ext2 filesystem driver

src/kernel/arch/i386/fs/ext2.c · include/kernel/ext2.h

A read and write ext2 driver, mirroring the FAT32 driver’s public API so the VFS can dispatch the hard-disk mount table to either backend. The intended layout is FAT32 for the kernel + bootloader modules + root (EFI-partition style) and ext2 for apps / user directories.

Supported on-disk formats

Aspect Support
Revision rev 0 and rev 1 (dynamic)
Block size 1024 / 2048 / 4096 bytes
Inode size 128 (rev 0) or the superblock’s s_inode_size (rev 1)
Incompat features FILETYPE only — a volume with any other incompat feature (journal, extents, meta_bg, 64bit) is refused at mount so Makar never corrupts an ext3/ext4 volume
Block maps direct (0–11) + single-indirect + double-indirect (triple-indirect is not implemented)
Directories linear variable-length entries; dir_index (HTree) volumes should be created with ^dir_index so the linear walk sees every entry

Public API

Mount/format:

  • ext2_probe(drive, lba) — cheap superblock-magic test (0xEF53); no state change.
  • ext2_mount(drive, lba) — 0, -2 (not ext2), -3 (unsupported incompat), -1 (I/O).
  • ext2_unmount() — flush metadata + release the cached group-descriptor table.
  • ext2_mkfs(drive, lba, sectors) — format as ext2 (1 KiB blocks, rev 1, FILETYPE): superblock + per-group descriptors / bitmaps / inode tables (super + GDT backup in every group), root directory and lost+found.

Operations (paths are volume-relative, case-sensitive, /-separated):

ext2_ls · ext2_cd · ext2_mkdir · ext2_read_file · ext2_write_file · ext2_file_exists · ext2_delete_file · ext2_delete_dir (non-empty refused) · ext2_rename_file / ext2_rename_dir · ext2_complete (tab completion).

Write path

Block and inode allocation scan the per-group bitmaps for a free bit, set it, decrement the group and superblock free counts, and flush the affected metadata. ext2_write_file truncates an existing file (freeing its block map) before rewriting, or allocates a fresh inode + directory entry. ext2_mkdir seeds a new directory block with . / .. and bumps the parent’s link count. Directory-entry insertion reuses slack in existing entries or appends a new directory block.

VFS integration

The VFS keeps a small mount table; each /mnt/<name> maps to a backend. mount /dev/hdaN /mnt/<name> calls vfs_mount_hd, which prefers ext2 when a superblock is present and falls back to FAT32. Because the FAT32 and ext2 drivers each hold a single volume, one of each can be mounted simultaneously at different mountpoints. mkfs.ext2 /dev/hdaN formats a partition; umount /mnt/<name> flushes and detaches it.

Limitations / not yet done

  • Single volume per backend (no two ext2 mounts at once).
  • No triple-indirect blocks, no HTree (dir_index) traversal, no journalling.
  • 128-byte inodes use pre-2038 timestamps (mkfs default).
  • Clean-room implementation; not derived from the Linux ext2 source.