Re: [PATCH 0/2] btrfs-progs: prevent mkfs from aborting with small volume

[Hidetoshi Seto <seto.hidetoshi@jp.fujitsu.com>]

(2013/08/26 23:23), Eric Sandeen wrote:
> Thanks for looking into this - how small of a device did you test?
> 
> I tried a 2MB device w/ these 2 patches and still got:
> 
> [btrfs-progs]# truncate --size=2m testfile
> [btrfs-progs]# ./mkfs.btrfs testfile
> 
> WARNING! - Btrfs v0.20-rc1-360-geeeb4e9 IS EXPERIMENTAL
> WARNING! - see http://btrfs.wiki.kernel.org before using
> 
> SMALL VOLUME: forcing mixed metadata/data groups
> mkfs.btrfs: volumes.c:857: btrfs_alloc_chunk: Assertion `!(ret)' failed.
> Aborted (core dumped)
> 
> which was at:
> 
>                 ret = btrfs_alloc_dev_extent(trans, device,
>                              info->chunk_root->root_key.objectid,
>                              BTRFS_FIRST_CHUNK_TREE_OBJECTID, key.offset,
>                              calc_size, &dev_offset);
>                 BUG_ON(ret);
> 
> :(

Wow...
It seems that this abort is different problem from the bug which
my patches are going to fix.  I'll try to make new patch to fix this
problem.

> 
> Also, I'm curious - I know the code existed before your patch 2/2,
> but I don't understand why it reserves 1MB for the first superblock 
> when the first superblock is actually at 64k.  Any idea?
> 
> -Eric

I'm not sure... According to the git-log, this 1M trick is in 
the following old commit by Chris:

  commit a6de0bd778475504f42a142c83b8077993cbddfe
  Author: Chris Mason <chris.mason@oracle.com>
  Date:   Thu Apr 3 16:35:48 2008 -0400

     Add mirroring support across multiple drives


Thanks,
H.Seto