Allow for raw marshalling of BackingIds

Cargo.toml

@@ -85,3 +85,6 @@ name = "ioctl"
 
 [[example]]
 name = "passthrough"
+
+[[example]]
+name = "passthrough_fork"

Makefile

@@ -63,8 +63,9 @@ mount_tests:
 	 fuser:mount_tests_libfuse3 bash -c "cd /code/fuser && cargo run -p fuser-tests -- linux-mount-libfuse3"
 
 test_passthrough:
-	cargo build --example passthrough
+	cargo build --example passthrough --example passthrough_fork
 	sudo tests/test_passthrough.sh target/debug/examples/passthrough
+	sudo tests/test_passthrough.sh target/debug/examples/passthrough_fork
 
 test: pre mount_tests pjdfs_tests xfstests
 	cargo test

examples/passthrough_fork.rs

@@ -0,0 +1,335 @@
+// This example requires fuse 7.40 or later. Run with:
+//
+//   cargo run --example passthrough_fork /tmp/foobar
+
+mod common;
+
+use std::collections::HashMap;
+use std::ffi::OsStr;
+use std::fs::File;
+use std::os::fd::AsFd;
+use std::os::fd::AsRawFd;
+use std::os::fd::FromRawFd;
+use std::os::fd::OwnedFd;
+use std::os::unix::ffi::OsStrExt;
+use std::os::unix::net::UnixDatagram;
+use std::sync::Arc;
+use std::sync::Weak;
+use std::time::Duration;
+use std::time::UNIX_EPOCH;
+
+use clap::Parser;
+use fuser::BackingId;
+use fuser::Errno;
+use fuser::FileAttr;
+use fuser::FileHandle;
+use fuser::FileType;
+use fuser::Filesystem;
+use fuser::FopenFlags;
+use fuser::INodeNo;
+use fuser::InitFlags;
+use fuser::KernelConfig;
+use fuser::LockOwner;
+use fuser::MountOption;
+use fuser::OpenFlags;
+use fuser::ReplyAttr;
+use fuser::ReplyDirectory;
+use fuser::ReplyEmpty;
+use fuser::ReplyEntry;
+use fuser::ReplyOpen;
+use fuser::Request;
+use nix::sys::socket;
+use nix::sys::socket::MsgFlags;
+use parking_lot::Mutex;
+
+#[derive(Parser)]
+#[command(version)]
+struct Args {
+    #[clap(flatten)]
+    common_args: CommonArgs,
+}
+
+const TTL: Duration = Duration::from_secs(1); // 1 second
+
+use crate::common::args::CommonArgs;
+
+// See [./passthrough.rs]
+#[derive(Debug, Default)]
+struct BackingCache {
+    by_handle: HashMap<u64, Arc<BackingId>>,
+    by_inode: HashMap<INodeNo, Weak<BackingId>>,
+    next_fh: u64,
+}
+
+impl BackingCache {
+    fn next_fh(&mut self) -> u64 {
+        self.next_fh += 1;
+        self.next_fh
+    }
+
+    fn get_or(
+        &mut self,
+        ino: INodeNo,
+        callback: impl Fn() -> std::io::Result<BackingId>,
+    ) -> std::io::Result<(u64, Arc<BackingId>)> {
+        let fh = self.next_fh();
+
+        let id = if let Some(id) = self.by_inode.get(&ino).and_then(Weak::upgrade) {
+            eprintln!("HIT! reusing {id:?}");
+            id
+        } else {
+            let id = Arc::new(callback()?);
+            self.by_inode.insert(ino, Arc::downgrade(&id));
+            eprintln!("MISS! new {id:?}");
+            id
+        };
+
+        self.by_handle.insert(fh, Arc::clone(&id));
+        Ok((fh, id))
+    }
+
+    fn put(&mut self, fh: u64) {
+        eprintln!("Put fh {fh}");
+        match self.by_handle.remove(&fh) {
+            None => eprintln!("ERROR: Put fh {fh} but it wasn't found in cache!!\n"),
+            Some(id) => eprintln!("Put fh {fh}, was {id:?}\n"),
+        }
+    }
+}
+
+#[derive(Debug)]
+struct ForkPassthroughFs {
+    root_attr: FileAttr,
+    passthrough_file_attr: FileAttr,
+    socket: UnixDatagram,
+    backing_cache: Mutex<BackingCache>,
+}
+
+impl ForkPassthroughFs {
+    fn new(socket: UnixDatagram) -> Self {
+        let uid = nix::unistd::getuid().into();
+        let gid = nix::unistd::getgid().into();
+
+        let root_attr = FileAttr {
+            ino: INodeNo::ROOT,
+            size: 0,
+            blocks: 0,
+            atime: UNIX_EPOCH, // 1970-01-01 00:00:00
+            mtime: UNIX_EPOCH,
+            ctime: UNIX_EPOCH,
+            crtime: UNIX_EPOCH,
+            kind: FileType::Directory,
+            perm: 0o755,
+            nlink: 2,
+            uid,
+            gid,
+            rdev: 0,
+            flags: 0,
+            blksize: 512,
+        };
+
+        let passthrough_file_attr = FileAttr {
+            ino: INodeNo(2),
+            size: 123_456,
+            blocks: 1,
+            atime: UNIX_EPOCH, // 1970-01-01 00:00:00
+            mtime: UNIX_EPOCH,
+            ctime: UNIX_EPOCH,
+            crtime: UNIX_EPOCH,
+            kind: FileType::RegularFile,
+            perm: 0o644,
+            nlink: 1,
+            uid: 333,
+            gid: 333,
+            rdev: 0,
+            flags: 0,
+            blksize: 512,
+        };
+
+        Self {
+            root_attr,
+            passthrough_file_attr,
+            socket,
+            backing_cache: Mutex::default(),
+        }
+    }
+}
+
+impl Filesystem for ForkPassthroughFs {
+    fn init(&mut self, _req: &Request, config: &mut KernelConfig) -> std::io::Result<()> {
+        config
+            .add_capabilities(InitFlags::FUSE_PASSTHROUGH)
+            .unwrap();
+        config.set_max_stack_depth(2).unwrap();
+        Ok(())
+    }
+
+    fn destroy(&mut self) {
+        // Tell the parent process to shut down
+        self.socket.send(&[]).unwrap();
+    }
+
+    fn lookup(&self, _req: &Request, parent: INodeNo, name: &OsStr, reply: ReplyEntry) {
+        if parent == INodeNo::ROOT && name.to_str() == Some("passthrough") {
+            reply.entry(&TTL, &self.passthrough_file_attr, fuser::Generation(0));
+        } else {
+            reply.error(Errno::ENOENT);
+        }
+    }
+
+    fn getattr(&self, _req: &Request, ino: INodeNo, _fh: Option<FileHandle>, reply: ReplyAttr) {
+        match ino.0 {
+            1 => reply.attr(&TTL, &self.root_attr),
+            2 => reply.attr(&TTL, &self.passthrough_file_attr),
+            _ => reply.error(Errno::ENOENT),
+        }
+    }
+
+    fn open(&self, _req: &Request, ino: INodeNo, _flags: OpenFlags, reply: ReplyOpen) {
+        if ino != INodeNo(2) {
+            reply.error(Errno::ENOENT);
+            return;
+        }
+
+        let (fh, id) = self
+            .backing_cache
+            .lock()
+            .get_or(ino, || {
+                // Ask the parent process to open a backing ID for us (concurrency is left as an exercise for the reader)
+                const FILE: &str = "/etc/profile";
+                eprintln!("Asking server to open backing file for {FILE:?}");
+
+                let mut buf = [0u8; 4];
+                self.socket.send(FILE.as_bytes())?;
+                self.socket.recv(&mut buf)?;
+                Ok(unsafe { reply.wrap_backing(u32::from_ne_bytes(buf)) })
+            })
+            .unwrap();
+
+        eprintln!("  -> opened_passthrough({fh:?}, 0, {id:?});\n");
+        reply.opened_passthrough(FileHandle(fh), FopenFlags::empty(), &id);
+    }
+
+    fn release(
+        &self,
+        _req: &Request,
+        _ino: INodeNo,
+        fh: FileHandle,
+        _flags: OpenFlags,
+        _lock_owner: Option<LockOwner>,
+        _flush: bool,
+        reply: ReplyEmpty,
+    ) {
+        self.backing_cache.lock().put(fh.into());
+        reply.ok();
+    }
+
+    fn readdir(
+        &self,
+        _req: &Request,
+        ino: INodeNo,
+        _fh: FileHandle,
+        offset: u64,
+        mut reply: ReplyDirectory,
+    ) {
+        if ino != INodeNo::ROOT {
+            reply.error(Errno::ENOENT);
+            return;
+        }
+
+        let entries = vec![
+            (1, FileType::Directory, "."),
+            (1, FileType::Directory, ".."),
+            (2, FileType::RegularFile, "passthrough"),
+        ];
+
+        for (i, entry) in entries.into_iter().enumerate().skip(offset as usize) {
+            // i + 1 means the index of the next entry
+            if reply.add(INodeNo(entry.0), (i + 1) as u64, entry.1, entry.2) {
+                break;
+            }
+        }
+        reply.ok();
+    }
+}
+
+fn main() {
+    // Fork and handle opening backing IDs in the parent process
+    // Do this early since forking isn't guaranteed to be a safe operation once e.g. libraries get involved
+    // You may also choose to use `std::process::Command` / etc instead
+    let (parent_sock, child_sock) = UnixDatagram::pair().unwrap();
+    match unsafe { nix::unistd::fork().unwrap() } {
+        nix::unistd::ForkResult::Parent { .. } => {
+            drop(child_sock);
+            backing_id_server(parent_sock);
+            return;
+        }
+        nix::unistd::ForkResult::Child => {
+            drop(parent_sock);
+        }
+    }
+
+    // You may wish to drop privileges (i.e. CAP_SYS_ADMIN / root) here
+
+    // Mount the FS as usual
+    let args = Args::parse();
+    env_logger::init();
+
+    let mut cfg = args.common_args.config();
+    cfg.mount_options
+        .extend([MountOption::FSName("passthrough".to_string())]);
+    let fs = ForkPassthroughFs::new(child_sock.try_clone().unwrap());
+    let session = fuser::Session::new(fs, &args.common_args.mount_point, &cfg).unwrap();
+
+    // Send the FUSE FD to the parent process so it may open backing files
+    let fds = [session.as_fd().as_raw_fd()];
+
+    socket::sendmsg::<()>(
+        child_sock.as_raw_fd(),
+        &[],
+        &[socket::ControlMessage::ScmRights(&fds)],
+        MsgFlags::empty(),
+        None,
+    )
+    .unwrap();
+
+    // Run the FS
+    session.run().unwrap();
+}
+
+fn backing_id_server(socket: UnixDatagram) {
+    let mut buf = [0u8; 1024];
+
+    // Receive the FUSE FD over the Unix socket
+    let msg = socket::recvmsg::<()>(
+        socket.as_fd().as_raw_fd(),
+        &mut [],
+        Some(&mut buf),
+        MsgFlags::empty(),
+    )
+    .unwrap();
+
+    let fuse_fd = 'fd: {
+        for cmsg in msg.cmsgs().unwrap() {
+            if let socket::ControlMessageOwned::ScmRights(fds) = cmsg {
+                let fd = unsafe { OwnedFd::from_raw_fd(fds[0]) };
+                break 'fd fd;
+            }
+        }
+        unreachable!();
+    };
+
+    // Handle backing ID requests (in real world scenarios, you may wish to perform input validation / etc here)
+    loop {
+        let sz = socket.recv(&mut buf).unwrap();
+        if sz == 0 {
+            return;
+        }
+
+        let path = OsStr::from_bytes(&buf[..sz]);
+        eprintln!("[SERVER] opening backing file for {path:?}");
+
+        let id = BackingId::create_raw(&fuse_fd, File::open(path).unwrap()).unwrap();
+        socket.send(&u32::to_ne_bytes(id)).unwrap();
+    }
+}

src/channel.rs

@@ -95,4 +95,8 @@ impl ChannelSender {
     pub(crate) fn open_backing(&self, fd: BorrowedFd<'_>) -> std::io::Result<BackingId> {
         BackingId::create(&self.0, fd)
     }
+
+    pub(crate) unsafe fn wrap_backing(&self, id: u32) -> BackingId {
+        unsafe { BackingId::wrap_raw(&self.0, id) }
+    }
 }

src/lib.rs

@@ -49,7 +49,6 @@ pub use crate::ll::request::LockOwner;
 pub use crate::ll::request::Version;
 pub use crate::mnt::mount_options::Config;
 pub use crate::mnt::mount_options::MountOption;
-use crate::mnt::mount_options::check_option_conflicts;
 pub use crate::notify::Notifier;
 pub use crate::notify::PollHandle;
 pub use crate::notify::PollNotifier;
@@ -1054,8 +1053,7 @@ pub fn mount<FS: Filesystem, P: AsRef<Path>>(
     mountpoint: P,
     options: &Config,
 ) -> io::Result<()> {
-    check_option_conflicts(options)?;
-    Session::new(filesystem, mountpoint.as_ref(), options).and_then(|se| se.run())
+    Session::new(filesystem, mountpoint.as_ref(), options).and_then(session::Session::run)
 }
 
 /// Mount the given filesystem to the given mountpoint. This function spawns
@@ -1072,6 +1070,5 @@ pub fn spawn_mount<'a, FS: Filesystem + Send + 'static + 'a, P: AsRef<Path>>(
     mountpoint: P,
     options: &Config,
 ) -> io::Result<BackgroundSession> {
-    check_option_conflicts(options)?;
     Session::new(filesystem, mountpoint.as_ref(), options).and_then(session::Session::spawn)
 }