[DRAFT] Support new IGVMAgent contract with Skip HW Unsealing signal by mingweishih · Pull Request #2921 · microsoft/openvmm

mingweishih · 2026-03-10T06:44:40Z

No description provided.

Signed-off-by: Ming-Wei Shih <mishih@microsoft.com>

This reverts commit 7bdcc4d.

Signed-off-by: Ming-Wei Shih <mishih@microsoft.com>

github-actions · 2026-03-10T06:44:58Z

⚠️ Unsafe Code Detected

This PR modifies files containing unsafe Rust code. Extra scrutiny is required during review.

For more on why we check whole files, instead of just diffs, check out the Rustonomicon

Copilot

Pull request overview

This PR adds support for the skip_hw_unsealing signal in the new IGVMAgent contract. When the IGVM agent fails a key release request with the skip_hw_unsealing bit set in the response, OpenHCL should skip the hardware unsealing fallback path entirely instead of attempting it. This is important for security scenarios where the agent explicitly signals that hardware unsealing should not be attempted.

Changes:

Adds the skip_hw_unsealing signal to the IGVM attestation protocol (IgvmCapabilityBitMap and IgvmSignal bitfields in get.rs)
Implements handling in underhill_attestation to detect and propagate the skip_hw_unsealing signal through the key release flow, blocking hardware unsealing fallback and logging a DEK_HARDWARE_UNSEALING_SKIPPED event
Adds per-VM test agent infrastructure in test_igvm_agent_rpc_server, replacing the shared singleton with a HashMap-keyed registry, plus new integration and unit tests for the new contract

Reviewed changes

Copilot reviewed 11 out of 11 changed files in this pull request and generated 2 comments.

Show a summary per file

File	Description
`openhcl/openhcl_attestation_protocol/src/igvm_attest/get.rs`	Adds `skip_hw_unsealing` bit to `IgvmCapabilityBitMap` and `IgvmSignal` protocol structs
`openhcl/underhill_attestation/src/igvm_attest/mod.rs`	Advertises the new capability bit in requests; parses and propagates `skip_hw_unsealing` from responses; adds unit tests
`openhcl/underhill_attestation/src/secure_key_release.rs`	Logs and propagates the `skip_hw_unsealing` flag from key release error responses
`openhcl/underhill_attestation/src/lib.rs`	Extracts `skip_hw_unsealing` signal and blocks hardware unsealing fallback when set; adds unit test `init_sec_secure_key_release_skip_hw_unsealing`
`vm/devices/get/get_protocol/src/lib.rs`	Adds `DEK_HARDWARE_UNSEALING_SKIPPED = 23` to `EventLogId` enum
`vm/devices/get/get_resources/src/lib.rs`	Adds `KeyReleaseFailureSkipHwUnsealing` and `KeyReleaseFailure` to `IgvmAttestTestConfig` enum
`vm/devices/get/test_igvm_agent_lib/src/lib.rs`	Adds `RespondFailureSkipHwUnsealing` action and corresponding plans for new test configs
`vm/devices/get/test_igvm_agent_rpc_server/src/rpc/igvm_agent.rs`	Replaces singleton agent with per-VM registry keyed by VM name; adds `config_for_vm_name` pattern matching
`vm/devices/get/test_igvm_agent_rpc_server/src/rpc/handlers.rs`	Passes VM name to `process_igvm_attest` for per-VM routing
`vm/devices/get/test_igvm_agent_rpc_server/src/main.rs`	Adds new `KeyReleaseFailureSkipHwUnsealing` and `KeyReleaseFailure` CLI test config variants
`vmm_tests/vmm_tests/tests/tests/multiarch/tpm.rs`	Adds four new integration tests: `ak_cert_cache`, `ak_cert_retry`, `skip_hw_unseal`, and `use_hw_unseal`

vmm_tests/vmm_tests/tests/tests/multiarch/tpm.rs

vm/devices/get/test_igvm_agent_rpc_server/src/rpc/igvm_agent.rs

Signed-off-by: Ming-Wei Shih <mishih@microsoft.com>

github-actions · 2026-03-10T08:16:35Z

At least one Petri test failed.

Signed-off-by: Ming-Wei Shih <mishih@microsoft.com>

github-actions · 2026-03-10T19:54:51Z

At least one Petri test failed.

Signed-off-by: Ming-Wei Shih <mishih@microsoft.com>

github-actions · 2026-03-10T22:44:59Z

At least one Petri test failed.

Signed-off-by: Ming-Wei Shih <mishih@microsoft.com>

github-actions · 2026-03-11T02:55:31Z

At least one Petri test failed.

Signed-off-by: Ming-Wei Shih <mishih@microsoft.com>

Copilot

Pull request overview

Copilot reviewed 12 out of 12 changed files in this pull request and generated 5 comments.

Copilot · 2026-03-11T20:03:35Z

vmm_tests/vmm_tests/tests/tests/multiarch/tpm.rs

+    tracing::info!("first boot");
+
+    agent.reboot().await?;
+
+    tracing::info!("second boot");
+
+    let host_binary_path = tpm_guest_tests_artifact.get();
+    let tpm_guest_tests =
+        TpmGuestTests::send_tpm_guest_tests(&agent, host_binary_path, guest_binary_path, os_flavor)
+            .await?;


After calling agent.reboot().await?, the current PipetteClient instance is no longer guaranteed to be usable (it only sends the reboot request; it does not wait for reset/reconnect). The subsequent send_tpm_guest_tests and AK cert reads are likely to fail against a stale connection. Update this flow to wait for the VM reset and obtain a fresh agent (e.g., make vm/agent mutable and call agent = vm.wait_for_reset().await? after reboot) before continuing.

Copilot · 2026-03-11T20:03:36Z

vm/devices/get/test_igvm_agent_rpc_server/src/rpc/igvm_agent.rs

+///
+/// The substring must be short enough to survive Hyper-V's 100-char name
+/// limit even on the longest image prefixes (~85 chars).  Keep patterns
+/// ≤ 15 characters.
+///


config_for_vm_name documents that patterns must be short (≤ 15 chars) to survive Hyper-V VM name truncation, but the current matching strategy uses long substrings that include the image name and other prefix components. Because petri::vm::make_vm_safe_name truncates long VM names to 100 chars, these long patterns are likely to be truncated away and never match, causing the wrong/default config to be applied. Prefer matching only on the (short) test function name portion (e.g., "ak_cert_retry", "ak_cert_cache", "skip_hw_unseal", "use_hw_unseal") or another guaranteed-to-survive unique suffix.

Copilot · 2026-03-11T20:03:36Z

vm/devices/get/test_igvm_agent_rpc_server/src/rpc/igvm_agent.rs

+            "ubuntu_2504_server_x64_ak_cert_retry",
+            IgvmAttestTestConfig::AkCertRequestFailureAndRetry,
+        ),
+        (
+            "windows_datacenter_core_2022_x64_ak_cert_retry",
+            IgvmAttestTestConfig::AkCertRequestFailureAndRetry,
+        ),
+        (
+            "ubuntu_2504_server_x64_vbs_ak_cert_retry",
+            IgvmAttestTestConfig::AkCertRequestFailureAndRetry,
+        ),
+        (
+            "windows_datacenter_core_2025_x64_prepped_vbs_ak_cert_retry",
+            IgvmAttestTestConfig::AkCertRequestFailureAndRetry,
+        ),
+        (
+            "ubuntu_2504_server_x64_snp_ak_cert_retry",
+            IgvmAttestTestConfig::AkCertRequestFailureAndRetry,
+        ),
+        (
+            "windows_datacenter_core_2025_x64_prepped_snp_ak_cert_retry",


Example of the truncation risk: these KNOWN_TEST_CONFIGS patterns are far longer than the ≤15-character guideline stated above, so they are unlikely to be present in Hyper-V’s truncated (100-char) VM names. Shorten these patterns to something that is guaranteed to appear near the end of the VM name (typically just the test fn name suffix).

Suggested change

"ubuntu_2504_server_x64_ak_cert_retry",

IgvmAttestTestConfig::AkCertRequestFailureAndRetry,

),

(

"windows_datacenter_core_2022_x64_ak_cert_retry",

IgvmAttestTestConfig::AkCertRequestFailureAndRetry,

),

(

"ubuntu_2504_server_x64_vbs_ak_cert_retry",

IgvmAttestTestConfig::AkCertRequestFailureAndRetry,

),

(

"windows_datacenter_core_2025_x64_prepped_vbs_ak_cert_retry",

IgvmAttestTestConfig::AkCertRequestFailureAndRetry,

),

(

"ubuntu_2504_server_x64_snp_ak_cert_retry",

IgvmAttestTestConfig::AkCertRequestFailureAndRetry,

),

(

"windows_datacenter_core_2025_x64_prepped_snp_ak_cert_retry",

"ak_cert_retry",

IgvmAttestTestConfig::AkCertRequestFailureAndRetry,

),

(

"ak_cert_retry",

IgvmAttestTestConfig::AkCertRequestFailureAndRetry,

),

(

"ak_cert_retry",

IgvmAttestTestConfig::AkCertRequestFailureAndRetry,

),

(

"ak_cert_retry",

IgvmAttestTestConfig::AkCertRequestFailureAndRetry,

),

(

"ak_cert_retry",

IgvmAttestTestConfig::AkCertRequestFailureAndRetry,

),

(

"ak_cert_retry",

Copilot · 2026-03-11T20:03:36Z

vm/devices/get/test_igvm_agent_lib/src/lib.rs

+        } else if self.plan.is_some() {
+            // A plan is installed but has no more actions for this request type.
+            // Return NoResponse to avoid silently falling back to normal mode.
+            tracing::info!(?request.header.request_type, "Plan exhausted: NoResponse");
+            (vec![], 0)


The new else if self.plan.is_some() branch makes the agent return NoResponse for any request type that is not explicitly present in the plan (or whose queue is empty). This breaks existing tests that install a plan for only one request type and rely on default behavior for others (e.g., plans that only override WRAPPED_KEY_REQUEST or only KEY_RELEASE_REQUEST). Adjust the logic to distinguish between "no entry for this request type" (should fall back to default behavior) vs "entry exists but queue exhausted" (could reasonably become NoResponse).

Copilot · 2026-03-11T20:03:36Z

vmm_tests/vmm_tests/tests/tests/multiarch/tpm.rs

+            }
+        }
+        Ok(other) => {
+            anyhow::bail!("Expected Reset or VM start failure, got {other:?}");


In the Ok(other) branch, the test bails out without tearing down the VM. Since PetriVm does not have a Drop teardown, an unexpected halt reason could leak a VM/resource in the test environment. Consider tearing down the VM before returning the error (or restructuring to ensure teardown happens on all branches).

Suggested change

anyhow::bail!("Expected Reset or VM start failure, got {other:?}");

let error = anyhow::anyhow!(

"Expected Reset or VM start failure, got {other:?}"

);

vm.teardown().await?;

return Err(error);

Signed-off-by: Ming-Wei Shih <mishih@microsoft.com>

github-actions · 2026-03-11T21:21:05Z

At least one Petri test failed.

github-actions · 2026-03-11T22:38:33Z

At least one Petri test failed.

mingweishih added 5 commits March 7, 2026 00:05

init implementation

a8c01ba

Signed-off-by: Ming-Wei Shih <mishih@microsoft.com>

vmm tests

2520c65

Signed-off-by: Ming-Wei Shih <mishih@microsoft.com>

support statless + hw sealing

7bdcc4d

Signed-off-by: Ming-Wei Shih <mishih@microsoft.com>

Revert "support statless + hw sealing"

a7b3430

This reverts commit 7bdcc4d.

update

2f11da2

Signed-off-by: Ming-Wei Shih <mishih@microsoft.com>

mingweishih requested a review from a team as a code owner March 10, 2026 06:44

Copilot AI review requested due to automatic review settings March 10, 2026 06:44

github-actions bot added the unsafe Related to unsafe code label Mar 10, 2026

Copilot started reviewing on behalf of mingweishih March 10, 2026 06:45 View session

Copilot AI reviewed Mar 10, 2026

View reviewed changes

vmm_tests/vmm_tests/tests/tests/multiarch/tpm.rs Outdated Show resolved Hide resolved

vm/devices/get/test_igvm_agent_rpc_server/src/rpc/igvm_agent.rs Outdated Show resolved Hide resolved

x

1c7c745

Signed-off-by: Ming-Wei Shih <mishih@microsoft.com>

x

f0db87d

Signed-off-by: Ming-Wei Shih <mishih@microsoft.com>

debug

22ad3e3

Signed-off-by: Ming-Wei Shih <mishih@microsoft.com>

mingweishih added 2 commits March 10, 2026 23:56

test

f55d7cf

Signed-off-by: Ming-Wei Shih <mishih@microsoft.com>

x

97b8119

Signed-off-by: Ming-Wei Shih <mishih@microsoft.com>

mingweishih added 2 commits March 11, 2026 18:34

x

4461cfb

Signed-off-by: Ming-Wei Shih <mishih@microsoft.com>

update

ba08178

Signed-off-by: Ming-Wei Shih <mishih@microsoft.com>

Copilot AI review requested due to automatic review settings March 11, 2026 19:58

Copilot started reviewing on behalf of mingweishih March 11, 2026 19:59 View session

Copilot AI reviewed Mar 11, 2026

View reviewed changes

x

d6a1926

Signed-off-by: Ming-Wei Shih <mishih@microsoft.com>

-            anyhow::bail!("Expected Reset or VM start failure, got {other:?}");
+            let error = anyhow::anyhow!(
+                "Expected Reset or VM start failure, got {other:?}"
+            );
+            vm.teardown().await?;
+            return Err(error);

Conversation

mingweishih commented Mar 10, 2026

Uh oh!

github-actions bot commented Mar 10, 2026

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Uh oh!

Uh oh!

github-actions bot commented Mar 10, 2026

Uh oh!

github-actions bot commented Mar 10, 2026

Uh oh!

github-actions bot commented Mar 10, 2026

Uh oh!

github-actions bot commented Mar 11, 2026

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Copilot AI Mar 11, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Mar 11, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Mar 11, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Mar 11, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Mar 11, 2026

Choose a reason for hiding this comment

Uh oh!

github-actions bot commented Mar 11, 2026

Uh oh!

github-actions bot commented Mar 11, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants