Re: [PATCH v2 9/9] btrfs: enable experimental bs > ps support

[Qu Wenruo <wqu@suse.com>]

在 2025/9/22 20:12, David Sterba 写道:
> On Mon, Sep 22, 2025 at 07:57:40PM +0930, Qu Wenruo wrote:
>>
>>
>> 在 2025/9/22 19:51, David Sterba 写道:
>>> On Mon, Sep 22, 2025 at 08:10:51AM +0930, Qu Wenruo wrote:
>>>> +	if (fs_info->sectorsize > PAGE_SIZE) {
>>>> +		ret = -ENOTTY;
>>>
>>> I did a minor fixup of the error code to -EOPNOTSUPP,
>>
>> Please don't.
>>
>> The error code is to co-operate with btrfs-progs, which doesn't check
>> EOPNOTSUPP.
> 
> Well then we should fix btrfs-progs because this breaks the informal
> error code protocol

It's the egg-and-chicken compatibility problem.

> 
>> And the encoded receive only falls back to regular one when got a ENOTTY.
> 
> But this is still only for the bs > ps case, right? So it may be
> temporarily broken until we sync the kernel code and progs but would not
> otherwise break the other cases.

I'd prefer to remove those exceptions when the sub-block handling is 
added, and keep the compatibility so that I don't need to run tests with 
a special version of btrfs-progs.

The fix of btrfs-progs will take time to reach end users.

And that full bs > ps support will not be that far away, considering how 
many cycles (in fact, just two cycles, some preparation in v6.17, then 
v6.18) we spent to reach the current state (full fstests run, 6 known 
failures, all in btrfs group), the removal of those checks should be 
pretty close.

Especially I already have an idea on how to support sub-block checksum 
handling.

Thanks,
Qu