fix(drivers): validate vendor-agnostic CDI device names

crates/openshell-core/src/gpu.rs

@@ -302,6 +302,62 @@ fn cdi_nvidia_gpu_suffix(id: &str) -> Option<&str> {
     id.strip_prefix(CDI_NVIDIA_GPU_PREFIX)
 }
 
+// Local equivalent of the upstream CDI parser until its public API is released:
+// https://github.com/cncf-tags/container-device-interface-rs/pull/177
+fn is_qualified_name(device: &str) -> bool {
+    let Some((kind, name)) = device.split_once('=') else {
+        return false;
+    };
+    let Some((vendor, class)) = kind.split_once('/') else {
+        return false;
+    };
+    let valid_component = |value: &str, device_name: bool| {
+        let bytes = value.as_bytes();
+        bytes.first().is_some_and(|byte| {
+            if device_name {
+                byte.is_ascii_alphanumeric()
+            } else {
+                byte.is_ascii_alphabetic()
+            }
+        }) && bytes.last().is_some_and(u8::is_ascii_alphanumeric)
+            && bytes.iter().all(|byte| {
+                byte.is_ascii_alphanumeric()
+                    || matches!(byte, b'_' | b'-' | b'.')
+                    || (device_name && *byte == b':')
+            })
+    };
+    valid_component(vendor, false) && valid_component(class, false) && valid_component(name, true)
+}
+
+/// Validate vendor-agnostic CDI qualified names without resolving or authorizing devices.
+///
+/// # Errors
+/// Returns an error for host paths or malformed `<vendor>/<class>=<name>` selectors.
+pub fn validate_cdi_device_names(devices: &[String], field: &str) -> Result<(), String> {
+    if devices.iter().any(|device| !is_qualified_name(device)) {
+        return Err(format!(
+            "{field} must contain CDI qualified names (<vendor>/<class>=<name>)"
+        ));
+    }
+    Ok(())
+}
+
+/// Validate explicit CDI selectors and their GPU request requirements.
+///
+/// Docker and Podman use this helper; drivers with other identifier formats use
+/// [`validate_specific_gpu_device_request`] directly.
+///
+/// # Errors
+/// Returns an error for malformed CDI names or invalid GPU request requirements.
+pub fn validate_cdi_gpu_device_request(
+    gpu: Option<&DriverGpuResourceRequirements>,
+    devices: &[String],
+    field: &str,
+) -> Result<(), String> {
+    validate_cdi_device_names(devices, field)?;
+    validate_specific_gpu_device_request(gpu, devices, field)
+}
+
 /// Validate a compute-driver GPU request against driver-owned specific devices.
 ///
 /// Drivers call this when a sandbox request combines portable GPU requirements
@@ -347,6 +403,64 @@ pub fn validate_specific_gpu_device_request(
 mod tests {
     use super::*;
 
+    #[test]
+    fn cdi_device_names_accept_vendor_agnostic_selectors() {
+        for device in [
+            "nvidia.com/gpu=0",
+            "nvidia.com/gpu=all",
+            "nvidia.com/gpu=GPU-5b2d",
+            "nvidia.com/gpu=MIG-5b2d",
+            "nvidia.com/gpu=0:1",
+            "example.com/accelerator=card_0.1",
+            "intel.com/gpu=0",
+            "amd.com/gpu=0",
+            "v/c=0",
+        ] {
+            let devices = vec![device.to_string()];
+            validate_cdi_device_names(&devices, "driver_config.cdi_devices")
+                .unwrap_or_else(|error| panic!("{device:?}: {error}"));
+        }
+    }
+
+    #[test]
+    fn cdi_device_names_reject_host_paths_and_malformed_selectors() {
+        for device in [
+            "",
+            "/dev/sda",
+            "/dev",
+            "/dev/sda:/dev/sda:rwm",
+            "dev/sda",
+            "../dev/sda",
+            "vendor/gpu",
+            "/gpu=0",
+            "vendor/=0",
+            "vendor/gpu=",
+            "vendor/gpu=/dev/sda",
+            "vendor/gpu=0/1",
+            "vendor/gpu=0=1",
+            "vendor/gpu=0,1",
+            "vendor/gpu=-0",
+            "vendor/gpu=0:",
+            "vendor/gpu=0 ",
+            " vendor/gpu=0",
+            "vendor/gpu=0\n",
+            "vendor/gpu=0\0",
+            "vendor/gpu=é",
+            "véndor/gpu=0",
+            "vendor/gpü=0",
+            "0vendor/gpu=0",
+            "vendor/0gpu=0",
+            "vendor./gpu=0",
+            "vendor/gpu-=0",
+            "vendor/gpu/other=0",
+        ] {
+            let error =
+                validate_cdi_device_names(&[device.to_string()], "driver_config.cdi_devices")
+                    .unwrap_err();
+            assert!(error.contains("CDI qualified names"), "{device:?}: {error}");
+        }
+    }
+
     #[test]
     fn effective_driver_gpu_count_normalizes_missing_count() {
         let gpu = DriverGpuResourceRequirements { count: None };
@@ -547,6 +661,18 @@ mod tests {
         );
     }
 
+    #[test]
+    fn specific_gpu_device_request_preserves_non_cdi_identifiers() {
+        let gpu = DriverGpuResourceRequirements { count: Some(1) };
+        let devices = vec!["0000:2d:00.0".to_string()];
+        validate_specific_gpu_device_request(Some(&gpu), &devices, "driver_config.gpu_device_ids")
+            .expect("generic GPU request validation must accept PCI device identifiers");
+        let error =
+            validate_cdi_gpu_device_request(Some(&gpu), &devices, "driver_config.cdi_devices")
+                .expect_err("CDI drivers must reject non-CDI identifiers");
+        assert!(error.contains("CDI qualified names"));
+    }
+
     #[test]
     fn validate_specific_gpu_device_request_ignores_empty_devices() {
         validate_specific_gpu_device_request(None, &[], "driver_config.cdi_devices")

crates/openshell-driver-docker/src/lib.rs

@@ -33,7 +33,7 @@ use openshell_core::driver_utils::{
 };
 use openshell_core::gpu::{
     CdiGpuDefaultSelector, CdiGpuInventory, CdiGpuSelectionError, driver_gpu_requirements,
-    effective_driver_gpu_count, validate_specific_gpu_device_request,
+    effective_driver_gpu_count, validate_cdi_device_names, validate_cdi_gpu_device_request,
 };
 use openshell_core::progress::{
     PROGRESS_STEP_PULLING_IMAGE, PROGRESS_STEP_REQUESTING_SANDBOX, PROGRESS_STEP_STARTING_SANDBOX,
@@ -1126,7 +1126,7 @@ impl DockerComputeDriver {
         }
 
         if let Some(cdi_devices) = driver_config.cdi_devices.as_deref() {
-            validate_specific_gpu_device_request(
+            validate_cdi_gpu_device_request(
                 gpu_requirements,
                 cdi_devices,
                 "driver_config.cdi_devices",
@@ -1309,7 +1309,7 @@ impl DockerComputeDriver {
         ) -> Result<Vec<String>, CdiGpuSelectionError>,
     ) -> Result<Option<Vec<String>>, Status> {
         if let Some(cdi_devices) = driver_config.cdi_devices.as_deref() {
-            validate_specific_gpu_device_request(
+            validate_cdi_gpu_device_request(
                 gpu_requirements,
                 cdi_devices,
                 "driver_config.cdi_devices",
@@ -5618,12 +5618,8 @@ fn build_container_create_body(
         .as_ref()
         .and_then(|spec| driver_gpu_requirements(spec.resource_requirements.as_ref()));
     let cdi_devices = if let Some(cdi_devices) = driver_config.cdi_devices.as_ref() {
-        validate_specific_gpu_device_request(
-            gpu_requirements,
-            cdi_devices,
-            "driver_config.cdi_devices",
-        )
-        .map_err(Status::invalid_argument)?;
+        validate_cdi_gpu_device_request(gpu_requirements, cdi_devices, "driver_config.cdi_devices")
+            .map_err(Status::invalid_argument)?;
         Some(cdi_devices.as_slice())
     } else {
         None
@@ -5674,6 +5670,10 @@ fn build_container_create_body_for_image(
     image: &DockerImageMetadata,
     workload_identity: &ResolvedWorkloadIdentity,
 ) -> Result<ContainerCreateBody, Status> {
+    if let Some(device_ids) = gpu_device_ids {
+        validate_cdi_device_names(device_ids, "driver_config.cdi_devices")
+            .map_err(Status::invalid_argument)?;
+    }
     let spec = sandbox
         .spec
         .as_ref()

crates/openshell-driver-docker/src/tests.rs

@@ -2719,6 +2719,91 @@ fn build_container_create_body_omits_devices_without_resolved_default_cdi_device
     );
 }
 
+#[test]
+fn container_spec_accepts_non_nvidia_explicit_and_resolved_cdi_devices() {
+    let mut config = runtime_config();
+    config.gpu.cdi_supported = true;
+    let mut sandbox = test_sandbox();
+    sandbox.spec.as_mut().unwrap().resource_requirements = Some(gpu_resources(None));
+    for device in ["example.com/gpu=0", "intel.com/gpu=0", "amd.com/gpu=0"] {
+        sandbox
+            .spec
+            .as_mut()
+            .unwrap()
+            .template
+            .as_mut()
+            .unwrap()
+            .driver_config = Some(cdi_devices_config(&[device]));
+        let explicit = build_container_create_body(&sandbox, &config).unwrap();
+        sandbox
+            .spec
+            .as_mut()
+            .unwrap()
+            .template
+            .as_mut()
+            .unwrap()
+            .driver_config = None;
+        let resolved = build_container_create_body_with_gpu_devices(
+            &sandbox,
+            &config,
+            &DockerSandboxDriverConfig::default(),
+            Some(&[device.to_string()]),
+        )
+        .unwrap();
+        for body in [explicit, resolved] {
+            let requests = body.host_config.unwrap().device_requests.unwrap();
+            assert_eq!(requests[0].driver.as_deref(), Some("cdi"));
+            assert_eq!(
+                requests[0].device_ids.as_ref().unwrap(),
+                &[device.to_string()]
+            );
+        }
+    }
+}
+
+#[test]
+fn container_spec_rejects_host_paths_in_explicit_and_resolved_cdi_devices() {
+    let mut config = runtime_config();
+    config.gpu.cdi_supported = true;
+    let mut sandbox = test_sandbox();
+    sandbox.spec.as_mut().unwrap().resource_requirements = Some(gpu_resources(None));
+    for device in [
+        "/dev/sda",
+        "/dev",
+        "/dev/sda:/dev/sda:rwm",
+        "vendor/gpu=0/1",
+    ] {
+        sandbox
+            .spec
+            .as_mut()
+            .unwrap()
+            .template
+            .as_mut()
+            .unwrap()
+            .driver_config = Some(cdi_devices_config(&[device]));
+        let explicit_error = build_container_create_body(&sandbox, &config).unwrap_err();
+        sandbox
+            .spec
+            .as_mut()
+            .unwrap()
+            .template
+            .as_mut()
+            .unwrap()
+            .driver_config = None;
+        let resolved_error = build_container_create_body_with_gpu_devices(
+            &sandbox,
+            &config,
+            &DockerSandboxDriverConfig::default(),
+            Some(&[device.to_string()]),
+        )
+        .unwrap_err();
+        for error in [explicit_error, resolved_error] {
+            assert_eq!(error.code(), tonic::Code::InvalidArgument);
+            assert!(error.message().contains("CDI qualified names"));
+        }
+    }
+}
+
 #[test]
 fn build_container_create_body_passes_explicit_cdi_device_id_through() {
     let mut config = runtime_config();