Ext3-devel/ext3minutes-20060621

From cmm@us.ibm.com Fri Jun 23 16:34:17 2006 Date: Fri, 23 Jun 2006 16:29:48 -0700 From: Mingming Cao  To: ext2-devel@lists.sourceforge.net Subject: [Ext2-devel] ext3 developers interlock meeting

Hello!!

We had weekly ext3 developers interlock meetings between some ext3 developers for a few weeks. A lot of discussions about the 48 bit ext3 filesystem were covered during last few meetings. All the previous meeting minutes are kept at ext3 wiki page http://fedoraproject.org/wiki/ext3-devel

The interlock meeting is open for everyone when are interested in the ext3/ext4 design/development/testing/or general interest. The meeting is hosted on every Wednesday at 8:00am PCT. Please contact me directly for the calling infos.

Here is the meeting minutes of ext3 developers interlock meeting on June 21st, 2006:

Attendees:

Stephen Tweedie, Mingming Cao, Theodore Ts'o, Andreas Dilger, Badari Pulavarty, Suparna Bhattacharya, Takashi Sato, Laurent Vivier, Jean-Noel Cordenner, Alexandre Ratchov, Don Howard, Avantika Mathur

1. The lastes 48bit ext3 patch series is still at: http://ext2.sourceforge.net/48bitext3/patches/patches-2.6.17-rc6-06082006/

Stephen pointed bugs that breaks current non-extent 32 bit ext3 filesystem in this series:


 * Wrong superblock fields.

The ext3_fs.h superblock definition is missing some new fields defined by e2fsprogs, specifically mkfs-time and the journal inode backup. Action: Alexandra seems posted a patch for this, Stephen to review.


 * Split read of journal tags on recovery.

Reading a journal block during recovery uses be64_read_split, which (if CONFIG_LBD is set) reads 64-bits, even when the journal is using 32-bit entries! This leads to garbage in the top 32-bits of the block numbers during recovery, and recovery fails.

Action: Badari to ping Stephen to get it fixed.
 * revoke.c still has an undefined construct:

int hash = (int)block ^ (int)(block >> 32);

where ">> 32" is not defined if "block" is already 32-bits wide

Action: Mingming fixed this. to be included in next release.

compatibility flags for the available block device max size;
 * there is also missing basic sanity checking for superblock/journal

Action: Stephen will look at adding that.

2. Response to lkml's discussion about time for ext4

People who called in this week seems agree forking a new filesystem to address the overall concern about stability impact to current stable ext3 filesystem. The current plan is to start a new filesystem, called ext3-devel at this moment. Once all it's stabilized, call it ext4. This is to prevent some of the confusion about the versions of the this new filesystem

This meeting also discussed whether to turn on extent by default in this ext3-devel filesystem. People probably will expect to see extent when they using ext4 filesystem, but on that other hand, this also means the new

The current plan is not the split the JBD code.Current changes to JBD layer is not so intrusive, just a few "ifs" to determine a 64 bit branch handling. Andreas mentioned there are some patches that added checksum to JBD which we wanted but relatively intrusive. We probably need to split JBD code when we really get there.

Action: Stephen and Ted to draft a mail about proposed next steps and potential issues for setting up the next "ext" filesystem -ext4, and send it to ext2-devel for discussion before expose it to lklm. The projected plan to stabilize ext4 at least take a year.

3. Currently the patches are hosted at ext2.sf.net/48bitext3/patches Plan to keep the quilt styled patch series on ext2.sf.net. There is a wiki page for this project created at Fedora website: http://fedoraproject.org/wiki/ext3-devel

5. 32 bit ext3 The current ext3_fsblk_t patches that enables true 32 bit ext3 (8TB -> 16TB) are in mm tree. Hopefully it get into 2.6.18 kernel and we could start to fork the 48 bit ext4 filesystem from there.

6. extents We ought to support extent based directory, otherwise we can't address directories located on disk beyond 2^32 blocks.

So we should move the extent patch prior to everything turn on 48 bit ext3. and get it tested and integrated to the new filesystem first.

Action: Mingming to adjust the order of the series in next release when all the known easy bugs fixed.

5. tools Alexandra from BULL posted the changes to enable extents in e2fsprogs (changes originally from luster filesystem). Ted started to review them.

Thanks, Mingming