Move virtmgr files to their own directory
Create virtualizationmanager/ folder, move the corresponding files into
it and create build rules.
We also remove now unused dependencies of virtualizationservice.
Bug: 245727626
Test: atest -p packages/modules/Virtualization:avf-presubmit
Change-Id: If33e88bdcdac89e1d451589463ce21e11d6aafc8
diff --git a/virtualizationmanager/Android.bp b/virtualizationmanager/Android.bp
new file mode 100644
index 0000000..a436cea
--- /dev/null
+++ b/virtualizationmanager/Android.bp
@@ -0,0 +1,82 @@
+package {
+ default_applicable_licenses: ["Android-Apache-2.0"],
+}
+
+rust_defaults {
+ name: "virtualizationmanager_defaults",
+ crate_name: "virtualizationmanager",
+ edition: "2021",
+ // Only build on targets which crosvm builds on.
+ enabled: false,
+ target: {
+ android64: {
+ compile_multilib: "64",
+ enabled: true,
+ },
+ linux_bionic_arm64: {
+ enabled: true,
+ },
+ },
+ prefer_rlib: true,
+ rustlibs: [
+ "android.system.virtualizationcommon-rust",
+ "android.system.virtualizationservice-rust",
+ "android.system.virtualizationservice_internal-rust",
+ "android.system.virtualmachineservice-rust",
+ "android.os.permissions_aidl-rust",
+ "libandroid_logger",
+ "libanyhow",
+ "libapkverify",
+ "libbase_rust",
+ "libbinder_rs",
+ "libclap",
+ "libcommand_fds",
+ "libdisk",
+ "liblazy_static",
+ "liblibc",
+ "liblog_rust",
+ "libmicrodroid_metadata",
+ "libmicrodroid_payload_config",
+ "libnested_virt",
+ "libnix",
+ "libonce_cell",
+ "libregex",
+ "librpcbinder_rs",
+ "librustutils",
+ "libsemver",
+ "libselinux_bindgen",
+ "libserde",
+ "libserde_json",
+ "libserde_xml_rs",
+ "libshared_child",
+ "libstatslog_virtualization_rust",
+ "libtombstoned_client_rust",
+ "libvm_control",
+ "libvmconfig",
+ "libzip",
+ "libvsock",
+ // TODO(b/202115393) stabilize the interface
+ "packagemanager_aidl-rust",
+ ],
+ shared_libs: [
+ "libbinder_rpc_unstable",
+ "libselinux",
+ ],
+}
+
+rust_binary {
+ name: "virtmgr",
+ defaults: ["virtualizationmanager_defaults"],
+ srcs: ["src/main.rs"],
+ apex_available: ["com.android.virt"],
+}
+
+rust_test {
+ name: "virtualizationmanager_device_test",
+ srcs: ["src/main.rs"],
+ defaults: ["virtualizationmanager_defaults"],
+ rustlibs: [
+ "libtempfile",
+ ],
+ test_suites: ["general-tests"],
+}
diff --git a/virtualizationmanager/TEST_MAPPING b/virtualizationmanager/TEST_MAPPING
new file mode 100644
index 0000000..a680f6d
--- /dev/null
+++ b/virtualizationmanager/TEST_MAPPING
@@ -0,0 +1,7 @@
+{
+ "avf-presubmit": [
+ {
+ "name": "virtualizationmanager_device_test"
+ }
+ ]
+}
diff --git a/virtualizationmanager/src/aidl.rs b/virtualizationmanager/src/aidl.rs
new file mode 100644
index 0000000..c827c2e
--- /dev/null
+++ b/virtualizationmanager/src/aidl.rs
@@ -0,0 +1,1178 @@
+// Copyright 2021, The Android Open Source Project
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+// http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+//! Implementation of the AIDL interface of the VirtualizationService.
+
+use crate::{get_calling_pid, get_calling_uid};
+use crate::atom::{
+ write_vm_booted_stats, write_vm_creation_stats};
+use crate::composite::make_composite_image;
+use crate::crosvm::{CrosvmConfig, DiskFile, PayloadState, VmContext, VmInstance, VmState};
+use crate::payload::{add_microdroid_payload_images, add_microdroid_system_images};
+use crate::selinux::{getfilecon, SeContext};
+use android_os_permissions_aidl::aidl::android::os::IPermissionController;
+use android_system_virtualizationcommon::aidl::android::system::virtualizationcommon::{
+ DeathReason::DeathReason,
+ ErrorCode::ErrorCode,
+};
+use android_system_virtualizationservice::aidl::android::system::virtualizationservice::{
+ DiskImage::DiskImage,
+ IVirtualMachine::{BnVirtualMachine, IVirtualMachine},
+ IVirtualMachineCallback::IVirtualMachineCallback,
+ IVirtualizationService::IVirtualizationService,
+ MemoryTrimLevel::MemoryTrimLevel,
+ Partition::Partition,
+ PartitionType::PartitionType,
+ VirtualMachineAppConfig::{DebugLevel::DebugLevel, Payload::Payload, VirtualMachineAppConfig},
+ VirtualMachineConfig::VirtualMachineConfig,
+ VirtualMachineDebugInfo::VirtualMachineDebugInfo,
+ VirtualMachinePayloadConfig::VirtualMachinePayloadConfig,
+ VirtualMachineRawConfig::VirtualMachineRawConfig,
+ VirtualMachineState::VirtualMachineState,
+};
+use android_system_virtualizationservice_internal::aidl::android::system::virtualizationservice_internal::IVirtualizationServiceInternal::IVirtualizationServiceInternal;
+use android_system_virtualmachineservice::aidl::android::system::virtualmachineservice::IVirtualMachineService::{
+ BnVirtualMachineService, IVirtualMachineService,
+};
+use anyhow::{bail, Context, Result};
+use apkverify::{HashAlgorithm, V4Signature};
+use binder::{
+ self, wait_for_interface, BinderFeatures, ExceptionCode, Interface, ParcelFileDescriptor,
+ Status, StatusCode, Strong,
+};
+use disk::QcowFile;
+use lazy_static::lazy_static;
+use log::{debug, error, info, warn};
+use microdroid_payload_config::{OsConfig, Task, TaskType, VmPayloadConfig};
+use nix::unistd::pipe;
+use rpcbinder::RpcServer;
+use semver::VersionReq;
+use std::convert::TryInto;
+use std::ffi::CStr;
+use std::fs::{read_dir, remove_file, File, OpenOptions};
+use std::io::{BufRead, BufReader, Error, ErrorKind, Write};
+use std::num::NonZeroU32;
+use std::os::unix::io::{FromRawFd, IntoRawFd};
+use std::os::unix::raw::pid_t;
+use std::path::{Path, PathBuf};
+use std::sync::{Arc, Mutex, Weak};
+use vmconfig::VmConfig;
+use vsock::VsockStream;
+use zip::ZipArchive;
+
+/// The unique ID of a VM used (together with a port number) for vsock communication.
+pub type Cid = u32;
+
+pub const BINDER_SERVICE_IDENTIFIER: &str = "android.system.virtualizationservice";
+
+/// The size of zero.img.
+/// Gaps in composite disk images are filled with a shared zero.img.
+const ZERO_FILLER_SIZE: u64 = 4096;
+
+/// Magic string for the instance image
+const ANDROID_VM_INSTANCE_MAGIC: &str = "Android-VM-instance";
+
+/// Version of the instance image format
+const ANDROID_VM_INSTANCE_VERSION: u16 = 1;
+
+const MICRODROID_OS_NAME: &str = "microdroid";
+
+const UNFORMATTED_STORAGE_MAGIC: &str = "UNFORMATTED-STORAGE";
+
+lazy_static! {
+ pub static ref GLOBAL_SERVICE: Strong<dyn IVirtualizationServiceInternal> =
+ wait_for_interface(BINDER_SERVICE_IDENTIFIER)
+ .expect("Could not connect to VirtualizationServiceInternal");
+}
+
+fn create_or_update_idsig_file(
+ input_fd: &ParcelFileDescriptor,
+ idsig_fd: &ParcelFileDescriptor,
+) -> Result<()> {
+ let mut input = clone_file(input_fd)?;
+ let metadata = input.metadata().context("failed to get input metadata")?;
+ if !metadata.is_file() {
+ bail!("input is not a regular file");
+ }
+ let mut sig = V4Signature::create(&mut input, 4096, &[], HashAlgorithm::SHA256)
+ .context("failed to create idsig")?;
+
+ let mut output = clone_file(idsig_fd)?;
+ output.set_len(0).context("failed to set_len on the idsig output")?;
+ sig.write_into(&mut output).context("failed to write idsig")?;
+ Ok(())
+}
+
+pub fn remove_temporary_files(path: &PathBuf) -> Result<()> {
+ for dir_entry in read_dir(path)? {
+ remove_file(dir_entry?.path())?;
+ }
+ Ok(())
+}
+
+/// Implementation of `IVirtualizationService`, the entry point of the AIDL service.
+#[derive(Debug, Default)]
+pub struct VirtualizationService {
+ state: Arc<Mutex<State>>,
+}
+
+impl Interface for VirtualizationService {
+ fn dump(&self, mut file: &File, _args: &[&CStr]) -> Result<(), StatusCode> {
+ check_permission("android.permission.DUMP").or(Err(StatusCode::PERMISSION_DENIED))?;
+ let state = &mut *self.state.lock().unwrap();
+ let vms = state.vms();
+ writeln!(file, "Running {0} VMs:", vms.len()).or(Err(StatusCode::UNKNOWN_ERROR))?;
+ for vm in vms {
+ writeln!(file, "VM CID: {}", vm.cid).or(Err(StatusCode::UNKNOWN_ERROR))?;
+ writeln!(file, "\tState: {:?}", vm.vm_state.lock().unwrap())
+ .or(Err(StatusCode::UNKNOWN_ERROR))?;
+ writeln!(file, "\tPayload state {:?}", vm.payload_state())
+ .or(Err(StatusCode::UNKNOWN_ERROR))?;
+ writeln!(file, "\tProtected: {}", vm.protected).or(Err(StatusCode::UNKNOWN_ERROR))?;
+ writeln!(file, "\ttemporary_directory: {}", vm.temporary_directory.to_string_lossy())
+ .or(Err(StatusCode::UNKNOWN_ERROR))?;
+ writeln!(file, "\trequester_uid: {}", vm.requester_uid)
+ .or(Err(StatusCode::UNKNOWN_ERROR))?;
+ writeln!(file, "\trequester_debug_pid: {}", vm.requester_debug_pid)
+ .or(Err(StatusCode::UNKNOWN_ERROR))?;
+ }
+ Ok(())
+ }
+}
+
+impl IVirtualizationService for VirtualizationService {
+ /// Creates (but does not start) a new VM with the given configuration, assigning it the next
+ /// available CID.
+ ///
+ /// Returns a binder `IVirtualMachine` object referring to it, as a handle for the client.
+ fn createVm(
+ &self,
+ config: &VirtualMachineConfig,
+ console_fd: Option<&ParcelFileDescriptor>,
+ log_fd: Option<&ParcelFileDescriptor>,
+ ) -> binder::Result<Strong<dyn IVirtualMachine>> {
+ let mut is_protected = false;
+ let ret = self.create_vm_internal(config, console_fd, log_fd, &mut is_protected);
+ write_vm_creation_stats(config, is_protected, &ret);
+ ret
+ }
+
+ /// Initialise an empty partition image of the given size to be used as a writable partition.
+ fn initializeWritablePartition(
+ &self,
+ image_fd: &ParcelFileDescriptor,
+ size: i64,
+ partition_type: PartitionType,
+ ) -> binder::Result<()> {
+ check_manage_access()?;
+ let size = size.try_into().map_err(|e| {
+ Status::new_exception_str(
+ ExceptionCode::ILLEGAL_ARGUMENT,
+ Some(format!("Invalid size {}: {:?}", size, e)),
+ )
+ })?;
+ let image = clone_file(image_fd)?;
+ // initialize the file. Any data in the file will be erased.
+ image.set_len(0).map_err(|e| {
+ Status::new_service_specific_error_str(
+ -1,
+ Some(format!("Failed to reset a file: {:?}", e)),
+ )
+ })?;
+ let mut part = QcowFile::new(image, size).map_err(|e| {
+ Status::new_service_specific_error_str(
+ -1,
+ Some(format!("Failed to create QCOW2 image: {:?}", e)),
+ )
+ })?;
+
+ match partition_type {
+ PartitionType::RAW => Ok(()),
+ PartitionType::ANDROID_VM_INSTANCE => format_as_android_vm_instance(&mut part),
+ PartitionType::ENCRYPTEDSTORE => format_as_encryptedstore(&mut part),
+ _ => Err(Error::new(
+ ErrorKind::Unsupported,
+ format!("Unsupported partition type {:?}", partition_type),
+ )),
+ }
+ .map_err(|e| {
+ Status::new_service_specific_error_str(
+ -1,
+ Some(format!("Failed to initialize partition as {:?}: {:?}", partition_type, e)),
+ )
+ })?;
+
+ Ok(())
+ }
+
+ /// Creates or update the idsig file by digesting the input APK file.
+ fn createOrUpdateIdsigFile(
+ &self,
+ input_fd: &ParcelFileDescriptor,
+ idsig_fd: &ParcelFileDescriptor,
+ ) -> binder::Result<()> {
+ // TODO(b/193504400): do this only when (1) idsig_fd is empty or (2) the APK digest in
+ // idsig_fd is different from APK digest in input_fd
+
+ check_manage_access()?;
+
+ create_or_update_idsig_file(input_fd, idsig_fd)
+ .map_err(|e| Status::new_service_specific_error_str(-1, Some(format!("{:?}", e))))?;
+ Ok(())
+ }
+
+ /// Get a list of all currently running VMs. This method is only intended for debug purposes,
+ /// and as such is only permitted from the shell user.
+ fn debugListVms(&self) -> binder::Result<Vec<VirtualMachineDebugInfo>> {
+ // Delegate to the global service, including checking the debug permission.
+ GLOBAL_SERVICE.debugListVms()
+ }
+}
+
+impl VirtualizationService {
+ pub fn init() -> VirtualizationService {
+ VirtualizationService::default()
+ }
+
+ fn create_vm_context(
+ &self,
+ requester_debug_pid: pid_t,
+ ) -> binder::Result<(VmContext, Cid, PathBuf)> {
+ const NUM_ATTEMPTS: usize = 5;
+
+ for _ in 0..NUM_ATTEMPTS {
+ let vm_context = GLOBAL_SERVICE.allocateGlobalVmContext(requester_debug_pid as i32)?;
+ let cid = vm_context.getCid()? as Cid;
+ let temp_dir: PathBuf = vm_context.getTemporaryDirectory()?.into();
+ let service = VirtualMachineService::new_binder(self.state.clone(), cid).as_binder();
+
+ // Start VM service listening for connections from the new CID on port=CID.
+ let port = cid;
+ match RpcServer::new_vsock(service, cid, port) {
+ Ok(vm_server) => {
+ vm_server.start();
+ return Ok((VmContext::new(vm_context, vm_server), cid, temp_dir));
+ }
+ Err(err) => {
+ warn!("Could not start RpcServer on port {}: {}", port, err);
+ }
+ }
+ }
+ Err(Status::new_service_specific_error_str(
+ -1,
+ Some("Too many attempts to create VM context failed."),
+ ))
+ }
+
+ fn create_vm_internal(
+ &self,
+ config: &VirtualMachineConfig,
+ console_fd: Option<&ParcelFileDescriptor>,
+ log_fd: Option<&ParcelFileDescriptor>,
+ is_protected: &mut bool,
+ ) -> binder::Result<Strong<dyn IVirtualMachine>> {
+ let requester_uid = get_calling_uid();
+ let requester_debug_pid = get_calling_pid();
+
+ // Allocating VM context checks the MANAGE_VIRTUAL_MACHINE permission.
+ let (vm_context, cid, temporary_directory) = self.create_vm_context(requester_debug_pid)?;
+
+ let is_custom = match config {
+ VirtualMachineConfig::RawConfig(_) => true,
+ VirtualMachineConfig::AppConfig(config) => {
+ // Some features are reserved for platform apps only, even when using
+ // VirtualMachineAppConfig:
+ // - controlling CPUs;
+ // - specifying a config file in the APK.
+ !config.taskProfiles.is_empty() || matches!(config.payload, Payload::ConfigPath(_))
+ }
+ };
+ if is_custom {
+ check_use_custom_virtual_machine()?;
+ }
+
+ let state = &mut *self.state.lock().unwrap();
+ let console_fd =
+ clone_or_prepare_logger_fd(config, console_fd, format!("Console({})", cid))?;
+ let log_fd = clone_or_prepare_logger_fd(config, log_fd, format!("Log({})", cid))?;
+
+ // Counter to generate unique IDs for temporary image files.
+ let mut next_temporary_image_id = 0;
+ // Files which are referred to from composite images. These must be mapped to the crosvm
+ // child process, and not closed before it is started.
+ let mut indirect_files = vec![];
+
+ let (is_app_config, config) = match config {
+ VirtualMachineConfig::RawConfig(config) => (false, BorrowedOrOwned::Borrowed(config)),
+ VirtualMachineConfig::AppConfig(config) => {
+ let config = load_app_config(config, &temporary_directory).map_err(|e| {
+ *is_protected = config.protectedVm;
+ let message = format!("Failed to load app config: {:?}", e);
+ error!("{}", message);
+ Status::new_service_specific_error_str(-1, Some(message))
+ })?;
+ (true, BorrowedOrOwned::Owned(config))
+ }
+ };
+ let config = config.as_ref();
+ *is_protected = config.protectedVm;
+
+ // Check if partition images are labeled incorrectly. This is to prevent random images
+ // which are not protected by the Android Verified Boot (e.g. bits downloaded by apps) from
+ // being loaded in a pVM. This applies to everything in the raw config, and everything but
+ // the non-executable, generated partitions in the app config.
+ config
+ .disks
+ .iter()
+ .flat_map(|disk| disk.partitions.iter())
+ .filter(|partition| {
+ if is_app_config {
+ !is_safe_app_partition(&partition.label)
+ } else {
+ true // all partitions are checked
+ }
+ })
+ .try_for_each(check_label_for_partition)
+ .map_err(|e| Status::new_service_specific_error_str(-1, Some(format!("{:?}", e))))?;
+
+ let kernel = maybe_clone_file(&config.kernel)?;
+ let initrd = maybe_clone_file(&config.initrd)?;
+
+ // In a protected VM, we require custom kernels to come from a trusted source (b/237054515).
+ if config.protectedVm {
+ check_label_for_kernel_files(&kernel, &initrd).map_err(|e| {
+ Status::new_service_specific_error_str(-1, Some(format!("{:?}", e)))
+ })?;
+ }
+
+ let zero_filler_path = temporary_directory.join("zero.img");
+ write_zero_filler(&zero_filler_path).map_err(|e| {
+ error!("Failed to make composite image: {:?}", e);
+ Status::new_service_specific_error_str(
+ -1,
+ Some(format!("Failed to make composite image: {:?}", e)),
+ )
+ })?;
+
+ // Assemble disk images if needed.
+ let disks = config
+ .disks
+ .iter()
+ .map(|disk| {
+ assemble_disk_image(
+ disk,
+ &zero_filler_path,
+ &temporary_directory,
+ &mut next_temporary_image_id,
+ &mut indirect_files,
+ )
+ })
+ .collect::<Result<Vec<DiskFile>, _>>()?;
+
+ // Creating this ramdump file unconditionally is not harmful as ramdump will be created
+ // only when the VM is configured as such. `ramdump_write` is sent to crosvm and will
+ // be the backing store for the /dev/hvc1 where VM will emit ramdump to. `ramdump_read`
+ // will be sent back to the client (i.e. the VM owner) for readout.
+ let ramdump_path = temporary_directory.join("ramdump");
+ let ramdump = prepare_ramdump_file(&ramdump_path).map_err(|e| {
+ error!("Failed to prepare ramdump file: {:?}", e);
+ Status::new_service_specific_error_str(
+ -1,
+ Some(format!("Failed to prepare ramdump file: {:?}", e)),
+ )
+ })?;
+
+ // Actually start the VM.
+ let crosvm_config = CrosvmConfig {
+ cid,
+ name: config.name.clone(),
+ bootloader: maybe_clone_file(&config.bootloader)?,
+ kernel,
+ initrd,
+ disks,
+ params: config.params.to_owned(),
+ protected: *is_protected,
+ memory_mib: config.memoryMib.try_into().ok().and_then(NonZeroU32::new),
+ cpus: config.numCpus.try_into().ok().and_then(NonZeroU32::new),
+ task_profiles: config.taskProfiles.clone(),
+ console_fd,
+ log_fd,
+ ramdump: Some(ramdump),
+ indirect_files,
+ platform_version: parse_platform_version_req(&config.platformVersion)?,
+ detect_hangup: is_app_config,
+ };
+ let instance = Arc::new(
+ VmInstance::new(
+ crosvm_config,
+ temporary_directory,
+ requester_uid,
+ requester_debug_pid,
+ vm_context,
+ )
+ .map_err(|e| {
+ error!("Failed to create VM with config {:?}: {:?}", config, e);
+ Status::new_service_specific_error_str(
+ -1,
+ Some(format!("Failed to create VM: {:?}", e)),
+ )
+ })?,
+ );
+ state.add_vm(Arc::downgrade(&instance));
+ Ok(VirtualMachine::create(instance))
+ }
+}
+
+fn write_zero_filler(zero_filler_path: &Path) -> Result<()> {
+ let file = OpenOptions::new()
+ .create_new(true)
+ .read(true)
+ .write(true)
+ .open(zero_filler_path)
+ .with_context(|| "Failed to create zero.img")?;
+ file.set_len(ZERO_FILLER_SIZE)?;
+ Ok(())
+}
+
+fn format_as_android_vm_instance(part: &mut dyn Write) -> std::io::Result<()> {
+ part.write_all(ANDROID_VM_INSTANCE_MAGIC.as_bytes())?;
+ part.write_all(&ANDROID_VM_INSTANCE_VERSION.to_le_bytes())?;
+ part.flush()
+}
+
+fn format_as_encryptedstore(part: &mut dyn Write) -> std::io::Result<()> {
+ part.write_all(UNFORMATTED_STORAGE_MAGIC.as_bytes())?;
+ part.flush()
+}
+
+fn prepare_ramdump_file(ramdump_path: &Path) -> Result<File> {
+ File::create(ramdump_path).context(format!("Failed to create ramdump file {:?}", &ramdump_path))
+}
+
+/// Given the configuration for a disk image, assembles the `DiskFile` to pass to crosvm.
+///
+/// This may involve assembling a composite disk from a set of partition images.
+fn assemble_disk_image(
+ disk: &DiskImage,
+ zero_filler_path: &Path,
+ temporary_directory: &Path,
+ next_temporary_image_id: &mut u64,
+ indirect_files: &mut Vec<File>,
+) -> Result<DiskFile, Status> {
+ let image = if !disk.partitions.is_empty() {
+ if disk.image.is_some() {
+ warn!("DiskImage {:?} contains both image and partitions.", disk);
+ return Err(Status::new_exception_str(
+ ExceptionCode::ILLEGAL_ARGUMENT,
+ Some("DiskImage contains both image and partitions."),
+ ));
+ }
+
+ let composite_image_filenames =
+ make_composite_image_filenames(temporary_directory, next_temporary_image_id);
+ let (image, partition_files) = make_composite_image(
+ &disk.partitions,
+ zero_filler_path,
+ &composite_image_filenames.composite,
+ &composite_image_filenames.header,
+ &composite_image_filenames.footer,
+ )
+ .map_err(|e| {
+ error!("Failed to make composite image with config {:?}: {:?}", disk, e);
+ Status::new_service_specific_error_str(
+ -1,
+ Some(format!("Failed to make composite image: {:?}", e)),
+ )
+ })?;
+
+ // Pass the file descriptors for the various partition files to crosvm when it
+ // is run.
+ indirect_files.extend(partition_files);
+
+ image
+ } else if let Some(image) = &disk.image {
+ clone_file(image)?
+ } else {
+ warn!("DiskImage {:?} didn't contain image or partitions.", disk);
+ return Err(Status::new_exception_str(
+ ExceptionCode::ILLEGAL_ARGUMENT,
+ Some("DiskImage didn't contain image or partitions."),
+ ));
+ };
+
+ Ok(DiskFile { image, writable: disk.writable })
+}
+
+fn load_app_config(
+ config: &VirtualMachineAppConfig,
+ temporary_directory: &Path,
+) -> Result<VirtualMachineRawConfig> {
+ let apk_file = clone_file(config.apk.as_ref().unwrap())?;
+ let idsig_file = clone_file(config.idsig.as_ref().unwrap())?;
+ let instance_file = clone_file(config.instanceImage.as_ref().unwrap())?;
+
+ let storage_image = if let Some(file) = config.encryptedStorageImage.as_ref() {
+ Some(clone_file(file)?)
+ } else {
+ None
+ };
+
+ let vm_payload_config = match &config.payload {
+ Payload::ConfigPath(config_path) => {
+ load_vm_payload_config_from_file(&apk_file, config_path.as_str())
+ .with_context(|| format!("Couldn't read config from {}", config_path))?
+ }
+ Payload::PayloadConfig(payload_config) => create_vm_payload_config(payload_config)?,
+ };
+
+ // For now, the only supported OS is Microdroid
+ let os_name = vm_payload_config.os.name.as_str();
+ if os_name != MICRODROID_OS_NAME {
+ bail!("Unknown OS \"{}\"", os_name);
+ }
+
+ // It is safe to construct a filename based on the os_name because we've already checked that it
+ // is one of the allowed values.
+ let vm_config_path = PathBuf::from(format!("/apex/com.android.virt/etc/{}.json", os_name));
+ let vm_config_file = File::open(vm_config_path)?;
+ let mut vm_config = VmConfig::load(&vm_config_file)?.to_parcelable()?;
+
+ if config.memoryMib > 0 {
+ vm_config.memoryMib = config.memoryMib;
+ }
+
+ vm_config.name = config.name.clone();
+ vm_config.protectedVm = config.protectedVm;
+ vm_config.numCpus = config.numCpus;
+ vm_config.taskProfiles = config.taskProfiles.clone();
+
+ // Microdroid takes additional init ramdisk & (optionally) storage image
+ add_microdroid_system_images(config, instance_file, storage_image, &mut vm_config)?;
+
+ // Include Microdroid payload disk (contains apks, idsigs) in vm config
+ add_microdroid_payload_images(
+ config,
+ temporary_directory,
+ apk_file,
+ idsig_file,
+ &vm_payload_config,
+ &mut vm_config,
+ )?;
+
+ Ok(vm_config)
+}
+
+fn load_vm_payload_config_from_file(apk_file: &File, config_path: &str) -> Result<VmPayloadConfig> {
+ let mut apk_zip = ZipArchive::new(apk_file)?;
+ let config_file = apk_zip.by_name(config_path)?;
+ Ok(serde_json::from_reader(config_file)?)
+}
+
+fn create_vm_payload_config(
+ payload_config: &VirtualMachinePayloadConfig,
+) -> Result<VmPayloadConfig> {
+ // There isn't an actual config file. Construct a synthetic VmPayloadConfig from the explicit
+ // parameters we've been given. Microdroid will do something equivalent inside the VM using the
+ // payload config that we send it via the metadata file.
+
+ let payload_binary_name = &payload_config.payloadBinaryName;
+ if payload_binary_name.contains('/') {
+ bail!("Payload binary name must not specify a path: {payload_binary_name}");
+ }
+
+ let task = Task { type_: TaskType::MicrodroidLauncher, command: payload_binary_name.clone() };
+ Ok(VmPayloadConfig {
+ os: OsConfig { name: MICRODROID_OS_NAME.to_owned() },
+ task: Some(task),
+ apexes: vec![],
+ extra_apks: vec![],
+ prefer_staged: false,
+ export_tombstones: false,
+ enable_authfs: false,
+ })
+}
+
+/// Generates a unique filename to use for a composite disk image.
+fn make_composite_image_filenames(
+ temporary_directory: &Path,
+ next_temporary_image_id: &mut u64,
+) -> CompositeImageFilenames {
+ let id = *next_temporary_image_id;
+ *next_temporary_image_id += 1;
+ CompositeImageFilenames {
+ composite: temporary_directory.join(format!("composite-{}.img", id)),
+ header: temporary_directory.join(format!("composite-{}-header.img", id)),
+ footer: temporary_directory.join(format!("composite-{}-footer.img", id)),
+ }
+}
+
+/// Filenames for a composite disk image, including header and footer partitions.
+#[derive(Clone, Debug, Eq, PartialEq)]
+struct CompositeImageFilenames {
+ /// The composite disk image itself.
+ composite: PathBuf,
+ /// The header partition image.
+ header: PathBuf,
+ /// The footer partition image.
+ footer: PathBuf,
+}
+
+/// Checks whether the caller has a specific permission
+fn check_permission(perm: &str) -> binder::Result<()> {
+ let calling_pid = get_calling_pid();
+ let calling_uid = get_calling_uid();
+ // Root can do anything
+ if calling_uid == 0 {
+ return Ok(());
+ }
+ let perm_svc: Strong<dyn IPermissionController::IPermissionController> =
+ binder::get_interface("permission")?;
+ if perm_svc.checkPermission(perm, calling_pid, calling_uid as i32)? {
+ Ok(())
+ } else {
+ Err(Status::new_exception_str(
+ ExceptionCode::SECURITY,
+ Some(format!("does not have the {} permission", perm)),
+ ))
+ }
+}
+
+/// Check whether the caller of the current Binder method is allowed to manage VMs
+fn check_manage_access() -> binder::Result<()> {
+ check_permission("android.permission.MANAGE_VIRTUAL_MACHINE")
+}
+
+/// Check whether the caller of the current Binder method is allowed to create custom VMs
+fn check_use_custom_virtual_machine() -> binder::Result<()> {
+ check_permission("android.permission.USE_CUSTOM_VIRTUAL_MACHINE")
+}
+
+/// Return whether a partition is exempt from selinux label checks, because we know that it does
+/// not contain code and is likely to be generated in an app-writable directory.
+fn is_safe_app_partition(label: &str) -> bool {
+ // See add_microdroid_system_images & add_microdroid_payload_images in payload.rs.
+ label == "vm-instance"
+ || label == "encryptedstore"
+ || label == "microdroid-apk-idsig"
+ || label == "payload-metadata"
+ || label.starts_with("extra-idsig-")
+}
+
+/// Check that a file SELinux label is acceptable.
+///
+/// We only want to allow code in a VM to be sourced from places that apps, and the
+/// system, do not have write access to.
+///
+/// Note that sepolicy must also grant read access for these types to both virtualization
+/// service and crosvm.
+///
+/// App private data files are deliberately excluded, to avoid arbitrary payloads being run on
+/// user devices (W^X).
+fn check_label_is_allowed(context: &SeContext) -> Result<()> {
+ match context.selinux_type()? {
+ | "system_file" // immutable dm-verity protected partition
+ | "apk_data_file" // APKs of an installed app
+ | "staging_data_file" // updated/staged APEX images
+ | "shell_data_file" // test files created via adb shell
+ => Ok(()),
+ _ => bail!("Label {} is not allowed", context),
+ }
+}
+
+fn check_label_for_partition(partition: &Partition) -> Result<()> {
+ let file = partition.image.as_ref().unwrap().as_ref();
+ check_label_is_allowed(&getfilecon(file)?)
+ .with_context(|| format!("Partition {} invalid", &partition.label))
+}
+
+fn check_label_for_kernel_files(kernel: &Option<File>, initrd: &Option<File>) -> Result<()> {
+ if let Some(f) = kernel {
+ check_label_for_file(f, "kernel")?;
+ }
+ if let Some(f) = initrd {
+ check_label_for_file(f, "initrd")?;
+ }
+ Ok(())
+}
+fn check_label_for_file(file: &File, name: &str) -> Result<()> {
+ check_label_is_allowed(&getfilecon(file)?).with_context(|| format!("{} file invalid", name))
+}
+
+/// Implementation of the AIDL `IVirtualMachine` interface. Used as a handle to a VM.
+#[derive(Debug)]
+struct VirtualMachine {
+ instance: Arc<VmInstance>,
+}
+
+impl VirtualMachine {
+ fn create(instance: Arc<VmInstance>) -> Strong<dyn IVirtualMachine> {
+ BnVirtualMachine::new_binder(VirtualMachine { instance }, BinderFeatures::default())
+ }
+}
+
+impl Interface for VirtualMachine {}
+
+impl IVirtualMachine for VirtualMachine {
+ fn getCid(&self) -> binder::Result<i32> {
+ // Don't check permission. The owner of the VM might have passed this binder object to
+ // others.
+ Ok(self.instance.cid as i32)
+ }
+
+ fn getState(&self) -> binder::Result<VirtualMachineState> {
+ // Don't check permission. The owner of the VM might have passed this binder object to
+ // others.
+ Ok(get_state(&self.instance))
+ }
+
+ fn registerCallback(
+ &self,
+ callback: &Strong<dyn IVirtualMachineCallback>,
+ ) -> binder::Result<()> {
+ // Don't check permission. The owner of the VM might have passed this binder object to
+ // others.
+ //
+ // TODO: Should this give an error if the VM is already dead?
+ self.instance.callbacks.add(callback.clone());
+ Ok(())
+ }
+
+ fn start(&self) -> binder::Result<()> {
+ self.instance.start().map_err(|e| {
+ error!("Error starting VM with CID {}: {:?}", self.instance.cid, e);
+ Status::new_service_specific_error_str(-1, Some(e.to_string()))
+ })
+ }
+
+ fn stop(&self) -> binder::Result<()> {
+ self.instance.kill().map_err(|e| {
+ error!("Error stopping VM with CID {}: {:?}", self.instance.cid, e);
+ Status::new_service_specific_error_str(-1, Some(e.to_string()))
+ })
+ }
+
+ fn onTrimMemory(&self, level: MemoryTrimLevel) -> binder::Result<()> {
+ self.instance.trim_memory(level).map_err(|e| {
+ error!("Error trimming VM with CID {}: {:?}", self.instance.cid, e);
+ Status::new_service_specific_error_str(-1, Some(e.to_string()))
+ })
+ }
+
+ fn connectVsock(&self, port: i32) -> binder::Result<ParcelFileDescriptor> {
+ if !matches!(&*self.instance.vm_state.lock().unwrap(), VmState::Running { .. }) {
+ return Err(Status::new_service_specific_error_str(-1, Some("VM is not running")));
+ }
+ let port = port as u32;
+ if port < 1024 {
+ return Err(Status::new_service_specific_error_str(
+ -1,
+ Some(format!("Can't connect to privileged port {port}")),
+ ));
+ }
+ let stream = VsockStream::connect_with_cid_port(self.instance.cid, port).map_err(|e| {
+ Status::new_service_specific_error_str(-1, Some(format!("Failed to connect: {:?}", e)))
+ })?;
+ Ok(vsock_stream_to_pfd(stream))
+ }
+}
+
+impl Drop for VirtualMachine {
+ fn drop(&mut self) {
+ debug!("Dropping {:?}", self);
+ if let Err(e) = self.instance.kill() {
+ debug!("Error stopping dropped VM with CID {}: {:?}", self.instance.cid, e);
+ }
+ }
+}
+
+/// A set of Binders to be called back in response to various events on the VM, such as when it
+/// dies.
+#[derive(Debug, Default)]
+pub struct VirtualMachineCallbacks(Mutex<Vec<Strong<dyn IVirtualMachineCallback>>>);
+
+impl VirtualMachineCallbacks {
+ /// Call all registered callbacks to notify that the payload has started.
+ pub fn notify_payload_started(&self, cid: Cid) {
+ let callbacks = &*self.0.lock().unwrap();
+ for callback in callbacks {
+ if let Err(e) = callback.onPayloadStarted(cid as i32) {
+ error!("Error notifying payload start event from VM CID {}: {:?}", cid, e);
+ }
+ }
+ }
+
+ /// Call all registered callbacks to notify that the payload is ready to serve.
+ pub fn notify_payload_ready(&self, cid: Cid) {
+ let callbacks = &*self.0.lock().unwrap();
+ for callback in callbacks {
+ if let Err(e) = callback.onPayloadReady(cid as i32) {
+ error!("Error notifying payload ready event from VM CID {}: {:?}", cid, e);
+ }
+ }
+ }
+
+ /// Call all registered callbacks to notify that the payload has finished.
+ pub fn notify_payload_finished(&self, cid: Cid, exit_code: i32) {
+ let callbacks = &*self.0.lock().unwrap();
+ for callback in callbacks {
+ if let Err(e) = callback.onPayloadFinished(cid as i32, exit_code) {
+ error!("Error notifying payload finish event from VM CID {}: {:?}", cid, e);
+ }
+ }
+ }
+
+ /// Call all registered callbacks to say that the VM encountered an error.
+ pub fn notify_error(&self, cid: Cid, error_code: ErrorCode, message: &str) {
+ let callbacks = &*self.0.lock().unwrap();
+ for callback in callbacks {
+ if let Err(e) = callback.onError(cid as i32, error_code, message) {
+ error!("Error notifying error event from VM CID {}: {:?}", cid, e);
+ }
+ }
+ }
+
+ /// Call all registered callbacks to say that the VM has died.
+ pub fn callback_on_died(&self, cid: Cid, reason: DeathReason) {
+ let callbacks = &*self.0.lock().unwrap();
+ for callback in callbacks {
+ if let Err(e) = callback.onDied(cid as i32, reason) {
+ error!("Error notifying exit of VM CID {}: {:?}", cid, e);
+ }
+ }
+ }
+
+ /// Add a new callback to the set.
+ fn add(&self, callback: Strong<dyn IVirtualMachineCallback>) {
+ self.0.lock().unwrap().push(callback);
+ }
+}
+
+/// The mutable state of the VirtualizationService. There should only be one instance of this
+/// struct.
+#[derive(Debug, Default)]
+struct State {
+ /// The VMs which have been started. When VMs are started a weak reference is added to this list
+ /// while a strong reference is returned to the caller over Binder. Once all copies of the
+ /// Binder client are dropped the weak reference here will become invalid, and will be removed
+ /// from the list opportunistically the next time `add_vm` is called.
+ vms: Vec<Weak<VmInstance>>,
+}
+
+impl State {
+ /// Get a list of VMs which still have Binder references to them.
+ fn vms(&self) -> Vec<Arc<VmInstance>> {
+ // Attempt to upgrade the weak pointers to strong pointers.
+ self.vms.iter().filter_map(Weak::upgrade).collect()
+ }
+
+ /// Add a new VM to the list.
+ fn add_vm(&mut self, vm: Weak<VmInstance>) {
+ // Garbage collect any entries from the stored list which no longer exist.
+ self.vms.retain(|vm| vm.strong_count() > 0);
+
+ // Actually add the new VM.
+ self.vms.push(vm);
+ }
+
+ /// Get a VM that corresponds to the given cid
+ fn get_vm(&self, cid: Cid) -> Option<Arc<VmInstance>> {
+ self.vms().into_iter().find(|vm| vm.cid == cid)
+ }
+}
+
+/// Gets the `VirtualMachineState` of the given `VmInstance`.
+fn get_state(instance: &VmInstance) -> VirtualMachineState {
+ match &*instance.vm_state.lock().unwrap() {
+ VmState::NotStarted { .. } => VirtualMachineState::NOT_STARTED,
+ VmState::Running { .. } => match instance.payload_state() {
+ PayloadState::Starting => VirtualMachineState::STARTING,
+ PayloadState::Started => VirtualMachineState::STARTED,
+ PayloadState::Ready => VirtualMachineState::READY,
+ PayloadState::Finished => VirtualMachineState::FINISHED,
+ PayloadState::Hangup => VirtualMachineState::DEAD,
+ },
+ VmState::Dead => VirtualMachineState::DEAD,
+ VmState::Failed => VirtualMachineState::DEAD,
+ }
+}
+
+/// Converts a `&ParcelFileDescriptor` to a `File` by cloning the file.
+pub fn clone_file(file: &ParcelFileDescriptor) -> Result<File, Status> {
+ file.as_ref().try_clone().map_err(|e| {
+ Status::new_exception_str(
+ ExceptionCode::BAD_PARCELABLE,
+ Some(format!("Failed to clone File from ParcelFileDescriptor: {:?}", e)),
+ )
+ })
+}
+
+/// Converts an `&Option<ParcelFileDescriptor>` to an `Option<File>` by cloning the file.
+fn maybe_clone_file(file: &Option<ParcelFileDescriptor>) -> Result<Option<File>, Status> {
+ file.as_ref().map(clone_file).transpose()
+}
+
+/// Converts a `VsockStream` to a `ParcelFileDescriptor`.
+fn vsock_stream_to_pfd(stream: VsockStream) -> ParcelFileDescriptor {
+ // SAFETY: ownership is transferred from stream to f
+ let f = unsafe { File::from_raw_fd(stream.into_raw_fd()) };
+ ParcelFileDescriptor::new(f)
+}
+
+/// Parses the platform version requirement string.
+fn parse_platform_version_req(s: &str) -> Result<VersionReq, Status> {
+ VersionReq::parse(s).map_err(|e| {
+ Status::new_exception_str(
+ ExceptionCode::BAD_PARCELABLE,
+ Some(format!("Invalid platform version requirement {}: {:?}", s, e)),
+ )
+ })
+}
+
+fn is_debuggable(config: &VirtualMachineConfig) -> bool {
+ match config {
+ VirtualMachineConfig::AppConfig(config) => config.debugLevel != DebugLevel::NONE,
+ _ => false,
+ }
+}
+
+fn clone_or_prepare_logger_fd(
+ config: &VirtualMachineConfig,
+ fd: Option<&ParcelFileDescriptor>,
+ tag: String,
+) -> Result<Option<File>, Status> {
+ if let Some(fd) = fd {
+ return Ok(Some(clone_file(fd)?));
+ }
+
+ if !is_debuggable(config) {
+ return Ok(None);
+ }
+
+ let (raw_read_fd, raw_write_fd) = pipe().map_err(|e| {
+ Status::new_service_specific_error_str(-1, Some(format!("Failed to create pipe: {:?}", e)))
+ })?;
+
+ // SAFETY: We are the sole owners of these fds as they were just created.
+ let mut reader = BufReader::new(unsafe { File::from_raw_fd(raw_read_fd) });
+ let write_fd = unsafe { File::from_raw_fd(raw_write_fd) };
+
+ std::thread::spawn(move || loop {
+ let mut buf = vec![];
+ match reader.read_until(b'\n', &mut buf) {
+ Ok(0) => {
+ // EOF
+ return;
+ }
+ Ok(size) => {
+ if buf[size - 1] == b'\n' {
+ buf.pop();
+ }
+ info!("{}: {}", &tag, &String::from_utf8_lossy(&buf));
+ }
+ Err(e) => {
+ error!("Could not read console pipe: {:?}", e);
+ return;
+ }
+ };
+ });
+
+ Ok(Some(write_fd))
+}
+
+/// Simple utility for referencing Borrowed or Owned. Similar to std::borrow::Cow, but
+/// it doesn't require that T implements Clone.
+enum BorrowedOrOwned<'a, T> {
+ Borrowed(&'a T),
+ Owned(T),
+}
+
+impl<'a, T> AsRef<T> for BorrowedOrOwned<'a, T> {
+ fn as_ref(&self) -> &T {
+ match self {
+ Self::Borrowed(b) => b,
+ Self::Owned(o) => o,
+ }
+ }
+}
+
+/// Implementation of `IVirtualMachineService`, the entry point of the AIDL service.
+#[derive(Debug, Default)]
+struct VirtualMachineService {
+ state: Arc<Mutex<State>>,
+ cid: Cid,
+}
+
+impl Interface for VirtualMachineService {}
+
+impl IVirtualMachineService for VirtualMachineService {
+ fn notifyPayloadStarted(&self) -> binder::Result<()> {
+ let cid = self.cid;
+ if let Some(vm) = self.state.lock().unwrap().get_vm(cid) {
+ info!("VM with CID {} started payload", cid);
+ vm.update_payload_state(PayloadState::Started).map_err(|e| {
+ Status::new_exception_str(ExceptionCode::ILLEGAL_STATE, Some(e.to_string()))
+ })?;
+ vm.callbacks.notify_payload_started(cid);
+
+ let vm_start_timestamp = vm.vm_metric.lock().unwrap().start_timestamp;
+ write_vm_booted_stats(vm.requester_uid as i32, &vm.name, vm_start_timestamp);
+ Ok(())
+ } else {
+ error!("notifyPayloadStarted is called from an unknown CID {}", cid);
+ Err(Status::new_service_specific_error_str(
+ -1,
+ Some(format!("cannot find a VM with CID {}", cid)),
+ ))
+ }
+ }
+
+ fn notifyPayloadReady(&self) -> binder::Result<()> {
+ let cid = self.cid;
+ if let Some(vm) = self.state.lock().unwrap().get_vm(cid) {
+ info!("VM with CID {} reported payload is ready", cid);
+ vm.update_payload_state(PayloadState::Ready).map_err(|e| {
+ Status::new_exception_str(ExceptionCode::ILLEGAL_STATE, Some(e.to_string()))
+ })?;
+ vm.callbacks.notify_payload_ready(cid);
+ Ok(())
+ } else {
+ error!("notifyPayloadReady is called from an unknown CID {}", cid);
+ Err(Status::new_service_specific_error_str(
+ -1,
+ Some(format!("cannot find a VM with CID {}", cid)),
+ ))
+ }
+ }
+
+ fn notifyPayloadFinished(&self, exit_code: i32) -> binder::Result<()> {
+ let cid = self.cid;
+ if let Some(vm) = self.state.lock().unwrap().get_vm(cid) {
+ info!("VM with CID {} finished payload", cid);
+ vm.update_payload_state(PayloadState::Finished).map_err(|e| {
+ Status::new_exception_str(ExceptionCode::ILLEGAL_STATE, Some(e.to_string()))
+ })?;
+ vm.callbacks.notify_payload_finished(cid, exit_code);
+ Ok(())
+ } else {
+ error!("notifyPayloadFinished is called from an unknown CID {}", cid);
+ Err(Status::new_service_specific_error_str(
+ -1,
+ Some(format!("cannot find a VM with CID {}", cid)),
+ ))
+ }
+ }
+
+ fn notifyError(&self, error_code: ErrorCode, message: &str) -> binder::Result<()> {
+ let cid = self.cid;
+ if let Some(vm) = self.state.lock().unwrap().get_vm(cid) {
+ info!("VM with CID {} encountered an error", cid);
+ vm.update_payload_state(PayloadState::Finished).map_err(|e| {
+ Status::new_exception_str(ExceptionCode::ILLEGAL_STATE, Some(e.to_string()))
+ })?;
+ vm.callbacks.notify_error(cid, error_code, message);
+ Ok(())
+ } else {
+ error!("notifyError is called from an unknown CID {}", cid);
+ Err(Status::new_service_specific_error_str(
+ -1,
+ Some(format!("cannot find a VM with CID {}", cid)),
+ ))
+ }
+ }
+}
+
+impl VirtualMachineService {
+ fn new_binder(state: Arc<Mutex<State>>, cid: Cid) -> Strong<dyn IVirtualMachineService> {
+ BnVirtualMachineService::new_binder(
+ VirtualMachineService { state, cid },
+ BinderFeatures::default(),
+ )
+ }
+}
+
+#[cfg(test)]
+mod tests {
+ use super::*;
+
+ #[test]
+ fn test_is_allowed_label_for_partition() -> Result<()> {
+ let expected_results = vec![
+ ("u:object_r:system_file:s0", true),
+ ("u:object_r:apk_data_file:s0", true),
+ ("u:object_r:app_data_file:s0", false),
+ ("u:object_r:app_data_file:s0:c512,c768", false),
+ ("u:object_r:privapp_data_file:s0:c512,c768", false),
+ ("invalid", false),
+ ("user:role:apk_data_file:severity:categories", true),
+ ("user:role:apk_data_file:severity:categories:extraneous", false),
+ ];
+
+ for (label, expected_valid) in expected_results {
+ let context = SeContext::new(label)?;
+ let result = check_label_is_allowed(&context);
+ if expected_valid {
+ assert!(result.is_ok(), "Expected label {} to be allowed, got {:?}", label, result);
+ } else if result.is_ok() {
+ bail!("Expected label {} to be disallowed", label);
+ }
+ }
+ Ok(())
+ }
+
+ #[test]
+ fn test_create_or_update_idsig_file_empty_apk() -> Result<()> {
+ let apk = tempfile::tempfile().unwrap();
+ let idsig = tempfile::tempfile().unwrap();
+
+ let ret = create_or_update_idsig_file(
+ &ParcelFileDescriptor::new(apk),
+ &ParcelFileDescriptor::new(idsig),
+ );
+ assert!(ret.is_err(), "should fail");
+ Ok(())
+ }
+
+ #[test]
+ fn test_create_or_update_idsig_dir_instead_of_file_for_apk() -> Result<()> {
+ let tmp_dir = tempfile::TempDir::new().unwrap();
+ let apk = File::open(tmp_dir.path()).unwrap();
+ let idsig = tempfile::tempfile().unwrap();
+
+ let ret = create_or_update_idsig_file(
+ &ParcelFileDescriptor::new(apk),
+ &ParcelFileDescriptor::new(idsig),
+ );
+ assert!(ret.is_err(), "should fail");
+ Ok(())
+ }
+
+ /// Verifies that create_or_update_idsig_file won't oom if a fd that corresponds to a directory
+ /// on ext4 filesystem is passed.
+ /// On ext4 lseek on a directory fd will return (off_t)-1 (see:
+ /// https://bugzilla.kernel.org/show_bug.cgi?id=200043), which will result in
+ /// create_or_update_idsig_file ooming while attempting to allocate petabytes of memory.
+ #[test]
+ fn test_create_or_update_idsig_does_not_crash_dir_on_ext4() -> Result<()> {
+ // APEXes are backed by the ext4.
+ let apk = File::open("/apex/com.android.virt/").unwrap();
+ let idsig = tempfile::tempfile().unwrap();
+
+ let ret = create_or_update_idsig_file(
+ &ParcelFileDescriptor::new(apk),
+ &ParcelFileDescriptor::new(idsig),
+ );
+ assert!(ret.is_err(), "should fail");
+ Ok(())
+ }
+}
diff --git a/virtualizationmanager/src/atom.rs b/virtualizationmanager/src/atom.rs
new file mode 100644
index 0000000..c33f262
--- /dev/null
+++ b/virtualizationmanager/src/atom.rs
@@ -0,0 +1,186 @@
+// Copyright 2022, The Android Open Source Project
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+// http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+//! Functions for creating and collecting atoms.
+
+use crate::aidl::{clone_file, GLOBAL_SERVICE};
+use crate::crosvm::VmMetric;
+use crate::get_calling_uid;
+use android_system_virtualizationcommon::aidl::android::system::virtualizationcommon::DeathReason::DeathReason;
+use android_system_virtualizationservice::aidl::android::system::virtualizationservice::{
+ IVirtualMachine::IVirtualMachine,
+ VirtualMachineAppConfig::{Payload::Payload, VirtualMachineAppConfig},
+ VirtualMachineConfig::VirtualMachineConfig,
+};
+use android_system_virtualizationservice::binder::{Status, Strong};
+use android_system_virtualizationservice_internal::aidl::android::system::virtualizationservice_internal::{
+ AtomVmBooted::AtomVmBooted,
+ AtomVmCreationRequested::AtomVmCreationRequested,
+ AtomVmExited::AtomVmExited,
+};
+use anyhow::{anyhow, Result};
+use binder::ParcelFileDescriptor;
+use log::warn;
+use microdroid_payload_config::VmPayloadConfig;
+use statslog_virtualization_rust::vm_creation_requested;
+use std::thread;
+use std::time::{Duration, SystemTime};
+use zip::ZipArchive;
+
+fn get_apex_list(config: &VirtualMachineAppConfig) -> String {
+ match &config.payload {
+ Payload::PayloadConfig(_) => String::new(),
+ Payload::ConfigPath(config_path) => {
+ let vm_payload_config = get_vm_payload_config(&config.apk, config_path);
+ if let Ok(vm_payload_config) = vm_payload_config {
+ vm_payload_config
+ .apexes
+ .iter()
+ .map(|x| x.name.clone())
+ .collect::<Vec<String>>()
+ .join(":")
+ } else {
+ "INFO: Can't get VmPayloadConfig".to_owned()
+ }
+ }
+ }
+}
+
+fn get_vm_payload_config(
+ apk_fd: &Option<ParcelFileDescriptor>,
+ config_path: &str,
+) -> Result<VmPayloadConfig> {
+ let apk = apk_fd.as_ref().ok_or_else(|| anyhow!("APK is none"))?;
+ let apk_file = clone_file(apk)?;
+ let mut apk_zip = ZipArchive::new(&apk_file)?;
+ let config_file = apk_zip.by_name(config_path)?;
+ let vm_payload_config: VmPayloadConfig = serde_json::from_reader(config_file)?;
+ Ok(vm_payload_config)
+}
+
+fn get_duration(vm_start_timestamp: Option<SystemTime>) -> Duration {
+ match vm_start_timestamp {
+ Some(vm_start_timestamp) => vm_start_timestamp.elapsed().unwrap_or_default(),
+ None => Duration::default(),
+ }
+}
+
+/// Write the stats of VMCreation to statsd
+pub fn write_vm_creation_stats(
+ config: &VirtualMachineConfig,
+ is_protected: bool,
+ ret: &binder::Result<Strong<dyn IVirtualMachine>>,
+) {
+ let creation_succeeded;
+ let binder_exception_code;
+ match ret {
+ Ok(_) => {
+ creation_succeeded = true;
+ binder_exception_code = Status::ok().exception_code() as i32;
+ }
+ Err(ref e) => {
+ creation_succeeded = false;
+ binder_exception_code = e.exception_code() as i32;
+ }
+ }
+ let (vm_identifier, config_type, num_cpus, memory_mib, apexes) = match config {
+ VirtualMachineConfig::AppConfig(config) => (
+ config.name.clone(),
+ vm_creation_requested::ConfigType::VirtualMachineAppConfig,
+ config.numCpus,
+ config.memoryMib,
+ get_apex_list(config),
+ ),
+ VirtualMachineConfig::RawConfig(config) => (
+ config.name.clone(),
+ vm_creation_requested::ConfigType::VirtualMachineRawConfig,
+ config.numCpus,
+ config.memoryMib,
+ String::new(),
+ ),
+ };
+
+ let atom = AtomVmCreationRequested {
+ uid: get_calling_uid() as i32,
+ vmIdentifier: vm_identifier,
+ isProtected: is_protected,
+ creationSucceeded: creation_succeeded,
+ binderExceptionCode: binder_exception_code,
+ configType: config_type as i32,
+ numCpus: num_cpus,
+ memoryMib: memory_mib,
+ apexes,
+ };
+
+ thread::spawn(move || {
+ GLOBAL_SERVICE.atomVmCreationRequested(&atom).unwrap_or_else(|e| {
+ warn!("Failed to write VmCreationRequested atom: {e}");
+ });
+ });
+}
+
+/// Write the stats of VM boot to statsd
+/// The function creates a separate thread which waits fro statsd to start to push atom
+pub fn write_vm_booted_stats(
+ uid: i32,
+ vm_identifier: &str,
+ vm_start_timestamp: Option<SystemTime>,
+) {
+ let vm_identifier = vm_identifier.to_owned();
+ let duration = get_duration(vm_start_timestamp);
+
+ let atom = AtomVmBooted {
+ uid,
+ vmIdentifier: vm_identifier,
+ elapsedTimeMillis: duration.as_millis() as i64,
+ };
+
+ thread::spawn(move || {
+ GLOBAL_SERVICE.atomVmBooted(&atom).unwrap_or_else(|e| {
+ warn!("Failed to write VmCreationRequested atom: {e}");
+ });
+ });
+}
+
+/// Write the stats of VM exit to statsd
+/// The function creates a separate thread which waits fro statsd to start to push atom
+pub fn write_vm_exited_stats(
+ uid: i32,
+ vm_identifier: &str,
+ reason: DeathReason,
+ exit_signal: Option<i32>,
+ vm_metric: &VmMetric,
+) {
+ let vm_identifier = vm_identifier.to_owned();
+ let elapsed_time_millis = get_duration(vm_metric.start_timestamp).as_millis() as i64;
+ let guest_time_millis = vm_metric.cpu_guest_time.unwrap_or_default();
+ let rss = vm_metric.rss.unwrap_or_default();
+
+ let atom = AtomVmExited {
+ uid,
+ vmIdentifier: vm_identifier,
+ elapsedTimeMillis: elapsed_time_millis,
+ deathReason: reason,
+ guestTimeMillis: guest_time_millis,
+ rssVmKb: rss.vm,
+ rssCrosvmKb: rss.crosvm,
+ exitSignal: exit_signal.unwrap_or_default(),
+ };
+
+ thread::spawn(move || {
+ GLOBAL_SERVICE.atomVmExited(&atom).unwrap_or_else(|e| {
+ warn!("Failed to write VmExited atom: {e}");
+ });
+ });
+}
diff --git a/virtualizationmanager/src/composite.rs b/virtualizationmanager/src/composite.rs
new file mode 100644
index 0000000..fe17ff4
--- /dev/null
+++ b/virtualizationmanager/src/composite.rs
@@ -0,0 +1,133 @@
+// Copyright 2021, The Android Open Source Project
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+// http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+//! Functions for creating a composite disk image.
+
+use android_system_virtualizationservice::aidl::android::system::virtualizationservice::Partition::Partition;
+use anyhow::{anyhow, Context, Error};
+use disk::{
+ create_composite_disk, create_disk_file, ImagePartitionType, PartitionInfo, MAX_NESTING_DEPTH,
+};
+use std::fs::{File, OpenOptions};
+use std::os::unix::io::AsRawFd;
+use std::path::{Path, PathBuf};
+
+/// Constructs a composite disk image for the given list of partitions, and opens it ready to use.
+///
+/// Returns the composite disk image file, and a list of files whose file descriptors must be passed
+/// to any process which wants to use it. This is necessary because the composite image contains
+/// paths of the form `/proc/self/fd/N` for the partition images.
+pub fn make_composite_image(
+ partitions: &[Partition],
+ zero_filler_path: &Path,
+ output_path: &Path,
+ header_path: &Path,
+ footer_path: &Path,
+) -> Result<(File, Vec<File>), Error> {
+ let (partitions, mut files) = convert_partitions(partitions)?;
+
+ let mut composite_image = OpenOptions::new()
+ .create_new(true)
+ .read(true)
+ .write(true)
+ .open(output_path)
+ .with_context(|| format!("Failed to create composite image {:?}", output_path))?;
+ let mut header_file =
+ OpenOptions::new().create_new(true).read(true).write(true).open(header_path).with_context(
+ || format!("Failed to create composite image header {:?}", header_path),
+ )?;
+ let mut footer_file =
+ OpenOptions::new().create_new(true).read(true).write(true).open(footer_path).with_context(
+ || format!("Failed to create composite image header {:?}", footer_path),
+ )?;
+ let zero_filler_file = File::open(zero_filler_path).with_context(|| {
+ format!("Failed to open composite image zero filler {:?}", zero_filler_path)
+ })?;
+
+ create_composite_disk(
+ &partitions,
+ &fd_path_for_file(&zero_filler_file),
+ &fd_path_for_file(&header_file),
+ &mut header_file,
+ &fd_path_for_file(&footer_file),
+ &mut footer_file,
+ &mut composite_image,
+ )?;
+
+ // Re-open the composite image as read-only.
+ let composite_image = File::open(output_path)
+ .with_context(|| format!("Failed to open composite image {:?}", output_path))?;
+
+ files.push(header_file);
+ files.push(footer_file);
+ files.push(zero_filler_file);
+
+ Ok((composite_image, files))
+}
+
+/// Given the AIDL config containing a list of partitions, with a [`ParcelFileDescriptor`] for each
+/// partition, returns the corresponding list of PartitionInfo and the list of files whose file
+/// descriptors must be passed to any process using the composite image.
+fn convert_partitions(partitions: &[Partition]) -> Result<(Vec<PartitionInfo>, Vec<File>), Error> {
+ // File descriptors to pass to child process.
+ let mut files = vec![];
+
+ let partitions = partitions
+ .iter()
+ .map(|partition| {
+ // TODO(b/187187765): This shouldn't be an Option.
+ let file = partition
+ .image
+ .as_ref()
+ .context("Invalid partition image file descriptor")?
+ .as_ref()
+ .try_clone()
+ .context("Failed to clone partition image file descriptor")?;
+ let path = fd_path_for_file(&file);
+ let size = get_partition_size(&file, &path)?;
+ files.push(file);
+
+ Ok(PartitionInfo {
+ label: partition.label.to_owned(),
+ path,
+ partition_type: ImagePartitionType::LinuxFilesystem,
+ writable: partition.writable,
+ size,
+ })
+ })
+ .collect::<Result<_, Error>>()?;
+
+ Ok((partitions, files))
+}
+
+fn fd_path_for_file(file: &File) -> PathBuf {
+ let fd = file.as_raw_fd();
+ format!("/proc/self/fd/{}", fd).into()
+}
+
+/// Find the size of the partition image in the given file by parsing the header.
+///
+/// This will work for raw, QCOW2, composite and Android sparse images.
+fn get_partition_size(partition: &File, path: &Path) -> Result<u64, Error> {
+ // TODO: Use `context` once disk::Error implements std::error::Error.
+ // TODO: Add check for is_sparse_file
+ Ok(create_disk_file(
+ partition.try_clone()?,
+ /* is_sparse_file */ false,
+ MAX_NESTING_DEPTH,
+ path,
+ )
+ .map_err(|e| anyhow!("Failed to open partition image: {}", e))?
+ .get_len()?)
+}
diff --git a/virtualizationmanager/src/crosvm.rs b/virtualizationmanager/src/crosvm.rs
new file mode 100644
index 0000000..98e7d99
--- /dev/null
+++ b/virtualizationmanager/src/crosvm.rs
@@ -0,0 +1,861 @@
+// Copyright 2021, The Android Open Source Project
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+// http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+//! Functions for running instances of `crosvm`.
+
+use crate::aidl::{remove_temporary_files, Cid, VirtualMachineCallbacks};
+use crate::atom::write_vm_exited_stats;
+use anyhow::{anyhow, bail, Context, Error, Result};
+use command_fds::CommandFdExt;
+use lazy_static::lazy_static;
+use libc::{sysconf, _SC_CLK_TCK};
+use log::{debug, error, info};
+use semver::{Version, VersionReq};
+use nix::{fcntl::OFlag, unistd::pipe2, unistd::Uid, unistd::User};
+use regex::{Captures, Regex};
+use rustutils::system_properties;
+use shared_child::SharedChild;
+use std::borrow::Cow;
+use std::cmp::max;
+use std::fmt;
+use std::fs::{read_to_string, File};
+use std::io::{self, Read};
+use std::mem;
+use std::num::NonZeroU32;
+use std::os::unix::io::{AsRawFd, RawFd, FromRawFd};
+use std::os::unix::process::ExitStatusExt;
+use std::path::{Path, PathBuf};
+use std::process::{Command, ExitStatus};
+use std::sync::{Arc, Condvar, Mutex};
+use std::time::{Duration, SystemTime};
+use std::thread::{self, JoinHandle};
+use android_system_virtualizationcommon::aidl::android::system::virtualizationcommon::DeathReason::DeathReason;
+use android_system_virtualizationservice::aidl::android::system::virtualizationservice::MemoryTrimLevel::MemoryTrimLevel;
+use android_system_virtualizationservice_internal::aidl::android::system::virtualizationservice_internal::IGlobalVmContext::IGlobalVmContext;
+use binder::Strong;
+use android_system_virtualmachineservice::aidl::android::system::virtualmachineservice::IVirtualMachineService::IVirtualMachineService;
+use tombstoned_client::{TombstonedConnection, DebuggerdDumpType};
+use rpcbinder::RpcServer;
+
+/// external/crosvm
+use base::UnixSeqpacketListener;
+use vm_control::{BalloonControlCommand, VmRequest, VmResponse};
+
+const CROSVM_PATH: &str = "/apex/com.android.virt/bin/crosvm";
+
+/// Version of the platform that crosvm currently implements. The format follows SemVer. This
+/// should be updated when there is a platform change in the crosvm side. Having this value here is
+/// fine because virtualizationservice and crosvm are supposed to be updated together in the virt
+/// APEX.
+const CROSVM_PLATFORM_VERSION: &str = "1.0.0";
+
+/// The exit status which crosvm returns when it has an error starting a VM.
+const CROSVM_START_ERROR_STATUS: i32 = 1;
+/// The exit status which crosvm returns when a VM requests a reboot.
+const CROSVM_REBOOT_STATUS: i32 = 32;
+/// The exit status which crosvm returns when it crashes due to an error.
+const CROSVM_CRASH_STATUS: i32 = 33;
+/// The exit status which crosvm returns when vcpu is stalled.
+const CROSVM_WATCHDOG_REBOOT_STATUS: i32 = 36;
+
+const MILLIS_PER_SEC: i64 = 1000;
+
+const SYSPROP_CUSTOM_PVMFW_PATH: &str = "hypervisor.pvmfw.path";
+
+lazy_static! {
+ /// If the VM doesn't move to the Started state within this amount time, a hang-up error is
+ /// triggered.
+ static ref BOOT_HANGUP_TIMEOUT: Duration = if nested_virt::is_nested_virtualization().unwrap() {
+ // Nested virtualization is slow, so we need a longer timeout.
+ Duration::from_secs(100)
+ } else {
+ Duration::from_secs(10)
+ };
+}
+
+/// Configuration for a VM to run with crosvm.
+#[derive(Debug)]
+pub struct CrosvmConfig {
+ pub cid: Cid,
+ pub name: String,
+ pub bootloader: Option<File>,
+ pub kernel: Option<File>,
+ pub initrd: Option<File>,
+ pub disks: Vec<DiskFile>,
+ pub params: Option<String>,
+ pub protected: bool,
+ pub memory_mib: Option<NonZeroU32>,
+ pub cpus: Option<NonZeroU32>,
+ pub task_profiles: Vec<String>,
+ pub console_fd: Option<File>,
+ pub log_fd: Option<File>,
+ pub ramdump: Option<File>,
+ pub indirect_files: Vec<File>,
+ pub platform_version: VersionReq,
+ pub detect_hangup: bool,
+}
+
+/// A disk image to pass to crosvm for a VM.
+#[derive(Debug)]
+pub struct DiskFile {
+ pub image: File,
+ pub writable: bool,
+}
+
+/// The lifecycle state which the payload in the VM has reported itself to be in.
+///
+/// Note that the order of enum variants is significant; only forward transitions are allowed by
+/// [`VmInstance::update_payload_state`].
+#[derive(Copy, Clone, Debug, Eq, Ord, PartialEq, PartialOrd)]
+pub enum PayloadState {
+ Starting,
+ Started,
+ Ready,
+ Finished,
+ Hangup, // Hasn't reached to Ready before timeout expires
+}
+
+/// The current state of the VM itself.
+#[derive(Debug)]
+pub enum VmState {
+ /// The VM has not yet tried to start.
+ NotStarted {
+ ///The configuration needed to start the VM, if it has not yet been started.
+ config: CrosvmConfig,
+ },
+ /// The VM has been started.
+ Running {
+ /// The crosvm child process.
+ child: Arc<SharedChild>,
+ /// The thread waiting for crosvm to finish.
+ monitor_vm_exit_thread: Option<JoinHandle<()>>,
+ },
+ /// The VM died or was killed.
+ Dead,
+ /// The VM failed to start.
+ Failed,
+}
+
+/// RSS values of VM and CrosVM process itself.
+#[derive(Copy, Clone, Debug, Default)]
+pub struct Rss {
+ pub vm: i64,
+ pub crosvm: i64,
+}
+
+/// Metrics regarding the VM.
+#[derive(Debug, Default)]
+pub struct VmMetric {
+ /// Recorded timestamp when the VM is started.
+ pub start_timestamp: Option<SystemTime>,
+ /// Update most recent guest_time periodically from /proc/[crosvm pid]/stat while VM is running.
+ pub cpu_guest_time: Option<i64>,
+ /// Update maximum RSS values periodically from /proc/[crosvm pid]/smaps while VM is running.
+ pub rss: Option<Rss>,
+}
+
+impl VmState {
+ /// Tries to start the VM, if it is in the `NotStarted` state.
+ ///
+ /// Returns an error if the VM is in the wrong state, or fails to start.
+ fn start(&mut self, instance: Arc<VmInstance>) -> Result<(), Error> {
+ let state = mem::replace(self, VmState::Failed);
+ if let VmState::NotStarted { config } = state {
+ let detect_hangup = config.detect_hangup;
+ let (failure_pipe_read, failure_pipe_write) = create_pipe()?;
+
+ // If this fails and returns an error, `self` will be left in the `Failed` state.
+ let child =
+ Arc::new(run_vm(config, &instance.crosvm_control_socket_path, failure_pipe_write)?);
+
+ let instance_monitor_status = instance.clone();
+ let child_monitor_status = child.clone();
+ thread::spawn(move || {
+ instance_monitor_status.clone().monitor_vm_status(child_monitor_status);
+ });
+
+ let child_clone = child.clone();
+ let instance_clone = instance.clone();
+ let monitor_vm_exit_thread = Some(thread::spawn(move || {
+ instance_clone.monitor_vm_exit(child_clone, failure_pipe_read);
+ }));
+
+ if detect_hangup {
+ let child_clone = child.clone();
+ thread::spawn(move || {
+ instance.monitor_payload_hangup(child_clone);
+ });
+ }
+
+ // If it started correctly, update the state.
+ *self = VmState::Running { child, monitor_vm_exit_thread };
+ Ok(())
+ } else {
+ *self = state;
+ bail!("VM already started or failed")
+ }
+ }
+}
+
+/// Internal struct that holds the handles to globally unique resources of a VM.
+#[derive(Debug)]
+pub struct VmContext {
+ #[allow(dead_code)] // Keeps the global context alive
+ global_context: Strong<dyn IGlobalVmContext>,
+ #[allow(dead_code)] // Keeps the server alive
+ vm_server: RpcServer,
+}
+
+impl VmContext {
+ /// Construct new VmContext.
+ pub fn new(global_context: Strong<dyn IGlobalVmContext>, vm_server: RpcServer) -> VmContext {
+ VmContext { global_context, vm_server }
+ }
+}
+
+/// Information about a particular instance of a VM which may be running.
+#[derive(Debug)]
+pub struct VmInstance {
+ /// The current state of the VM.
+ pub vm_state: Mutex<VmState>,
+ /// Global resources allocated for this VM.
+ #[allow(dead_code)] // Keeps the context alive
+ vm_context: VmContext,
+ /// The CID assigned to the VM for vsock communication.
+ pub cid: Cid,
+ /// Path to crosvm control socket
+ crosvm_control_socket_path: PathBuf,
+ /// The name of the VM.
+ pub name: String,
+ /// Whether the VM is a protected VM.
+ pub protected: bool,
+ /// Directory of temporary files used by the VM while it is running.
+ pub temporary_directory: PathBuf,
+ /// The UID of the process which requested the VM.
+ pub requester_uid: u32,
+ /// The PID of the process which requested the VM. Note that this process may no longer exist
+ /// and the PID may have been reused for a different process, so this should not be trusted.
+ pub requester_debug_pid: i32,
+ /// Callbacks to clients of the VM.
+ pub callbacks: VirtualMachineCallbacks,
+ /// VirtualMachineService binder object for the VM.
+ pub vm_service: Mutex<Option<Strong<dyn IVirtualMachineService>>>,
+ /// Recorded metrics of VM such as timestamp or cpu / memory usage.
+ pub vm_metric: Mutex<VmMetric>,
+ /// The latest lifecycle state which the payload reported itself to be in.
+ payload_state: Mutex<PayloadState>,
+ /// Represents the condition that payload_state was updated
+ payload_state_updated: Condvar,
+ /// The human readable name of requester_uid
+ requester_uid_name: String,
+}
+
+impl fmt::Display for VmInstance {
+ fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
+ let adj = if self.protected { "Protected" } else { "Non-protected" };
+ write!(
+ f,
+ "{} virtual machine \"{}\" (owner: {}, cid: {})",
+ adj, self.name, self.requester_uid_name, self.cid
+ )
+ }
+}
+
+impl VmInstance {
+ /// Validates the given config and creates a new `VmInstance` but doesn't start running it.
+ pub fn new(
+ config: CrosvmConfig,
+ temporary_directory: PathBuf,
+ requester_uid: u32,
+ requester_debug_pid: i32,
+ vm_context: VmContext,
+ ) -> Result<VmInstance, Error> {
+ validate_config(&config)?;
+ let cid = config.cid;
+ let name = config.name.clone();
+ let protected = config.protected;
+ let requester_uid_name = User::from_uid(Uid::from_raw(requester_uid))
+ .ok()
+ .flatten()
+ .map_or_else(|| format!("{}", requester_uid), |u| u.name);
+ let instance = VmInstance {
+ vm_state: Mutex::new(VmState::NotStarted { config }),
+ vm_context,
+ cid,
+ crosvm_control_socket_path: temporary_directory.join("crosvm.sock"),
+ name,
+ protected,
+ temporary_directory,
+ requester_uid,
+ requester_debug_pid,
+ callbacks: Default::default(),
+ vm_service: Mutex::new(None),
+ vm_metric: Mutex::new(Default::default()),
+ payload_state: Mutex::new(PayloadState::Starting),
+ payload_state_updated: Condvar::new(),
+ requester_uid_name,
+ };
+ info!("{} created", &instance);
+ Ok(instance)
+ }
+
+ /// Starts an instance of `crosvm` to manage the VM. The `crosvm` instance will be killed when
+ /// the `VmInstance` is dropped.
+ pub fn start(self: &Arc<Self>) -> Result<(), Error> {
+ let mut vm_metric = self.vm_metric.lock().unwrap();
+ vm_metric.start_timestamp = Some(SystemTime::now());
+ let ret = self.vm_state.lock().unwrap().start(self.clone());
+ if ret.is_ok() {
+ info!("{} started", &self);
+ }
+ ret.with_context(|| format!("{} failed to start", &self))
+ }
+
+ /// Monitors the exit of the VM (i.e. termination of the `child` process). When that happens,
+ /// handles the event by updating the state, noityfing the event to clients by calling
+ /// callbacks, and removing temporary files for the VM.
+ fn monitor_vm_exit(&self, child: Arc<SharedChild>, mut failure_pipe_read: File) {
+ let result = child.wait();
+ match &result {
+ Err(e) => error!("Error waiting for crosvm({}) instance to die: {}", child.id(), e),
+ Ok(status) => {
+ info!("crosvm({}) exited with status {}", child.id(), status);
+ if let Some(exit_status_code) = status.code() {
+ if exit_status_code == CROSVM_WATCHDOG_REBOOT_STATUS {
+ info!("detected vcpu stall on crosvm");
+ }
+ }
+ }
+ }
+
+ let mut vm_state = self.vm_state.lock().unwrap();
+ *vm_state = VmState::Dead;
+ // Ensure that the mutex is released before calling the callbacks.
+ drop(vm_state);
+ info!("{} exited", &self);
+
+ // Read the pipe to see if any failure reason is written
+ let mut failure_reason = String::new();
+ match failure_pipe_read.read_to_string(&mut failure_reason) {
+ Err(e) => error!("Error reading VM failure reason from pipe: {}", e),
+ Ok(len) if len > 0 => info!("VM returned failure reason '{}'", &failure_reason),
+ _ => (),
+ };
+
+ // In case of hangup, the pipe doesn't give us any information because the hangup can't be
+ // detected on the VM side (otherwise, it isn't a hangup), but in the
+ // monitor_payload_hangup function below which updates the payload state to Hangup.
+ let failure_reason =
+ if failure_reason.is_empty() && self.payload_state() == PayloadState::Hangup {
+ Cow::from("HANGUP")
+ } else {
+ Cow::from(failure_reason)
+ };
+
+ self.handle_ramdump().unwrap_or_else(|e| error!("Error handling ramdump: {}", e));
+
+ let death_reason = death_reason(&result, &failure_reason);
+ let exit_signal = exit_signal(&result);
+
+ self.callbacks.callback_on_died(self.cid, death_reason);
+
+ let vm_metric = self.vm_metric.lock().unwrap();
+ write_vm_exited_stats(
+ self.requester_uid as i32,
+ &self.name,
+ death_reason,
+ exit_signal,
+ &*vm_metric,
+ );
+
+ // Delete temporary files. The folder itself is removed by VirtualizationServiceInternal.
+ remove_temporary_files(&self.temporary_directory).unwrap_or_else(|e| {
+ error!("Error removing temporary files from {:?}: {}", self.temporary_directory, e);
+ });
+ }
+
+ /// Waits until payload is started, or timeout expires. When timeout occurs, kill
+ /// the VM to prevent indefinite hangup and update the payload_state accordingly.
+ fn monitor_payload_hangup(&self, child: Arc<SharedChild>) {
+ debug!("Starting to monitor hangup for Microdroid({})", child.id());
+ let (_, result) = self
+ .payload_state_updated
+ .wait_timeout_while(self.payload_state.lock().unwrap(), *BOOT_HANGUP_TIMEOUT, |s| {
+ *s < PayloadState::Started
+ })
+ .unwrap();
+ let child_still_running = child.try_wait().ok() == Some(None);
+ if result.timed_out() && child_still_running {
+ error!(
+ "Microdroid({}) failed to start payload within {} secs timeout. Shutting down.",
+ child.id(),
+ BOOT_HANGUP_TIMEOUT.as_secs()
+ );
+ self.update_payload_state(PayloadState::Hangup).unwrap();
+ if let Err(e) = self.kill() {
+ error!("Error stopping timed-out VM with CID {}: {:?}", child.id(), e);
+ }
+ }
+ }
+
+ fn monitor_vm_status(&self, child: Arc<SharedChild>) {
+ let pid = child.id();
+
+ loop {
+ {
+ // Check VM state
+ let vm_state = &*self.vm_state.lock().unwrap();
+ if let VmState::Dead = vm_state {
+ break;
+ }
+
+ let mut vm_metric = self.vm_metric.lock().unwrap();
+
+ // Get CPU Information
+ if let Ok(guest_time) = get_guest_time(pid) {
+ vm_metric.cpu_guest_time = Some(guest_time);
+ } else {
+ error!("Failed to parse /proc/[pid]/stat");
+ }
+
+ // Get Memory Information
+ if let Ok(rss) = get_rss(pid) {
+ vm_metric.rss = match &vm_metric.rss {
+ Some(x) => Some(Rss::extract_max(x, &rss)),
+ None => Some(rss),
+ }
+ } else {
+ error!("Failed to parse /proc/[pid]/smaps");
+ }
+ }
+
+ thread::sleep(Duration::from_secs(1));
+ }
+ }
+
+ /// Returns the last reported state of the VM payload.
+ pub fn payload_state(&self) -> PayloadState {
+ *self.payload_state.lock().unwrap()
+ }
+
+ /// Updates the payload state to the given value, if it is a valid state transition.
+ pub fn update_payload_state(&self, new_state: PayloadState) -> Result<(), Error> {
+ let mut state_locked = self.payload_state.lock().unwrap();
+ // Only allow forward transitions, e.g. from starting to started or finished, not back in
+ // the other direction.
+ if new_state > *state_locked {
+ *state_locked = new_state;
+ self.payload_state_updated.notify_all();
+ Ok(())
+ } else {
+ bail!("Invalid payload state transition from {:?} to {:?}", *state_locked, new_state)
+ }
+ }
+
+ /// Kills the crosvm instance, if it is running.
+ pub fn kill(&self) -> Result<(), Error> {
+ let monitor_vm_exit_thread = {
+ let vm_state = &mut *self.vm_state.lock().unwrap();
+ if let VmState::Running { child, monitor_vm_exit_thread } = vm_state {
+ let id = child.id();
+ debug!("Killing crosvm({})", id);
+ // TODO: Talk to crosvm to shutdown cleanly.
+ child.kill().with_context(|| format!("Error killing crosvm({id}) instance"))?;
+ monitor_vm_exit_thread.take()
+ } else {
+ bail!("VM is not running")
+ }
+ };
+
+ // Wait for monitor_vm_exit() to finish. Must release vm_state lock
+ // first, as monitor_vm_exit() takes it as well.
+ monitor_vm_exit_thread.map(JoinHandle::join);
+
+ Ok(())
+ }
+
+ /// Responds to memory-trimming notifications by inflating the virtio
+ /// balloon to reclaim guest memory.
+ pub fn trim_memory(&self, level: MemoryTrimLevel) -> Result<(), Error> {
+ let request = VmRequest::BalloonCommand(BalloonControlCommand::Stats {});
+ match vm_control::client::handle_request(&request, &self.crosvm_control_socket_path) {
+ Ok(VmResponse::BalloonStats { stats, balloon_actual: _ }) => {
+ if let Some(total_memory) = stats.total_memory {
+ // Reclaim up to 50% of total memory assuming worst case
+ // most memory is anonymous and must be swapped to zram
+ // with an approximate 2:1 compression ratio.
+ let pct = match level {
+ MemoryTrimLevel::TRIM_MEMORY_RUNNING_CRITICAL => 50,
+ MemoryTrimLevel::TRIM_MEMORY_RUNNING_LOW => 30,
+ MemoryTrimLevel::TRIM_MEMORY_RUNNING_MODERATE => 10,
+ _ => bail!("Invalid memory trim level {:?}", level),
+ };
+ let command =
+ BalloonControlCommand::Adjust { num_bytes: total_memory * pct / 100 };
+ if let Err(e) = vm_control::client::handle_request(
+ &VmRequest::BalloonCommand(command),
+ &self.crosvm_control_socket_path,
+ ) {
+ bail!("Error sending balloon adjustment: {:?}", e);
+ }
+ }
+ }
+ Ok(VmResponse::Err(e)) => {
+ // ENOTSUP is returned when the balloon protocol is not initialised. This
+ // can occur for numerous reasons: Guest is still booting, guest doesn't
+ // support ballooning, host doesn't support ballooning. We don't log or
+ // raise an error in this case: trim is just a hint and we can ignore it.
+ if e.errno() != libc::ENOTSUP {
+ bail!("Errno return when requesting balloon stats: {}", e.errno())
+ }
+ }
+ e => bail!("Error requesting balloon stats: {:?}", e),
+ }
+ Ok(())
+ }
+
+ /// Checks if ramdump has been created. If so, send it to tombstoned.
+ fn handle_ramdump(&self) -> Result<(), Error> {
+ let ramdump_path = self.temporary_directory.join("ramdump");
+ if std::fs::metadata(&ramdump_path)?.len() > 0 {
+ Self::send_ramdump_to_tombstoned(&ramdump_path)?;
+ }
+ Ok(())
+ }
+
+ fn send_ramdump_to_tombstoned(ramdump_path: &Path) -> Result<(), Error> {
+ let mut input = File::open(ramdump_path)
+ .context(format!("Failed to open ramdump {:?} for reading", ramdump_path))?;
+
+ let pid = std::process::id() as i32;
+ let conn = TombstonedConnection::connect(pid, DebuggerdDumpType::Tombstone)
+ .context("Failed to connect to tombstoned")?;
+ let mut output = conn
+ .text_output
+ .as_ref()
+ .ok_or_else(|| anyhow!("Could not get file to write the tombstones on"))?;
+
+ std::io::copy(&mut input, &mut output).context("Failed to send ramdump to tombstoned")?;
+ info!("Ramdump {:?} sent to tombstoned", ramdump_path);
+
+ conn.notify_completion()?;
+ Ok(())
+ }
+}
+
+impl Rss {
+ fn extract_max(x: &Rss, y: &Rss) -> Rss {
+ Rss { vm: max(x.vm, y.vm), crosvm: max(x.crosvm, y.crosvm) }
+ }
+}
+
+// Get guest time from /proc/[crosvm pid]/stat
+fn get_guest_time(pid: u32) -> Result<i64> {
+ let file = read_to_string(format!("/proc/{}/stat", pid))?;
+ let data_list: Vec<_> = file.split_whitespace().collect();
+
+ // Information about guest_time is at 43th place of the file split with the whitespace.
+ // Example of /proc/[pid]/stat :
+ // 6603 (kworker/104:1H-kblockd) I 2 0 0 0 -1 69238880 0 0 0 0 0 88 0 0 0 -20 1 0 1845 0 0
+ // 18446744073709551615 0 0 0 0 0 0 0 2147483647 0 0 0 0 17 104 0 0 0 0 0 0 0 0 0 0 0 0 0
+ if data_list.len() < 43 {
+ bail!("Failed to parse command result for getting guest time : {}", file);
+ }
+
+ let guest_time_ticks = data_list[42].parse::<i64>()?;
+ // SAFETY : It just returns an integer about CPU tick information.
+ let ticks_per_sec = unsafe { sysconf(_SC_CLK_TCK) } as i64;
+ Ok(guest_time_ticks * MILLIS_PER_SEC / ticks_per_sec)
+}
+
+// Get rss from /proc/[crosvm pid]/smaps
+fn get_rss(pid: u32) -> Result<Rss> {
+ let file = read_to_string(format!("/proc/{}/smaps", pid))?;
+ let lines: Vec<_> = file.split('\n').collect();
+
+ let mut rss_vm_total = 0i64;
+ let mut rss_crosvm_total = 0i64;
+ let mut is_vm = false;
+ for line in lines {
+ if line.contains("crosvm_guest") {
+ is_vm = true;
+ } else if line.contains("Rss:") {
+ let data_list: Vec<_> = line.split_whitespace().collect();
+ if data_list.len() < 2 {
+ bail!("Failed to parse command result for getting rss :\n{}", line);
+ }
+ let rss = data_list[1].parse::<i64>()?;
+
+ if is_vm {
+ rss_vm_total += rss;
+ is_vm = false;
+ }
+ rss_crosvm_total += rss;
+ }
+ }
+
+ Ok(Rss { vm: rss_vm_total, crosvm: rss_crosvm_total })
+}
+
+fn death_reason(result: &Result<ExitStatus, io::Error>, mut failure_reason: &str) -> DeathReason {
+ if let Some(position) = failure_reason.find('|') {
+ // Separator indicates extra context information is present after the failure name.
+ error!("Failure info: {}", &failure_reason[(position + 1)..]);
+ failure_reason = &failure_reason[..position];
+ }
+ if let Ok(status) = result {
+ match failure_reason {
+ "PVM_FIRMWARE_PUBLIC_KEY_MISMATCH" => {
+ return DeathReason::PVM_FIRMWARE_PUBLIC_KEY_MISMATCH
+ }
+ "PVM_FIRMWARE_INSTANCE_IMAGE_CHANGED" => {
+ return DeathReason::PVM_FIRMWARE_INSTANCE_IMAGE_CHANGED
+ }
+ "BOOTLOADER_PUBLIC_KEY_MISMATCH" => return DeathReason::BOOTLOADER_PUBLIC_KEY_MISMATCH,
+ "BOOTLOADER_INSTANCE_IMAGE_CHANGED" => {
+ return DeathReason::BOOTLOADER_INSTANCE_IMAGE_CHANGED
+ }
+ "MICRODROID_FAILED_TO_CONNECT_TO_VIRTUALIZATION_SERVICE" => {
+ return DeathReason::MICRODROID_FAILED_TO_CONNECT_TO_VIRTUALIZATION_SERVICE
+ }
+ "MICRODROID_PAYLOAD_HAS_CHANGED" => return DeathReason::MICRODROID_PAYLOAD_HAS_CHANGED,
+ "MICRODROID_PAYLOAD_VERIFICATION_FAILED" => {
+ return DeathReason::MICRODROID_PAYLOAD_VERIFICATION_FAILED
+ }
+ "MICRODROID_INVALID_PAYLOAD_CONFIG" => {
+ return DeathReason::MICRODROID_INVALID_PAYLOAD_CONFIG
+ }
+ "MICRODROID_UNKNOWN_RUNTIME_ERROR" => {
+ return DeathReason::MICRODROID_UNKNOWN_RUNTIME_ERROR
+ }
+ "HANGUP" => return DeathReason::HANGUP,
+ _ => {}
+ }
+ match status.code() {
+ None => DeathReason::KILLED,
+ Some(0) => DeathReason::SHUTDOWN,
+ Some(CROSVM_START_ERROR_STATUS) => DeathReason::START_FAILED,
+ Some(CROSVM_REBOOT_STATUS) => DeathReason::REBOOT,
+ Some(CROSVM_CRASH_STATUS) => DeathReason::CRASH,
+ Some(CROSVM_WATCHDOG_REBOOT_STATUS) => DeathReason::WATCHDOG_REBOOT,
+ Some(_) => DeathReason::UNKNOWN,
+ }
+ } else {
+ DeathReason::INFRASTRUCTURE_ERROR
+ }
+}
+
+fn exit_signal(result: &Result<ExitStatus, io::Error>) -> Option<i32> {
+ match result {
+ Ok(status) => status.signal(),
+ Err(_) => None,
+ }
+}
+
+/// Starts an instance of `crosvm` to manage a new VM.
+fn run_vm(
+ config: CrosvmConfig,
+ crosvm_control_socket_path: &Path,
+ failure_pipe_write: File,
+) -> Result<SharedChild, Error> {
+ validate_config(&config)?;
+
+ let mut command = Command::new(CROSVM_PATH);
+ // TODO(qwandor): Remove --disable-sandbox.
+ command
+ .arg("--extended-status")
+ // Configure the logger for the crosvm process to silence logs from the disk crate which
+ // don't provide much information to us (but do spamming us).
+ .arg("--log-level")
+ .arg("info,disk=off")
+ .arg("run")
+ .arg("--disable-sandbox")
+ .arg("--cid")
+ .arg(config.cid.to_string());
+
+ if system_properties::read_bool("hypervisor.memory_reclaim.supported", false)? {
+ command.arg("--balloon-page-reporting");
+ } else {
+ command.arg("--no-balloon");
+ }
+
+ if config.protected {
+ match system_properties::read(SYSPROP_CUSTOM_PVMFW_PATH)? {
+ Some(pvmfw_path) if !pvmfw_path.is_empty() => {
+ command.arg("--protected-vm-with-firmware").arg(pvmfw_path)
+ }
+ _ => command.arg("--protected-vm"),
+ };
+
+ // 3 virtio-console devices + vsock = 4.
+ let virtio_pci_device_count = 4 + config.disks.len();
+ // crosvm virtio queue has 256 entries, so 2 MiB per device (2 pages per entry) should be
+ // enough.
+ let swiotlb_size_mib = 2 * virtio_pci_device_count;
+ command.arg("--swiotlb").arg(swiotlb_size_mib.to_string());
+ }
+
+ if let Some(memory_mib) = config.memory_mib {
+ command.arg("--mem").arg(memory_mib.to_string());
+ }
+
+ if let Some(cpus) = config.cpus {
+ command.arg("--cpus").arg(cpus.to_string());
+ }
+
+ if !config.task_profiles.is_empty() {
+ command.arg("--task-profiles").arg(config.task_profiles.join(","));
+ }
+
+ // Keep track of what file descriptors should be mapped to the crosvm process.
+ let mut preserved_fds = config.indirect_files.iter().map(|file| file.as_raw_fd()).collect();
+
+ // Setup the serial devices.
+ // 1. uart device: used as the output device by bootloaders and as early console by linux
+ // 2. uart device: used to report the reason for the VM failing.
+ // 3. virtio-console device: used as the console device where kmsg is redirected to
+ // 4. virtio-console device: used as the ramdump output
+ // 5. virtio-console device: used as the logcat output
+ //
+ // When [console|log]_fd is not specified, the devices are attached to sink, which means what's
+ // written there is discarded.
+ let console_arg = format_serial_arg(&mut preserved_fds, &config.console_fd);
+ let log_arg = format_serial_arg(&mut preserved_fds, &config.log_fd);
+ let failure_serial_path = add_preserved_fd(&mut preserved_fds, &failure_pipe_write);
+ let ramdump_arg = format_serial_arg(&mut preserved_fds, &config.ramdump);
+
+ // Warning: Adding more serial devices requires you to shift the PCI device ID of the boot
+ // disks in bootconfig.x86_64. This is because x86 crosvm puts serial devices and the block
+ // devices in the same PCI bus and serial devices comes before the block devices. Arm crosvm
+ // doesn't have the issue.
+ // /dev/ttyS0
+ command.arg(format!("--serial={},hardware=serial,num=1", &console_arg));
+ // /dev/ttyS1
+ command.arg(format!("--serial=type=file,path={},hardware=serial,num=2", &failure_serial_path));
+ // /dev/hvc0
+ command.arg(format!("--serial={},hardware=virtio-console,num=1", &console_arg));
+ // /dev/hvc1
+ command.arg(format!("--serial={},hardware=virtio-console,num=2", &ramdump_arg));
+ // /dev/hvc2
+ command.arg(format!("--serial={},hardware=virtio-console,num=3", &log_arg));
+
+ if let Some(bootloader) = &config.bootloader {
+ command.arg("--bios").arg(add_preserved_fd(&mut preserved_fds, bootloader));
+ }
+
+ if let Some(initrd) = &config.initrd {
+ command.arg("--initrd").arg(add_preserved_fd(&mut preserved_fds, initrd));
+ }
+
+ if let Some(params) = &config.params {
+ command.arg("--params").arg(params);
+ }
+
+ for disk in &config.disks {
+ command
+ .arg(if disk.writable { "--rwdisk" } else { "--disk" })
+ .arg(add_preserved_fd(&mut preserved_fds, &disk.image));
+ }
+
+ if let Some(kernel) = &config.kernel {
+ command.arg(add_preserved_fd(&mut preserved_fds, kernel));
+ }
+
+ let control_server_socket = UnixSeqpacketListener::bind(crosvm_control_socket_path)
+ .context("failed to create control server")?;
+ command.arg("--socket").arg(add_preserved_fd(&mut preserved_fds, &control_server_socket));
+
+ debug!("Preserving FDs {:?}", preserved_fds);
+ command.preserved_fds(preserved_fds);
+
+ command.arg("--params").arg("crashkernel=17M");
+ print_crosvm_args(&command);
+
+ let result = SharedChild::spawn(&mut command)?;
+ debug!("Spawned crosvm({}).", result.id());
+ Ok(result)
+}
+
+/// Ensure that the configuration has a valid combination of fields set, or return an error if not.
+fn validate_config(config: &CrosvmConfig) -> Result<(), Error> {
+ if config.bootloader.is_none() && config.kernel.is_none() {
+ bail!("VM must have either a bootloader or a kernel image.");
+ }
+ if config.bootloader.is_some() && (config.kernel.is_some() || config.initrd.is_some()) {
+ bail!("Can't have both bootloader and kernel/initrd image.");
+ }
+ let version = Version::parse(CROSVM_PLATFORM_VERSION).unwrap();
+ if !config.platform_version.matches(&version) {
+ bail!(
+ "Incompatible platform version. The config is compatible with platform version(s) \
+ {}, but the actual platform version is {}",
+ config.platform_version,
+ version
+ );
+ }
+
+ Ok(())
+}
+
+/// Print arguments of the crosvm command. In doing so, /proc/self/fd/XX is annotated with the
+/// actual file path if the FD is backed by a regular file. If not, the /proc path is printed
+/// unmodified.
+fn print_crosvm_args(command: &Command) {
+ let re = Regex::new(r"/proc/self/fd/[\d]+").unwrap();
+ info!(
+ "Running crosvm with args: {:?}",
+ command
+ .get_args()
+ .map(|s| s.to_string_lossy())
+ .map(|s| {
+ re.replace_all(&s, |caps: &Captures| {
+ let path = &caps[0];
+ if let Ok(realpath) = std::fs::canonicalize(path) {
+ format!("{} ({})", path, realpath.to_string_lossy())
+ } else {
+ path.to_owned()
+ }
+ })
+ .into_owned()
+ })
+ .collect::<Vec<_>>()
+ );
+}
+
+/// Adds the file descriptor for `file` to `preserved_fds`, and returns a string of the form
+/// "/proc/self/fd/N" where N is the file descriptor.
+fn add_preserved_fd(preserved_fds: &mut Vec<RawFd>, file: &dyn AsRawFd) -> String {
+ let fd = file.as_raw_fd();
+ preserved_fds.push(fd);
+ format!("/proc/self/fd/{}", fd)
+}
+
+/// Adds the file descriptor for `file` (if any) to `preserved_fds`, and returns the appropriate
+/// string for a crosvm `--serial` flag. If `file` is none, creates a dummy sink device.
+fn format_serial_arg(preserved_fds: &mut Vec<RawFd>, file: &Option<File>) -> String {
+ if let Some(file) = file {
+ format!("type=file,path={}", add_preserved_fd(preserved_fds, file))
+ } else {
+ "type=sink".to_string()
+ }
+}
+
+/// Creates a new pipe with the `O_CLOEXEC` flag set, and returns the read side and write side.
+fn create_pipe() -> Result<(File, File), Error> {
+ let (raw_read, raw_write) = pipe2(OFlag::O_CLOEXEC)?;
+ // SAFETY: We are the sole owners of these fds as they were just created.
+ let read_fd = unsafe { File::from_raw_fd(raw_read) };
+ let write_fd = unsafe { File::from_raw_fd(raw_write) };
+ Ok((read_fd, write_fd))
+}
diff --git a/virtualizationmanager/src/main.rs b/virtualizationmanager/src/main.rs
new file mode 100644
index 0000000..dca64cb
--- /dev/null
+++ b/virtualizationmanager/src/main.rs
@@ -0,0 +1,145 @@
+// Copyright 2022, The Android Open Source Project
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+// http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+//! Android Virtualization Manager
+
+mod aidl;
+mod atom;
+mod composite;
+mod crosvm;
+mod payload;
+mod selinux;
+
+use crate::aidl::{GLOBAL_SERVICE, VirtualizationService};
+use android_system_virtualizationservice::aidl::android::system::virtualizationservice::IVirtualizationService::BnVirtualizationService;
+use anyhow::{bail, Context};
+use binder::{BinderFeatures, ProcessState};
+use lazy_static::lazy_static;
+use log::{info, Level};
+use rpcbinder::{FileDescriptorTransportMode, RpcServer};
+use std::os::unix::io::{FromRawFd, OwnedFd, RawFd};
+use clap::Parser;
+use nix::fcntl::{fcntl, F_GETFD, F_SETFD, FdFlag};
+use nix::unistd::{Pid, Uid};
+use std::os::unix::raw::{pid_t, uid_t};
+use rustutils::system_properties;
+
+const LOG_TAG: &str = "virtmgr";
+
+lazy_static! {
+ static ref PID_PARENT: Pid = Pid::parent();
+ static ref UID_CURRENT: Uid = Uid::current();
+}
+
+fn get_calling_pid() -> pid_t {
+ // The caller is the parent of this process.
+ PID_PARENT.as_raw()
+}
+
+fn get_calling_uid() -> uid_t {
+ // The caller and this process share the same UID.
+ UID_CURRENT.as_raw()
+}
+
+#[derive(Parser)]
+struct Args {
+ /// File descriptor inherited from the caller to run RpcBinder server on.
+ /// This should be one end of a socketpair() compatible with RpcBinder's
+ /// UDS bootstrap transport.
+ #[clap(long)]
+ rpc_server_fd: RawFd,
+ /// File descriptor inherited from the caller to signal RpcBinder server
+ /// readiness. This should be one end of pipe() and the caller should be
+ /// waiting for HUP on the other end.
+ #[clap(long)]
+ ready_fd: RawFd,
+}
+
+fn take_fd_ownership(raw_fd: RawFd, owned_fds: &mut Vec<RawFd>) -> Result<OwnedFd, anyhow::Error> {
+ // Basic check that the integer value does correspond to a file descriptor.
+ fcntl(raw_fd, F_GETFD).with_context(|| format!("Invalid file descriptor {raw_fd}"))?;
+
+ // The file descriptor had CLOEXEC disabled to be inherited from the parent.
+ // Re-enable it to make sure it is not accidentally inherited further.
+ fcntl(raw_fd, F_SETFD(FdFlag::FD_CLOEXEC))
+ .with_context(|| format!("Could not set CLOEXEC on file descriptor {raw_fd}"))?;
+
+ // Creating OwnedFd for stdio FDs is not safe.
+ if [libc::STDIN_FILENO, libc::STDOUT_FILENO, libc::STDERR_FILENO].contains(&raw_fd) {
+ bail!("File descriptor {raw_fd} is standard I/O descriptor");
+ }
+
+ // Reject RawFds that already have a corresponding OwnedFd.
+ if owned_fds.contains(&raw_fd) {
+ bail!("File descriptor {raw_fd} already owned");
+ }
+ owned_fds.push(raw_fd);
+
+ // SAFETY - Initializing OwnedFd for a RawFd provided in cmdline arguments.
+ // We checked that the integer value corresponds to a valid FD and that this
+ // is the first argument to claim its ownership.
+ Ok(unsafe { OwnedFd::from_raw_fd(raw_fd) })
+}
+
+fn is_property_set(name: &str) -> bool {
+ system_properties::read_bool(name, false)
+ .unwrap_or_else(|e| panic!("Failed to read {name}: {e:?}"))
+}
+
+fn main() {
+ android_logger::init_once(
+ android_logger::Config::default()
+ .with_tag(LOG_TAG)
+ .with_min_level(Level::Info)
+ .with_log_id(android_logger::LogId::System),
+ );
+
+ let non_protected_vm_supported = is_property_set("ro.boot.hypervisor.vm.supported");
+ let protected_vm_supported = is_property_set("ro.boot.hypervisor.protected_vm.supported");
+ if !non_protected_vm_supported && !protected_vm_supported {
+ // This should never happen, it indicates a misconfigured device where the virt APEX
+ // is present but VMs are not supported. If it does happen, fail fast to avoid wasting
+ // resources trying.
+ panic!("Device doesn't support protected or unprotected VMs");
+ }
+
+ let args = Args::parse();
+
+ let mut owned_fds = vec![];
+ let rpc_server_fd = take_fd_ownership(args.rpc_server_fd, &mut owned_fds)
+ .expect("Failed to take ownership of rpc_server_fd");
+ let ready_fd = take_fd_ownership(args.ready_fd, &mut owned_fds)
+ .expect("Failed to take ownership of ready_fd");
+
+ // Start thread pool for kernel Binder connection to VirtualizationServiceInternal.
+ ProcessState::start_thread_pool();
+
+ GLOBAL_SERVICE.removeMemlockRlimit().expect("Failed to remove memlock rlimit");
+
+ let service = VirtualizationService::init();
+ let service =
+ BnVirtualizationService::new_binder(service, BinderFeatures::default()).as_binder();
+
+ let server = RpcServer::new_unix_domain_bootstrap(service, rpc_server_fd)
+ .expect("Failed to start RpcServer");
+ server.set_supported_file_descriptor_transport_modes(&[FileDescriptorTransportMode::Unix]);
+
+ info!("Started VirtualizationService RpcServer. Ready to accept connections");
+
+ // Signal readiness to the caller by closing our end of the pipe.
+ drop(ready_fd);
+
+ server.join();
+ info!("Shutting down VirtualizationService RpcServer");
+}
diff --git a/virtualizationmanager/src/payload.rs b/virtualizationmanager/src/payload.rs
new file mode 100644
index 0000000..02e8f8e
--- /dev/null
+++ b/virtualizationmanager/src/payload.rs
@@ -0,0 +1,684 @@
+// Copyright 2021, The Android Open Source Project
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+// http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+//! Payload disk image
+
+use android_system_virtualizationservice::aidl::android::system::virtualizationservice::{
+ DiskImage::DiskImage,
+ Partition::Partition,
+ VirtualMachineAppConfig::DebugLevel::DebugLevel,
+ VirtualMachineAppConfig::{Payload::Payload, VirtualMachineAppConfig},
+ VirtualMachineRawConfig::VirtualMachineRawConfig,
+};
+use anyhow::{anyhow, bail, Context, Result};
+use binder::{wait_for_interface, ParcelFileDescriptor};
+use log::{info, warn};
+use microdroid_metadata::{ApexPayload, ApkPayload, Metadata, PayloadConfig, PayloadMetadata};
+use microdroid_payload_config::{ApexConfig, VmPayloadConfig};
+use once_cell::sync::OnceCell;
+use packagemanager_aidl::aidl::android::content::pm::{
+ IPackageManagerNative::IPackageManagerNative, StagedApexInfo::StagedApexInfo,
+};
+use regex::Regex;
+use serde::Deserialize;
+use serde_xml_rs::from_reader;
+use std::collections::HashSet;
+use std::fs::{metadata, File, OpenOptions};
+use std::path::{Path, PathBuf};
+use std::process::Command;
+use std::time::SystemTime;
+use vmconfig::open_parcel_file;
+
+/// The list of APEXes which microdroid requires.
+// TODO(b/192200378) move this to microdroid.json?
+const MICRODROID_REQUIRED_APEXES: [&str; 1] = ["com.android.os.statsd"];
+const MICRODROID_REQUIRED_APEXES_DEBUG: [&str; 1] = ["com.android.adbd"];
+
+const APEX_INFO_LIST_PATH: &str = "/apex/apex-info-list.xml";
+
+const PACKAGE_MANAGER_NATIVE_SERVICE: &str = "package_native";
+
+/// Represents the list of APEXes
+#[derive(Clone, Debug, Deserialize, Eq, PartialEq)]
+struct ApexInfoList {
+ #[serde(rename = "apex-info")]
+ list: Vec<ApexInfo>,
+}
+
+#[derive(Clone, Debug, Default, Deserialize, Eq, PartialEq)]
+struct ApexInfo {
+ #[serde(rename = "moduleName")]
+ name: String,
+ #[serde(rename = "versionCode")]
+ version: u64,
+ #[serde(rename = "modulePath")]
+ path: PathBuf,
+
+ #[serde(default)]
+ has_classpath_jar: bool,
+
+ // The field claims to be milliseconds but is actually seconds.
+ #[serde(rename = "lastUpdateMillis")]
+ last_update_seconds: u64,
+
+ #[serde(rename = "isFactory")]
+ is_factory: bool,
+
+ #[serde(rename = "isActive")]
+ is_active: bool,
+
+ #[serde(rename = "provideSharedApexLibs")]
+ provide_shared_apex_libs: bool,
+}
+
+impl ApexInfoList {
+ /// Loads ApexInfoList
+ fn load() -> Result<&'static ApexInfoList> {
+ static INSTANCE: OnceCell<ApexInfoList> = OnceCell::new();
+ INSTANCE.get_or_try_init(|| {
+ let apex_info_list = File::open(APEX_INFO_LIST_PATH)
+ .context(format!("Failed to open {}", APEX_INFO_LIST_PATH))?;
+ let mut apex_info_list: ApexInfoList = from_reader(apex_info_list)
+ .context(format!("Failed to parse {}", APEX_INFO_LIST_PATH))?;
+
+ // For active APEXes, we run derive_classpath and parse its output to see if it
+ // contributes to the classpath(s). (This allows us to handle any new classpath env
+ // vars seamlessly.)
+ let classpath_vars = run_derive_classpath()?;
+ let classpath_apexes = find_apex_names_in_classpath(&classpath_vars)?;
+
+ for apex_info in apex_info_list.list.iter_mut() {
+ apex_info.has_classpath_jar = classpath_apexes.contains(&apex_info.name);
+ }
+
+ Ok(apex_info_list)
+ })
+ }
+
+ // Override apex info with the staged one
+ fn override_staged_apex(&mut self, staged_apex_info: &StagedApexInfo) -> Result<()> {
+ let mut need_to_add: Option<ApexInfo> = None;
+ for apex_info in self.list.iter_mut() {
+ if staged_apex_info.moduleName == apex_info.name {
+ if apex_info.is_active && apex_info.is_factory {
+ // Copy the entry to the end as factory/non-active after the loop
+ // to keep the factory version. Typically this step is unncessary,
+ // but some apexes (like sharedlibs) need to be kept even if it's inactive.
+ need_to_add.replace(ApexInfo { is_active: false, ..apex_info.clone() });
+ // And make this one as non-factory. Note that this one is still active
+ // and overridden right below.
+ apex_info.is_factory = false;
+ }
+ // Active one is overridden with the staged one.
+ if apex_info.is_active {
+ apex_info.version = staged_apex_info.versionCode as u64;
+ apex_info.path = PathBuf::from(&staged_apex_info.diskImagePath);
+ apex_info.has_classpath_jar = staged_apex_info.hasClassPathJars;
+ apex_info.last_update_seconds = last_updated(&apex_info.path)?;
+ }
+ }
+ }
+ if let Some(info) = need_to_add {
+ self.list.push(info);
+ }
+ Ok(())
+ }
+}
+
+fn last_updated<P: AsRef<Path>>(path: P) -> Result<u64> {
+ let metadata = metadata(path)?;
+ Ok(metadata.modified()?.duration_since(SystemTime::UNIX_EPOCH)?.as_secs())
+}
+
+impl ApexInfo {
+ fn matches(&self, apex_config: &ApexConfig) -> bool {
+ // Match with pseudo name "{CLASSPATH}" which represents APEXes contributing
+ // to any derive_classpath environment variable
+ if apex_config.name == "{CLASSPATH}" && self.has_classpath_jar {
+ return true;
+ }
+ if apex_config.name == self.name {
+ return true;
+ }
+ false
+ }
+}
+
+struct PackageManager {
+ apex_info_list: &'static ApexInfoList,
+}
+
+impl PackageManager {
+ fn new() -> Result<Self> {
+ let apex_info_list = ApexInfoList::load()?;
+ Ok(Self { apex_info_list })
+ }
+
+ fn get_apex_list(&self, prefer_staged: bool) -> Result<ApexInfoList> {
+ // get the list of active apexes
+ let mut list = self.apex_info_list.clone();
+ // When prefer_staged, we override ApexInfo by consulting "package_native"
+ if prefer_staged {
+ let pm =
+ wait_for_interface::<dyn IPackageManagerNative>(PACKAGE_MANAGER_NATIVE_SERVICE)
+ .context("Failed to get service when prefer_staged is set.")?;
+ let staged =
+ pm.getStagedApexModuleNames().context("getStagedApexModuleNames failed")?;
+ for name in staged {
+ if let Some(staged_apex_info) =
+ pm.getStagedApexInfo(&name).context("getStagedApexInfo failed")?
+ {
+ list.override_staged_apex(&staged_apex_info)?;
+ }
+ }
+ }
+ Ok(list)
+ }
+}
+
+fn make_metadata_file(
+ app_config: &VirtualMachineAppConfig,
+ apex_infos: &[&ApexInfo],
+ temporary_directory: &Path,
+) -> Result<ParcelFileDescriptor> {
+ let payload_metadata = match &app_config.payload {
+ Payload::PayloadConfig(payload_config) => PayloadMetadata::config(PayloadConfig {
+ payload_binary_name: payload_config.payloadBinaryName.clone(),
+ ..Default::default()
+ }),
+ Payload::ConfigPath(config_path) => {
+ PayloadMetadata::config_path(format!("/mnt/apk/{}", config_path))
+ }
+ };
+
+ let metadata = Metadata {
+ version: 1,
+ apexes: apex_infos
+ .iter()
+ .enumerate()
+ .map(|(i, apex_info)| {
+ Ok(ApexPayload {
+ name: apex_info.name.clone(),
+ partition_name: format!("microdroid-apex-{}", i),
+ last_update_seconds: apex_info.last_update_seconds,
+ is_factory: apex_info.is_factory,
+ ..Default::default()
+ })
+ })
+ .collect::<Result<_>>()?,
+ apk: Some(ApkPayload {
+ name: "apk".to_owned(),
+ payload_partition_name: "microdroid-apk".to_owned(),
+ idsig_partition_name: "microdroid-apk-idsig".to_owned(),
+ ..Default::default()
+ })
+ .into(),
+ payload: Some(payload_metadata),
+ ..Default::default()
+ };
+
+ // Write metadata to file.
+ let metadata_path = temporary_directory.join("metadata");
+ let mut metadata_file = OpenOptions::new()
+ .create_new(true)
+ .read(true)
+ .write(true)
+ .open(&metadata_path)
+ .with_context(|| format!("Failed to open metadata file {:?}", metadata_path))?;
+ microdroid_metadata::write_metadata(&metadata, &mut metadata_file)?;
+
+ // Re-open the metadata file as read-only.
+ open_parcel_file(&metadata_path, false)
+}
+
+/// Creates a DiskImage with partitions:
+/// payload-metadata: metadata
+/// microdroid-apex-0: apex 0
+/// microdroid-apex-1: apex 1
+/// ..
+/// microdroid-apk: apk
+/// microdroid-apk-idsig: idsig
+/// extra-apk-0: additional apk 0
+/// extra-idsig-0: additional idsig 0
+/// extra-apk-1: additional apk 1
+/// extra-idsig-1: additional idsig 1
+/// ..
+fn make_payload_disk(
+ app_config: &VirtualMachineAppConfig,
+ apk_file: File,
+ idsig_file: File,
+ vm_payload_config: &VmPayloadConfig,
+ temporary_directory: &Path,
+) -> Result<DiskImage> {
+ if vm_payload_config.extra_apks.len() != app_config.extraIdsigs.len() {
+ bail!(
+ "payload config has {} apks, but app config has {} idsigs",
+ vm_payload_config.extra_apks.len(),
+ app_config.extraIdsigs.len()
+ );
+ }
+
+ let pm = PackageManager::new()?;
+ let apex_list = pm.get_apex_list(vm_payload_config.prefer_staged)?;
+
+ // collect APEXes from config
+ let mut apex_infos =
+ collect_apex_infos(&apex_list, &vm_payload_config.apexes, app_config.debugLevel);
+
+ // Pass sorted list of apexes. Sorting key shouldn't use `path` because it will change after
+ // reboot with prefer_staged. `last_update_seconds` is added to distinguish "samegrade"
+ // update.
+ apex_infos.sort_by_key(|info| (&info.name, &info.version, &info.last_update_seconds));
+ info!("Microdroid payload APEXes: {:?}", apex_infos.iter().map(|ai| &ai.name));
+
+ let metadata_file = make_metadata_file(app_config, &apex_infos, temporary_directory)?;
+ // put metadata at the first partition
+ let mut partitions = vec![Partition {
+ label: "payload-metadata".to_owned(),
+ image: Some(metadata_file),
+ writable: false,
+ }];
+
+ for (i, apex_info) in apex_infos.iter().enumerate() {
+ let apex_file = open_parcel_file(&apex_info.path, false)?;
+ partitions.push(Partition {
+ label: format!("microdroid-apex-{}", i),
+ image: Some(apex_file),
+ writable: false,
+ });
+ }
+ partitions.push(Partition {
+ label: "microdroid-apk".to_owned(),
+ image: Some(ParcelFileDescriptor::new(apk_file)),
+ writable: false,
+ });
+ partitions.push(Partition {
+ label: "microdroid-apk-idsig".to_owned(),
+ image: Some(ParcelFileDescriptor::new(idsig_file)),
+ writable: false,
+ });
+
+ // we've already checked that extra_apks and extraIdsigs are in the same size.
+ let extra_apks = &vm_payload_config.extra_apks;
+ let extra_idsigs = &app_config.extraIdsigs;
+ for (i, (extra_apk, extra_idsig)) in extra_apks.iter().zip(extra_idsigs.iter()).enumerate() {
+ partitions.push(Partition {
+ label: format!("extra-apk-{}", i),
+ image: Some(ParcelFileDescriptor::new(
+ File::open(PathBuf::from(&extra_apk.path)).with_context(|| {
+ format!("Failed to open the extra apk #{} {}", i, extra_apk.path)
+ })?,
+ )),
+ writable: false,
+ });
+
+ partitions.push(Partition {
+ label: format!("extra-idsig-{}", i),
+ image: Some(ParcelFileDescriptor::new(
+ extra_idsig
+ .as_ref()
+ .try_clone()
+ .with_context(|| format!("Failed to clone the extra idsig #{}", i))?,
+ )),
+ writable: false,
+ });
+ }
+
+ Ok(DiskImage { image: None, partitions, writable: false })
+}
+
+fn run_derive_classpath() -> Result<String> {
+ let result = Command::new("/apex/com.android.sdkext/bin/derive_classpath")
+ .arg("/proc/self/fd/1")
+ .output()
+ .context("Failed to run derive_classpath")?;
+
+ if !result.status.success() {
+ bail!("derive_classpath returned {}", result.status);
+ }
+
+ String::from_utf8(result.stdout).context("Converting derive_classpath output")
+}
+
+fn find_apex_names_in_classpath(classpath_vars: &str) -> Result<HashSet<String>> {
+ // Each line should be in the format "export <var name> <paths>", where <paths> is a
+ // colon-separated list of paths to JARs. We don't care about the var names, and we're only
+ // interested in paths that look like "/apex/<apex name>/<anything>" so we know which APEXes
+ // contribute to at least one var.
+ let mut apexes = HashSet::new();
+
+ let pattern = Regex::new(r"^export [^ ]+ ([^ ]+)$").context("Failed to construct Regex")?;
+ for line in classpath_vars.lines() {
+ if let Some(captures) = pattern.captures(line) {
+ if let Some(paths) = captures.get(1) {
+ apexes.extend(paths.as_str().split(':').filter_map(|path| {
+ let path = path.strip_prefix("/apex/")?;
+ Some(path[..path.find('/')?].to_owned())
+ }));
+ continue;
+ }
+ }
+ warn!("Malformed line from derive_classpath: {}", line);
+ }
+
+ Ok(apexes)
+}
+
+// Collect ApexInfos from VM config
+fn collect_apex_infos<'a>(
+ apex_list: &'a ApexInfoList,
+ apex_configs: &[ApexConfig],
+ debug_level: DebugLevel,
+) -> Vec<&'a ApexInfo> {
+ let mut additional_apexes: Vec<&str> = MICRODROID_REQUIRED_APEXES.to_vec();
+ if debug_level != DebugLevel::NONE {
+ additional_apexes.extend(MICRODROID_REQUIRED_APEXES_DEBUG.to_vec());
+ }
+
+ apex_list
+ .list
+ .iter()
+ .filter(|ai| {
+ apex_configs.iter().any(|cfg| ai.matches(cfg) && ai.is_active)
+ || additional_apexes.iter().any(|name| name == &ai.name && ai.is_active)
+ || ai.provide_shared_apex_libs
+ })
+ .collect()
+}
+
+pub fn add_microdroid_system_images(
+ config: &VirtualMachineAppConfig,
+ instance_file: File,
+ storage_image: Option<File>,
+ vm_config: &mut VirtualMachineRawConfig,
+) -> Result<()> {
+ let debug_suffix = match config.debugLevel {
+ DebugLevel::NONE => "normal",
+ DebugLevel::FULL => "debuggable",
+ _ => return Err(anyhow!("unsupported debug level: {:?}", config.debugLevel)),
+ };
+ let initrd = format!("/apex/com.android.virt/etc/microdroid_initrd_{}.img", debug_suffix);
+ vm_config.initrd = Some(open_parcel_file(Path::new(&initrd), false)?);
+
+ let mut writable_partitions = vec![Partition {
+ label: "vm-instance".to_owned(),
+ image: Some(ParcelFileDescriptor::new(instance_file)),
+ writable: true,
+ }];
+
+ if let Some(file) = storage_image {
+ writable_partitions.push(Partition {
+ label: "encryptedstore".to_owned(),
+ image: Some(ParcelFileDescriptor::new(file)),
+ writable: true,
+ });
+ }
+
+ vm_config.disks.push(DiskImage {
+ image: None,
+ partitions: writable_partitions,
+ writable: true,
+ });
+
+ Ok(())
+}
+
+pub fn add_microdroid_payload_images(
+ config: &VirtualMachineAppConfig,
+ temporary_directory: &Path,
+ apk_file: File,
+ idsig_file: File,
+ vm_payload_config: &VmPayloadConfig,
+ vm_config: &mut VirtualMachineRawConfig,
+) -> Result<()> {
+ vm_config.disks.push(make_payload_disk(
+ config,
+ apk_file,
+ idsig_file,
+ vm_payload_config,
+ temporary_directory,
+ )?);
+
+ Ok(())
+}
+
+#[cfg(test)]
+mod tests {
+ use super::*;
+ use tempfile::NamedTempFile;
+
+ #[test]
+ fn test_find_apex_names_in_classpath() {
+ let vars = r#"
+export FOO /apex/unterminated
+export BAR /apex/valid.apex/something
+wrong
+export EMPTY
+export OTHER /foo/bar:/baz:/apex/second.valid.apex/:gibberish:"#;
+ let expected = vec!["valid.apex", "second.valid.apex"];
+ let expected: HashSet<_> = expected.into_iter().map(ToString::to_string).collect();
+
+ assert_eq!(find_apex_names_in_classpath(vars).unwrap(), expected);
+ }
+
+ #[test]
+ fn test_collect_apexes() {
+ let apex_info_list = ApexInfoList {
+ list: vec![
+ ApexInfo {
+ // 0
+ name: "com.android.adbd".to_string(),
+ path: PathBuf::from("adbd"),
+ has_classpath_jar: false,
+ last_update_seconds: 12345678,
+ is_factory: true,
+ is_active: true,
+ ..Default::default()
+ },
+ ApexInfo {
+ // 1
+ name: "com.android.os.statsd".to_string(),
+ path: PathBuf::from("statsd"),
+ has_classpath_jar: false,
+ last_update_seconds: 12345678,
+ is_factory: true,
+ is_active: false,
+ ..Default::default()
+ },
+ ApexInfo {
+ // 2
+ name: "com.android.os.statsd".to_string(),
+ path: PathBuf::from("statsd/updated"),
+ has_classpath_jar: false,
+ last_update_seconds: 12345678 + 1,
+ is_factory: false,
+ is_active: true,
+ ..Default::default()
+ },
+ ApexInfo {
+ // 3
+ name: "no_classpath".to_string(),
+ path: PathBuf::from("no_classpath"),
+ has_classpath_jar: false,
+ last_update_seconds: 12345678,
+ is_factory: true,
+ is_active: true,
+ ..Default::default()
+ },
+ ApexInfo {
+ // 4
+ name: "has_classpath".to_string(),
+ path: PathBuf::from("has_classpath"),
+ has_classpath_jar: true,
+ last_update_seconds: 87654321,
+ is_factory: true,
+ is_active: false,
+ ..Default::default()
+ },
+ ApexInfo {
+ // 5
+ name: "has_classpath".to_string(),
+ path: PathBuf::from("has_classpath/updated"),
+ has_classpath_jar: true,
+ last_update_seconds: 87654321 + 1,
+ is_factory: false,
+ is_active: true,
+ ..Default::default()
+ },
+ ApexInfo {
+ // 6
+ name: "apex-foo".to_string(),
+ path: PathBuf::from("apex-foo"),
+ has_classpath_jar: false,
+ last_update_seconds: 87654321,
+ is_factory: true,
+ is_active: false,
+ ..Default::default()
+ },
+ ApexInfo {
+ // 7
+ name: "apex-foo".to_string(),
+ path: PathBuf::from("apex-foo/updated"),
+ has_classpath_jar: false,
+ last_update_seconds: 87654321 + 1,
+ is_factory: false,
+ is_active: true,
+ ..Default::default()
+ },
+ ApexInfo {
+ // 8
+ name: "sharedlibs".to_string(),
+ path: PathBuf::from("apex-foo"),
+ last_update_seconds: 87654321,
+ is_factory: true,
+ provide_shared_apex_libs: true,
+ ..Default::default()
+ },
+ ApexInfo {
+ // 9
+ name: "sharedlibs".to_string(),
+ path: PathBuf::from("apex-foo/updated"),
+ last_update_seconds: 87654321 + 1,
+ is_active: true,
+ provide_shared_apex_libs: true,
+ ..Default::default()
+ },
+ ],
+ };
+ let apex_configs = vec![
+ ApexConfig { name: "apex-foo".to_string() },
+ ApexConfig { name: "{CLASSPATH}".to_string() },
+ ];
+ assert_eq!(
+ collect_apex_infos(&apex_info_list, &apex_configs, DebugLevel::FULL),
+ vec![
+ // Pass active/required APEXes
+ &apex_info_list.list[0],
+ &apex_info_list.list[2],
+ // Pass active APEXes specified in the config
+ &apex_info_list.list[5],
+ &apex_info_list.list[7],
+ // Pass both preinstalled(inactive) and updated(active) for "sharedlibs" APEXes
+ &apex_info_list.list[8],
+ &apex_info_list.list[9],
+ ]
+ );
+ }
+
+ #[test]
+ fn test_prefer_staged_apex_with_factory_active_apex() {
+ let single_apex = ApexInfo {
+ name: "foo".to_string(),
+ version: 1,
+ path: PathBuf::from("foo.apex"),
+ is_factory: true,
+ is_active: true,
+ ..Default::default()
+ };
+ let mut apex_info_list = ApexInfoList { list: vec![single_apex.clone()] };
+
+ let staged = NamedTempFile::new().unwrap();
+ apex_info_list
+ .override_staged_apex(&StagedApexInfo {
+ moduleName: "foo".to_string(),
+ versionCode: 2,
+ diskImagePath: staged.path().to_string_lossy().to_string(),
+ ..Default::default()
+ })
+ .expect("should be ok");
+
+ assert_eq!(
+ apex_info_list,
+ ApexInfoList {
+ list: vec![
+ ApexInfo {
+ version: 2,
+ is_factory: false,
+ path: staged.path().to_owned(),
+ last_update_seconds: last_updated(staged.path()).unwrap(),
+ ..single_apex.clone()
+ },
+ ApexInfo { is_active: false, ..single_apex },
+ ],
+ }
+ );
+ }
+
+ #[test]
+ fn test_prefer_staged_apex_with_factory_and_inactive_apex() {
+ let factory_apex = ApexInfo {
+ name: "foo".to_string(),
+ version: 1,
+ path: PathBuf::from("foo.apex"),
+ is_factory: true,
+ ..Default::default()
+ };
+ let active_apex = ApexInfo {
+ name: "foo".to_string(),
+ version: 2,
+ path: PathBuf::from("foo.downloaded.apex"),
+ is_active: true,
+ ..Default::default()
+ };
+ let mut apex_info_list =
+ ApexInfoList { list: vec![factory_apex.clone(), active_apex.clone()] };
+
+ let staged = NamedTempFile::new().unwrap();
+ apex_info_list
+ .override_staged_apex(&StagedApexInfo {
+ moduleName: "foo".to_string(),
+ versionCode: 3,
+ diskImagePath: staged.path().to_string_lossy().to_string(),
+ ..Default::default()
+ })
+ .expect("should be ok");
+
+ assert_eq!(
+ apex_info_list,
+ ApexInfoList {
+ list: vec![
+ // factory apex isn't touched
+ factory_apex,
+ // update active one
+ ApexInfo {
+ version: 3,
+ path: staged.path().to_owned(),
+ last_update_seconds: last_updated(staged.path()).unwrap(),
+ ..active_apex
+ },
+ ],
+ }
+ );
+ }
+}
diff --git a/virtualizationmanager/src/selinux.rs b/virtualizationmanager/src/selinux.rs
new file mode 100644
index 0000000..0485943
--- /dev/null
+++ b/virtualizationmanager/src/selinux.rs
@@ -0,0 +1,120 @@
+// Copyright 2021, The Android Open Source Project
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+// http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+//! Wrapper to libselinux
+
+use anyhow::{anyhow, bail, Context, Result};
+use std::ffi::{CStr, CString};
+use std::fmt;
+use std::fs::File;
+use std::io;
+use std::ops::Deref;
+use std::os::raw::c_char;
+use std::os::unix::io::AsRawFd;
+use std::ptr;
+
+// Partially copied from system/security/keystore2/selinux/src/lib.rs
+/// SeContext represents an SELinux context string. It can take ownership of a raw
+/// s-string as allocated by `getcon` or `selabel_lookup`. In this case it uses
+/// `freecon` to free the resources when dropped. In its second variant it stores
+/// an `std::ffi::CString` that can be initialized from a Rust string slice.
+#[derive(Debug)]
+#[allow(dead_code)] // CString variant is used in tests
+pub enum SeContext {
+ /// Wraps a raw context c-string as returned by libselinux.
+ Raw(*mut ::std::os::raw::c_char),
+ /// Stores a context string as `std::ffi::CString`.
+ CString(CString),
+}
+
+impl PartialEq for SeContext {
+ fn eq(&self, other: &Self) -> bool {
+ // We dereference both and thereby delegate the comparison
+ // to `CStr`'s implementation of `PartialEq`.
+ **self == **other
+ }
+}
+
+impl Eq for SeContext {}
+
+impl fmt::Display for SeContext {
+ fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
+ write!(f, "{}", self.to_str().unwrap_or("Invalid context"))
+ }
+}
+
+impl Drop for SeContext {
+ fn drop(&mut self) {
+ if let Self::Raw(p) = self {
+ // SAFETY: SeContext::Raw is created only with a pointer that is set by libselinux and
+ // has to be freed with freecon.
+ unsafe { selinux_bindgen::freecon(*p) };
+ }
+ }
+}
+
+impl Deref for SeContext {
+ type Target = CStr;
+
+ fn deref(&self) -> &Self::Target {
+ match self {
+ // SAFETY: the non-owned C string pointed by `p` is guaranteed to be valid (non-null
+ // and shorter than i32::MAX). It is freed when SeContext is dropped.
+ Self::Raw(p) => unsafe { CStr::from_ptr(*p) },
+ Self::CString(cstr) => cstr,
+ }
+ }
+}
+
+impl SeContext {
+ /// Initializes the `SeContext::CString` variant from a Rust string slice.
+ #[allow(dead_code)] // Used in tests
+ pub fn new(con: &str) -> Result<Self> {
+ Ok(Self::CString(
+ CString::new(con)
+ .with_context(|| format!("Failed to create SeContext with \"{}\"", con))?,
+ ))
+ }
+
+ pub fn selinux_type(&self) -> Result<&str> {
+ let context = self.deref().to_str().context("Label is not valid UTF8")?;
+
+ // The syntax is user:role:type:sensitivity[:category,...],
+ // ignoring security level ranges, which don't occur on Android. See
+ // https://github.com/SELinuxProject/selinux-notebook/blob/main/src/security_context.md
+ // We only want the type.
+ let fields: Vec<_> = context.split(':').collect();
+ if fields.len() < 4 || fields.len() > 5 {
+ bail!("Syntactically invalid label {}", self);
+ }
+ Ok(fields[2])
+ }
+}
+
+pub fn getfilecon(file: &File) -> Result<SeContext> {
+ let fd = file.as_raw_fd();
+ let mut con: *mut c_char = ptr::null_mut();
+ // SAFETY: the returned pointer `con` is wrapped in SeContext::Raw which is freed with
+ // `freecon` when it is dropped.
+ match unsafe { selinux_bindgen::fgetfilecon(fd, &mut con) } {
+ 1.. => {
+ if !con.is_null() {
+ Ok(SeContext::Raw(con))
+ } else {
+ Err(anyhow!("fgetfilecon returned a NULL context"))
+ }
+ }
+ _ => Err(anyhow!(io::Error::last_os_error())).context("fgetfilecon failed"),
+ }
+}