On 2019/7/29 下午9:52, Swâmi Petaramesh wrote:
> On 7/29/19 3:47 PM, Qu Wenruo wrote:
>> Although there are some bug fixes queued for stable, it doesn't look
>> like related to such CoW breakage.
>>
>> Thus we need to rule out lower layer bugs to make sure it's btrfs
>> causing the problem.
>
> Please tell me how I could help.
Full history of the backup device please.
Including the mount/usage before and after 5.2 kernel.
>
> This machine was extremely stable (for years) before upgrading from
> kernel 5.1 to 5.2 so unless the hardware is failing, I can hardly
> imagine what else could be the problem...
You know, LUKS/LVM all uses device mapper, which adds an extra layer for
the storage stack, and it may affects how FLUSH/FUA is handled, and
break the fragile CoW used in btrfs.
And device mapper code is also upgraded with kernel.
(Although I don't believe that's the case, but we still need to wipe out
all possibilities)
Despite that, testing btrfs without LUKS/LVM on the same backup disk
(after you have restored needed data) would help us to determine if it's
the disk to blame.
(It's possible the disk itself doesn't handle FUA/FLUSH correctly thus
it's just a problem of time to hit such problem)
Thanks,
Qu
>
> Both FSes are BTRFS over LUKS (one using an LVM, the other not).
>
> Kind regards.
>
