nvx-rs: parse fake OpenVMM --mount-child values with one function

[Pedro Henrique Penna (ppenna)]

Summary

The fake OpenVMM (nvx::testing::fake_openvmm) parsed --mount-child NAME,HOST,MODE[,file] values with two copies of the same parser: one in sandbox_shares, for the children of a sandbox's aggregate at /run/nvx/shares, and one in mapped_children, for the children of the host mapping table's export. This change keeps that parser once, in the private function mount_child_host, so that the fake's two option paths accept the same syntax by construction. The production --mount-child producers and OpenVMM itself are untouched.

Changes

All in nvx-rs/src/testing/fake_openvmm.rs:

• mount_child_host(child, index) returns the host path of a --mount-child value, or fails with invalid --mount-child VALUE unless the host path is a directory or, with the file flag, a regular file, the mode is ro or rw, and, if index is given, the name is that index.
• The sites that now share it:
  • sandbox_shares calls it without an index, for the children of a sandbox's aggregate, whose names are digests;
  • mapped_children calls it with each child's index, for the numbered children of the host mapping table.
• The unit test mount_children_resolve_to_hosts_of_their_kind covers the accepted directory and file children, a child without an index, and each refusal: another name, a file without the flag, a directory with it, a missing host path, another mode, a missing mode, and a value with only a name.

Behavior

Nothing observable changes:

• Every refusal keeps its message, invalid --mount-child VALUE, and every other message of the fake is untouched.
• The parser takes the same steps in the same order: it splits off the name, strips the file flag, splits off the mode, probes the host path with is_file or is_dir, and then checks the name, the kind, and the mode, as the condition of mapped_children did.
• mapped_children still requires the names 0, 1, ... in order, and sandbox_shares still accepts any name.
• The fake is test-only, behind the testing feature. No public API, ErrorKind, protocol byte, golden fixture, guest binary, or nvx-rs CLI changes; the CLI build (--features cli) doesn't compile this module.

Existing tests drive both callers: the CLI golden lifecycle scenarios lifecycle: several shares and lifecycle: file share, in both the POSIX and Windows sections of tests/fixtures/cli.json, run the fake with directory and file children of a sandbox aggregate, and aci_edge_sandboxes/tests/openvmm_fake.rs maps directories and files through the host mapping table.

Follow-ups

The survey turned these up, and this change leaves them alone.

• Suspected bug: sandbox_shares accepts --mount-aggregate /run/nvx/shares without any --mount-child, and repeated child names, while OpenVMM's MicrovmFilesystemConfig::new_aggregate refuses both (InvalidChildCount and DuplicateChildName), and mapped_children refuses an aggregate without children. The fake's sandbox path is thus more lenient than OpenVMM. Making it strict changes what the fake accepts, so it wants its own change.
• In host/session.rs, features, map_host_paths, and next_exec_event repeat the request-ID check with the same message, map_host_paths and next_exec_event truncate an error category to 64 characters the same way, and ping and stop differ only in their request and reply kinds and their message.
• In host/sandbox/mount.rs, MountAccess, MountOwner, and MountKind repeat from_name with the same message form, unsupported sandbox mount WHAT 'NAME'; choose A or B, and host/sandbox/mount/tests.rs and host/sandbox/tests.rs define the same policy test helper.
• These follow-ups of earlier runs remain:
  • Instance::runtime and Instance::launch in host/instance.rs;
  • the option-values test helper of host/filesystem.rs, host/network/mod.rs, and host/sandbox/options/tests.rs;
  • the fixture of a Python process whose main thread exits first, in host/process.rs and host/cli/sandbox/identity.rs;
  • the managed-agent test fixtures, the entry convention of the nine guest programs, the failure-reporting closures of console_pending.rs and mmio_write.rs, run_file and run_network in device_io.rs, the device_io helper in tests/guest_tools.rs, and the time tools' watcher-violation sites and port-window lock failures.

Validation

• BASE: 2fc63e7465decee7a7764e937472d0f7416d9c82, origin/dev when the run started.
• TREE: b976a1f3140a207ae31bcd82ddb439e72150c20a, the tree of this commit, applied as a patch on BASE on each host.
• Release: v0.1.0-dev.ff95973ba9da, which nvx.py download installed on each host as the nearest released first-parent ancestor of BASE, with OpenVMM d0bdbed8f6e3b86271277dc22658109e2e633b2d, the pinned revision.
• Guest artifacts: prometheus30 rebuilt the guest binaries from TREE with build-guest-binaries, ran test-guest-binaries on them, and built the Alpine initramfs from them with NVX_GUEST_BINARIES. All three hosts ran the same initramfs.cpio.gz (SHA-256 b5832af0e8e9985ed4d0e684253be3ec03956f960b36c133ac26daa2e900598b), package manifest, and provenance. Each host also rebuilt the nvx-rs CLI from TREE and recorded its provenance before its hypervisor tests.
• Local checks, on Windows: cargo fmt --check; Clippy with -D warnings for all targets and features and for the library without default features; cargo test with all features and with the defaults; and rustdoc with -D warnings. This machine also ran the steps for which the hosts lack a toolchain or target: the Rust 1.89 MSRV check of both crates on Windows, and the aarch64-apple-darwin check of both crates, which CI runs on Linux.

Host profile	Backend	Host type	check-nvx-rs	check-aci-edge-sandboxes	Guest binaries and initramfs	test-aci-edge-sandboxes	test-microvm (default suite)
prometheus32	kvm	baremetal	Linux steps pass, including MSRV 1.89, the bundled-release test, and check-guest-binaries	Linux steps pass, including MSRV 1.89 and agent bundling	—	pass, 18 of 18	pass
prometheus30	mshv	baremetal	—	—	build-guest-binaries, test-guest-binaries, and build-initramfs pass	pass, 18 of 18	pass
prometheus28	whp	baremetal	Windows steps pass, including the bundled-release test; MSRV ran locally	Windows steps pass, including agent bundling; MSRV ran locally	—	pass, 18 of 18	pass

directories_on_another_volume_are_mapped passes by skipping on hosts without a second volume, as on all three.

Left to CI: the test-microvm scenarios that run only when named (smp-lapic, managed-exec-config, and sandbox-lifecycle), the Ubuntu and Azure Linux guests, the OpenVMM suites, guest determinism, benchmarks, and check-quality and validate-nvx, which this change doesn't affect: it changes no Python, shell, or Markdown file, and adds, moves, or deletes none.

[Copilot]

🟢 Approval recommended

The focused refactor preserves existing behavior, with regression coverage and no identified blocking issues.

No open issues 🎉

What changed in this PR

Consolidates duplicate --mount-child parsing in the test-only fake OpenVMM while preserving both callers’ validation and error messages.

Changes:

• Extracts mount_child_host and uses it for sandbox shares and mapped children.
• Adds regression coverage for valid inputs and validation failures.

File	Description
nvx-rs/​src/​testing/​fake_openvmm.rs	Shares mount-child parsing between callers and adds focused tests.

_{Review effort: Balanced}