Struct JournalWriter 

pub struct JournalWriter { /* private fields */ }

Owns the live-write side of the JBD2 journal. Built once at mount time (when the FS has a journal); reused across every mutating capi call.

Cheap to construct (one inode read + extent walk). Not thread-safe; the outer Filesystem lock must serialize mutating ops anyway.

Implementations

impl JournalWriter

pub fn open(fs: &Filesystem) -> Result<Option<Self>>

Open the writer for a mounted filesystem.

Returns Ok(None) when the FS has no journal (s_journal_inum == 0) — callers should fall back to unjournaled writes in that case. Returns Err when the journal is misconfigured (e.g. the journal inode is missing or unparsable).

Flavor-aware: walks the journal inode’s block tree via crate::indirect::map_logical_any, so ext3 journals (legacy indirect block pointers) work the same as ext4 ones (extent tree).

pub fn begin(&self) -> Transaction

Begin a new transaction with the next sequence number. Caller adds writes, then calls Self::commit to publish.

pub fn max_blocks_per_transaction(&self) -> usize

The capacity (in fs blocks) the caller can fit into one transaction. One block is reserved for the JBD2 superblock; the remaining max_len - 1 carry [descriptor, data…, (revoke), commit].

pub fn commit(&mut self, dev: &dyn BlockDevice, tx: &Transaction) -> Result<()>

Publish a transaction crash-safely (see module docs for the four- fence ordering). On success, the in-memory JSB is up-to-date and the on-disk JSB has been written back twice (dirty marker + clean marker).

pub fn is_healthy(&self) -> bool

False once a commit failed after its first device write: the log and the final locations are then in a state only a fresh replay can judge, so this writer refuses every later commit. A refusal before any write (read-only device, stale sequence, oversized transaction) leaves the device untouched and the writer usable.