virtio/fs/macos: implement DAX support

Bring virtiofs DAX support into macOS relying on the mechanisms added for supporting virtio-gpu blobs, which allow us to request HVF the injection of memory regions. Signed-off-by: Sergio Lopez <[email protected]>
containers · Aug 28, 2024 · ca9ade8 · ca9ade8
1 parent f448d63
commit ca9ade8
Show file tree

Hide file tree

Showing 6 changed files with 227 additions and 9 deletions.
diff --git a/src/devices/src/virtio/fs/device.rs b/src/devices/src/virtio/fs/device.rs
@@ -1,9 +1,13 @@
+#[cfg(target_os = "macos")]
+use crossbeam_channel::Sender;
 use std::cmp;
 use std::io::Write;
 use std::sync::atomic::AtomicUsize;
 use std::sync::{Arc, Mutex};
 use std::thread::JoinHandle;
 
+#[cfg(target_os = "macos")]
+use hvf::MemoryMapping;
 use utils::eventfd::{EventFd, EFD_NONBLOCK};
 use virtio_bindings::{virtio_config::VIRTIO_F_VERSION_1, virtio_ring::VIRTIO_RING_F_EVENT_IDX};
 use vm_memory::{ByteValued, GuestMemoryMmap};
@@ -49,6 +53,8 @@ pub struct Fs {
     passthrough_cfg: passthrough::Config,
     worker_thread: Option<JoinHandle<()>>,
     worker_stopfd: EventFd,
+    #[cfg(target_os = "macos")]
+    map_sender: Option<Sender<MemoryMapping>>,
 }
 
 impl Fs {
@@ -90,6 +96,8 @@ impl Fs {
             passthrough_cfg: fs_cfg,
             worker_thread: None,
             worker_stopfd: EventFd::new(EFD_NONBLOCK).map_err(FsError::EventFd)?,
+            #[cfg(target_os = "macos")]
+            map_sender: None,
         })
     }
 
@@ -112,6 +120,11 @@ impl Fs {
     pub fn set_shm_region(&mut self, shm_region: VirtioShmRegion) {
         self.shm_region = Some(shm_region);
     }
+
+    #[cfg(target_os = "macos")]
+    pub fn set_map_sender(&mut self, map_sender: Sender<MemoryMapping>) {
+        self.map_sender = Some(map_sender);
+    }
 }
 
 impl VirtioDevice for Fs {
@@ -202,6 +215,8 @@ impl VirtioDevice for Fs {
             self.shm_region.clone(),
             self.passthrough_cfg.clone(),
             self.worker_stopfd.try_clone().unwrap(),
+            #[cfg(target_os = "macos")]
+            self.map_sender.clone(),
         );
         self.worker_thread = Some(worker.run());
 

diff --git a/src/devices/src/virtio/fs/filesystem.rs b/src/devices/src/virtio/fs/filesystem.rs
@@ -2,6 +2,11 @@
 // Use of this source code is governed by a BSD-style license that can be
 // found in the LICENSE file.
 
+#[cfg(target_os = "macos")]
+use crossbeam_channel::Sender;
+#[cfg(target_os = "macos")]
+use hvf::MemoryMapping;
+
 use std::convert::TryInto;
 use std::ffi::{CStr, CString};
 use std::fs::File;
@@ -1121,6 +1126,7 @@ pub trait FileSystem {
         moffset: u64,
         host_shm_base: u64,
         shm_size: u64,
+        #[cfg(target_os = "macos")] map_sender: &Option<Sender<MemoryMapping>>,
     ) -> io::Result<()> {
         Err(io::Error::from_raw_os_error(libc::ENOSYS))
     }
@@ -1131,6 +1137,7 @@ pub trait FileSystem {
         requests: Vec<RemovemappingOne>,
         host_shm_base: u64,
         shm_size: u64,
+        #[cfg(target_os = "macos")] map_sender: &Option<Sender<MemoryMapping>>,
     ) -> io::Result<()> {
         Err(io::Error::from_raw_os_error(libc::ENOSYS))
     }

diff --git a/src/devices/src/virtio/fs/macos/passthrough.rs b/src/devices/src/virtio/fs/macos/passthrough.rs
@@ -4,18 +4,22 @@
 
 use std::collections::btree_map;
 use std::collections::BTreeMap;
+use std::collections::HashMap;
 use std::ffi::{CStr, CString};
 use std::fs::File;
 use std::io;
 #[cfg(not(feature = "efi"))]
 use std::mem;
 use std::mem::MaybeUninit;
 use std::os::unix::io::{AsRawFd, FromRawFd, RawFd};
+use std::ptr::null_mut;
 use std::str::FromStr;
 use std::sync::atomic::{AtomicBool, AtomicU64, Ordering};
 use std::sync::{Arc, Mutex, RwLock};
 use std::time::Duration;
 
+use crossbeam_channel::{unbounded, Sender};
+use hvf::MemoryMapping;
 use vm_memory::ByteValued;
 
 use crate::virtio::fs::filesystem::SecContext;
@@ -411,6 +415,8 @@ pub struct PassthroughFs {
     next_handle: AtomicU64,
     init_handle: u64,
 
+    map_windows: Mutex<HashMap<u64, u64>>,
+
     // Whether writeback caching is enabled for this directory. This will only be true when
     // `cfg.writeback` is true and `init` was called with `FsOptions::WRITEBACK_CACHE`.
     writeback: AtomicBool,
@@ -446,6 +452,8 @@ impl PassthroughFs {
             next_handle: AtomicU64::new(1),
             init_handle: 0,
 
+            map_windows: Mutex::new(HashMap::new()),
+
             writeback: AtomicBool::new(false),
             cfg,
         })
@@ -1802,4 +1810,135 @@ impl FileSystem for PassthroughFs {
             Ok(res as u64)
         }
     }
+
+    fn setupmapping(
+        &self,
+        _ctx: Context,
+        inode: Inode,
+        _handle: Handle,
+        foffset: u64,
+        len: u64,
+        flags: u64,
+        moffset: u64,
+        guest_shm_base: u64,
+        shm_size: u64,
+        map_sender: &Option<Sender<MemoryMapping>>,
+    ) -> io::Result<()> {
+        if map_sender.is_none() {
+            return Err(linux_error(io::Error::from_raw_os_error(libc::ENOSYS)));
+        }
+
+        let prot_flags = if (flags & fuse::SetupmappingFlags::WRITE.bits()) != 0 {
+            libc::PROT_READ | libc::PROT_WRITE
+        } else {
+            libc::PROT_READ
+        };
+
+        if (moffset + len) > shm_size {
+            return Err(linux_error(io::Error::from_raw_os_error(libc::EINVAL)));
+        }
+
+        let guest_addr = guest_shm_base + moffset;
+
+        debug!(
+            "setupmapping: ino {:?} guest_addr={:x} len={}",
+            inode, guest_addr, len
+        );
+
+        let file = self.open_inode(inode, libc::O_RDWR)?;
+        let fd = file.as_raw_fd();
+
+        let host_addr = unsafe {
+            libc::mmap(
+                null_mut(),
+                len as usize,
+                prot_flags,
+                libc::MAP_SHARED,
+                fd,
+                foffset as libc::off_t,
+            )
+        };
+        if host_addr == libc::MAP_FAILED {
+            return Err(linux_error(io::Error::last_os_error()));
+        }
+
+        let ret = unsafe { libc::close(fd) };
+        if ret == -1 {
+            return Err(linux_error(io::Error::last_os_error()));
+        }
+
+        // We've checked that map_sender is something above.
+        let sender = map_sender.as_ref().unwrap();
+        let (reply_sender, reply_receiver) = unbounded();
+        sender
+            .send(MemoryMapping::AddMapping(
+                reply_sender,
+                host_addr as u64,
+                guest_addr,
+                len,
+            ))
+            .unwrap();
+        if !reply_receiver.recv().unwrap() {
+            error!("Error requesting HVF the addition of a DAX window");
+            unsafe { libc::munmap(host_addr, len as usize) };
+            return Err(linux_error(io::Error::from_raw_os_error(libc::EINVAL)));
+        }
+
+        self.map_windows
+            .lock()
+            .unwrap()
+            .insert(guest_addr, host_addr as u64);
+
+        Ok(())
+    }
+
+    fn removemapping(
+        &self,
+        _ctx: Context,
+        requests: Vec<fuse::RemovemappingOne>,
+        guest_shm_base: u64,
+        shm_size: u64,
+        map_sender: &Option<Sender<MemoryMapping>>,
+    ) -> io::Result<()> {
+        if map_sender.is_none() {
+            return Err(linux_error(io::Error::from_raw_os_error(libc::ENOSYS)));
+        }
+
+        for req in requests {
+            let guest_addr = guest_shm_base + req.moffset;
+            if (req.moffset + req.len) > shm_size {
+                return Err(linux_error(io::Error::from_raw_os_error(libc::EINVAL)));
+            }
+            let host_addr = match self.map_windows.lock().unwrap().remove(&guest_addr) {
+                Some(a) => a,
+                None => return Err(linux_error(io::Error::from_raw_os_error(libc::EINVAL))),
+            };
+            debug!(
+                "removemapping: guest_addr={:x} len={:?}",
+                guest_addr, req.len
+            );
+
+            let sender = map_sender.as_ref().unwrap();
+            let (reply_sender, reply_receiver) = unbounded();
+            sender
+                .send(MemoryMapping::RemoveMapping(
+                    reply_sender,
+                    guest_addr,
+                    req.len,
+                ))
+                .unwrap();
+            if !reply_receiver.recv().unwrap() {
+                error!("Error requesting HVF the removal of a DAX window");
+                return Err(linux_error(io::Error::from_raw_os_error(libc::EINVAL)));
+            }
+
+            let ret = unsafe { libc::munmap(host_addr as *mut libc::c_void, req.len as usize) };
+            if ret == -1 {
+                error!("Error unmapping DAX window");
+                return Err(linux_error(io::Error::last_os_error()));
+            }
+        }
+
+        Ok(())
+    }
 }
diff --git a/src/devices/src/virtio/fs/server.rs b/src/devices/src/virtio/fs/server.rs
@@ -2,6 +2,11 @@
 // Use of this source code is governed by a BSD-style license that can be
 // found in the LICENSE file.
 
+#[cfg(target_os = "macos")]
+use crossbeam_channel::Sender;
+#[cfg(target_os = "macos")]
+use hvf::MemoryMapping;
+
 use std::convert::TryInto;
 use std::ffi::{CStr, CString};
 use std::fs::File;
@@ -78,6 +83,7 @@ impl<F: FileSystem + Sync> Server<F> {
         mut r: Reader,
         w: Writer,
         shm_region: &Option<VirtioShmRegion>,
+        #[cfg(target_os = "macos")] map_sender: &Option<Sender<MemoryMapping>>,
     ) -> Result<usize> {
         let in_header: InHeader = r.read_obj().map_err(Error::DecodeMessage)?;
 
@@ -141,15 +147,31 @@ impl<F: FileSystem + Sync> Server<F> {
                 let shm_base_addr = shm.host_addr;
                 #[cfg(target_os = "macos")]
                 let shm_base_addr = shm.guest_addr;
-                self.setupmapping(in_header, r, w, shm_base_addr, shm.size as u64)
+                self.setupmapping(
+                    in_header,
+                    r,
+                    w,
+                    shm_base_addr,
+                    shm.size as u64,
+                    #[cfg(target_os = "macos")]
+                    map_sender,
+                )
             }
             x if (x == Opcode::RemoveMapping as u32) && shm_region.is_some() => {
                 let shm = shm_region.as_ref().unwrap();
                 #[cfg(target_os = "linux")]
                 let shm_base_addr = shm.host_addr;
                 #[cfg(target_os = "macos")]
                 let shm_base_addr = shm.guest_addr;
-                self.removemapping(in_header, r, w, shm_base_addr, shm.size as u64)
+                self.removemapping(
+                    in_header,
+                    r,
+                    w,
+                    shm_base_addr,
+                    shm.size as u64,
+                    #[cfg(target_os = "macos")]
+                    map_sender,
+                )
             }
             _ => reply_error(
                 linux_error(io::Error::from_raw_os_error(libc::ENOSYS)),
@@ -1309,6 +1331,7 @@ impl<F: FileSystem + Sync> Server<F> {
         w: Writer,
         host_shm_base: u64,
         shm_size: u64,
+        #[cfg(target_os = "macos")] map_sender: &Option<Sender<MemoryMapping>>,
     ) -> Result<usize> {
         let SetupmappingIn {
             fh,
@@ -1328,6 +1351,8 @@ impl<F: FileSystem + Sync> Server<F> {
             moffset,
             host_shm_base,
             shm_size,
+            #[cfg(target_os = "macos")]
+            map_sender,
         ) {
             Ok(()) => reply_ok(None::<u8>, None, in_header.unique, w),
             Err(e) => reply_error(e, in_header.unique, w),
@@ -1341,6 +1366,7 @@ impl<F: FileSystem + Sync> Server<F> {
         w: Writer,
         host_shm_base: u64,
         shm_size: u64,
+        #[cfg(target_os = "macos")] map_sender: &Option<Sender<MemoryMapping>>,
     ) -> Result<usize> {
         let RemovemappingIn { count } = r.read_obj().map_err(Error::DecodeMessage)?;
 
@@ -1368,10 +1394,14 @@ impl<F: FileSystem + Sync> Server<F> {
             );
         }
 
-        match self
-            .fs
-            .removemapping(Context::from(in_header), requests, host_shm_base, shm_size)
-        {
+        match self.fs.removemapping(
+            Context::from(in_header),
+            requests,
+            host_shm_base,
+            shm_size,
+            #[cfg(target_os = "macos")]
+            map_sender,
+        ) {
             Ok(()) => reply_ok(None::<u8>, None, in_header.unique, w),
             Err(e) => reply_error(e, in_header.unique, w),
         }