Btrfs devel - Jun 2007 - Btrfs v0.4 available (warning: disk format change)

Hello everyone,

I wasn't planning on sending out v0.4 until ENOSPC was fixed, but Frank
Groeneveld narrowed down a bug with the file data checksumming code on
highmem 32bit machines.

Basically the way I was using cryptomgr made it incorrectly calculate
checksums on highmem pages.

So to simplify things, I've switched over to plain libcrc32c.  Anyone
running on a 32 bit with highmem is going to hit this bug.  Now you only
need to compile:

CONFIG_LIBCRC32C=m

Into the main kernel (static is also fine).

v0.4 has my first round of enospc fixes, which tries much harder to find
free blocks when initial allocation fails.  Things will still go very
badly when the FS is full, but now full is defined as pretty close to
100%.

v0.4 does have a disk format change.  I added some fields to the tree
root items so that I could detect and remove dead roots after a crash.
This recovers space taken up in tree roots that were either being
pruned or scheduled for pruning at the time of the crash.  Any
filesystems made with either earlier versions will not be readable by
v0.4.

If you update, please make sure to get both the userland progs and the
kernel modules.  If they don't match you will crash in strange and
unhappy ways.

-chris

On Mon, Jun 25, 2007 at 10:42:36PM +0200, Christian Hesse
wrote:
> On Friday 22 June 2007, Chris Mason wrote:
> > Hello everyone,
> >
> > I wasn't planning on sending out v0.4 until ENOSPC was fixed, but
Frank
> > Groeneveld narrowed down a bug with the file data checksumming code on
> > highmem 32bit machines.
> 
> Hi everyone!
> 
> I wonder why nobody answered to this message, so I will do now.
> 
> I'm not a coder but one of the guys who love to play with new code. 
> Additionally I I tested most filesystems available for linux today and 
> returned to ext3 some time ago, but it is sill not the right thing [tm].
> 
> The feature list is really great. Said this I had a lot of fun testing
btrfs.
> Everything worked smoothly: I untared a kernel tree, snapshotted, patched
the
> tree, snapshotted again, patched some other things, diffed the snapshots,
...
> No problems so far, anything works reliable and without crashes.
> 
> Oh, one more thing: I test a lot of kernel stuff and hate to recompile
every
> single module for a new kernel so I put it into my kernel tree. If anybody
is
> interested: Untar the archive to linux/fs/, more btrfs-${VERSION}/ to
btrfs/,
> apply the attached patch and run make oldconfig.
> Btw, compiling btrfs into the kernel statically results in a kernel panic
when
> trying to mount the root fs, so the module is always built as module.
Whoops, could you please paste that panic here?
> 
> So please keep up the good work! I will keep an eye on btrfs and test the
code
> from time to time.
Great, thanks for the testing!

-chris