Re: [PATCH 3/3] btrfs: use timespec64 for i_otime

[Nikolay Borisov <nborisov@suse.com>]

On 20.06.2018 22:41, Arnd Bergmann wrote:
> On Wed, Jun 20, 2018 at 9:34 PM, Nikolay Borisov <nborisov@suse.com> wrote:
>>
>>
>> On 20.06.2018 19:38, David Sterba wrote:
>>> On Wed, Jun 20, 2018 at 04:34:34PM +0200, Arnd Bergmann wrote:
>>>> While the regular inode timestamps all use timespec64 now, the
>>>> i_otime field is btrfs specific and still needs to be converted
>>>> to correctly represent times beyond 2038.
>>>>
>>>> Signed-off-by: Arnd Bergmann <arnd@arndb.de>
>>>
>>> This patch addresses the remaining type conversions, so I'm going to
>>> merge it, thanks.
>>>
>>
>> Actually for the sake of consistency we might want to merge this series
>> altogether. As it stands we now use ktime_get_seconds and
>> ktime_get_real_seconds (from Allen's patch). I haven't dug to see what's
>> the difference (if any) between the two .
> 
> I just checked again and see that Allen's patch addresses the first two
> of my three patches, but he picked a different approach for
> transaction_kthread(): My patch moved to CLOCK_MONOTONIC,
> while his version only changed the to time64_t but kept the
> CLOCK_REALTIME behavior. It's a small difference, but I think
> my version is slightly better. My patch 2/3 is identical to his version.

I agree, in the transaction_kthread we are only interested in knowing
whether a fixed time windows (commit_internval) has passed. So monotonic
makes more sense here.
> 
> If you like, I can also rebase my patch 1/3 on top of his patch and
> change it to CLOCK_MONOTONIC.

Please do.

> 
>          Arnd
>