Relax layout requirements for MultiUseSandbox::restore

[ludfjig]

This makes it possible to restore snapshots from different guest binaries and memory layouts into a sandbox, provided the registered host functions are compatible.

closes #1592

[Copilot]

Pull request overview

This PR relaxes MultiUseSandbox::restore compatibility rules so snapshots can be restored across different guest binaries and sandbox memory layouts, as long as the target sandbox’s registered host functions satisfy the snapshot’s requirements.

Changes:

• Remove layout-compatibility validation from snapshot restore, and drop the SnapshotLayoutMismatch error path.
• Update restore to fully reset VM mappings during restore and clear crashdump state that would otherwise refer to the pre-restore guest.
• Add/expand tests covering cross-guest restore (WIT/Rust/C) and cross-layout restore, including persisted snapshots with non-default layouts.

Reviewed changes

Copilot reviewed 8 out of 8 changed files in this pull request and generated 1 comment.

Show a summary per file

File	Description
src/hyperlight_host/tests/wit_test.rs	Adds integration tests restoring WIT snapshots over Rust/C guests and vice-versa.
src/hyperlight_host/src/sandbox/snapshot/mod.rs	Removes Snapshot::validate_compatibility (layout + host funcs) helper.
src/hyperlight_host/src/sandbox/snapshot/file_tests.rs	Adds persisted non-default layout test; updates restore semantics comment.
src/hyperlight_host/src/sandbox/initialized_multi_use.rs	Implements relaxed restore (host-func-only validation) and remaps/unmaps regions during restore; updates docs/tests.
src/hyperlight_host/src/mem/layout.rs	Removes SandboxMemoryLayout::is_compatible_with and its tests.
src/hyperlight_host/src/hypervisor/hyperlight_vm/mod.rs	Adds clear_crashdump_binary_path for post-restore crashdump correctness.
src/hyperlight_host/src/error.rs	Removes HyperlightError::SnapshotLayoutMismatch variant.
CHANGELOG.md	Notes the new restore flexibility.

Suppressed comments (1)

src/hyperlight_host/src/sandbox/initialized_multi_use.rs:469

• restore now unmaps all VM regions returned by self.vm.get_mapped_regions() before applying the snapshot. The rustdoc for restore does not mention that mappings created via map_region / map_file_cow will be removed, which is a user-visible behavior change and can surprise callers relying on those mappings persisting across restores.

    /// signature). Extras on the sandbox are allowed. The registry
    /// itself is left unchanged. A mismatch returns
    /// [`SnapshotHostFunctionMismatch`](crate::HyperlightError::SnapshotHostFunctionMismatch)
    /// carrying the missing names and signature differences.
    ///

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[jprendes]

LGTM

The merge-base changed after approval.

[syntactically]

Does this also need to make the sandbox unrecoverable if it fails? Or is it intended that this just re-poisoning the sandbox and letting the user retry is fine? I would worry a little that an error here is likely to be about the health of the system as a whole, and I feel like this encourages a retry loop while sandbox.status().is_poisoned() { sandbox.restore(...); } which might end up spinning.

[syntactically]

Until we properly add the root finder to the snapshot, I think it is defensible to leave it in here and say "the user should write a root finder which works with any guest that they load", since (a) most users probably only use 1 or 2 different "base" guests and (b) the root finder has the opportunity to inspect the guest state and dispatch on what guest it thinks it is.

[ludfjig]

agree