More journal work

Author: nicholasbishop

src/error.rs

@@ -214,6 +214,18 @@ pub enum Corrupt {
         u32,
     ),
 
+    /// Journal size is invalid.
+    JournalSize,
+
+    /// Journal magic is invalid.
+    JournalMagic,
+
+    /// Journal contains a block of unknown type.
+    JournalBlockType(
+        /// Raw block type.
+        u32,
+    ),
+
     /// An inode's checksum is invalid.
     InodeChecksum(
         /// Inode number.
@@ -324,6 +336,18 @@ impl Display for Corrupt {
                 f,
                 "invalid checksum for block group descriptor {block_group_num}"
             ),
+            Self::JournalSize => {
+                write!(f, "journal size is invalid")
+            }
+            Self::JournalMagic => {
+                write!(f, "journal magic is invalid")
+            }
+            Self::JournalBlockType(block_type) => {
+                write!(
+                    f,
+                    "journal contains an unknown block type: {block_type}"
+                )
+            }
             Self::InodeChecksum(inode) => {
                 write!(f, "invalid checksum for inode {inode}")
             }
@@ -422,6 +446,12 @@ pub enum Incompatible {
         /// Inode number.
         u32,
     ),
+
+    /// Journal superblock type is not supported.
+    JournalSuperblockType(
+        /// Raw journal block type.
+        u32,
+    ),
 }
 
 impl Display for Incompatible {
@@ -442,6 +472,9 @@ impl Display for Incompatible {
             Self::DirectoryEncrypted(inode) => {
                 write!(f, "directory in inode {inode} is encrypted")
             }
+            Self::JournalSuperblockType(val) => {
+                write!(f, "journal superblock type is not supported: {val}")
+            }
         }
     }
 }

src/journal.rs

@@ -6,29 +6,318 @@
 // option. This file may not be copied, modified, or distributed
 // except according to those terms.
 
-use crate::{Ext4, Ext4Error};
+// TODO
+#![expect(dead_code)]
+
+use crate::inode::Inode;
+use crate::iters::file_blocks::FileBlocks;
+use crate::util::{read_u32be, u64_from_hilo};
+use crate::{Corrupt, Ext4, Ext4Error, Incompatible};
+use alloc::collections::BTreeMap;
+use alloc::vec;
+use alloc::vec::Vec;
+use bitflags::bitflags;
 
 #[derive(Debug)]
 pub(crate) struct Journal {
-    // TODO: add journal data.
+    blocks: BTreeMap<u64, u64>,
 }
 
 impl Journal {
-    /// Create an empty journal.
     pub(crate) fn empty() -> Self {
-        Self {}
+        Self {
+            blocks: BTreeMap::new(),
+        }
     }
 
-    /// Load a journal from the filesystem.
     pub(crate) fn load(fs: &Ext4) -> Result<Self, Ext4Error> {
-        let Some(_journal_inode) = fs.0.superblock.journal_inode else {
+        // Note: ext4 is all little-endian, except for the journal,
+        // which is all big-endian. 😬
+
+        let Some(journal_inode) = fs.0.superblock.journal_inode else {
             // Return an empty journal if this filesystem does not have
             // a journal.
             return Ok(Self::empty());
         };
 
-        // TODO: actually load the journal.
+        let journal_inode = Inode::read(fs, journal_inode)?;
+
+        let mut journal_block_iter =
+            FileBlocks::new(fs.clone(), &journal_inode)?;
+        let block_index =
+            journal_block_iter.next().ok_or(Corrupt::JournalSize)??;
+
+        let block_size = fs.0.superblock.block_size;
+        let mut block = vec![0; block_size.to_usize()];
+        fs.read_from_block(block_index, 0, &mut block)?;
+
+        let header = JournalBlockHeader::read_bytes(&block)?;
+
+        // Check superblock type.
+        if header.block_type != JournalBlockType::SUPERBLOCK_V2 {
+            return Err(Incompatible::JournalSuperblockType(
+                header.block_type.0,
+            )
+            .into());
+        }
+
+        let s_blocksize = read_u32be(&block, 0xc);
+        let _s_maxlen = read_u32be(&block, 0x10);
+        // TODO: what's the difference between first and start?
+        let _s_first = read_u32be(&block, 0x14);
+        let _s_sequence = read_u32be(&block, 0x18);
+        let s_start = read_u32be(&block, 0x1c);
+        let s_feature_compat = read_u32be(&block, 0x24);
+        let s_feature_incompat = read_u32be(&block, 0x28);
+
+        // TODO: check features.
+        // TODO: checksum type
+
+        // TODO
+        assert_eq!(s_blocksize, block_size);
+
+        let compat_features =
+            JournalCompatibleFeatures::from_bits_retain(s_feature_compat);
+        let incompat_features =
+            JournalIncompatibleFeatures::from_bits_retain(s_feature_incompat);
+
+        // TODO
+        assert_eq!(compat_features, JournalCompatibleFeatures::empty());
+        assert!(
+            incompat_features.contains(JournalIncompatibleFeatures::IS_64BIT)
+        );
+        //JournalIncompatibleFeatures::BLOCK_REVOCATIONS |
+        //| JournalIncompatibleFeatures::CHECKSUM_V3
+
+        let mut blocks = BTreeMap::new();
+        // TODO... minus 1 because already read the journal superblock
+        for _ in 1..s_start {
+            // TODO: unwrap
+            journal_block_iter.next().unwrap()?;
+        }
+        while let Some(block_index) = journal_block_iter.next() {
+            let block_index = block_index?;
+
+            // TODO: not all blocks need to be read...
+            fs.read_from_block(block_index, 0, &mut block)?;
+
+            // TODO: validate checksums.
+
+            let h_magic = read_u32be(&block, 0x0);
+            if h_magic != 0xc03b3998 {
+                // No magic.
+                // dbg!("no magic");
+                break;
+            }
+
+            let header = JournalBlockHeader::read_bytes(&block)?;
+
+            if header.block_type == JournalBlockType::DESCRIPTOR {
+                let tags =
+                    JournalDescriptorBlockTag::read_bytes_to_vec(&block[12..])
+                        .unwrap();
+
+                // TODO: are these blocks the size of the filesystem or
+                // of the journal? Or always the same?
+
+                for tag in &tags {
+                    // TODO: unwrap
+                    let block_index = journal_block_iter.next().unwrap()?;
+
+                    blocks.insert(tag.block_number, block_index);
+                }
+            } else if header.block_type == JournalBlockType::COMMIT {
+                // TODO: do stuff with the commit block
+            } else {
+                todo!()
+            }
+        }
+
+        Ok(Self { blocks })
+    }
+
+    pub(crate) fn map_block_index(&self, block_index: u64) -> u64 {
+        *self.blocks.get(&block_index).unwrap_or(&block_index)
+    }
+}
+
+#[derive(Clone, Copy, Debug, Eq, PartialEq)]
+struct JournalBlockType(u32);
+
+impl JournalBlockType {
+    const DESCRIPTOR: Self = Self(1);
+    const COMMIT: Self = Self(2);
+    const SUPERBLOCK_V1: Self = Self(3);
+    const SUPERBLOCK_V2: Self = Self(4);
+    const REVOCATION: Self = Self(5);
+}
+
+#[derive(Debug)]
+struct JournalBlockHeader {
+    block_type: JournalBlockType,
+    sequence: u32,
+}
+
+impl JournalBlockHeader {
+    fn read_bytes(bytes: &[u8]) -> Result<Self, Ext4Error> {
+        assert!(bytes.len() >= 12);
+
+        let h_magic = read_u32be(bytes, 0x0);
+        let h_blocktype = read_u32be(bytes, 0x4);
+        let h_sequence = read_u32be(bytes, 0x8);
+
+        // Check journal magic.
+        if h_magic != 0xc03b3998 {
+            return Err(Corrupt::JournalMagic.into());
+        }
+
+        let block_type = JournalBlockType(h_blocktype);
+
+        Ok(Self {
+            block_type,
+            sequence: h_sequence,
+        })
+    }
+}
+
+bitflags! {
+    #[derive(Clone, Copy, Debug, Eq, PartialEq, Ord, PartialOrd, Hash)]
+    pub struct JournalCompatibleFeatures: u32 {
+        const CHECKSUMS = 0x1;
+    }
+}
+
+bitflags! {
+    #[derive(Clone, Copy, Debug, Eq, PartialEq, Ord, PartialOrd, Hash)]
+    pub struct JournalIncompatibleFeatures: u32 {
+        const BLOCK_REVOCATIONS = 0x1;
+        const IS_64BIT = 0x2;
+        const ASYNC_COMMITS = 0x4;
+        const CHECKSUM_V2 = 0x8;
+        const CHECKSUM_V3 = 0x10;
+        const FAST_COMMITS = 0x20;
+    }
+}
+
+enum JournalChecksumType {
+    Crc32 = 1,
+    Md5 = 2,
+    Sha1 = 3,
+    Crc32c = 4,
+}
+
+// TODO: the kernel docs for this are a mess
+#[derive(Debug)]
+struct JournalDescriptorBlockTag {
+    block_number: u64,
+    flags: JournalDescriptorBlockTagFlags,
+    checksum: u32,
+    uuid: [u8; 16],
+}
+
+impl JournalDescriptorBlockTag {
+    fn read_bytes(bytes: &[u8]) -> (Self, usize) {
+        // TODO: for now assuming the `incompat_features` assert above.
+
+        let t_blocknr = read_u32be(bytes, 0);
+        let t_flags = read_u32be(bytes, 4);
+        let t_blocknr_high = read_u32be(bytes, 8);
+        let t_checksum = read_u32be(bytes, 12);
+
+        let flags = JournalDescriptorBlockTagFlags::from_bits_retain(t_flags);
+        let mut size: usize = 16;
+
+        let mut uuid = [0; 16];
+        if !flags.contains(JournalDescriptorBlockTagFlags::UUID_OMITTED) {
+            // OK to unwrap: length is 16.
+            uuid = bytes[16..32].try_into().unwrap();
+            // TODO: unwrap
+            size = size.checked_add(16).unwrap();
+        }
+
+        (
+            Self {
+                block_number: u64_from_hilo(t_blocknr_high, t_blocknr),
+                flags,
+                checksum: t_checksum,
+                uuid,
+            },
+            size,
+        )
+    }
+
+    fn read_bytes_to_vec(mut bytes: &[u8]) -> Result<Vec<Self>, Ext4Error> {
+        let mut v = Vec::new();
+        while !bytes.is_empty() {
+            let (tag, size) = Self::read_bytes(bytes);
+            let is_end =
+                tag.flags.contains(JournalDescriptorBlockTagFlags::LAST_TAG);
+            v.push(tag);
+
+            if is_end {
+                return Ok(v);
+            }
+
+            bytes = &bytes[size..];
+        }
+        // TODO: return a Corrupt error.
+        todo!("missing end tag")
+    }
+}
+
+bitflags! {
+    #[derive(Clone, Copy, Debug, Eq, PartialEq, Ord, PartialOrd, Hash)]
+    pub struct JournalDescriptorBlockTagFlags: u32 {
+        const ESCAPED = 0x1;
+        const UUID_OMITTED = 0x2;
+        const DELETED = 0x4;
+        const LAST_TAG = 0x8;
+    }
+}
+
+// TODO
+#[cfg(feature = "std")]
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use crate::test_util::load_compressed_filesystem;
+    use alloc::rc::Rc;
+
+    // TODO
+    #[test]
+    fn test_journal() {
+        let mut fs =
+            load_compressed_filesystem("test_disk_4k_block_journal.bin.zst");
+
+        let block_size = fs.0.superblock.block_size;
+        let mut b1 = vec![0; block_size.to_usize()];
+        let mut b2 = vec![0; block_size.to_usize()];
+        println!("looking for mismatches...");
+        for (dst, src) in &fs.0.journal.blocks {
+            fs.read_from_block(*dst, 0, &mut b1).unwrap();
+            fs.read_from_block(*src, 0, &mut b2).unwrap();
+            if b1 != b2 {
+                dbg!(dst, src);
+            }
+            //dbg!(dst, src, b1 == b2);
+        }
+        println!("done looking");
+
+        let entries = fs
+            .read_dir("/")
+            .unwrap()
+            .map(|e| e.unwrap().file_name().as_str().unwrap().to_owned())
+            .collect::<Vec<_>>();
+        dbg!(entries);
+        //todo!();
+
+        let test_dir = "/dir500";
+
+        // With the journal in place, this directory exists.
+        assert!(fs.exists(test_dir).unwrap());
 
-        Ok(Self {})
+        // Clear the journal, and verify that the directory no longer exists.
+        Rc::get_mut(&mut fs.0).unwrap().journal.blocks.clear();
+        assert!(!fs.exists(test_dir).unwrap());
     }
 }

src/lib.rs

@@ -285,6 +285,8 @@ impl Ext4 {
             })
         };
 
+        let block_index = self.0.journal.map_block_index(block_index);
+
         // The first 1024 bytes are reserved for non-filesystem
         // data. This conveniently allows for something like a null
         // pointer check.