oci测试及runc

net/ax-net/src/unix/dgram.rs

@@ -72,6 +72,8 @@ impl Bind {
 pub struct DgramTransport {
     /// Receiver installed when the socket is bound or paired.
     data_rx: Mutex<Option<(async_channel::Receiver<Packet>, Arc<PollSet>)>>,
+    /// Datagram retained by `MSG_PEEK` until a consuming receive.
+    peeked: Mutex<Option<Packet>>,
     /// Direct peer channel for connected datagram sockets.
     connected: RwLock<Option<Channel>>,
     /// Address reported as sender on outgoing datagrams.
@@ -88,6 +90,7 @@ impl DgramTransport {
     pub fn new(pid: u32) -> Self {
         DgramTransport {
             data_rx: Mutex::new(None),
+            peeked: Mutex::new(None),
             connected: RwLock::new(None),
             local_addr: RwLock::new(UnixSocketAddr::Unnamed),
             poll_state: Arc::default(),
@@ -103,6 +106,7 @@ impl DgramTransport {
     ) -> Self {
         DgramTransport {
             data_rx: Mutex::new(Some(data_rx)),
+            peeked: Mutex::new(None),
             connected: RwLock::new(Some(connected)),
             local_addr: RwLock::new(UnixSocketAddr::Unnamed),
             poll_state: Arc::default(),
@@ -265,18 +269,46 @@ impl TransportOps for DgramTransport {
     fn recv(&self, mut dst: impl Write, mut options: RecvOptions) -> AxResult<usize> {
         let extra_nb = options.flags.contains(RecvFlags::DONTWAIT);
         self.general.recv_poller_with(self, extra_nb, move || {
+            let mut peeked = self.peeked.lock();
             let mut guard = self.data_rx.lock();
             let Some((rx, _)) = guard.as_mut() else {
                 return Err(AxError::NotConnected);
             };
 
-            let Packet { data, cmsg, sender } = match rx.try_recv() {
-                Ok(packet) => packet,
-                Err(TryRecvError::Empty) => {
-                    return Err(AxError::WouldBlock);
+            if options.flags.contains(RecvFlags::PEEK) {
+                if peeked.is_none() {
+                    *peeked = Some(match rx.try_recv() {
+                        Ok(packet) => packet,
+                        Err(TryRecvError::Empty) => return Err(AxError::WouldBlock),
+                        Err(TryRecvError::Closed) => return Ok(0),
+                    });
                 }
-                Err(TryRecvError::Closed) => {
-                    return Ok(0);
+                let packet = peeked.as_ref().unwrap();
+                let count = dst.write(&packet.data)?;
+                if count < packet.data.len() {
+                    warn!(
+                        "UDP message truncated: {} -> {} bytes",
+                        packet.data.len(),
+                        count
+                    );
+                }
+                if let Some(from) = options.from.as_mut() {
+                    **from = SocketAddrEx::Unix(packet.sender.clone());
+                }
+                return Ok(if options.flags.contains(RecvFlags::TRUNCATE) {
+                    packet.data.len()
+                } else {
+                    count
+                });
+            }
+
+            let Packet { data, cmsg, sender } = if let Some(packet) = peeked.take() {
+                packet
+            } else {
+                match rx.try_recv() {
+                    Ok(packet) => packet,
+                    Err(TryRecvError::Empty) => return Err(AxError::WouldBlock),
+                    Err(TryRecvError::Closed) => return Ok(0),
                 }
             };
             let count = dst.write(&data)?;
@@ -303,8 +335,9 @@ impl TransportOps for DgramTransport {
 impl Pollable for DgramTransport {
     fn poll(&self) -> IoEvents {
         let mut events = IoEvents::OUT;
+        let has_peeked = self.peeked.lock().is_some();
         if let Some((rx, _)) = self.data_rx.lock().as_ref() {
-            events.set(IoEvents::IN, !rx.is_empty());
+            events.set(IoEvents::IN, has_peeked || !rx.is_empty());
         }
         events
     }

os/StarryOS/axnsproxy/src/mnt.rs

@@ -1,5 +1,8 @@
 use alloc::sync::Arc;
-use core::ffi::c_char;
+use core::{
+    ffi::c_char,
+    sync::atomic::{AtomicU64, Ordering},
+};
 
 use ax_kspin::SpinNoIrq;
 
@@ -8,6 +11,8 @@ use ax_kspin::SpinNoIrq;
 pub static ROOT_MNT_NS: spin::LazyLock<Arc<SpinNoIrq<MntNamespace>>> =
     spin::LazyLock::new(|| Arc::new(SpinNoIrq::new(MntNamespace::new_root())));
 
+static NEXT_MNT_NS_ID: AtomicU64 = AtomicU64::new(2);
+
 const fn pad_mnt_root(root: &str) -> [c_char; 256] {
     let mut data: [c_char; 256] = [0; 256];
     unsafe {
@@ -16,21 +21,28 @@ const fn pad_mnt_root(root: &str) -> [c_char; 256] {
     data
 }
 
-/// Per-process mount namespace.
-/// Isolates the set of filesystem mount points seen by a process.
-/// In the root namespace `root` starts as `"/"`.
+/// Mount namespace identity metadata.
+///
+/// The namespace-local mount tree is owned by `ax-fs-ng`'s `FsContext`;
+/// this object supplies the stable identity shared by nsproxy and namespace
+/// file descriptors.
 pub struct MntNamespace {
+    pub ns_id: u64,
     pub root: [c_char; 256],
 }
 
 impl MntNamespace {
     pub fn new_root() -> Self {
         Self {
+            ns_id: 1,
             root: pad_mnt_root("/"),
         }
     }
 
     pub fn clone_ns(&self) -> Self {
-        Self { root: self.root }
+        Self {
+            ns_id: NEXT_MNT_NS_ID.fetch_add(1, Ordering::Relaxed),
+            root: self.root,
+        }
     }
 }

os/StarryOS/docs/mount-umount2-linux-compat.md

@@ -662,10 +662,11 @@ pub fn detach(self: &Arc<Self>) -> VfsResult<()> {
 |---|---|---|
 | `MS_SHARED | MS_PRIVATE` | `EINVAL` | 已支持 |
 | `MS_SHARED | MS_BIND` | `EINVAL` | 已支持 |
-| `MS_PRIVATE` | 修改已有 mount 的传播属性 | 已支持当前测试覆盖的 private 语义 |
-| `MS_SHARED` | 修改已有 mount 的传播属性 | 已支持当前测试覆盖的 shared peer 传播语义 |
-| `MS_SLAVE` | 修改已有 mount 的传播属性 | 已支持当前测试覆盖的 slave 单向传播语义 |
+| `MS_PRIVATE` | 修改已有 mount 的传播属性 | 已支持当前测试覆盖的 private 语义，停止后续传播 |
+| `MS_SHARED` | 修改已有 mount 的传播属性 | 已支持当前测试覆盖的 shared peer 双向传播语义 |
+| `MS_SLAVE` | 修改已有 mount 的传播属性 | 已支持当前测试覆盖的 master → slave 单向传播语义 |
 | `MS_UNBINDABLE` | 修改已有 mount 的传播属性 | 已支持 bind 禁止与 recursive bind prune |
+| propagation flag + `MS_REC` | 递归修改 mount subtree | 已支持 |
 | `MS_BIND` | 普通 bind，不带 nested submount | 已支持 |
 | `MS_BIND` on subdirectory | bind mount 内部子目录 | 已支持 |
 | `MS_BIND|MS_REC` | recursive bind，带 nested submount | 已支持 |
@@ -715,11 +716,15 @@ pub fn detach(self: &Arc<Self>) -> VfsResult<()> {
 
 ## 8. 当前仍未覆盖或未完全实现的点
 
-虽然这次 `util-linux` 测试已经全绿，但仍有一些 Linux mount/namespace 语义没有在本轮实现：
+虽然 shared subtree 的核心传播路径已经实现，但仍有一些更深层 Linux
+mount/namespace 语义没有完全覆盖：
 
 - 更完整的 shared subtree propagation corner case
 - 更复杂 namespace 拓扑下的传播行为
 - 真实 peer group 生命周期管理
+- 复杂多级 shared+slave 拓扑中的 `propagate_from` 展示与边界行为
+- `MS_MOVE` 在 shared subtree 中的全部 Linux 限制和传播规则
+- detached mount、过挂载堆栈和传播失败时的完整事务回滚
 - 其他普通 mount flags，如 `MS_NODEV`、`MS_NOSUID`、`MS_NOEXEC` 等
 - `MNT_FORCE` 的真实强制卸载语义