Ext4 devel interlock meeting minutes (July 30, 2007)

Ext4 devel interlock meeting minutes (July 30, 2007)

[Avantika Mathur]

Ext4 Developer Interlock Call: July 30, 2007 Meeting Minutes

Minutes can be accessed at: 
http://ext4.wiki.kernel.org/index.php/Ext4_Developer%27s_Conference_Call

Attendees: Mingming Cao, Andreas Dilger, Jean-Pierre Dion, Jean-Noel 
Cordenner, Valérie Clement, Aneesh Veetil, Dave Kleikamp, Jose Santos, 
Ted T'so, Avantika Mathur

PATCH STATUS

Delalloc:
- Alex posted his vfs delayed allocation patches.
- Comments are that there is not sufficient generic support for XFS.  
The code that Christoph Hellwig had posted which does support XFS is 
very complex.
- Christoph would be ok to take most of Alex's patch as an ext4 patch.  
In the short term we can put the delalloc support in ext4 and then later 
work on generic vfs support.
- Mingming will respond to the thread asking what Andrew Morton's 
preference is.

Multiple Block Allocation:
- Aneesh has been hitting a bug when creating a filesystem with 
different number of blocks/block group then default.  the problem 
occurred during online migration with mballoc enabled.
- A possible temporary solution to this problem is to disable mballoc if 
the filesystem created with not the default blocks/block group ratio.
- Aneesh will write-up the problem he is seeing and send an e-mail to Alex.
- Aneesh will split mballoc into more patches, to be added to the 
ext4-patch-queue.  This will make it easier to review.

Inode Version:
- Jean-Noel sent the mount option patch to linux-ext4.  We had discussed 
adding a flag to the superblock indicating that i_version is enabled, 
but there are no bits available in f_flags.
- Mingming will reply to the post saying the overwrite case needs to be 
addressed first

Uninitialized Block Groups:
- Avantika will look into why the checksum is not consistent - and why 
the check fails even when the feature is not enabled.

Flexible Block Groups:
- Jose is trying to see why fsck is not working.  once this is resolved, 
the feature will be complete.  
- He has seen a problem with fsck when meta_bg feature is enabled.  
Andreas saw a similar problem, and the uninitialized block groups patch 
fixes the issue.  Jose will look back at the fix which is in the series 
of e2fsprogs patches sent by Girish last week.

- Mingming asked about exporting JBD stat info to debugfs; as well as 
procfs.  The consensus had been to just export to procfs as not everyone 
uses debugfs.

Re: Ext4 devel interlock meeting minutes (July 30, 2007)

[Jose R. Santos]

On Fri, 03 Aug 2007 15:56:37 -0700
Avantika Mathur <mathur@linux.vnet.ibm.com> wrote:
> Flexible Block Groups:
> - Jose is trying to see why fsck is not working.  once this is resolved, 
> the feature will be complete. 

Not quite true.  Solving the fsck issue means that we can do more
thorough testing of the performance impacts of packaging meta-data
closer together.  The patches do not currently remove the 512TB limit
imposed by the number of block groups, which is something that I also
plan to address on this feature.

-JRS

Ext4 devel interlock meeting minutes (July 30, 2007)

[Avantika Mathur]

Ext4 Developer Interlock Call:  August 6, 2007 Meeting Minutes

Attendees: Mingming Cao, Ted T'so, Andreas Dilger, Jean Pierre Dion, 
Valerie Clement, Jose Santos, Aneesh Veetil, Avantika Mathur

- Valerie has developed a series of patches to e2fsprogs, preparing it 
for 64-bit support.  She still needs to re-write some of the patches, 
taking Ted and Andreas' comments into account.  
    -- The current patches are against e2fsprogs-1.40, in which there is 
no extents support.  Ted will first work on designing the extents 
interface before extents will supported in e2fsprogs.

- Aneesh has been working on an undo I/O manager. Currently his this is 
only for mke2fs, Andreas said it would be useful to have this 
functionality available in e2fsck as well.  Aneesh will try to add this 
as he gets time.

- Aneesh's enlarge inode patches are desired for addition in e2fsprogs.  
Current interface is through tunefs.  Andreas wonders if it might make 
sense to make this function part of resizefs rather than adding 
complexity to tunefs.  Also, since we will also have the ability to 
change the number of inodes, resizefs would be a good place for this 
functionality.
- Aneesh will also try to post some testcases for the feature next time 
he posts the patches

There was much discussion on the Flexible block groups feature under 
development by Jose; including the desired function and how it is 
implemented.  Jose proposed implementing dynamic inode tables on top of 
the flex_bg feature.
- Jose will send an e-mail to the mailing list with the purpose of the 
feature and current plans, to make sure there is consensus before moving 
forward.