On Tue, Sep 29, 2009 at 6:17 AM, Peter Teoh <htmldeveloper@xxxxxxxxx> wrote:
> On Tue, Sep 29, 2009 at 2:51 AM, Venkatesh Srinivas <me@xxxxxxxxxxx> wrote:
>> Hi,
>>
>> Thanks for the reply!
>>
>> I was looking for more information on data=journal, not data=ordered
>> or data=writeback; I didn't see comments on it on the ext wiki page,
>> or two of the three links. Do you know anyplace I could look for more
>> on that?
>>
>> Thanks,
>> -- vs
>>
>
> I guessed fs/ext3/fsync.c:ext3_sync_file() best summarized it:
>
> 53 /*
> 54 * data=writeback:
> 55 * The caller's filemap_fdatawrite()/wait will sync the data.
> 56 * sync_inode() will sync the metadata
> 57 *
> 58 * data=ordered:
> 59 * The caller's filemap_fdatawrite() will write the data and
> 60 * sync_inode() will write the inode if it is dirty.
> Then the caller's
> 61 * filemap_fdatawait() will wait on the pages.
> 62 *
> 63 * data=journal:
> 64 * filemap_fdatawrite won't do anything (the buffers
> are clean).
> 65 * ext3_force_commit will write the file data into
> the journal and
> 66 * will wait on that.
> 67 * filemap_fdatawait() will encounter a ton of
> newly-dirtied pages
> 68 * (they were dirtied by commit). But that's OK -
> the blocks are
> 69 * safe in-journal, which is all fsync() needs to ensure.
> 70 */
> 71 if (ext3_should_journal_data(inode)) {
> 72 ret = ext3_force_commit(inode->i_sb);
> 73 goto out;
> 74 }
> 75
>
> as indicated above, ext3_force_commit() is called, which essentially
> writes to the journal file. And ext3_should_journal_data() is YES
> when in data=journal mode.
>
> now, that is indeed an extra overhead, as later in the function,
> sync_inode() will be called to write the inode to its data block as
> well. so data is duplicated both in journal + data block.
>
> make sense?
>
i must also add that there are many other parts of ext3 fs that used
ext3_should_journal_data() logic to increase reliability at the
expense of performance as well.
eg, inode.c:
84
85 if (test_opt(inode->i_sb, DATA_FLAGS) ==
EXT3_MOUNT_JOURNAL_DATA ||
86 (!is_metadata && !ext3_should_journal_data(inode))) {
87 if (bh) {
88 BUFFER_TRACE(bh, "call journal_forget");
89 return ext3_journal_forget(handle, bh);
90 }
91 return 0;
92 }
93
94 /*
95 * data!=journal && (is_metadata || should_journal_data(inode))
96 */
97 BUFFER_TRACE(bh, "call ext3_journal_revoke");
and then super.c ...... etc etc.
--
Regards,
Peter Teoh
--
To unsubscribe from this list: send an email with
"unsubscribe kernelnewbies" to ecartis@xxxxxxxxxxxx
Please read the FAQ at http://kernelnewbies.org/FAQ
