vmm: return seccomp rules according to hypervisors

That requires stashing the hypervisor type into various places.

Signed-off-by: Wei Liu <liuwe@microsoft.com>

vmm/src/api/http.rs

@@ -7,6 +7,7 @@ use crate::api::http_endpoint::{VmActionHandler, VmCreate, VmInfo, VmmPing, VmmS
 use crate::api::{ApiError, ApiRequest, VmAction};
 use crate::seccomp_filters::{get_seccomp_filter, Thread};
 use crate::{Error as VmmError, Result};
+use hypervisor::HypervisorType;
 use micro_http::{Body, HttpServer, MediaType, Method, Request, Response, StatusCode, Version};
 use once_cell::sync::Lazy;
 use seccompiler::{apply_filter, SeccompAction};
@@ -278,10 +279,11 @@ fn start_http_thread(
     api_sender: Sender<ApiRequest>,
     seccomp_action: &SeccompAction,
     exit_evt: EventFd,
+    hypervisor_type: HypervisorType,
 ) -> Result<thread::JoinHandle<Result<()>>> {
     // Retrieve seccomp filter for API thread
-    let api_seccomp_filter =
+    let api_seccomp_filter = get_seccomp_filter(seccomp_action, Thread::Api, hypervisor_type)
-        get_seccomp_filter(seccomp_action, Thread::Api).map_err(VmmError::CreateSeccompFilter)?;
+        .map_err(VmmError::CreateSeccompFilter)?;
 
     thread::Builder::new()
         .name("http-server".to_string())
@@ -336,12 +338,20 @@ pub fn start_http_path_thread(
     api_sender: Sender<ApiRequest>,
     seccomp_action: &SeccompAction,
     exit_evt: EventFd,
+    hypervisor_type: HypervisorType,
 ) -> Result<thread::JoinHandle<Result<()>>> {
     let socket_path = PathBuf::from(path);
     let socket_fd = UnixListener::bind(socket_path).map_err(VmmError::CreateApiServerSocket)?;
     let server =
         HttpServer::new_from_fd(socket_fd.into_raw_fd()).map_err(VmmError::CreateApiServer)?;
-    start_http_thread(server, api_notifier, api_sender, seccomp_action, exit_evt)
+    start_http_thread(
+        server,
+        api_notifier,
+        api_sender,
+        seccomp_action,
+        exit_evt,
+        hypervisor_type,
+    )
 }
 
 pub fn start_http_fd_thread(
@@ -350,7 +360,15 @@ pub fn start_http_fd_thread(
     api_sender: Sender<ApiRequest>,
     seccomp_action: &SeccompAction,
     exit_evt: EventFd,
+    hypervisor_type: HypervisorType,
 ) -> Result<thread::JoinHandle<Result<()>>> {
     let server = HttpServer::new_from_fd(fd).map_err(VmmError::CreateApiServer)?;
-    start_http_thread(server, api_notifier, api_sender, seccomp_action, exit_evt)
+    start_http_thread(
+        server,
+        api_notifier,
+        api_sender,
+        seccomp_action,
+        exit_evt,
+        hypervisor_type,
+    )
 }

vmm/src/cpu.rs

@@ -46,7 +46,7 @@ use hypervisor::arch::x86::{SpecialRegisters, StandardRegisters};
 use hypervisor::kvm::kvm_bindings;
 #[cfg(feature = "tdx")]
 use hypervisor::kvm::{TdxExitDetails, TdxExitStatus};
-use hypervisor::{CpuState, HypervisorCpuError, VmExit, VmOps};
+use hypervisor::{CpuState, HypervisorCpuError, HypervisorType, VmExit, VmOps};
 use libc::{c_void, siginfo_t};
 #[cfg(feature = "guest_debug")]
 use linux_loader::elf::Elf64_Nhdr;
@@ -412,6 +412,7 @@ impl Snapshottable for Vcpu {
 }
 
 pub struct CpuManager {
+    hypervisor_type: HypervisorType,
     config: CpusConfig,
     #[cfg_attr(target_arch = "aarch64", allow(dead_code))]
     interrupt_controller: Option<Arc<Mutex<dyn InterruptController>>>,
@@ -589,6 +590,7 @@ impl CpuManager {
         let guest_memory = memory_manager.lock().unwrap().guest_memory();
         let mut vcpu_states = Vec::with_capacity(usize::from(config.max_vcpus));
         vcpu_states.resize_with(usize::from(config.max_vcpus), VcpuState::default);
+        let hypervisor_type = hypervisor.hypervisor_type();
 
         #[cfg(target_arch = "x86_64")]
         let sgx_epc_sections = memory_manager
@@ -688,6 +690,7 @@ impl CpuManager {
         };
 
         let cpu_manager = Arc::new(Mutex::new(CpuManager {
+            hypervisor_type,
             config: config.clone(),
             interrupt_controller: device_manager.interrupt_controller().clone(),
             vm_memory: guest_memory,
@@ -836,8 +839,9 @@ impl CpuManager {
         });
 
         // Retrieve seccomp filter for vcpu thread
-        let vcpu_seccomp_filter = get_seccomp_filter(&self.seccomp_action, Thread::Vcpu)
+        let vcpu_seccomp_filter =
-            .map_err(Error::CreateSeccompFilter)?;
+            get_seccomp_filter(&self.seccomp_action, Thread::Vcpu, self.hypervisor_type)
+                .map_err(Error::CreateSeccompFilter)?;
 
         #[cfg(target_arch = "x86_64")]
         let interrupt_controller_clone = self.interrupt_controller.as_ref().cloned();

vmm/src/device_manager.rs

@@ -50,7 +50,7 @@ use devices::legacy::Serial;
 use devices::{
     interrupt_controller, interrupt_controller::InterruptController, AcpiNotificationFlags,
 };
-use hypervisor::{HypervisorVmError, IoEventAddress};
+use hypervisor::{HypervisorType, HypervisorVmError, IoEventAddress};
 use libc::{
     cfmakeraw, isatty, tcgetattr, tcsetattr, termios, MAP_NORESERVE, MAP_PRIVATE, MAP_SHARED,
     O_TMPFILE, PROT_READ, PROT_WRITE, TCSANOW,
@@ -807,6 +807,9 @@ struct MetaVirtioDevice {
 }
 
 pub struct DeviceManager {
+    // The underlying hypervisor
+    hypervisor_type: HypervisorType,
+
     // Manage address space related to devices
     address_manager: Arc<AddressManager>,
 
@@ -945,6 +948,7 @@ pub struct DeviceManager {
 impl DeviceManager {
     #[allow(clippy::too_many_arguments)]
     pub fn new(
+        hypervisor_type: HypervisorType,
         vm: Arc<dyn hypervisor::Vm>,
         config: Arc<Mutex<VmConfig>>,
         memory_manager: Arc<Mutex<MemoryManager>>,
@@ -1035,6 +1039,7 @@ impl DeviceManager {
         }
 
         let device_manager = DeviceManager {
+            hypervisor_type,
             address_manager: Arc::clone(&address_manager),
             console: Arc::new(Console::default()),
             interrupt_controller: None,
@@ -1803,8 +1808,12 @@ impl DeviceManager {
     }
 
     fn listen_for_sigwinch_on_tty(&mut self, pty: &File) -> std::io::Result<()> {
-        let seccomp_filter =
+        let seccomp_filter = get_seccomp_filter(
-            get_seccomp_filter(&self.seccomp_action, Thread::PtyForeground).unwrap();
+            &self.seccomp_action,
+            Thread::PtyForeground,
+            self.hypervisor_type,
+        )
+        .unwrap();
 
         match start_sigwinch_listener(seccomp_filter, pty) {
             Ok(pipe) => {

vmm/src/lib.rs

@@ -280,10 +280,11 @@ pub fn start_vmm_thread(
     let gdb_vm_debug_event = vm_debug_event.try_clone().map_err(Error::EventFdClone)?;
 
     let http_api_event = api_event.try_clone().map_err(Error::EventFdClone)?;
+    let hypervisor_type = hypervisor.hypervisor_type();
 
     // Retrieve seccomp filter
-    let vmm_seccomp_filter =
+    let vmm_seccomp_filter = get_seccomp_filter(seccomp_action, Thread::Vmm, hypervisor_type)
-        get_seccomp_filter(seccomp_action, Thread::Vmm).map_err(Error::CreateSeccompFilter)?;
+        .map_err(Error::CreateSeccompFilter)?;
 
     let vmm_seccomp_action = seccomp_action.clone();
     let exit_evt = EventFd::new(EFD_NONBLOCK).map_err(Error::EventFdCreate)?;
@@ -328,6 +329,7 @@ pub fn start_vmm_thread(
             api_sender,
             seccomp_action,
             exit_evt,
+            hypervisor_type,
         )?;
     } else if let Some(http_fd) = http_fd {
         api::start_http_fd_thread(
@@ -336,6 +338,7 @@ pub fn start_vmm_thread(
             api_sender,
             seccomp_action,
             exit_evt,
+            hypervisor_type,
         )?;
     }
 
@@ -413,9 +416,12 @@ impl Vmm {
                 let exit_evt = self.exit_evt.try_clone().map_err(Error::EventFdClone)?;
                 let on_tty = unsafe { libc::isatty(libc::STDIN_FILENO as i32) } != 0;
 
-                let signal_handler_seccomp_filter =
+                let signal_handler_seccomp_filter = get_seccomp_filter(
-                    get_seccomp_filter(&self.seccomp_action, Thread::SignalHandler)
+                    &self.seccomp_action,
-                        .map_err(Error::CreateSeccompFilter)?;
+                    Thread::SignalHandler,
+                    self.hypervisor.hypervisor_type(),
+                )
+                .map_err(Error::CreateSeccompFilter)?;
                 self.threads.push(
                     thread::Builder::new()
                         .name("vmm_signal_handler".to_string())

vmm/src/seccomp_filters.rs

@@ -4,6 +4,7 @@
 //
 // SPDX-License-Identifier: Apache-2.0
 
+use hypervisor::HypervisorType;
 use seccompiler::{
     BackendError, BpfProgram, Error, SeccompAction, SeccompCmpArgLen as ArgLen, SeccompCmpOp::Eq,
     SeccompCondition as Cond, SeccompFilter, SeccompRule,
@@ -226,17 +227,22 @@ fn create_vmm_ioctl_seccomp_rule_common_kvm() -> Result<Vec<SeccompRule>, Backen
     ])
 }
 
-fn create_vmm_ioctl_seccomp_rule_hypervisor() -> Result<Vec<SeccompRule>, BackendError> {
+fn create_vmm_ioctl_seccomp_rule_hypervisor(
-    #[cfg(feature = "kvm")]
+    hypervisor_type: HypervisorType,
-    let rules = create_vmm_ioctl_seccomp_rule_common_kvm();
+) -> Result<Vec<SeccompRule>, BackendError> {
-
+    match hypervisor_type {
-    #[cfg(feature = "mshv")]
+        #[cfg(feature = "kvm")]
-    let rules = create_vmm_ioctl_seccomp_rule_common_mshv();
+        HypervisorType::Kvm => create_vmm_ioctl_seccomp_rule_common_kvm(),
-
+        #[cfg(feature = "mshv")]
-    rules
+        HypervisorType::Mshv => create_vmm_ioctl_seccomp_rule_common_mshv(),
+        #[allow(unreachable_patterns)]
+        _ => panic!("Invalid hypervisor {:?}", hypervisor_type),
+    }
 }
 
-fn create_vmm_ioctl_seccomp_rule_common() -> Result<Vec<SeccompRule>, BackendError> {
+fn create_vmm_ioctl_seccomp_rule_common(
+    hypervisor_type: HypervisorType,
+) -> Result<Vec<SeccompRule>, BackendError> {
     let mut common_rules = or![
         and![Cond::new(1, ArgLen::Dword, Eq, BLKSSZGET)?],
         and![Cond::new(1, ArgLen::Dword, Eq, BLKPBSZGET)?],
@@ -308,7 +314,7 @@ fn create_vmm_ioctl_seccomp_rule_common() -> Result<Vec<SeccompRule>, BackendErr
         and![Cond::new(1, ArgLen::Dword, Eq, VHOST_VDPA_GET_IOVA_RANGE)?],
     ];
 
-    let hypervisor_rules = create_vmm_ioctl_seccomp_rule_hypervisor()?;
+    let hypervisor_rules = create_vmm_ioctl_seccomp_rule_hypervisor(hypervisor_type)?;
 
     common_rules.extend(hypervisor_rules);
 
@@ -341,7 +347,7 @@ fn create_vmm_ioctl_seccomp_rule_kvm() -> Result<Vec<SeccompRule>, BackendError>
     const KVM_SET_GUEST_DEBUG: u64 = 0x4048_ae9b;
     const KVM_TRANSLATE: u64 = 0xc018_ae85;
 
-    let common_rules = create_vmm_ioctl_seccomp_rule_common()?;
+    let common_rules = create_vmm_ioctl_seccomp_rule_common(HypervisorType::Kvm)?;
     let mut arch_rules = or![
         and![Cond::new(1, ArgLen::Dword, Eq, KVM_CREATE_PIT2)?],
         and![Cond::new(1, ArgLen::Dword, Eq, KVM_GET_CLOCK,)?],
@@ -377,7 +383,7 @@ fn create_vmm_ioctl_seccomp_rule_kvm() -> Result<Vec<SeccompRule>, BackendError>
     const KVM_ARM_PREFERRED_TARGET: u64 = 0x8020_aeaf;
     const KVM_ARM_VCPU_INIT: u64 = 0x4020_aeae;
 
-    let common_rules = create_vmm_ioctl_seccomp_rule_common()?;
+    let common_rules = create_vmm_ioctl_seccomp_rule_common(HypervisorType::Kvm)?;
     let mut arch_rules = or![
         and![Cond::new(1, ArgLen::Dword, Eq, KVM_ARM_PREFERRED_TARGET,)?],
         and![Cond::new(1, ArgLen::Dword, Eq, KVM_ARM_VCPU_INIT,)?],
@@ -389,17 +395,20 @@ fn create_vmm_ioctl_seccomp_rule_kvm() -> Result<Vec<SeccompRule>, BackendError>
 
 #[cfg(all(target_arch = "x86_64", feature = "mshv"))]
 fn create_vmm_ioctl_seccomp_rule_mshv() -> Result<Vec<SeccompRule>, BackendError> {
-    create_vmm_ioctl_seccomp_rule_common()
+    create_vmm_ioctl_seccomp_rule_common(HypervisorType::Mshv)
 }
 
-fn create_vmm_ioctl_seccomp_rule() -> Result<Vec<SeccompRule>, BackendError> {
+fn create_vmm_ioctl_seccomp_rule(
-    #[cfg(feature = "kvm")]
+    hypervisor_type: HypervisorType,
-    let rules = create_vmm_ioctl_seccomp_rule_kvm();
+) -> Result<Vec<SeccompRule>, BackendError> {
-
+    match hypervisor_type {
-    #[cfg(feature = "mshv")]
+        #[cfg(feature = "kvm")]
-    let rules = create_vmm_ioctl_seccomp_rule_mshv();
+        HypervisorType::Kvm => create_vmm_ioctl_seccomp_rule_kvm(),
-
+        #[cfg(feature = "mshv")]
-    rules
+        HypervisorType::Mshv => create_vmm_ioctl_seccomp_rule_mshv(),
+        #[allow(unreachable_patterns)]
+        _ => panic!("Invalid hypervisor {:?}", hypervisor_type),
+    }
 }
 
 fn create_api_ioctl_seccomp_rule() -> Result<Vec<SeccompRule>, BackendError> {
@@ -465,7 +474,9 @@ fn pty_foreground_thread_rules() -> Result<Vec<(i64, Vec<SeccompRule>)>, Backend
 
 // The filter containing the white listed syscall rules required by the VMM to
 // function.
-fn vmm_thread_rules() -> Result<Vec<(i64, Vec<SeccompRule>)>, BackendError> {
+fn vmm_thread_rules(
+    hypervisor_type: HypervisorType,
+) -> Result<Vec<(i64, Vec<SeccompRule>)>, BackendError> {
     Ok(vec![
         (libc::SYS_accept4, vec![]),
         #[cfg(target_arch = "x86_64")]
@@ -506,7 +517,10 @@ fn vmm_thread_rules() -> Result<Vec<(i64, Vec<SeccompRule>)>, BackendError> {
         (libc::SYS_gettid, vec![]),
         (libc::SYS_gettimeofday, vec![]),
         (libc::SYS_getuid, vec![]),
-        (libc::SYS_ioctl, create_vmm_ioctl_seccomp_rule()?),
+        (
+            libc::SYS_ioctl,
+            create_vmm_ioctl_seccomp_rule(hypervisor_type)?,
+        ),
         (libc::SYS_io_uring_enter, vec![]),
         (libc::SYS_io_uring_setup, vec![]),
         (libc::SYS_io_uring_register, vec![]),
@@ -619,17 +633,22 @@ fn create_vcpu_ioctl_seccomp_rule_mshv() -> Result<Vec<SeccompRule>, BackendErro
     ])
 }
 
-fn create_vcpu_ioctl_seccomp_rule_hypervisor() -> Result<Vec<SeccompRule>, BackendError> {
+fn create_vcpu_ioctl_seccomp_rule_hypervisor(
-    #[cfg(feature = "kvm")]
+    hypervisor_type: HypervisorType,
-    let rules = create_vcpu_ioctl_seccomp_rule_kvm();
+) -> Result<Vec<SeccompRule>, BackendError> {
-
+    match hypervisor_type {
-    #[cfg(feature = "mshv")]
+        #[cfg(feature = "kvm")]
-    let rules = create_vcpu_ioctl_seccomp_rule_mshv();
+        HypervisorType::Kvm => create_vcpu_ioctl_seccomp_rule_kvm(),
-
+        #[cfg(feature = "mshv")]
-    rules
+        HypervisorType::Mshv => create_vcpu_ioctl_seccomp_rule_mshv(),
+        #[allow(unreachable_patterns)]
+        _ => panic!("Invalid hypervisor {:?}", hypervisor_type),
+    }
 }
 
-fn create_vcpu_ioctl_seccomp_rule() -> Result<Vec<SeccompRule>, BackendError> {
+fn create_vcpu_ioctl_seccomp_rule(
+    hypervisor_type: HypervisorType,
+) -> Result<Vec<SeccompRule>, BackendError> {
     let mut rules = or![
         and![Cond::new(1, ArgLen::Dword, Eq, VFIO_DEVICE_SET_IRQS)?],
         and![Cond::new(1, ArgLen::Dword, Eq, VFIO_GROUP_UNSET_CONTAINER)?],
@@ -645,14 +664,16 @@ fn create_vcpu_ioctl_seccomp_rule() -> Result<Vec<SeccompRule>, BackendError> {
         )?],
     ];
 
-    let hypervisor_rules = create_vcpu_ioctl_seccomp_rule_hypervisor()?;
+    let hypervisor_rules = create_vcpu_ioctl_seccomp_rule_hypervisor(hypervisor_type)?;
 
     rules.extend(hypervisor_rules);
 
     Ok(rules)
 }
 
-fn vcpu_thread_rules() -> Result<Vec<(i64, Vec<SeccompRule>)>, BackendError> {
+fn vcpu_thread_rules(
+    hypervisor_type: HypervisorType,
+) -> Result<Vec<(i64, Vec<SeccompRule>)>, BackendError> {
     Ok(vec![
         (libc::SYS_brk, vec![]),
         (libc::SYS_clock_gettime, vec![]),
@@ -665,7 +686,10 @@ fn vcpu_thread_rules() -> Result<Vec<(i64, Vec<SeccompRule>)>, BackendError> {
         (libc::SYS_futex, vec![]),
         (libc::SYS_getrandom, vec![]),
         (libc::SYS_getpid, vec![]),
-        (libc::SYS_ioctl, create_vcpu_ioctl_seccomp_rule()?),
+        (
+            libc::SYS_ioctl,
+            create_vcpu_ioctl_seccomp_rule(hypervisor_type)?,
+        ),
         (libc::SYS_lseek, vec![]),
         (libc::SYS_madvise, vec![]),
         (libc::SYS_mmap, vec![]),
@@ -727,12 +751,15 @@ fn api_thread_rules() -> Result<Vec<(i64, Vec<SeccompRule>)>, BackendError> {
     ])
 }
 
-fn get_seccomp_rules(thread_type: Thread) -> Result<Vec<(i64, Vec<SeccompRule>)>, BackendError> {
+fn get_seccomp_rules(
+    thread_type: Thread,
+    hypervisor_type: HypervisorType,
+) -> Result<Vec<(i64, Vec<SeccompRule>)>, BackendError> {
     match thread_type {
         Thread::Api => Ok(api_thread_rules()?),
         Thread::SignalHandler => Ok(signal_handler_thread_rules()?),
-        Thread::Vcpu => Ok(vcpu_thread_rules()?),
+        Thread::Vcpu => Ok(vcpu_thread_rules(hypervisor_type)?),
-        Thread::Vmm => Ok(vmm_thread_rules()?),
+        Thread::Vmm => Ok(vmm_thread_rules(hypervisor_type)?),
         Thread::PtyForeground => Ok(pty_foreground_thread_rules()?),
     }
 }
@@ -741,11 +768,12 @@ fn get_seccomp_rules(thread_type: Thread) -> Result<Vec<(i64, Vec<SeccompRule>)>
 pub fn get_seccomp_filter(
     seccomp_action: &SeccompAction,
     thread_type: Thread,
+    hypervisor_type: HypervisorType,
 ) -> Result<BpfProgram, Error> {
     match seccomp_action {
         SeccompAction::Allow => Ok(vec![]),
         SeccompAction::Log => SeccompFilter::new(
-            get_seccomp_rules(thread_type)
+            get_seccomp_rules(thread_type, hypervisor_type)
                 .map_err(Error::Backend)?
                 .into_iter()
                 .collect(),
@@ -756,7 +784,7 @@ pub fn get_seccomp_filter(
         .and_then(|filter| filter.try_into())
         .map_err(Error::Backend),
         _ => SeccompFilter::new(
-            get_seccomp_rules(thread_type)
+            get_seccomp_rules(thread_type, hypervisor_type)
                 .map_err(Error::Backend)?
                 .into_iter()
                 .collect(),

vmm/src/vm.rs

@@ -471,7 +471,6 @@ pub struct Vm {
     numa_nodes: NumaNodes,
     seccomp_action: SeccompAction,
     exit_evt: EventFd,
-    #[cfg(all(feature = "kvm", target_arch = "x86_64"))]
     hypervisor: Arc<dyn hypervisor::Hypervisor>,
     stop_on_boot: bool,
     #[cfg(target_arch = "x86_64")]
@@ -534,6 +533,7 @@ impl Vm {
         let stop_on_boot = false;
 
         let device_manager = DeviceManager::new(
+            hypervisor.hypervisor_type(),
             vm.clone(),
             config.clone(),
             memory_manager.clone(),
@@ -617,7 +617,6 @@ impl Vm {
             numa_nodes,
             seccomp_action: seccomp_action.clone(),
             exit_evt,
-            #[cfg(all(feature = "kvm", target_arch = "x86_64"))]
             hypervisor,
             stop_on_boot,
             #[cfg(target_arch = "x86_64")]
@@ -1969,9 +1968,12 @@ impl Vm {
             Ok(signals) => {
                 self.signals = Some(signals.handle());
                 let exit_evt = self.exit_evt.try_clone().map_err(Error::EventFdClone)?;
-                let signal_handler_seccomp_filter =
+                let signal_handler_seccomp_filter = get_seccomp_filter(
-                    get_seccomp_filter(&self.seccomp_action, Thread::SignalHandler)
+                    &self.seccomp_action,
-                        .map_err(Error::CreateSeccompFilter)?;
+                    Thread::SignalHandler,
+                    self.hypervisor.hypervisor_type(),
+                )
+                .map_err(Error::CreateSeccompFilter)?;
                 self.threads.push(
                     thread::Builder::new()
                         .name("vm_signal_handler".to_string())