Does Reiser3 really support data=journal mode?

Does Reiser3 really support data=journal mode?

[Jan Kara]

  Hello,

  I've just looked into the Reiserfs code in 2.6.25 kernel and from the code
it seems as data=journal mode isn't really handled any specially from
data=writeback mode. So does Reiserfs really support journaling of data?
If no, then we should remove the mount option. If yes, then I really wonder
where is the distinction from data=writeback mode... Thanks for an answer in
advance.

									Honza

PS: I'm looking into this because I'm fixing some problems with quota when
journaling data so I'd like to fix Reiserfs as well in case it's affected.

--
Jan Kara <jack@suse.cz>
SUSE Labs, CR

Re: Does Reiser3 really support data=journal mode?

[Jeff Mahoney]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Jan Kara wrote:
>   Hello,
> 
>   I've just looked into the Reiserfs code in 2.6.25 kernel and from the code
> it seems as data=journal mode isn't really handled any specially from
> data=writeback mode. So does Reiserfs really support journaling of data?
> If no, then we should remove the mount option. If yes, then I really wonder
> where is the distinction from data=writeback mode... Thanks for an answer in
> advance.

Check out reiserfs_commit_page() for the code where it is attached to a
transaction. Then it's handled identically to dirty metadata when the
transaction is committed.

- -Jeff

- --
Jeff Mahoney
SUSE Labs
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v2.0.4-svn0 (GNU/Linux)
Comment: Using GnuPG with SUSE - http://enigmail.mozdev.org

iD8DBQFIDi8qLPWxlyuTD7IRAgd3AJ4h6aJRexiVTVC5ykRsqnRIwjdT9QCfcgRV
oOXg9E+0h01zCQ5XjZQWJ24=
=wHK4
-----END PGP SIGNATURE-----

Re: Does Reiser3 really support data=journal mode?

[Jan Kara]

On Tue 22-04-08 14:32:11, Jeff Mahoney wrote:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
> 
> Jan Kara wrote:
> >   Hello,
> > 
> >   I've just looked into the Reiserfs code in 2.6.25 kernel and from the code
> > it seems as data=journal mode isn't really handled any specially from
> > data=writeback mode. So does Reiserfs really support journaling of data?
> > If no, then we should remove the mount option. If yes, then I really wonder
> > where is the distinction from data=writeback mode... Thanks for an answer in
> > advance.
> 
> Check out reiserfs_commit_page() for the code where it is attached to a
> transaction. Then it's handled identically to dirty metadata when the
> transaction is committed.
  Ah, I see, thanks. I was grepping for "reiserfs_data_log" but there is
"reiserfs_file_data_log" which I completely missed.
  Maybe you can help me with my original problem :) - I have a file (quota
file) and I need to make sure that all the dirty data and metadata of that
file is on disk in its final locations (i.e.,  not only in journal). Will
just starting an auxiliary transaction and then doing
  do_journal_end(th, sb, 1, FLUSH_ALL)

  do the work as I think? Thanks.

									Honza 
-- 
Jan Kara <jack@suse.cz>
SUSE Labs, CR

Re: Does Reiser3 really support data=journal mode?

[Jeff Mahoney]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Jan Kara wrote:
> On Tue 22-04-08 14:32:11, Jeff Mahoney wrote:
>> -----BEGIN PGP SIGNED MESSAGE-----
>> Hash: SHA1
>>
>> Jan Kara wrote:
>>>   Hello,
>>>
>>>   I've just looked into the Reiserfs code in 2.6.25 kernel and from the code
>>> it seems as data=journal mode isn't really handled any specially from
>>> data=writeback mode. So does Reiserfs really support journaling of data?
>>> If no, then we should remove the mount option. If yes, then I really wonder
>>> where is the distinction from data=writeback mode... Thanks for an answer in
>>> advance.
>> Check out reiserfs_commit_page() for the code where it is attached to a
>> transaction. Then it's handled identically to dirty metadata when the
>> transaction is committed.
>   Ah, I see, thanks. I was grepping for "reiserfs_data_log" but there is
> "reiserfs_file_data_log" which I completely missed.
>   Maybe you can help me with my original problem :) - I have a file (quota
> file) and I need to make sure that all the dirty data and metadata of that
> file is on disk in its final locations (i.e.,  not only in journal). Will
> just starting an auxiliary transaction and then doing
>   do_journal_end(th, sb, 1, FLUSH_ALL)
> 
>   do the work as I think? Thanks.

Are you sure that's what you want? That will essentially make quota
operations sync events. If it's in the journal and the system crashes,
then it will be flushed out to the file system on journal replay anyway.

- -Jeff

- --
Jeff Mahoney
SUSE Labs
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v2.0.4-svn0 (GNU/Linux)
Comment: Using GnuPG with SUSE - http://enigmail.mozdev.org

iD8DBQFIDzafLPWxlyuTD7IRAmp8AJ9qgewSiH/dHiCSfIkf1npHH9NkWwCgnE/2
iJEBJijZXnMwKesZ5TP+ckE=
=TLOA
-----END PGP SIGNATURE-----

Re: Does Reiser3 really support data=journal mode?

[Jan Kara]

On Wed 23-04-08 09:16:15, Jeff Mahoney wrote:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
> 
> Jan Kara wrote:
> > On Tue 22-04-08 14:32:11, Jeff Mahoney wrote:
> >> -----BEGIN PGP SIGNED MESSAGE-----
> >> Hash: SHA1
> >>
> >> Jan Kara wrote:
> >>>   Hello,
> >>>
> >>>   I've just looked into the Reiserfs code in 2.6.25 kernel and from the code
> >>> it seems as data=journal mode isn't really handled any specially from
> >>> data=writeback mode. So does Reiserfs really support journaling of data?
> >>> If no, then we should remove the mount option. If yes, then I really wonder
> >>> where is the distinction from data=writeback mode... Thanks for an answer in
> >>> advance.
> >> Check out reiserfs_commit_page() for the code where it is attached to a
> >> transaction. Then it's handled identically to dirty metadata when the
> >> transaction is committed.
> >   Ah, I see, thanks. I was grepping for "reiserfs_data_log" but there is
> > "reiserfs_file_data_log" which I completely missed.
> >   Maybe you can help me with my original problem :) - I have a file (quota
> > file) and I need to make sure that all the dirty data and metadata of that
> > file is on disk in its final locations (i.e.,  not only in journal). Will
> > just starting an auxiliary transaction and then doing
> >   do_journal_end(th, sb, 1, FLUSH_ALL)
> > 
> >   do the work as I think? Thanks.
> 
> Are you sure that's what you want? That will essentially make quota
> operations sync events. If it's in the journal and the system crashes,
> then it will be flushed out to the file system on journal replay anyway.
  I need this only during quotaon and that is one-time event so performance
doesn't really matter. The problem is that userspace could have modified
quota files and we need to see all the changes. And we bypass page-cache
when reading quota files (we use block device's buffer cache directly) so
at the time of quotaon, we need to get all the data to their final location
on disk...

									Honza
-- 
Jan Kara <jack@suse.cz>
SUSE Labs, CR