Shrinking Support

Latest revision as of 21:19, 21 December 2008

Currently XFS Filesystems can't be shrunk.

To support shrinking XFS filesystems a few things needs to be implemented, based on a list by Dave Chinner [1]:

• A way to check space is available for shrink

• An ioctl or similar interface to prevent new allocations from a given allocation group.

• A variant of the xfs_reno tool to support moving inodes out of filesystem areas that go away.

• A variant of the xfs_fsr tool to support moving data out of the filesystem areas that go away.

• Some way to move out orphan metadata out of the AGs truncated off

• A transaction to shrink the filesystem.

At that point, we'll have a "working" shrink that will allow shrinking to only 50% of the original size because the log (in the middle of the filesystem) will get in the way. To fix that, we'll need to implement transactions to move the log...

[edit] Available pieces

• A script from Ruben Porras to check if enough free space is available to support shrinking [2]

• A patch from Ruben Porras to allow / disallow allocation from an allocation group [3] plus userspace support for setting / clearing it [4]

• The xfs_fsr tool in xfsprogs

• The xfs_reno too, see Unfinished_work#The_xfs_reno_tool

• An untested patch from Dave Chinner for a xfs_swap_inodes ioctl that allows to not just defragment extents but moving the whole inode [5] and a patch to xfs_reno to use it from Ruben Porras [6]