Re: [PATCH] ext4: add regression tests for ^extents punch hole

[Omar Sandoval <osandov@osandov.com>]

On Tue, Feb 24, 2015 at 07:24:37PM -0500, Theodore Ts'o wrote:
> I suspect our current generic fsstress and fsx tests would catch this
> already, and what I need to do is to make sure I add ext3/1k to my
> test configurations (currently I test an ext3 configuration, and a 1k
> block configuration, but I don't currently test ext3/1k together).
> That would probably round out my full test iteration to a very
> pleasing 24 hours or so, which is fine, although I wouldn't want it to
> take much longer than that.
> 
Actually, this issue isn't specific to 1k blocks, so I wouldn't count on
it :) That block size in my tests is an artifact of the initial
reproducer Lukas had (generic/270 on ext3 with 1k blocks). It just so
happens that the start/len arguments fsstress chooses for punch hole
operations in that setup straddles the indirect tree levels in the wrong
ways.


On Tue, Feb 24, 2015 at 03:58:06PM +0100, Lukáš Czerner wrote:
> Omar, can you make the test generic and can this be reproduced on 4k
> block size ? If not, can you make a generic reproducer which does
> not depend on the block size ?
> 
> Also if we want to include the special case for ext4, we need to
> have a function which allows us to alter the mkfs options without
> completely overriding the user specified options. I think that there
> is something like that for xfs, Omar can you do that for ext4 as
> well ?
> 
> Thanks!
> -Lukas
> 
> 
No problem, I'll have a go at it.


On Wed, Feb 25, 2015 at 09:07:19AM +1100, Dave Chinner wrote:
> It's built into the _scratch_mkfs_xfs wrapper, where if the test
> supplies extra options and that conflicts with the CLI supplied
> options it drops the CLI specific options and just uses the test
> options.
> 
> I've mentioned this specificly in the past, too. i.e. that all
> _scratch_mkfs_$FSTYP wrappers should be handling CLI vs test
> specific options like this.... :/
> 
> Cheers,
> 
> Dave.
> 
I'll take that into account.


Thanks, everyone!
-- 
Omar