rust-kernel

//! FAT32 support, using the `fatfs` crate (no_std + alloc mode).
//!
//! Two adapters glue it in:
//!
//! 1. [`BlockIo`]: `fatfs` wants something that looks like a `std::fs::File`
//!    (byte-addressed Read + Write + Seek). Our disks only do whole 512-byte
//!    sectors, so `BlockIo` keeps a byte position and turns byte requests
//!    into sector requests: whole aligned sectors go straight to the device;
//!    partial sectors go through a one-sector cache (read-modify-write for
//!    writes). Writes are write-through, so nothing is ever lost on "power
//!    off" because of the cache.
//!
//! 2. [`FatInode`]: our VFS `Inode` for one FAT file/directory. `fatfs`'s
//!    own `File`/`Dir` objects *borrow* the FileSystem (they carry a lifetime)
//!    so we can't store them in a long-lived `Arc<dyn Inode>`. Instead a
//!    FatInode just remembers its **path** and re-opens it from the root on
//!    every operation. That's slower (each read walks the directories again)
//!    but very simple and obviously correct -- fine for a learning kernel.
//!    A faster design would cache directory entries (like Linux's dcache).
//!
//! Locking: the whole FileSystem sits behind one Mutex, so FAT operations
//! are serialized. The virtio driver underneath polls with interrupts off;
//! this lock is only taken from thread context, never from an interrupt.

use super::{DirEntry, FileSystem, Inode, InodeKind};
use crate::block::{BlockDevice, SECTOR_SIZE};
use crate::errno::Errno;
use alloc::{format, string::String, sync::Arc, vec::Vec};
use fatfs::{IoBase, Read, Seek, SeekFrom, Write};
use spin::Mutex;

// ---------------------------------------------------------------------------
// 1. Byte-stream view of a block device
// ---------------------------------------------------------------------------

/// The error type `fatfs` reports back to us: just our errno.
#[derive(Debug)]
pub struct IoErr(pub Errno);

impl fatfs::IoError for IoErr {
    fn is_interrupted(&self) -> bool {
        false
    }
    fn new_unexpected_eof_error() -> Self {
        IoErr(Errno::EIO)
    }
    fn new_write_zero_error() -> Self {
        IoErr(Errno::EIO)
    }
}

pub struct BlockIo {
    dev: Arc<dyn BlockDevice>,
    pos: u64,
    size: u64,
    /// One cached sector: (lba, contents).
    cache: Option<(u64, [u8; SECTOR_SIZE])>,
}

impl BlockIo {
    pub fn new(dev: Arc<dyn BlockDevice>) -> Self {
        let size = dev.sector_count() * SECTOR_SIZE as u64;
        BlockIo { dev, pos: 0, size, cache: None }
    }

    /// Get sector `lba` into the cache and return it.
    fn cached(&mut self, lba: u64) -> Result<&mut [u8; SECTOR_SIZE], IoErr> {
        if self.cache.map(|(l, _)| l) != Some(lba) {
            let mut data = [0u8; SECTOR_SIZE];
            self.dev.read_sectors(lba, &mut data).map_err(IoErr)?;
            self.cache = Some((lba, data));
        }
        Ok(&mut self.cache.as_mut().unwrap().1)
    }
}

impl IoBase for BlockIo {
    type Error = IoErr;
}

impl Read for BlockIo {
    fn read(&mut self, buf: &mut [u8]) -> Result<usize, IoErr> {
        let len = buf.len().min(self.size.saturating_sub(self.pos) as usize);
        if len == 0 {
            return Ok(0);
        }
        let lba = self.pos / SECTOR_SIZE as u64;
        let offset = (self.pos % SECTOR_SIZE as u64) as usize;
        let n = if offset == 0 && len >= SECTOR_SIZE {
            // Fast path: whole sectors straight into the caller's buffer.
            let n = len / SECTOR_SIZE * SECTOR_SIZE;
            self.dev.read_sectors(lba, &mut buf[..n]).map_err(IoErr)?;
            // Writes keep the cache up to date, so it can't be stale here.
            n
        } else {
            // Part of one sector, via the cache.
            let n = len.min(SECTOR_SIZE - offset);
            let sector = self.cached(lba)?;
            buf[..n].copy_from_slice(&sector[offset..offset + n]);
            n
        };
        self.pos += n as u64;
        Ok(n)
    }
}

impl Write for BlockIo {
    fn write(&mut self, buf: &[u8]) -> Result<usize, IoErr> {
        let len = buf.len().min(self.size.saturating_sub(self.pos) as usize);
        if len == 0 {
            return Ok(0); // fatfs turns this into WriteZero (disk full)
        }
        let lba = self.pos / SECTOR_SIZE as u64;
        let offset = (self.pos % SECTOR_SIZE as u64) as usize;
        let n = if offset == 0 && len >= SECTOR_SIZE {
            let n = len / SECTOR_SIZE * SECTOR_SIZE;
            self.dev.write_sectors(lba, &buf[..n]).map_err(IoErr)?;
            // Drop the cached sector if we just overwrote it.
            if let Some((cached, _)) = self.cache {
                if cached >= lba && cached < lba + (n / SECTOR_SIZE) as u64 {
                    self.cache = None;
                }
            }
            n
        } else {
            // Read-modify-write one sector.
            let n = len.min(SECTOR_SIZE - offset);
            let sector = self.cached(lba)?;
            sector[offset..offset + n].copy_from_slice(&buf[..n]);
            let copy = *sector;
            self.dev.write_sectors(lba, &copy).map_err(IoErr)?;
            n
        };
        self.pos += n as u64;
        Ok(n)
    }

    fn flush(&mut self) -> Result<(), IoErr> {
        Ok(()) // write-through: nothing buffered
    }
}

impl Seek for BlockIo {
    fn seek(&mut self, pos: SeekFrom) -> Result<u64, IoErr> {
        let new = match pos {
            SeekFrom::Start(n) => Some(n),
            SeekFrom::End(d) => self.size.checked_add_signed(d),
            SeekFrom::Current(d) => self.pos.checked_add_signed(d),
        };
        match new {
            Some(n) if n <= self.size => {
                self.pos = n;
                Ok(n)
            }
            _ => Err(IoErr(Errno::EINVAL)),
        }
    }
}

/// Translate a fatfs error into an errno.
fn errno(e: fatfs::Error<IoErr>) -> Errno {
    match e {
        fatfs::Error::Io(IoErr(e)) => e,
        fatfs::Error::NotFound => Errno::ENOENT,
        fatfs::Error::AlreadyExists => Errno::EEXIST,
        fatfs::Error::DirectoryIsNotEmpty => Errno::ENOTEMPTY,
        fatfs::Error::NotEnoughSpace | fatfs::Error::WriteZero => Errno::ENOSPC,
        fatfs::Error::InvalidFileNameLength => Errno::ENAMETOOLONG,
        fatfs::Error::InvalidInput | fatfs::Error::UnsupportedFileNameCharacter => Errno::EINVAL,
        _ => Errno::EIO, // corrupted file system, unexpected EOF, ...
    }
}

// ---------------------------------------------------------------------------
// 2. The VFS side
// ---------------------------------------------------------------------------

type Fs = fatfs::FileSystem<BlockIo, fatfs::NullTimeProvider, fatfs::LossyOemCpConverter>;
type FatDir<'a> = fatfs::Dir<'a, BlockIo, fatfs::NullTimeProvider, fatfs::LossyOemCpConverter>;

/// A mounted FAT volume. The fatfs `FileSystem` sits behind one Mutex
/// (shared via Arc with every FatInode), so FAT operations are serialized.
pub struct FatFs {
    fs: Arc<Mutex<Fs>>,
}

impl FatFs {
    /// Read the boot sector of `dev` and mount it if it's FAT.
    pub fn mount(dev: Arc<dyn BlockDevice>) -> Result<Arc<FatFs>, Errno> {
        // No real-time clock driver yet, so new files get the FAT epoch date
        // (1980-01-01) via NullTimeProvider.
        let options = fatfs::FsOptions::new().time_provider(fatfs::NullTimeProvider::new());
        let fs = fatfs::FileSystem::new(BlockIo::new(dev), options).map_err(errno)?;
        Ok(Arc::new(FatFs { fs: Arc::new(Mutex::new(fs)) }))
    }

    /// The volume label (from `mkfs.vfat -n`).
    pub fn label(&self) -> String {
        String::from(self.fs.lock().volume_label().trim_end())
    }
}

impl FileSystem for FatFs {
    fn name(&self) -> &'static str {
        "fat32"
    }
    fn root(&self) -> Arc<dyn Inode> {
        Arc::new(FatInode { fs: self.fs.clone(), path: String::new(), kind: InodeKind::Directory })
    }
}

/// A file or directory on a FAT volume, identified by its path from the
/// volume root ("" = the root, "bin/hello" = a file).
pub struct FatInode {
    fs: Arc<Mutex<Fs>>,
    path: String,
    kind: InodeKind,
}

impl FatInode {
    fn child_path(&self, name: &str) -> String {
        if self.path.is_empty() { String::from(name) } else { format!("{}/{}", self.path, name) }
    }

    /// Run `f` on this inode's directory (re-opened from the root).
    fn with_dir<T>(&self, f: impl FnOnce(&FatDir) -> Result<T, Errno>) -> Result<T, Errno> {
        let fs = self.fs.lock();
        let root = fs.root_dir();
        if self.path.is_empty() {
            f(&root)
        } else {
            f(&root.open_dir(&self.path).map_err(errno)?)
        }
    }

    /// Run `f` on this inode's file (re-opened from the root).
    fn with_file<T>(&self, f: impl FnOnce(&mut fatfs::File<BlockIo, fatfs::NullTimeProvider, fatfs::LossyOemCpConverter>) -> Result<T, Errno>) -> Result<T, Errno> {
        if self.kind == InodeKind::Directory {
            return Err(Errno::EISDIR);
        }
        let fs = self.fs.lock();
        let mut file = fs.root_dir().open_file(&self.path).map_err(errno)?;
        f(&mut file)
    }
}

impl Inode for FatInode {
    fn kind(&self) -> InodeKind {
        self.kind
    }

    fn size(&self) -> usize {
        match self.kind {
            InodeKind::File => self.with_file(|f| f.seek(SeekFrom::End(0)).map_err(errno)).unwrap_or(0) as usize,
            InodeKind::Directory => self.list().map(|l| l.len()).unwrap_or(0),
        }
    }

    fn read_at(&self, offset: usize, buf: &mut [u8]) -> Result<usize, Errno> {
        self.with_file(|file| {
            let size = file.seek(SeekFrom::End(0)).map_err(errno)? as usize;
            if offset >= size {
                return Ok(0);
            }
            file.seek(SeekFrom::Start(offset as u64)).map_err(errno)?;
            let len = buf.len().min(size - offset);
            // fatfs reads at most one cluster per call; loop to fill buf.
            let mut done = 0;
            while done < len {
                match file.read(&mut buf[done..len]).map_err(errno)? {
                    0 => break,
                    n => done += n,
                }
            }
            Ok(done)
        })
    }

    fn write_at(&self, offset: usize, buf: &[u8]) -> Result<usize, Errno> {
        self.with_file(|file| {
            let size = file.seek(SeekFrom::End(0)).map_err(errno)? as usize;
            if offset > size {
                // Writing past the end: fill the gap with zeros (FAT has no
                // sparse files).
                let zeros = [0u8; 512];
                let mut gap = offset - size;
                while gap > 0 {
                    let n = gap.min(zeros.len());
                    file.write_all(&zeros[..n]).map_err(errno)?;
                    gap -= n;
                }
            } else {
                file.seek(SeekFrom::Start(offset as u64)).map_err(errno)?;
            }
            file.write_all(buf).map_err(errno)?;
            Ok(buf.len())
        })
    }

    fn truncate(&self) -> Result<(), Errno> {
        self.with_file(|file| file.truncate().map_err(errno)) // truncates at the position: 0
    }

    fn lookup(&self, name: &str) -> Result<Arc<dyn Inode>, Errno> {
        if self.kind != InodeKind::Directory {
            return Err(Errno::ENOTDIR);
        }
        let kind = self.with_dir(|dir| {
            for entry in dir.iter() {
                let entry = entry.map_err(errno)?;
                // FAT names are case-insensitive (README.TXT == readme.txt).
                if entry.file_name().eq_ignore_ascii_case(name) {
                    return Ok(if entry.is_dir() { InodeKind::Directory } else { InodeKind::File });
                }
            }
            Err(Errno::ENOENT)
        })?;
        Ok(Arc::new(FatInode { fs: self.fs.clone(), path: self.child_path(name), kind }))
    }

    fn create(&self, name: &str, kind: InodeKind) -> Result<Arc<dyn Inode>, Errno> {
        if self.lookup(name).is_ok() {
            return Err(Errno::EEXIST);
        }
        self.with_dir(|dir| {
            match kind {
                InodeKind::File => dir.create_file(name).map(|_| ()),
                InodeKind::Directory => dir.create_dir(name).map(|_| ()),
            }
            .map_err(errno)
        })?;
        Ok(Arc::new(FatInode { fs: self.fs.clone(), path: self.child_path(name), kind }))
    }

    fn list(&self) -> Result<Vec<DirEntry>, Errno> {
        if self.kind != InodeKind::Directory {
            return Err(Errno::ENOTDIR);
        }
        self.with_dir(|dir| {
            let mut out = Vec::new();
            for entry in dir.iter() {
                let entry = entry.map_err(errno)?;
                let name = entry.file_name();
                if name == "." || name == ".." {
                    continue; // the VFS doesn't list these (like our ramfs)
                }
                let kind = if entry.is_dir() { InodeKind::Directory } else { InodeKind::File };
                out.push(DirEntry { name, kind });
            }
            Ok(out)
        })
    }

    fn remove(&self, name: &str) -> Result<(), Errno> {
        if self.kind != InodeKind::Directory {
            return Err(Errno::ENOTDIR);
        }
        self.with_dir(|dir| dir.remove(name).map_err(errno))
    }
}