Enforce exact recovery ownership before cold start

crates/lazaret-core/src/records/lifecycle.rs

@@ -1033,6 +1033,7 @@ pub struct HelperOwnership {
 pub struct SandboxOwnership {
     pub container_id: Option<ContainerId>,
     pub runtime_root: Option<HostPath>,
+    pub runtime_root_identity: Option<MountIdentity>,
     pub bundle_digest: Option<BundleDigest>,
     pub host_coordinator: PersistedHostProcessOwnership,
     pub runtime_supervisor: PersistedHostProcessOwnership,
@@ -1041,8 +1042,12 @@ pub struct SandboxOwnership {
 
 #[derive(Clone, Debug, Deserialize, Eq, PartialEq, Serialize)]
 pub struct PublicationOwnership {
+    pub staged_socket_path: Option<HostPath>,
+    pub staged_identity_path: Option<HostPath>,
     pub socket_path: Option<HostPath>,
     pub identity_path: Option<HostPath>,
+    pub socket_identity: Option<MountIdentity>,
+    pub identity_file_identity: Option<MountIdentity>,
 }
 
 impl PreparedLowerOwnership {
@@ -1309,6 +1314,7 @@ impl SandboxOwnership {
         Self {
             container_id: None,
             runtime_root: None,
+            runtime_root_identity: None,
             bundle_digest: None,
             host_coordinator: PersistedHostProcessOwnership::unassigned(),
             runtime_supervisor: PersistedHostProcessOwnership::unassigned(),
@@ -1319,6 +1325,7 @@ impl SandboxOwnership {
     fn is_empty(&self) -> bool {
         self.container_id.is_none()
             && self.runtime_root.is_none()
+            && self.runtime_root_identity.is_none()
             && self.bundle_digest.is_none()
             && self.host_coordinator.is_unassigned()
             && self.runtime_supervisor.is_unassigned()
@@ -1328,6 +1335,7 @@ impl SandboxOwnership {
     fn is_committed(&self) -> bool {
         self.container_id.is_some()
             && self.runtime_root.is_some()
+            && self.runtime_root_identity.is_some()
             && self.bundle_digest.is_some()
             && self.host_coordinator.is_current_committed()
             && self.runtime_supervisor.is_current_committed()
@@ -1337,6 +1345,10 @@ impl SandboxOwnership {
     fn validate_progress_to(&self, next: &Self) -> Result<(), RecordTransitionError> {
         validate_option_progress(self.container_id.as_ref(), next.container_id.as_ref())?;
         validate_option_progress(self.runtime_root.as_ref(), next.runtime_root.as_ref())?;
+        validate_option_progress(
+            self.runtime_root_identity.as_ref(),
+            next.runtime_root_identity.as_ref(),
+        )?;
         validate_option_progress(self.bundle_digest.as_ref(), next.bundle_digest.as_ref())?;
         validate_existing_process_progress(&self.host_coordinator, &next.host_coordinator)?;
         validate_process_progress(&self.runtime_supervisor, &next.runtime_supervisor)?;
@@ -1347,22 +1359,49 @@ impl SandboxOwnership {
 impl PublicationOwnership {
     fn empty() -> Self {
         Self {
+            staged_socket_path: None,
+            staged_identity_path: None,
             socket_path: None,
             identity_path: None,
+            socket_identity: None,
+            identity_file_identity: None,
         }
     }
 
     fn is_empty(&self) -> bool {
-        self.socket_path.is_none() && self.identity_path.is_none()
+        self.staged_socket_path.is_none()
+            && self.staged_identity_path.is_none()
+            && self.socket_path.is_none()
+            && self.identity_path.is_none()
+            && self.socket_identity.is_none()
+            && self.identity_file_identity.is_none()
     }
 
     fn is_committed(&self) -> bool {
-        self.socket_path.is_some() && self.identity_path.is_some()
+        self.staged_socket_path.is_some()
+            && self.staged_identity_path.is_some()
+            && self.socket_path.is_some()
+            && self.identity_path.is_some()
+            && self.socket_identity.is_some()
+            && self.identity_file_identity.is_some()
     }
 
     fn validate_progress_to(&self, next: &Self) -> Result<(), RecordTransitionError> {
+        validate_option_progress(
+            self.staged_socket_path.as_ref(),
+            next.staged_socket_path.as_ref(),
+        )?;
+        validate_option_progress(
+            self.staged_identity_path.as_ref(),
+            next.staged_identity_path.as_ref(),
+        )?;
         validate_option_progress(self.socket_path.as_ref(), next.socket_path.as_ref())?;
-        validate_option_progress(self.identity_path.as_ref(), next.identity_path.as_ref())
+        validate_option_progress(self.identity_path.as_ref(), next.identity_path.as_ref())?;
+        validate_option_progress(self.socket_identity.as_ref(), next.socket_identity.as_ref())?;
+        validate_option_progress(
+            self.identity_file_identity.as_ref(),
+            next.identity_file_identity.as_ref(),
+        )
     }
 }
 

crates/lazaret-core/tests/schema_records.rs

@@ -457,14 +457,19 @@ fn complete_ownership(role: ParticipantRole, state: ParticipantState) -> Partici
         ParticipantRole::Sandbox => ParticipantOwnership::Sandbox(SandboxOwnership {
             container_id: Some(ContainerId::new(opaque(29))),
             runtime_root: Some(HostPath::new(b"runtime".to_vec())),
+            runtime_root_identity: Some(mount()),
             bundle_digest: Some(BundleDigest::new(opaque(43))),
             host_coordinator: owned_process(30),
             runtime_supervisor: owned_process(21),
             attached_runsc: owned_process(44),
         }),
         ParticipantRole::Publication => ParticipantOwnership::Publication(PublicationOwnership {
+            staged_socket_path: Some(HostPath::new(b"staged-socket".to_vec())),
+            staged_identity_path: Some(HostPath::new(b"staged-identity".to_vec())),
             socket_path: Some(HostPath::new(b"socket".to_vec())),
             identity_path: Some(HostPath::new(b"identity".to_vec())),
+            socket_identity: Some(mount()),
+            identity_file_identity: Some(mount()),
         }),
     }
 }
@@ -558,6 +563,7 @@ fn participant_edge_snapshot(
             ParticipantOwnership::Sandbox(SandboxOwnership {
                 container_id: Some(ContainerId::new(opaque(29))),
                 runtime_root: None,
+                runtime_root_identity: None,
                 bundle_digest: None,
                 host_coordinator: process_ownership,
                 runtime_supervisor: PersistedHostProcessOwnership::unassigned(),
@@ -569,6 +575,7 @@ fn participant_edge_snapshot(
             ParticipantOwnership::Sandbox(SandboxOwnership {
                 container_id: Some(ContainerId::new(opaque(29))),
                 runtime_root: None,
+                runtime_root_identity: None,
                 bundle_digest: None,
                 host_coordinator: owned_process(122),
                 runtime_supervisor: process_ownership,
@@ -580,6 +587,7 @@ fn participant_edge_snapshot(
             ParticipantOwnership::Sandbox(SandboxOwnership {
                 container_id: Some(ContainerId::new(opaque(29))),
                 runtime_root: None,
+                runtime_root_identity: None,
                 bundle_digest: None,
                 host_coordinator: owned_process(122),
                 runtime_supervisor: owned_process(123),

crates/lazaret-core/tests/schema_records/encoding.rs

@@ -375,6 +375,11 @@ fn every_schema_shape_round_trips_and_rejects_malformed_or_unknown_versions() {
         SandboxOwnership {
             container_id: Some(ContainerId::new(opaque(29))),
             runtime_root: Some(HostPath::new(b"runtime".to_vec())),
+            runtime_root_identity: Some(MountIdentity {
+                mount_id: MountId::new(68),
+                device: DeviceId::new(69),
+                inode: InodeId::new(70),
+            }),
             bundle_digest: Some(BundleDigest::new(opaque(66))),
             host_coordinator: owned_process(30),
             runtime_supervisor: owned_process(21),
@@ -384,8 +389,20 @@ fn every_schema_shape_round_trips_and_rejects_malformed_or_unknown_versions() {
     );
     verify_shape(
         PublicationOwnership {
+            staged_socket_path: Some(HostPath::new(b"staged-socket".to_vec())),
+            staged_identity_path: Some(HostPath::new(b"staged-identity".to_vec())),
             socket_path: Some(HostPath::new(b"socket".to_vec())),
             identity_path: Some(HostPath::new(b"identity".to_vec())),
+            socket_identity: Some(MountIdentity {
+                mount_id: MountId::new(71),
+                device: DeviceId::new(72),
+                inode: InodeId::new(73),
+            }),
+            identity_file_identity: Some(MountIdentity {
+                mount_id: MountId::new(74),
+                device: DeviceId::new(75),
+                inode: InodeId::new(76),
+            }),
         },
         &mut seen,
     );

crates/lazaret-host-coordinator/src/bootstrap.rs

@@ -7,7 +7,10 @@ use lazaret_core::configuration::ResolvedDomainConfiguration;
 use lazaret_core::identity::{BundleDigest, ContainerId, MountIdentity};
 use lazaret_core::network::{CanonicalDerivedNetworkPolicy, ReachabilityExpectation};
 use lazaret_core::records::Session;
-use lazaret_core::records::{Generation, GenerationIdentity, PreparedArtifact};
+use lazaret_core::records::{
+    Generation, GenerationIdentity, Participant, ParticipantOwnership, ParticipantRole,
+    ParticipantState, PreparedArtifact,
+};
 use lazaret_linux::process::current_process_identity;
 use lazaret_runtime_supervisor::network::NetworkProbeRequest;
 use lazaret_runtime_supervisor::sandbox::SandboxBundle;
@@ -20,7 +23,7 @@ use crate::control::{
     ControlError, NonInteractiveSessionExecutor, PublishedControlFace, SessionRegistry,
     StagedControlFace, TrustAnchor,
 };
-use crate::domain_store::HostCoordinatorDomainStore;
+use crate::domain_store::{DomainStoreMutation, HostCoordinatorDomainStore};
 use crate::lifecycle::{
     HostCoordinatorLifecycle, LifecycleAction, LifecycleAdapter, SessionAdmissionIdentity,
     SessionLifecycleAdapter,
@@ -240,6 +243,10 @@ impl<'a> ProductionLifecycleAdapter<'a> {
                 self.plan.generation_identity.generation_id,
                 self.plan.container_id,
                 &self.plan.runtime_root,
+                self.plan
+                    .runtime_staging
+                    .runsc_root_identity()
+                    .map_err(failure)?,
                 self.plan.bundle_digest,
                 current_process_identity().map_err(failure)?,
             )
@@ -347,8 +354,14 @@ impl<'a> ProductionLifecycleAdapter<'a> {
             .trust_anchor
             .take()
             .ok_or_else(|| ProductionLifecycleError("Trust Anchor is absent".into()))?;
-        self.staged_control = Some(
+        let generation_id = self.plan.generation_identity.generation_id;
-            StagedControlFace::stage_with_executor(
+        self.plan
+            .store
+            .apply(&[DomainStoreMutation::PutParticipant(
+                Participant::bare_intent(generation_id, ParticipantRole::Publication),
+            )])
+            .map_err(failure)?;
+        let staged = StagedControlFace::stage_with_executor(
             path_from_host(
                 self.plan
                     .configuration
@@ -361,22 +374,85 @@ impl<'a> ProductionLifecycleAdapter<'a> {
             Arc::clone(&self.registry),
             Arc::clone(&self.plan.non_interactive_executor),
         )
+        .map_err(failure)?;
+        let ownership = staged.recovery_ownership().map_err(failure)?;
+        self.plan
+            .store
+            .apply(&[DomainStoreMutation::PutParticipant(
+                Participant::snapshot(
+                    generation_id,
+                    ParticipantRole::Publication,
+                    ParticipantState::Intent,
+                    Some(ParticipantOwnership::Publication(ownership)),
+                )
                 .map_err(failure)?,
-        );
+            )])
+            .map_err(failure)?;
+        self.staged_control = Some(staged);
         Ok(())
     }
 
     fn expose_publication(&mut self) -> Result<(), ProductionLifecycleError> {
-        self.published_control = Some(Arc::new(
+        let published = self
-            self.staged_control
+            .staged_control
             .take()
             .ok_or_else(|| ProductionLifecycleError("staged control face is absent".into()))?
             .expose()
+            .map_err(failure)?;
+        let ownership = published.recovery_ownership().map_err(failure)?;
+        self.plan
+            .store
+            .apply(&[DomainStoreMutation::PutParticipant(
+                Participant::snapshot(
+                    self.plan.generation_identity.generation_id,
+                    ParticipantRole::Publication,
+                    ParticipantState::Committed,
+                    Some(ParticipantOwnership::Publication(ownership)),
+                )
                 .map_err(failure)?,
-        ));
+            )])
+            .map_err(failure)?;
+        self.published_control = Some(Arc::new(published));
         Ok(())
     }
 
+    fn clean_publication_after_withdrawal(&self) -> Result<(), ProductionLifecycleError> {
+        let generation_id = self.plan.generation_identity.generation_id;
+        let current = self
+            .plan
+            .store
+            .snapshot()
+            .map_err(failure)?
+            .participants
+            .into_iter()
+            .find(|participant| {
+                participant.generation_id() == generation_id
+                    && participant.role() == ParticipantRole::Publication
+            })
+            .ok_or_else(|| ProductionLifecycleError("Publication participant is absent".into()))?;
+        let release = Participant::snapshot(
+            generation_id,
+            ParticipantRole::Publication,
+            ParticipantState::ReleaseIntent,
+            current.ownership().cloned(),
+        )
+        .map_err(failure)?;
+        let clean = Participant::snapshot(
+            generation_id,
+            ParticipantRole::Publication,
+            ParticipantState::Clean,
+            Some(ParticipantOwnership::empty(ParticipantRole::Publication)),
+        )
+        .map_err(failure)?;
+        self.plan
+            .store
+            .apply(&[
+                DomainStoreMutation::PutParticipant(release),
+                DomainStoreMutation::CleanParticipantAfterAbsence(clean),
+            ])
+            .map_err(failure)
+    }
+
     fn stop_sandbox(&mut self) -> Result<(), ProductionLifecycleError> {
         let store = Arc::clone(&self.plan.store);
         let sandbox = SandboxCoordinator::new(&store);
@@ -517,6 +593,7 @@ impl LifecycleAdapter for ProductionLifecycleAdapter<'_> {
             LifecycleAction::BeginTeardown => {
                 if let Some(control) = self.published_control.as_ref() {
                     control.withdraw().map_err(failure)?;
+                    self.clean_publication_after_withdrawal()?;
                 }
                 self.published_control.take();
                 Ok(())

crates/lazaret-host-coordinator/src/control.rs

@@ -13,7 +13,8 @@ use ed25519_dalek::SigningKey;
 use lazaret_core::codec::{CodecError, decode, encode};
 use lazaret_core::configuration::ConfinedPathPlan;
 use lazaret_core::identity::{
-    CertificateDigest, ClientId, ContainerId, SessionAnchorId, SessionId, TrustManifestDigest,
+    CertificateDigest, ClientId, ContainerId, DeviceId, HostPath, InodeId, MountIdentity,
+    SessionAnchorId, SessionId, TrustManifestDigest,
 };
 use lazaret_core::protocol::{
     ProtocolValidationError, issue_protocol_credential, issue_session_capability,
@@ -23,9 +24,10 @@ use lazaret_core::records::{
     ClientEnrollmentRequest, ClientEnrollmentResponse, ControlRequest, ControlResponse,
     DurableDecision, GenerationIdentity, NonInteractiveSessionMessage,
     NonInteractiveSessionRequest, ProtocolCredential, ProtocolCredentialRole,
-    ProtocolPublicationIdentity, ProtocolSubjectId, ProtocolVersion, SandboxCredentialBundle,
+    ProtocolPublicationIdentity, ProtocolSubjectId, ProtocolVersion, PublicationOwnership,
-    Session, SessionBinding, SessionExit, SessionPhase, SessionRegistrationRequest,
+    SandboxCredentialBundle, Session, SessionBinding, SessionExit, SessionPhase,
-    SessionRegistrationResponse, TlsCredential, TlsRole, TlsSubjectId, TrustManifest,
+    SessionRegistrationRequest, SessionRegistrationResponse, TlsCredential, TlsRole, TlsSubjectId,
+    TrustManifest,
 };
 use rcgen::{
     BasicConstraints, CertificateParams, CertificateSigningRequestParams, CertifiedIssuer,
@@ -46,6 +48,7 @@ use crate::domain_store::{DomainStoreError, DomainStoreMutation, HostCoordinator
 use crate::lifecycle::{
     HostCoordinatorLifecycle, LifecycleAction, SessionAdmissionIdentity, SessionLifecycleAdapter,
 };
+use lazaret_linux::filesystem::mount_identity;
 use lazaret_linux::process::{ProcessExit, TargetCommand, child_broker};
 use lazaret_runtime_supervisor::sandbox::RunscCommand;
 
@@ -53,10 +56,10 @@ pub const CONTROL_ALPN: &[u8] = b"lazaret-control/1";
 pub const CONTROL_SERVER_NAME: &str = "coordinator.lazaret.invalid";
 pub const MAX_ENROLLMENT_BYTES: usize = 8_192;
 const MAX_CONTROL_BYTES: usize = 65_536;
-const CONTROL_SOCKET_NAME: &str = "c.sock";
+pub(crate) const CONTROL_SOCKET_NAME: &str = "c.sock";
-const PUBLICATION_IDENTITY_NAME: &str = "protocol-identity.postcard";
+pub(crate) const PUBLICATION_IDENTITY_NAME: &str = "protocol-identity.postcard";
-const STAGED_CONTROL_SOCKET_NAME: &str = ".c.staged";
+pub(crate) const STAGED_CONTROL_SOCKET_NAME: &str = ".c.staged";
-const STAGED_PUBLICATION_IDENTITY_NAME: &str = ".protocol-identity.staged";
+pub(crate) const STAGED_PUBLICATION_IDENTITY_NAME: &str = ".protocol-identity.staged";
 
 #[derive(Debug)]
 pub enum ControlError {
@@ -1009,6 +1012,10 @@ pub struct StagedControlFace {
 }
 
 impl StagedControlFace {
+    pub(crate) fn recovery_ownership(&self) -> Result<PublicationOwnership, ControlError> {
+        publication_recovery_ownership(&self.paths)
+    }
+
     /// Stages an undiscoverable private control face before durable publication.
     ///
     /// # Errors
@@ -1160,7 +1167,51 @@ impl Drop for StagedControlFace {
     }
 }
 
+fn publication_recovery_ownership(
+    paths: &ControlPaths,
+) -> Result<PublicationOwnership, ControlError> {
+    let socket_path = if paths.staged_socket.exists() {
+        &paths.staged_socket
+    } else {
+        &paths.socket
+    };
+    let identity_path = if paths.staged_identity.exists() {
+        &paths.staged_identity
+    } else {
+        &paths.identity
+    };
+    Ok(PublicationOwnership {
+        staged_socket_path: Some(HostPath::new(
+            STAGED_CONTROL_SOCKET_NAME.as_bytes().to_vec(),
+        )),
+        staged_identity_path: Some(HostPath::new(
+            STAGED_PUBLICATION_IDENTITY_NAME.as_bytes().to_vec(),
+        )),
+        socket_path: Some(HostPath::new(CONTROL_SOCKET_NAME.as_bytes().to_vec())),
+        identity_path: Some(HostPath::new(PUBLICATION_IDENTITY_NAME.as_bytes().to_vec())),
+        socket_identity: Some(publication_path_identity(paths, socket_path)?),
+        identity_file_identity: Some(publication_path_identity(paths, identity_path)?),
+    })
+}
+
+fn publication_path_identity(
+    paths: &ControlPaths,
+    path: &Path,
+) -> Result<MountIdentity, ControlError> {
+    let metadata = std::fs::symlink_metadata(path)?;
+    let root = mount_identity(&paths.root_file)?;
+    Ok(MountIdentity {
+        mount_id: root.mount_id,
+        device: DeviceId::new(metadata.dev()),
+        inode: InodeId::new(metadata.ino()),
+    })
+}
+
 impl PublishedControlFace {
+    pub(crate) fn recovery_ownership(&self) -> Result<PublicationOwnership, ControlError> {
+        publication_recovery_ownership(&self.paths)
+    }
+
     /// Selects blocking or polling admission for the publication listener.
     ///
     /// # Errors

crates/lazaret-host-coordinator/src/domain_store/tests.rs

@@ -112,8 +112,12 @@ fn publication_partial() -> Participant {
         ParticipantRole::Publication,
         ParticipantState::Intent,
         Some(ParticipantOwnership::Publication(PublicationOwnership {
+            staged_socket_path: None,
+            staged_identity_path: None,
             socket_path: Some(HostPath::new(b"control.sock".to_vec())),
             identity_path: None,
+            socket_identity: None,
+            identity_file_identity: None,
         })),
     )
     .unwrap()
@@ -364,8 +368,12 @@ fn lifecycle_updates_enforce_transition_and_absence_boundaries() {
                 ParticipantRole::Publication,
                 ParticipantState::Clean,
                 Some(ParticipantOwnership::Publication(PublicationOwnership {
+                    staged_socket_path: None,
+                    staged_identity_path: None,
                     socket_path: None,
                     identity_path: None,
+                    socket_identity: None,
+                    identity_file_identity: None,
                 })),
             )
             .unwrap(),

crates/lazaret-host-coordinator/src/production.rs

@@ -10,12 +10,12 @@ use std::path::{Path, PathBuf};
 use std::sync::{Arc, Mutex};
 
 use lazaret_core::configuration::{ConfinedPathPlan, ResolvedDomainConfiguration};
-use lazaret_core::identity::{BundleDigest, ContainerId};
+use lazaret_core::identity::{BundleDigest, ContainerId, MountIdentity};
 use lazaret_core::network::{ReachabilityExpectation, derive_ipv4_policy};
 use lazaret_core::records::{
     Generation, GenerationIdentity, HostProcessOwnershipSlot, IpAddress, PreparedArtifact, Session,
 };
-use lazaret_linux::filesystem::namespace_identity;
+use lazaret_linux::filesystem::{mount_identity, namespace_identity};
 use lazaret_linux::process::ProcessOwnershipContext;
 use lazaret_runtime_supervisor::network::NetworkProbeRequest;
 use lazaret_runtime_supervisor::sandbox::{
@@ -371,6 +371,7 @@ fn require_inputs(inputs: &ProductionColdStartInputs<'_>) -> Result<(), Producti
 }
 
 pub(crate) struct RuntimeStaging {
+    runsc_root: OwnedFd,
     bundles: OwnedFd,
     bundle: OwnedFd,
     controls: OwnedFd,
@@ -384,6 +385,10 @@ pub(crate) struct RuntimeStaging {
 }
 
 impl RuntimeStaging {
+    pub(crate) fn runsc_root_identity(&self) -> Result<MountIdentity, ProductionColdStartError> {
+        mount_identity(&self.runsc_root).map_err(failure)
+    }
+
     pub(crate) fn cleanup_overlay(&self) -> Result<(), ProductionColdStartError> {
         let retained = fstat(&self.overlay).map_err(failure)?;
         if !self.overlay_identity.matches(&retained) {
@@ -418,12 +423,7 @@ impl RuntimeStaging {
             open_or_create_private_directory(&root, "sandbox-control", invoking_uid, false)?;
         let overlays =
             open_or_create_private_directory(&root, "runsc-overlay", invoking_uid, false)?;
-        drop(open_or_create_private_directory(
+        let runsc_root = open_or_create_private_directory(&root, "runsc", invoking_uid, false)?;
-            &root,
-            "runsc",
-            invoking_uid,
-            false,
-        )?);
         let name = OsString::from(hex(container_id.as_bytes()));
         let bundle = open_or_create_private_directory(&bundles, &name, invoking_uid, true)?;
         let bundle_identity = FileIdentity::capture(&bundle)?;
@@ -445,6 +445,7 @@ impl RuntimeStaging {
         };
         let overlay_identity = FileIdentity::capture(&overlay)?;
         let staging = Self {
+            runsc_root,
             bundles,
             bundle,
             controls,

crates/lazaret-host-coordinator/src/runtime.rs

@@ -107,6 +107,7 @@ impl<'store> SandboxCoordinator<'store> {
         generation_id: GenerationId,
         container_id: ContainerId,
         runtime_root: &Path,
+        runtime_root_identity: MountIdentity,
         bundle_digest: BundleDigest,
         host_coordinator: PidfsProcessIdentityV1,
     ) -> Result<(), SandboxCoordinatorError> {
@@ -123,6 +124,7 @@ impl<'store> SandboxCoordinator<'store> {
         let ownership = SandboxOwnership {
             container_id: Some(container_id),
             runtime_root: Some(HostPath::new(runtime_root.as_os_str().as_bytes().to_vec())),
+            runtime_root_identity: Some(runtime_root_identity),
             bundle_digest: Some(bundle_digest),
             host_coordinator: PersistedHostProcessOwnership::committed(host_coordinator),
             runtime_supervisor: PersistedHostProcessOwnership::unassigned(),

crates/lazaret-host-coordinator/src/runtime/tests.rs

@@ -6,7 +6,7 @@ use lazaret_core::identity::{
     PidfsDeviceId, PidfsInodeId, PidfsProcessAuthorityV1, ProcessStartTimeTicks, SandboxId,
 };
 use lazaret_core::records::{Domain, DurableDecision, Generation};
-use lazaret_linux::filesystem::try_exclusive_flock;
+use lazaret_linux::filesystem::{current_mount_identity, try_exclusive_flock};
 use tempfile::TempDir;
 
 use super::*;
@@ -91,6 +91,7 @@ fn sandbox_staging_readiness_and_release_preserve_exact_ownership() {
             generation().generation_id,
             ContainerId::new(bytes(6)),
             directory.path(),
+            current_mount_identity(directory.path()).unwrap(),
             BundleDigest::new(bytes(7)),
             process(100),
         )
@@ -155,6 +156,7 @@ fn mismatched_and_skipped_acquisition_boundaries_refuse_without_state_change() {
             generation().generation_id,
             ContainerId::new(bytes(6)),
             directory.path(),
+            current_mount_identity(directory.path()).unwrap(),
             BundleDigest::new(bytes(7)),
             process(100),
         )

crates/lazaret-host-coordinator/src/state_anchor.rs

@@ -12,7 +12,7 @@ use lazaret_core::records::{
     ProcessOwnershipCommitAck, StagedProcessOwnership, StateAnchorOwnership,
     StateAnchorReleaseEvidence,
 };
-use lazaret_linux::process::{PidfsRecoveryMatch, match_recovery_process};
+use lazaret_linux::process::RetainedProcess;
 use lazaret_state_anchor::{StateAnchorError, StateAnchorLayout, VerifiedStateAnchorReadiness};
 use lazaret_state_anchor::{observe_mount_absence, recover_private_state_anchor_mount};
 use lazaret_storage_catalog::{Catalog, CatalogError};
@@ -80,6 +80,12 @@ pub struct StateAnchorReleaseAuthority {
     mount_namespace: NamespaceIdentity,
 }
 
+#[derive(Clone, Copy)]
+pub(crate) enum StateAnchorRecoveryOwner<'owner> {
+    Matched(&'owner RetainedProcess),
+    Absent(&'owner RuntimeOwnersAbsent),
+}
+
 impl StateAnchorReleaseAuthority {
     #[must_use]
     pub const fn generation_id(&self) -> GenerationId {
@@ -380,33 +386,28 @@ impl<'a> StateAnchorCoordinator<'a> {
     /// Refuses generation drift, pidfs/namespace/mount mismatch, unreadable absence, or helper
     /// failure. Boot mismatch and `ESRCH` reconcile physical absence without process action only
     /// under the caller-provided ordered runtime-owner absence witness.
-    pub fn recover_after_runtime_owner_loss(
+    pub(crate) fn recover_after_runtime_owner_loss(
         &self,
         generation_id: GenerationId,
         layout: &StateAnchorLayout<'_>,
-        owners_absent: Option<&RuntimeOwnersAbsent>,
+        owner: StateAnchorRecoveryOwner<'_>,
     ) -> Result<
         (StateAnchorReleaseAuthority, StateAnchorAbsenceObservation),
         StateAnchorCoordinatorError,
     > {
         let authority = self.begin_release(generation_id, layout)?;
-        let observed = match match_recovery_process(authority.daemon) {
+        let observed = match owner {
-            Ok(PidfsRecoveryMatch::Matched(holder)) => recover_private_state_anchor_mount(
+            StateAnchorRecoveryOwner::Matched(holder) => recover_private_state_anchor_mount(
                 layout,
-                &holder,
+                holder,
                 authority.user_namespace,
                 authority.mount_namespace,
                 authority.mount,
             )?,
-            Ok(PidfsRecoveryMatch::BootMismatch) => {
+            StateAnchorRecoveryOwner::Absent(owners_absent) => {
                 self.require_absence_witness(owners_absent, &authority)?;
                 observe_mount_absence(layout, authority.mount)?
             }
-            Err(error) if error.raw_os_error() == Some(rustix::io::Errno::SRCH.raw_os_error()) => {
-                self.require_absence_witness(owners_absent, &authority)?;
-                observe_mount_absence(layout, authority.mount)?
-            }
-            Err(error) => return Err(StateAnchorCoordinatorError::Physical(error.into())),
         };
         Ok((
             authority.clone(),
@@ -421,12 +422,9 @@ impl<'a> StateAnchorCoordinator<'a> {
 
     fn require_absence_witness(
         &self,
-        witness: Option<&RuntimeOwnersAbsent>,
+        witness: &RuntimeOwnersAbsent,
         authority: &StateAnchorReleaseAuthority,
     ) -> Result<(), StateAnchorCoordinatorError> {
-        let witness = witness.ok_or(StateAnchorCoordinatorError::InvalidState(
-            "complete runtime-owner absence proof is required",
-        ))?;
         let snapshot = self.domain_store.snapshot()?;
         let generation = snapshot
             .generations

crates/lazaret-host-coordinator/tests/recovery-process.rs

@@ -1,14 +1,22 @@
+use std::fs::{self, File};
 use std::io;
+use std::os::fd::OwnedFd;
+use std::os::unix::fs::PermissionsExt;
 use std::process::ExitCode;
 
 use lazaret_core::identity::{
-    ConfigurationFingerprint, CoordinatorId, DomainId, GenerationId, NetworkSidecarId, SandboxId,
+    BootId, BundleDigest, ConfigurationFingerprint, ContainerId, CoordinatorId, DomainId,
+    GenerationId, HostPath, HostPidLocator, MountId, MountIdentity, NetworkSidecarId,
+    PidfsDeviceId, PidfsInodeId, PidfsProcessAuthorityV1, PidfsProcessIdentityV1,
+    ProcessStartTimeTicks, SandboxId,
 };
 use lazaret_core::records::{
     DurableDecision, Generation, Participant, ParticipantOwnership, ParticipantRole,
-    ParticipantState, PersistedHostProcessOwnership, StateAnchorOwnership,
+    ParticipantState, PersistedHostProcessOwnership, SandboxOwnership, StateAnchorOwnership,
+};
+use lazaret_host_coordinator::recovery::{
+    delete_runsc_container, prove_runtime_owners_absent, runtime_recovery_ownership,
 };
-use lazaret_host_coordinator::recovery::{prove_runtime_owners_absent, runtime_recovery_ownership};
 use lazaret_linux::process::{TargetCommand, child_broker, dispatch_private_process_role};
 use rustix::process::Signal;
 
@@ -31,19 +39,61 @@ fn exit(result: io::Result<()>) -> ExitCode {
 }
 
 fn run() -> io::Result<()> {
+    verify_exact_runsc_delete()?;
     verify_exact_action(None)?;
     verify_exact_action(Some(Signal::STOP))
 }
 
+fn verify_exact_runsc_delete() -> io::Result<()> {
+    let tree = tempfile::tempdir()?;
+    let root_path = tree.path().join("runsc");
+    fs::create_dir(&root_path)?;
+    fs::create_dir(root_path.join("container"))?;
+    let program = tree.path().join("runsc-fixture");
+    fs::write(
+        &program,
+        b"#!/bin/sh\nroot=${1#-root=}\n[ \"$2\" = delete ] && [ \"$3\" = --force ] && rmdir \"$root/$4\"\n",
+    )?;
+    fs::set_permissions(&program, fs::Permissions::from_mode(0o700))?;
+    let root: OwnedFd = File::open(&root_path)?.into();
+    delete_runsc_container(&root, "container", program.as_os_str()).map_err(io::Error::other)?;
+    if root_path.join("container").exists() {
+        return Err(io::Error::other("exact runsc container remained"));
+    }
+    Ok(())
+}
+
 fn verify_exact_action(signal: Option<Signal>) -> io::Result<()> {
-    let captured =
+    let state =
         child_broker().run_transient_captured(&TargetCommand::new("/bin/sleep").argument("60"))?;
     if let Some(signal) = signal {
-        captured.process().send_signal(signal)?;
+        state.process().send_signal(signal)?;
     }
-    let process = captured.into_process();
+    let state = state.into_process();
+    let supervisor = child_broker()
+        .run_transient_captured(&TargetCommand::new("/bin/sleep").argument("60"))?
+        .into_process();
     let generation = generation();
     let participants = [
+        Participant::snapshot(
+            generation.generation_id,
+            ParticipantRole::Sandbox,
+            ParticipantState::Intent,
+            Some(ParticipantOwnership::Sandbox(SandboxOwnership {
+                container_id: Some(ContainerId::new([6; 32])),
+                runtime_root: Some(HostPath::new(b"/runtime/runsc".to_vec())),
+                runtime_root_identity: Some(MountIdentity {
+                    mount_id: MountId::new(7),
+                    device: lazaret_core::identity::DeviceId::new(8),
+                    inode: lazaret_core::identity::InodeId::new(9),
+                }),
+                bundle_digest: Some(BundleDigest::new([10; 32])),
+                host_coordinator: PersistedHostProcessOwnership::committed(absent_process()),
+                runtime_supervisor: PersistedHostProcessOwnership::staged(supervisor.identity()),
+                attached_runsc: PersistedHostProcessOwnership::unassigned(),
+            })),
+        )
+        .map_err(|_| io::Error::other("Sandbox participant is invalid"))?,
         Participant::bare_intent(generation.generation_id, ParticipantRole::PreparedLower),
         Participant::snapshot(
             generation.generation_id,
@@ -54,7 +104,7 @@ fn verify_exact_action(signal: Option<Signal>) -> io::Result<()> {
                 upper_path: None,
                 work_path: None,
                 merged_path: None,
-                daemon: PersistedHostProcessOwnership::staged(process.identity()),
+                daemon: PersistedHostProcessOwnership::staged(state.identity()),
                 user_namespace: None,
                 mount_namespace: None,
                 mount: None,
@@ -65,7 +115,20 @@ fn verify_exact_action(signal: Option<Signal>) -> io::Result<()> {
     let ownership =
         runtime_recovery_ownership(&participants, &generation).map_err(io::Error::other)?;
     prove_runtime_owners_absent(&ownership)?;
-    child_broker().reap(process)
+    child_broker().reap(state)?;
+    child_broker().reap(supervisor)
+}
+
+fn absent_process() -> PidfsProcessIdentityV1 {
+    PidfsProcessIdentityV1::new(
+        PidfsProcessAuthorityV1 {
+            boot_id: BootId::new([0xff; 16]),
+            pidfs_st_dev: PidfsDeviceId::new(1),
+            pidfs_st_ino: PidfsInodeId::new(2),
+        },
+        HostPidLocator::new(1),
+        ProcessStartTimeTicks::new(3),
+    )
 }
 
 fn generation() -> Generation {

crates/lazaret-host-coordinator/tests/runtime-composition.rs

@@ -33,7 +33,7 @@ use lazaret_host_coordinator::terminal::{
     NoActivePrompt, RunscSessionTerminalFactory, SessionTerminalEngine, TerminalAcquisition,
     TerminalRoute,
 };
-use lazaret_linux::filesystem::try_exclusive_flock;
+use lazaret_linux::filesystem::{current_mount_identity, try_exclusive_flock};
 use lazaret_linux::platform::{PlatformObservation, check_platform};
 use lazaret_linux::process::{
     ProcessOwnershipContext, RetainedProcess, TargetCommand, child_broker, current_process_identity,
@@ -1023,6 +1023,7 @@ impl Fixture {
                 GenerationId::new(7),
                 container_id,
                 &runtime_domain.join("runsc"),
+                current_mount_identity(&runtime_domain.join("runsc"))?,
                 BundleDigest::new([7; 32]),
                 current_process_identity()?,
             )
@@ -1306,6 +1307,7 @@ fn network_store(
             GenerationId::new(7),
             container_id,
             &runtime_domain.join("runsc"),
+            current_mount_identity(&runtime_domain.join("runsc"))?,
             BundleDigest::new([17; 32]),
             current_process_identity()?,
         )

docs/schema.md

@@ -1051,6 +1051,7 @@ classDiagram
         <<serialized>>
         optional container_id
         optional runtime_root
+        optional runtime_root_identity
         optional bundle_digest
         host_coordinator
         runtime_supervisor
@@ -1059,8 +1060,12 @@ classDiagram
 
     class PublicationOwnership {
         <<serialized>>
+        optional staged_socket_path
+        optional staged_identity_path
         optional socket_path
         optional identity_path
+        optional socket_identity
+        optional identity_file_identity
     }
 
     ParticipantOwnership --> PreparedLowerOwnership
@@ -1106,8 +1111,8 @@ The only serviceable role/state pairs are:
 | `GitRouting` | `Prepared` | three paths, mount, current committed daemon |
 | `Network` | `Committed` | policy digest, fingerprint, readiness evidence, namespace, current committed RS |
 | `Helper` | `Committed` | sidecar ID, current committed helper and RS |
-| `Sandbox` | `Committed` | container, runtime root, bundle digest, current committed HC, RS, and runsc |
+| `Sandbox` | `Committed` | container, runtime root and exact mount/device/inode identity, bundle digest, current committed HC, RS, and runsc |
-| `Publication` | `Committed` | socket and identity paths |
+| `Publication` | `Committed` | staged and exposed fixed labels plus exact socket and identity-file mount/device/inode identities |
 
 Network readiness evidence must name the enclosing Participant generation and
 match its policy digest, fingerprint, namespace, and committed RS identity.