On Fri, Oct 19, 2012 at 02:37:30PM -0300, Carlos Maiolino wrote:
> reviewing a patch to xfstests250, I ran it against newer kernels (3.6.0+ and
> 3.7.0-rc1) and noticed it is failing. i.e. btree is getting corrupted.
It's been failing on mainline kernels for a long time. In fact, i
think it's been failing since it was created. But it's not failing
due to btree corruption - it's failing because mkfs is not leaving
enough space in the AG that contains the log for sanity checks to
pass. i.e. that there are always a minimum of 4 blocks of freespace
in an AG.
This is not actually a problem - the log takes the entire AG, so
allocation will never occur in it, so having less than 4 blocks of
free space in the AG is just noise in this case. It's never bubbled
to the top of my list to fix...
> I'm going to take a look at it, but let me know if anybody has already
> found/fixed it.
If it is failing the check_scratch_fs stage, then it is most likely
the above issue. The corrupted btree problem that the test was
writen for caused the system to ASSERT fail or crash - i.e. it
didn't even run to the point of checking the fs....