feat(ovfs): support getattr and setattr (#4987)

* feat: support getattr and setattr

* typo

integrations/virtiofs/Cargo.toml

@@ -32,10 +32,12 @@ anyhow = { version = "1.0.86", features = ["std"] }
 libc = "0.2.139"
 log = "0.4.22"
 opendal = { version = "0.48.0", path = "../../core" }
-snafu = "0.8.3"
-vhost = "0.11.0"
-vhost-user-backend = "0.14.0"
-virtio-bindings = "0.2.1"
+sharded-slab = "0.1.7"
+snafu = "0.8.4"
+tokio = { version = "1.39.2", features = ["rt-multi-thread"] }
+vhost = "0.10.0"
+vhost-user-backend = "0.13.1"
+virtio-bindings = { version = "0.2.1", features = ["virtio-v5_0_0"] }
 virtio-queue = "0.11.0"
 vm-memory = { version = "0.14.0", features = [
   "backend-mmap",

integrations/virtiofs/src/error.rs

@@ -15,6 +15,7 @@
 // specific language governing permissions and limitations
 // under the License.
 
+use std::ffi::CStr;
 use std::io;
 
 use anyhow::Error as AnyError;
@@ -38,6 +39,22 @@ pub enum Error {
     },
 }
 
+impl From<libc::c_int> for Error {
+    fn from(errno: libc::c_int) -> Error {
+        let err_str = unsafe { libc::strerror(errno) };
+        let message = if err_str.is_null() {
+            format!("errno: {}", errno)
+        } else {
+            let c_str = unsafe { CStr::from_ptr(err_str) };
+            c_str.to_string_lossy().into_owned()
+        };
+        Error::VhostUserFsError {
+            message,
+            source: None,
+        }
+    }
+}
+
 impl From<Error> for io::Error {
     fn from(error: Error) -> io::Error {
         match error {

integrations/virtiofs/src/filesystem.rs

@@ -17,8 +17,14 @@
 
 use std::io::Write;
 use std::mem::size_of;
+use std::time::Duration;
 
+use log::debug;
+use opendal::ErrorKind;
 use opendal::Operator;
+use sharded_slab::Slab;
+use tokio::runtime::Builder;
+use tokio::runtime::Runtime;
 use vm_memory::ByteValued;
 
 use crate::error::*;
@@ -32,21 +38,99 @@ const KERNEL_MINOR_VERSION: u32 = 38;
 /// Minimum Minor version number supported.
 const MIN_KERNEL_MINOR_VERSION: u32 = 27;
 /// The length of the header part of the message.
-const BUFFER_HEADER_SIZE: u32 = 256;
+const BUFFER_HEADER_SIZE: u32 = 4096;
 /// The maximum length of the data part of the message, used for read/write data.
 const MAX_BUFFER_SIZE: u32 = 1 << 20;
+/// The default time to live of the attributes.
+const DEFAULT_TTL: Duration = Duration::from_secs(1);
+/// The default mode of the opened file.
+const DEFAULT_OPENED_FILE_MODE: u32 = 0o755;
+
+enum FileType {
+    Dir,
+    File,
+}
+
+struct OpenedFile {
+    path: String,
+    metadata: Attr,
+}
+
+impl OpenedFile {
+    fn new(file_type: FileType, path: &str, uid: u32, gid: u32) -> OpenedFile {
+        let mut attr: Attr = unsafe { std::mem::zeroed() };
+        attr.uid = uid;
+        attr.gid = gid;
+        match file_type {
+            FileType::Dir => {
+                attr.nlink = 2;
+                attr.mode = libc::S_IFDIR | DEFAULT_OPENED_FILE_MODE;
+            }
+            FileType::File => {
+                attr.nlink = 1;
+                attr.mode = libc::S_IFREG | DEFAULT_OPENED_FILE_MODE;
+            }
+        }
+        OpenedFile {
+            path: path.to_string(),
+            metadata: attr,
+        }
+    }
+}
+
+fn opendal_error2error(error: opendal::Error) -> Error {
+    match error.kind() {
+        ErrorKind::Unsupported => Error::from(libc::EOPNOTSUPP),
+        ErrorKind::IsADirectory => Error::from(libc::EISDIR),
+        ErrorKind::NotFound => Error::from(libc::ENOENT),
+        ErrorKind::PermissionDenied => Error::from(libc::EACCES),
+        ErrorKind::AlreadyExists => Error::from(libc::EEXIST),
+        ErrorKind::NotADirectory => Error::from(libc::ENOTDIR),
+        ErrorKind::RangeNotSatisfied => Error::from(libc::EINVAL),
+        ErrorKind::RateLimited => Error::from(libc::EBUSY),
+        _ => Error::from(libc::ENOENT),
+    }
+}
+
+fn opendal_metadata2opened_file(
+    path: &str,
+    metadata: &opendal::Metadata,
+    uid: u32,
+    gid: u32,
+) -> OpenedFile {
+    let file_type = match metadata.mode() {
+        opendal::EntryMode::DIR => FileType::Dir,
+        _ => FileType::File,
+    };
+    OpenedFile::new(file_type, path, uid, gid)
+}
 
 /// Filesystem is a filesystem implementation with opendal backend,
 /// and will decode and process messages from VMs.
 pub struct Filesystem {
-    // FIXME: #[allow(dead_code)] here should be removed in the future.
-    #[allow(dead_code)]
+    rt: Runtime,
     core: Operator,
+    uid: u32,
+    gid: u32,
+    opened_files: Slab<OpenedFile>,
 }
 
 impl Filesystem {
     pub fn new(core: Operator) -> Filesystem {
-        Filesystem { core }
+        let rt = Builder::new_multi_thread()
+            .worker_threads(4)
+            .enable_all()
+            .build()
+            .unwrap();
+
+        // Here we set the uid and gid to 1000, which is the default value.
+        Filesystem {
+            rt,
+            core,
+            uid: 1000,
+            gid: 1000,
+            opened_files: Slab::new(),
+        }
     }
 
     pub fn handle_message(&self, mut r: Reader, w: Writer) -> Result<usize> {
@@ -60,6 +144,9 @@ impl Filesystem {
         if let Ok(opcode) = Opcode::try_from(in_header.opcode) {
             match opcode {
                 Opcode::Init => self.init(in_header, r, w),
+                Opcode::Destroy => self.destroy(in_header, r, w),
+                Opcode::Getattr => self.getattr(in_header, r, w),
+                Opcode::Setattr => self.setattr(in_header, r, w),
             }
         } else {
             Filesystem::reply_error(in_header.unique, w)
@@ -134,4 +221,52 @@ impl Filesystem {
         };
         Filesystem::reply_ok(Some(out), None, in_header.unique, w)
     }
+
+    fn destroy(&self, _in_header: InHeader, _r: Reader, _w: Writer) -> Result<usize> {
+        // do nothing for destroy.
+        Ok(0)
+    }
+
+    fn getattr(&self, in_header: InHeader, _r: Reader, w: Writer) -> Result<usize> {
+        debug!("getattr: inode={}", in_header.nodeid);
+
+        let path = match self
+            .opened_files
+            .get(in_header.nodeid as usize)
+            .map(|f| f.path.clone())
+        {
+            Some(path) => path,
+            None => return Filesystem::reply_error(in_header.unique, w),
+        };
+
+        let mut metadata = match self.rt.block_on(self.do_get_metadata(&path)) {
+            Ok(metadata) => metadata,
+            Err(_) => return Filesystem::reply_error(in_header.unique, w),
+        };
+        metadata.metadata.ino = in_header.nodeid;
+
+        let out = AttrOut {
+            attr_valid: DEFAULT_TTL.as_secs(),
+            attr_valid_nsec: DEFAULT_TTL.subsec_nanos(),
+            attr: metadata.metadata,
+            ..Default::default()
+        };
+        Filesystem::reply_ok(Some(out), None, in_header.unique, w)
+    }
+
+    fn setattr(&self, in_header: InHeader, _r: Reader, w: Writer) -> Result<usize> {
+        debug!("setattr: inode={}", in_header.nodeid);
+
+        // do nothing for setattr.
+        self.getattr(in_header, _r, w)
+    }
+}
+
+impl Filesystem {
+    async fn do_get_metadata(&self, path: &str) -> Result<OpenedFile> {
+        let metadata = self.core.stat(path).await.map_err(opendal_error2error)?;
+        let attr = opendal_metadata2opened_file(path, &metadata, self.uid, self.gid);
+
+        Ok(attr)
+    }
 }

integrations/virtiofs/src/filesystem_message.rs

@@ -23,20 +23,52 @@ use crate::error::*;
 /// The corresponding value needs to be aligned with the specification.
 #[non_exhaustive]
 pub enum Opcode {
+    Getattr = 3,
+    Setattr = 4,
     Init = 26,
+    Destroy = 38,
 }
 
 impl TryFrom<u32> for Opcode {
     type Error = Error;
 
     fn try_from(value: u32) -> Result<Self, Self::Error> {
         match value {
+            3 => Ok(Opcode::Getattr),
+            4 => Ok(Opcode::Setattr),
             26 => Ok(Opcode::Init),
+            38 => Ok(Opcode::Destroy),
             _ => Err(new_vhost_user_fs_error("failed to decode opcode", None)),
         }
     }
 }
 
+/// Attr represents the file attributes in virtiofs.
+///
+/// The fields of the struct need to conform to the specific format of the virtiofs message.
+/// Currently, we only need to align them exactly with virtiofsd.
+/// Reference: https://gitlab.com/virtio-fs/virtiofsd/-/blob/main/src/fuse.rs?ref_type=heads#L577
+#[repr(C)]
+#[derive(Debug, Default, Clone, Copy)]
+pub struct Attr {
+    pub ino: u64,
+    pub size: u64,
+    pub blocks: u64,
+    pub atime: u64,
+    pub mtime: u64,
+    pub ctime: u64,
+    pub atimensec: u32,
+    pub mtimensec: u32,
+    pub ctimensec: u32,
+    pub mode: u32,
+    pub nlink: u32,
+    pub uid: u32,
+    pub gid: u32,
+    pub rdev: u32,
+    pub blksize: u32,
+    pub flags: u32,
+}
+
 /// InHeader represents the incoming message header in the filesystem call.
 ///
 /// The fields of the struct need to conform to the specific format of the virtiofs message.
@@ -105,9 +137,25 @@ pub struct InitOut {
     pub unused: [u32; 7],
 }
 
+/// AttrOut is used to return the file attributes in the filesystem call.
+///
+/// The fields of the struct need to conform to the specific format of the virtiofs message.
+/// Currently, we only need to align them exactly with virtiofsd.
+/// Reference: https://gitlab.com/virtio-fs/virtiofsd/-/blob/main/src/fuse.rs?ref_type=heads#L782
+#[repr(C)]
+#[derive(Debug, Default, Clone, Copy)]
+pub struct AttrOut {
+    pub attr_valid: u64,
+    pub attr_valid_nsec: u32,
+    pub dummy: u32,
+    pub attr: Attr,
+}
+
 /// We will use ByteValued to implement the encoding and decoding
 /// of these structures in shared memory.
+unsafe impl ByteValued for Attr {}
 unsafe impl ByteValued for InHeader {}
 unsafe impl ByteValued for OutHeader {}
 unsafe impl ByteValued for InitIn {}
 unsafe impl ByteValued for InitOut {}
+unsafe impl ByteValued for AttrOut {}