Re: [PATCH v2 2/2] btrfs: automatically remove the subvolume qgroup

[Qu Wenruo <quwenruo.btrfs@gmx.com>]

在 2024/5/3 00:30, David Sterba 写道:
> On Wed, May 01, 2024 at 07:35:09AM +0930, Qu Wenruo wrote:
>>
>>
>> 在 2024/4/30 20:29, David Sterba 写道:
>>> On Tue, Apr 30, 2024 at 07:35:11AM +0930, Qu Wenruo wrote:
>>>>
>>>>
>>>> 在 2024/4/30 02:01, David Sterba 写道:
>>>>> On Mon, Apr 29, 2024 at 06:13:33AM -0700, Boris Burkov wrote:
>>>>>> I support the auto deletion in the kernel as you propose, I think it
>>>>>> just makes sense. Who wants stale, empty qgroups around that aren't
>>>>>> attached to any subvol? I suppose that with the drop_thresh thing, it is
>>>>>> possible some parent qgroup still reflects the usage until the next full
>>>>>> scan?
>>>>>
>>>>> The stale qgroups have been out for a long time so removing them after
>>>>> subvolume deletion is changing default behaviour, this always breaks
>>>>> somebody's scripts or tools.
>>>>
>>>> If needed I can introduce a compat bit (the first one), to tell the
>>>> behavior difference.
>>>>
>>>> And if we go the compat bit way, I believe it can be the first example
>>>> on how to do a kernel behavior change correctly without breaking any
>>>> user space assumption.
>>>
>>> I don't see how a compat bit would work here, we use them for feature
>>> compatibility and for general access to data (full or read-only). What
>>> we do with individual behavioral changes are sysfs files. They're
>>> detectable by scripts and can be also used for configuration. In this
>>> case enabling/disabling autoclean of the qgroups.
>>
>> I mean the compat bit, which is fully empty now.
>
> Which compat bit and in which structure? What I mean is the super block
> incompat/compat bits but what you described below does not match it.

Superblock compat bit exactly.

I didn't see why my following description doesn't match it.

We have 3 compat bits:

- incompat
   Huge on-disk format change that older kernel can not understand and
   affects older kernels' ability to read files.
   This is the most abused bits from the past.
   A lot of features like skinny metadata should go compat_ro.

- compat_ro
   On-disk format changes, but older kernels should still be able to read
   the fs/csum tree etc.

- compat
   Not utilized for now. Older kernel can still do everything without any
   problem.
   Thus this should be the best candidate for runtime behavior changes
   that doesn't affect on-disk format at all.

Thanks,
Qu
>
>> The new bit would be something like BTRFS_QGROUP_AUTO_REMOVAL, with that
>> set, btrfs would auto remove any stale qgroups (for regular qgroups though).
>>
>> Without that, it would be as usual (no auto removal).