New new format branch ready for testing

Hello everyone,

Yan Zheng sent along an update for the new format patch today, and it slightly changed the disk format for backrefs used when balancing space across drives. We decided against supporting an intermediate format that was only a week old, and his update makes a few important features much much easier (snapshot deletion for one).

If you've tried the newformat branch and run btrfs-vol -b or run btrfs-vol -r, you'll have to backup your data and reformat before switching to the newest code.

If this is a problem for any testers of the newformat branch, please don't hesitate to send me an email. We'll find a way to help you transition to the new code. The main reason I'm willing to make this format switch is because these commands are not commonly run, and I doubt there has been a huge number of people trying out the newformat code.

If you've tried the newformat branch and haven't run btrfs-vol -b or btrfs-vol -r, you can safely switch to the updated code without any extra steps.

If you're not sure if you've run these commands, pull down the new code just for btrfs-progs, and run btrfsck. It will print an error if it finds any of the unsupported disk items.

I didn't want people to accidentally pull the new new format on top of the old new format without reading this email, so I gave the new new format a new branch name (newformat2).

For the kernel (still at 2.6.30-rc7, rebased today):

git pull git://git.kernel.org/pub/scm/linux/kernel/git/mason/btrfs-unstable.git newformat2

For the progs (also rebased today):

git://git.kernel.org/pub/scm/linux/kernel/git/mason/btrfs-progs-unstable.git newformat2

I deleted the newformat branch. Only newformat2 should be used for new testing now.

Since this is working well in testing so far, we plan on asking Linus to pull it for 2.6.31. We don't plan on any more format changes for this code.

-chris

Hello everyone,

Just a quick note that I've updated the newformat2 kernel branch to include some fixes for buffered IO stalls and a new mount -o ssd_spread mount option that should perform better than regular mount -o ssd when used on lower end SSDs.

The big difference is that mount -o ssd_spread forces allocations to find fully free areas, where mount -o ssd will find rough groupings of free blocks that might have allocated blocks mixed in.

Tomorrow I'll add some autodetection for devices that have the non-rotational flag set, which is a hint the block layer gives us that we're operating on an SSD.

-chris