Re: PROBLEM: repeatable lockup on RAID-6 with LUKS dm-crypt on NVMe devices when rsyncing many files

[Yu Kuai <yukuai1@huaweicloud.com>]

Hi,

在 2024/11/09 19:35, Xiao Ni 写道:
> Maybe it's not good to put the bitmap operation from raid5 to md which
> the new api is only used for raid5. And the bitmap region which raid5
> needs to handle is based on the member disk. It should be calculated
> rather than the bio address space. Because the bio address space is
> for the whole array.

Yes, this is just a patch to test if raid5 bitmap can be balanced,
not a formal version. raid1/raid10 can move to md.c much easier, and I'm
still trying to figure out why raid5 is specail and doesn't use the
array to indicate the bitmap. And if it's possible to use the array,
I'll prefer this way so that bitmap operations can be consistent for all
levels.

> 
> We have a customer who reports a similar problem. There is a patch
> from David. I put it in the attachment.

This patch looks good general and better for backport.

Thanks,
Kuai

> 
> @Christian, can you have a try with the patch? It can be applied
> cleanly on 6.11-rc6