linux-ext4 - JBD2 issue unnecessary journal commit mark I/O

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [thread-next>] [day] [month] [year] [list]

Message-ID: <CAJKSsyDG26b9n-me+U=nZw6ZNLzwXmGapU3J=8ejHjQjuRP1OQ@mail.gmail.com>
Date:	Wed, 28 Oct 2015 17:19:02 +0900
From:	Eunryoung Lim <eunryoung.lim@...il.com>
To:	linux-ext4@...r.kernel.org
Subject: JBD2 issue unnecessary journal commit mark I/O

Hello,

I am Eunryoung Lim and have a question about EXT4 journaling source.
I will be appreciated for your any comments.

* ordered journaling mode, disabled async_jounral_commit
On my understating, below steps is the simple process of journal
transaction commit (jbd2_journal_commit_transaction() ).
Step 1,3,4 are optional.

1. Write journal revoke block (if it required).
2. Write journal descriptor + journal buffer’s page.
3. Submit all the data buffers of inode associated with the
transaction (journal_submit_data_buffers() ) (if it required).
4. Wait for data submitted for writeback page
(journal_finish_inode_data_buffers() ) (if it required).
5. Write journal commit mark.

The problem is that JBD2 issue unnecessary journal commit mark I/O.
For example, JBD2 write journal commit mark (step 5), even though
transaction (running ->commit state) has no journal buffer yet.
The I/O for journal descriptor + journal buffer’s page (step 2) can be
skipped because JBD2 check whether the transaction has journal buffer
or not.
void jbd2_journal_commit_transaction(journal_t *journal)
{
            ….
            /* start to commit transaction */
            while (commit_transaction->t_buffers) {
            /* Find the next buffer to be journaled... */
            jh = commit_transaction->t_buffers;
            .…
}

I think, if the journal buffers do not exist in the transaction, JBD2
can skip step 5 to reduce amount of I/O.
Thus, the code can be modified as follows: (this is simple way, we can
optimize more)
(current)
        err = journal_submit_commit_record(journal, commit_transaction,
(after)
        if ( IS_ERR_OR_NULL(commit_transaction->t_buffers) ) {
        err = journal_submit_commit_record(journal, commit_transaction,
}
or we can simply return the function.
(add)if ( IS_ERR_OR_NULL(commit_transaction->t_buffers) )
            return ;

        /* start to commit transaction */
        while (commit_transaction->t_buffers) {
        ....

Note. Android uses SQLite, and it makes sure that directory entry
contains newly created journal file by calling fsync() to the
directory.
(However, SQLite call fsync() to directory even though the file is not created).
When fsync() is called to directory,
EXT4 calls ext4_sync_file() -> ext4_force_commit() ->
ext4_journal_force_commit() -> jbd2_journal_force_commit() -> ... and
finally, jbd2_journal_commit_transaction();
If journal transaction is empty, JBD2 only write journal commit mark
and it causes lots of I/Os.

Thanks,
Eunryoung Lim.
--
To unsubscribe from this list: send the line "unsubscribe linux-ext4" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html