Ext4 Developer Interlock Call: July 9, 2007 Meeting Minutes
Minutes can be accessed at:
Attendees: Mingming Cao, Takashi Sato, Dave Kleikamp, Jose Santos,
Andreas Dilger, Amit Arora, Suparna Bhattacharya, Aneesh Veetil, Ted
T'so, Avantika Mathur
Preallocation and fallocate:
- Amit will first be merging in Andreas' patch to fallocate, which
allows initialized extents to be the full 32768 blocks. Uninitialized
extents are limited to 32767 blocks. Amit will also add comments to
this, and have the update patches ready by tomorrow.
- The fallocate patches are already in the -mm tree. Ted will move the
fallocate patches to the front of the patch-queue.
- Once Amit has updated the preallocation, we will push fallocate
patches and the ext4 preallocation patches to Linus' tree.
- Once fallocate patches are merged in mainline, send patches to Ulrich
for glibc changes.
- Christoph Hellwig thinks there should be a separate system call for
deallocate functionality. - Mingming discussed an error case in current
preallocation. If the entire filesystem is full, When an uninitialized
extent is written to, there is no space to store the metadata blocks
needed to split it into unintialized and initialized extents. Andreas'
proposed solution is to fallback to writing zeroes out to the
preallocated blocks. This issue can be resolved as a bug fix after
preallocation is in mainline.
- Mingming asked about implementing preallocation in ext2/3. It was
decided that this is not desirable, as it is an advanced change, and our
current focus is ext4.
- Negative timestamps: Temporary solution will be to round small
negative timestamps within a close range to the epoch to zero. Large
negative timestamps would be converted to positive. This can later be
fixed to have signed timestamps.
- We will try to push the delalloc feature for ext4 to mainline.
- It is currently below the unstable patch marker in the queue, because
of possible vfs implementation of delalloc.
- After the rest of the ext4 patches are pushed; then we will work to
push delalloc. These can even be included after 2.6.23-rc1, since there
are no major format changes.
- If later there is a vfs level delalloc implemented, we can then take
out support for delalloc from within ext4.
- Trond has requested a flag in the superblock indicating i_version as
on or off, rather than using a mount option. - Plan is to remove the
last i_version patch, which supports disabling i_version through a
mount option. Mingming will remove this patch from the path queue.
The issue can be resolved once the feature is in mainline.
- Aneesh will resend a patch witha fix to the block reservation; to be
added to the patch queue.
- Aneesh had ported the mballoc patches to the current patch queue. The
patch was too large to send to the mailing list. He has found some
issues with the patches, and will work on fixing those before requesting
- Aneesh suggested adding 'from: author' to each patch to capture who
the real author of each feature is.
- Now that version 1.40 is out, Ted will focus on merging in the extetns
support patches, and trying to have full exents support.
- Ted will also start merging other small patches, such as large number
of subdirectory support, and nanosecond timestamps.