# ext4 jbd2
https://github.com/yuoo655/ext4_rs
https://github.com/yuoo655/jbd2_rs
# ext4 与 ext2/ext3 区别
| 日志记录 | ❌ | ✅ | ✅ |
| 目录索引 | ❌ | ✅ | ✅ |
| 在线扩容 | ❌ | ✅ | ✅ |
| 在线碎片整理 | ❌ | ❌ | ✅ |
| 校验和 | ❌ | ❌ | ✅ |
| 是否采用extent | ❌ | ❌ | ✅ |
| 无限数量的子目录 | ❌ | ❌ | ✅ |
| 预分配 | ❌ | ❌ | ✅ |
...
# ext4支持状态
| mount | ✅ |
| open | ✅ |
| lsdir | ✅ |
| mkdir | ✅ |
| read_file | ✅ |
| read_link | ✅ |
| create_file | ✅ |
| write_file | ✅ |
| link | ✅ |
| umount | ❌ |
| file_remove | ❌ |
| dir_remove | ❌ |
# jbd2支持状态
| load_journal | ✅ |
| journal_start | ✅ |
| transaction_start | ✅ |
| write_transaction | ✅ |
| transaction_stop | ✅ |
| journal_stop | ✅ |
| recover | ✅ |
| revoke block | ❌ |
| checksum | ❌ |
...
# 独立组件
```rust
pub trait BlockDevice: Send + Sync + Any + Debug {
fn read_offset(&self, offset: usize) -> Vec<u8>;
fn write_offset(&self, offset: usize, data: &[u8]);
}
pub trait Jbd2: Send + Sync + Any + Debug {
fn load_journal(&mut self);
fn journal_start(&mut self);
fn transaction_start(&mut self);
fn write_transaction(&mut self, block_id: usize, block_data: Vec<u8>);
fn transaction_stop(&mut self);
fn journal_stop(&mut self);
fn recover(&mut self);
}
```
# 打开文件
从挂载点开始ext4_dir_find_entry遍历目录来,对比文件名,找到目标文件，提取direntry中的inode号，这一步也就是查找文件路径到文件inode的过程。
```rust
fn ext4_generic_open(path){
loop {
ext4_dir_find_entry(path)
if is_goal {
file.inode = dir_search_result.dentry.inode;
return Ok(EOK);
}
}
}
```
# 读文件
由于ext4默认所有文件都使用extent。extent记录了文件逻辑块号对应磁盘存储的物理块号。读取文件的过程就是寻找文件所有extent的过程。找到extent之后便可从extent中获取物理块号读出数据
```rust
pub struct Ext4Inode {
...
pub block: [u32; 15],
...
}
```
```rust
pub fn ext4_file_read(&self, ext4_file: &mut Ext4File) -> Vec<u8> {
...
ext4_find_all_extent(&inode_ref, &mut extents);
// 遍历extents向量，对每个extent，计算它的物理块号，然后调用read_block函数来读取数据块
for extent in extents {
let block_data = inode_ref.fs().block_device.read_offset(block_num as usize * BLOCK_SIZE);
file_data.extend(block_data);
}
file_data
}
```
# 创建文件
- alloc inode
- init inode
- link
分配inode
```rust
r = ext4_fs_alloc_inode(&mut child_inode_ref, ftype);
```
寻找inode位图找到第一个可用的位
```rust
ext4_bmap_bit_find_clr(data, 0, inodes_in_bg, &mut idx_in_bg);
ext4_bmap_bit_set(&mut raw_data, idx_in_bg);
```
设置相应的inode计数
```rust
/* Modify filesystem counters */
bg.set_free_inodes_count(&super_block, free_inodes);
/* Increment used directories counter */
if is_dir {
used_dirs += 1;
bg.set_used_dirs_count(&super_block, used_dirs);
}
/* Decrease unused inodes count */
bg.set_itable_unused(&super_block, unused);
/* Update superblock */
super_block.decrease_free_inodes_count();
```
init inode设置inode基础信息
```rust
...
inode.ext4_inode_set_mode(mode);
inode.ext4_inode_set_links_cnt(0);
...
```
init inode extent 信息
```rust
pub fn ext4_extent_tree_init(inode_ref: &mut Ext4Inode) {
/* Initialize extent root header */
let mut header = unsafe { *ext4_inode_get_extent_header(inode_ref) };
ext4_extent_header_set_depth(&mut header, 0);
ext4_extent_header_set_entries_count(&mut header, 0);
ext4_extent_header_set_generation(&mut header, 0);
ext4_extent_header_set_magic(&mut header, EXT4_EXTENT_MAGIC);
ext4_extent_header_set_max_entries_count(&mut header, 4 as u16);
}
```
再接着link inode号到文件名,目录项，父目录，首先找到父目录的目录项，再把当前文件的目录项添加到父目录项的尾部。
```rust
ext4_link::<Hal>(&mp,&root_inode,&mut child_inode_ref,path,name_len,false){
ext4_dir_find_entry::<A>(&parent_inode, &path, len as u32, &mut dir_search_result);
/* Add entry to parent directory */
ext4_dir_add_entry::<A>(parent_inode, child_inode, path, len);
}
```
# 写文件
查找文件逻辑块对应的物理块，如果没有对应物理块则分配一个物理块。
```rust
pub fn ext4_file_write(&self, ext4_file: &mut Ext4File, data: &[u8], size: usize) {
...
let mut size = size;
while size >= block_size {
while iblk_idx < iblock_last {
if iblk_idx < ifile_blocks {
ext4_fs_append_inode_dblk(&mut inode_ref, &mut (iblk_idx as u32), &mut fblk);
}
iblk_idx += 1;
...
}
size -= block_size;
}
for i in 0..fblock_count {
...
self.block_device
.write_offset(offset, &data[idx..(idx + BLOCK_SIZE as usize)]);
}
...
inode_ref.write_back_inode();
}
```
分配物理块同样要从block bitmap中查询, 当分配完物理块后，就可以填写extent信息了。再把记录了逻辑块和物理块对应信息的extent插入extent树中。最后在相应的物理块中写入数据。
```rust
ext4_balloc_alloc_block()
ext4_ext_insert_extent()
```
# checksum
创建文件，写入文件都涉及对meta data的修改。所有meta data都有crc检验信息。修改元数据后，需设置校验信息，然后写入磁盘
```rust
例
pub fn sync_inode_to_disk_with_csum(
&mut self,
block_device: Arc<dyn BlockDevice>,
super_block: &Ext4Superblock,
inode_id: u32,
) -> Result<()> {
self.set_inode_checksum(super_block, inode_id);
self.sync_inode_to_disk(block_device, super_block, inode_id)
}
```
