made getdents64 stateful

Co-authored-by: Martin Kröning <martin.kroening@eonerc.rwth-aachen.de>

Author: jounathaen

src/fd/mod.rs

@@ -1,6 +1,5 @@
 use alloc::boxed::Box;
 use alloc::sync::Arc;
-use alloc::vec::Vec;
 use core::future::{self, Future};
 use core::mem::MaybeUninit;
 use core::task::Poll::{Pending, Ready};
@@ -12,7 +11,7 @@ use smoltcp::wire::{IpEndpoint, IpListenEndpoint};
 
 use crate::arch::kernel::core_local::core_scheduler;
 use crate::executor::block_on;
-use crate::fs::{DirectoryEntry, FileAttr, SeekWhence};
+use crate::fs::{FileAttr, SeekWhence};
 use crate::io;
 
 mod eventfd;
@@ -178,10 +177,10 @@ pub(crate) trait ObjectInterface: Sync + Send + core::fmt::Debug {
 		Err(io::Error::EINVAL)
 	}
 
-	/// 'readdir' returns a pointer to a dirent structure
-	/// representing the next directory entry in the directory stream
-	/// pointed to by the file descriptor
-	async fn readdir(&self) -> io::Result<Vec<DirectoryEntry>> {
+	/// `getdents` fills the given buffer `_buf` with [`Dirent64`](crate::syscalls::Dirent64)
+	/// formatted entries of a directory, imitating the Linux `getdents64` syscall.
+	/// On success, the number of bytes read is returned.  On end of directory, 0 is returned.  On error, -1 is returned
+	async fn getdents(&self, _buf: &mut [MaybeUninit<u8>]) -> io::Result<usize> {
 		Err(io::Error::EINVAL)
 	}
 

src/fs/fuse.rs

@@ -5,11 +5,12 @@ use alloc::string::String;
 use alloc::sync::Arc;
 use alloc::vec::Vec;
 use core::marker::PhantomData;
-use core::mem::MaybeUninit;
+use core::mem::{MaybeUninit, align_of, offset_of, size_of};
 use core::sync::atomic::{AtomicU64, Ordering};
 use core::task::Poll;
 use core::{future, mem};
 
+use align_address::Align;
 use async_lock::Mutex;
 use async_trait::async_trait;
 use fuse_abi::linux::*;
@@ -27,6 +28,7 @@ use crate::fs::{
 	SeekWhence, VfsNode,
 };
 use crate::mm::device_alloc::DeviceAlloc;
+use crate::syscalls::Dirent64;
 use crate::time::{time_t, timespec};
 use crate::{arch, io};
 
@@ -810,20 +812,24 @@ impl Clone for FuseFileHandle {
 	}
 }
 
-#[derive(Debug, Clone)]
+#[derive(Debug)]
 pub struct FuseDirectoryHandle {
 	name: Option<String>,
+	read_position: Mutex<usize>,
 }
 
 impl FuseDirectoryHandle {
 	pub fn new(name: Option<String>) -> Self {
-		Self { name }
+		Self {
+			name,
+			read_position: Mutex::new(0),
+		}
 	}
 }
 
 #[async_trait]
 impl ObjectInterface for FuseDirectoryHandle {
-	async fn readdir(&self) -> io::Result<Vec<DirectoryEntry>> {
+	async fn getdents(&self, buf: &mut [MaybeUninit<u8>]) -> io::Result<usize> {
 		let path: CString = if let Some(name) = &self.name {
 			CString::new("/".to_string() + name).unwrap()
 		} else {
@@ -848,7 +854,8 @@ impl ObjectInterface for FuseDirectoryHandle {
 
 		// Linux seems to allocate a single page to store the dirfile
 		let len = MAX_READ_LEN as u32;
-		let mut offset: usize = 0;
+		let rsp_offset: &mut usize = &mut *self.read_position.lock().await;
+		let mut buf_offset: usize = 0;
 
 		// read content of the directory
 		let (mut cmd, rsp_payload_len) = ops::Read::create(fuse_nid, fuse_fh, len, 0);
@@ -868,31 +875,53 @@ impl ObjectInterface for FuseDirectoryHandle {
 			return Err(io::Error::ENOENT);
 		}
 
-		let mut entries: Vec<DirectoryEntry> = Vec::new();
-		while (rsp.headers.out_header.len as usize) - offset > core::mem::size_of::<fuse_dirent>() {
+		let mut ret = 0;
+
+		while (rsp.headers.out_header.len as usize) - *rsp_offset > size_of::<fuse_dirent>() {
 			let dirent = unsafe {
 				&*rsp
 					.payload
 					.as_ref()
 					.unwrap()
 					.as_ptr()
-					.byte_add(offset)
+					.byte_add(*rsp_offset)
 					.cast::<fuse_dirent>()
 			};
 
-			offset += core::mem::size_of::<fuse_dirent>() + dirent.namelen as usize;
-			// Align to dirent struct
-			offset = ((offset) + U64_SIZE - 1) & (!(U64_SIZE - 1));
+			let dirent_len = offset_of!(Dirent64, d_name) + dirent.namelen as usize + 1;
+			let next_dirent = (buf_offset + dirent_len).align_up(align_of::<Dirent64>());
 
-			let name: &'static [u8] = unsafe {
-				core::slice::from_raw_parts(
-					dirent.name.as_ptr().cast(),
-					dirent.namelen.try_into().unwrap(),
-				)
-			};
-			entries.push(DirectoryEntry::new(unsafe {
-				core::str::from_utf8_unchecked(name).to_string()
-			}));
+			if next_dirent > buf.len() {
+				// target buffer full -> we return the nr. of bytes written (like linux does)
+				break;
+			}
+
+			// could be replaced with slice_as_ptr once maybe_uninit_slice is stabilized.
+			let target_dirent = buf[buf_offset].as_mut_ptr().cast::<Dirent64>();
+			unsafe {
+				target_dirent.write(Dirent64 {
+					d_ino: dirent.ino,
+					d_off: 0,
+					d_reclen: (dirent_len.align_up(align_of::<Dirent64>()))
+						.try_into()
+						.unwrap(),
+					d_type: (dirent.type_ as u8).try_into().unwrap(),
+					d_name: PhantomData {},
+				});
+				let nameptr = core::ptr::from_mut(&mut (*(target_dirent)).d_name).cast::<u8>();
+				core::ptr::copy_nonoverlapping(
+					dirent.name.as_ptr().cast::<u8>(),
+					nameptr,
+					dirent.namelen as usize,
+				);
+				nameptr.add(dirent.namelen as usize).write(0); // zero termination
+			}
+
+			*rsp_offset += core::mem::size_of::<fuse_dirent>() + dirent.namelen as usize;
+			// Align to dirent struct
+			*rsp_offset = ((*rsp_offset) + U64_SIZE - 1) & (!(U64_SIZE - 1));
+			buf_offset = next_dirent;
+			ret = buf_offset;
 		}
 
 		let (cmd, rsp_payload_len) = ops::Release::create(fuse_nid, fuse_fh);
@@ -901,7 +930,20 @@ impl ObjectInterface for FuseDirectoryHandle {
 			.lock()
 			.send_command(cmd, rsp_payload_len)?;
 
-		Ok(entries)
+		Ok(ret)
+	}
+
+	/// lseek for a directory entry is the equivalent for seekdir on linux. But on Hermit this is
+	/// logically the same operation, so we can just use the same fn in the backend.
+	/// Any other offset than 0 is not supported. (Mostly because it doesn't make any sense, as
+	/// userspace applications have no way of knowing valid offsets)
+	async fn lseek(&self, offset: isize, whence: SeekWhence) -> io::Result<isize> {
+		if whence != SeekWhence::Set && offset != 0 {
+			error!("Invalid offset for directory lseek ({offset})");
+			return Err(io::Error::EINVAL);
+		}
+		*self.read_position.lock().await = offset as usize;
+		Ok(offset)
 	}
 }
 

src/fs/mem.rs

@@ -14,14 +14,17 @@ use alloc::collections::BTreeMap;
 use alloc::string::String;
 use alloc::sync::Arc;
 use alloc::vec::Vec;
-use core::mem::MaybeUninit;
+use core::marker::PhantomData;
+use core::mem::{MaybeUninit, offset_of};
 
+use align_address::Align;
 use async_lock::{Mutex, RwLock};
 use async_trait::async_trait;
 
 use crate::executor::block_on;
 use crate::fd::{AccessPermission, ObjectInterface, OpenOption, PollEvent};
-use crate::fs::{DirectoryEntry, FileAttr, NodeKind, SeekWhence, VfsNode};
+use crate::fs::{DirectoryEntry, FileAttr, FileType, NodeKind, SeekWhence, VfsNode};
+use crate::syscalls::Dirent64;
 use crate::time::timespec;
 use crate::{arch, io};
 
@@ -374,11 +377,12 @@ impl RamFile {
 	}
 }
 
-#[derive(Debug, Clone)]
+#[derive(Debug)]
 pub struct MemDirectoryInterface {
 	/// Directory entries
 	inner:
 		Arc<RwLock<BTreeMap<String, Box<dyn VfsNode + core::marker::Send + core::marker::Sync>>>>,
+	read_idx: Mutex<usize>,
 }
 
 impl MemDirectoryInterface {
@@ -387,19 +391,71 @@ impl MemDirectoryInterface {
 			RwLock<BTreeMap<String, Box<dyn VfsNode + core::marker::Send + core::marker::Sync>>>,
 		>,
 	) -> Self {
-		Self { inner }
+		Self {
+			inner,
+			read_idx: Mutex::new(0),
+		}
 	}
 }
 
 #[async_trait]
 impl ObjectInterface for MemDirectoryInterface {
-	async fn readdir(&self) -> io::Result<Vec<DirectoryEntry>> {
-		let mut entries: Vec<DirectoryEntry> = Vec::new();
-		for name in self.inner.read().await.keys() {
-			entries.push(DirectoryEntry::new(name.clone()));
+	async fn getdents(&self, buf: &mut [MaybeUninit<u8>]) -> io::Result<usize> {
+		let mut buf_offset: usize = 0;
+		let mut ret = 0;
+		let mut read_idx = self.read_idx.lock().await;
+		for name in self.inner.read().await.keys().skip(*read_idx) {
+			let namelen = name.len();
+
+			let dirent_len = offset_of!(Dirent64, d_name) + namelen + 1;
+			let next_dirent = (buf_offset + dirent_len).align_up(align_of::<Dirent64>());
+
+			if next_dirent > buf.len() {
+				// target buffer full -> we return the nr. of bytes written (like linux does)
+				break;
+			}
+
+			*read_idx += 1;
+
+			// could be replaced with slice_as_ptr once maybe_uninit_slice is stabilized.
+			let target_dirent = buf[buf_offset].as_mut_ptr().cast::<Dirent64>();
+
+			unsafe {
+				target_dirent.write(Dirent64 {
+					d_ino: 1, // TODO: we don't have inodes in the mem filesystem. Maybe this could lead to problems
+					d_off: 0,
+					d_reclen: (dirent_len.align_up(align_of::<Dirent64>()))
+						.try_into()
+						.unwrap(),
+					d_type: FileType::Unknown, // TODO: Proper filetype
+					d_name: PhantomData {},
+				});
+				let nameptr = core::ptr::from_mut(&mut (*(target_dirent)).d_name).cast::<u8>();
+				core::ptr::copy_nonoverlapping(
+					name.as_bytes().as_ptr().cast::<u8>(),
+					nameptr,
+					namelen,
+				);
+				nameptr.add(namelen).write(0); // zero termination
+			}
+
+			buf_offset = next_dirent;
+			ret = buf_offset;
 		}
+		Ok(ret)
+	}
 
-		Ok(entries)
+	/// lseek for a directory entry is the equivalent for seekdir on linux. But on Hermit this is
+	/// logically the same operation, so we can just use the same fn in the backend.
+	/// Any other offset than 0 is not supported. (Mostly because it doesn't make any sense, as
+	/// userspace applications have no way of knowing valid offsets)
+	async fn lseek(&self, offset: isize, whence: SeekWhence) -> io::Result<isize> {
+		if whence != SeekWhence::Set && offset != 0 {
+			error!("Invalid offset for directory lseek ({offset})");
+			return Err(io::Error::EINVAL);
+		}
+		*self.read_idx.lock().await = offset as usize;
+		Ok(offset)
 	}
 }
 

src/fs/mod.rs

@@ -130,8 +130,9 @@ impl DirectoryReader {
 
 #[async_trait]
 impl ObjectInterface for DirectoryReader {
-	async fn readdir(&self) -> io::Result<Vec<DirectoryEntry>> {
-		Ok(self.0.clone())
+	async fn getdents(&self, _buf: &mut [core::mem::MaybeUninit<u8>]) -> io::Result<usize> {
+		let _ = &self.0; // Dummy statement to avoid warning for the moment
+		unimplemented!()
 	}
 }