the current parity state.
Stabilize stdio transport framing between sandboxd and the wasm runner, including robust envelope parsing and malformed-frame recovery under PTY load. Route wasm-node through the same FS RPC path as qemu/krun, tighten VFS bind resolution, and update startup flow so readiness and bind materialization are deterministic across backends. Add/refresh wasm-node parity and bridge tests covering memfs, hostfs, and custom provider round trips.
Document the intentionally unsupported wasm-node capabilities and include wasm-node in backend parity coverage where parity is expected. Skip tcp-forward, SSH, and ingress parity checks via explicit capability-based unsupported reasons and add wasm runtime preflight helpers for the parity suite.
Add a wasm-node profiling script plus lightweight per-process summaries and traces for the host bridge, runner bridge, and wasm runtime. Instrument the transport, bridge, and wasm runtime paths so boot, exec, PTY latency, throughput, framing, and runtime I/O costs can be measured directly while investigating wasm-node performance.
Always use the custom stdin bridge and poll_oneoff path so wasm-node exercises the same runtime I/O flow whether guest networking is enabled or not. Also avoid the old fixed 20ms idle wait by only retry-waiting for unresolved stdin or network subscriptions and reducing the wait to 1ms, which brings the real networking-enabled path much closer to the previous fast path.
Add wasm target-arch configuration and patch the container2wasm Dockerfile during build so wasm images get CAP_SYS_ADMIN, /dev/fuse, and FUSE-enabled riscv kernel configs. This makes gondolin.ts build produce wasm-node images that boot with sandboxfs/VFS mounts, preserve MITM cert injection, and carry baked environment variables as validated end-to-end.
Anchor relative clock subscriptions once per poll call and wait until the earliest deadline expires. This restores guest sleep/poll behavior in the wasm runner so idle shells stop busy-spinning the CPU, and adds regression coverage for relative clock waits.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
This PR adds a third
vmmbackend calledwasm-nodewhich runs a guest within WASM and executed by node's WASM support.This does not fork Gondolin's host control plane. Up until this point the host talked to
qemu/krunover Unix-socket virtio channels. As this is not an option for WASM, that transport layer is now abstracted behindServerTransport, andwasm-nodeswaps the Unix socket transport for a function based bridge while keeping the same framed protocol and mostly the same host-side server logic.Concretely:
qemu/krunboot a real Linux guest, and Gondolin talks to it over separate control/fs/ssh/ingress channels.vmm=wasm-nodestarts a Node child (WasmFunctionBridgeRunner/WasmFunctionBridgeController) instead of a VM process.The real wasm path works like this:
FunctionBridgeTransportchannels.wasm-runner.ts, which instantiates the wasm module with Node's WASI support.sandboxdnow supports--transport=stdio, so it can speak the same CBOR/framed protocol over stdin/stdout instead of a virtio socket.Currently stdout/stdin framing is wrapped in a line-safe base64 envelope so control traffic does not get mixed up with regular process stdio. That is a smell I want to remove.
Additionally the network support reuses the existing host stack: the wasm runner synthesizes socket-like WASI FDs backed by the existing host network backend, so
httpHooks, DNS policy, TLS MITM, etc. still go through the same enforcement path.Current intentional differences vs
main/ real VM backends:openTcpStream,enableSsh(),enableIngress())For builds, this branch also adds a path for custom images to emit
sandbox.wasm(via container2wasm), record it inmanifest.assets.wasm, and letvmm=wasm-noderesolve it automatically.