Re: block allocator issue with ext4+DAX

From: Jan Kara
Date: Thu Mar 31 2016 - 04:59:09 EST

On Wed 30-03-16 16:01:29, Ross Zwisler wrote:
> I've hit an issue in my testing which I believe to be related to the ext4
> block allocator when using the DAX mount option. I originally found this
> issue with the generic/102 xfstest, but have reduced it to the minimal
> reproducer at the bottom of this email. I've been able to reproduce this with
> both BRD and with PMEM as the underlying block device.
> For this test we're running in a very small filesystem, only 512 MiB. We
> fallocate() 400 MiB of that space, unlink the file, then try and rewrite that
> 400 MiB file one chunk at a time.
> What actually happens is that during the rewrite we run out of memory and the
> DAX call to get_block() in dax_io() fails with -ENOSPC.

Yes, I have already sent a fix for this bug here:

Ted, can you please pick it up? Thanks!

Jan Kara <jack@xxxxxxxx>