On 3/13/25 03:54, Li Wang wrote:
On Wed, Mar 12, 2025 at 9:59 PM Christoph Hellwig <hch@xxxxxx
<mailto:hch@xxxxxx>> wrote:
On Wed, Mar 12, 2025 at 05:19:36PM +0800, Li Wang wrote:
> Well, does that patch for ioctl_loop06 still make sense?
> Or any other workaround?
> https://lists.linux.it/pipermail/ltp/2025-March/042599.html
<https://lists.linux.it/pipermail/ltp/2025-March/042599.html>
The real question is what block sizes we want to support for the
loop driver. Because if it is larger than the physical block size
it can lead to torn writes. But I guess no one cared about those
on loop so far, so why care about this now..
That's because the kernel test-robot reports a LTP/ioctl_loop06 test
fail in kernel commit:
47dd67532303803 ("block/bdev: lift block size restrictions to 64k")
The ioctl_loop06 is a boundary testing and always fail with
LOOP_SET_BLOCK_SIZE set a value larger than PAGE_SIZE.
But now it's set successfully unexpectedly.
If you all believe the boundary test for loopback driver is redundant,
I can help remove that from LTP code.
I would remove it.
Yes, we might incur torn writes, but previously we hadn't cared about
that. And if we cared we should have a dedicated test for that; there's
no guarantee that we cannot have torn writes even with 4k blocks.
Cheers,
Hannes
--
Dr. Hannes Reinecke Kernel Storage Architect
hare@xxxxxxx +49 911 74053 688
SUSE Software Solutions GmbH, Frankenstr. 146, 90461 Nürnberg
HRB 36809 (AG Nürnberg), GF: I. Totev, A. McDonald, W. Knoblich
