[PATCH RFC] fs: erofs: support PG_mappedtodisk flag for folios with zero-filled

[Gao Xiang]

On 2024/10/17 16:09, Barry Song wrote:
> On Thu, Oct 17, 2024 at 9:00 PM Gao Xiang <hsiangkao at linux.alibaba.com> wrote:
>>
>>
>>
>> On 2024/10/17 15:58, Gao Xiang wrote:
>>> Hi Barry,
>>>
>>> On 2024/10/17 15:43, Barry Song wrote:
>>>> From: Barry Song <v-songbaohua at oppo.com>
>>>>
>>>> When a folio has never been zero-filled, mark it as mappedtodisk
>>>> to allow other software components to recognize and utilize the
>>>> flag.
>>>>
>>>> Signed-off-by: Barry Song <v-songbaohua at oppo.com>
>>>
>>> Thanks for this!
>>>
>>> It looks good to me as an improvement as long as PG_mappedtodisk
>>> is long-term lived and useful to users.
>>>
>>> Reviewed-by: Gao Xiang <hsiangkao at linux.alibaba.com>
>>
> 
> thanks!
> 
>> BTW, I wonder if iomap supports this since uncompressed EROFS
>> relies on iomap paths...
> 
> In the core layer, I only see fs/buffer.c's block_read_full_folio()
> and fs/mpage.c's mpage_readahead() and mpage_readahead()
> supporting this. I haven't found any code in iomap that sets the
> flag.
> 
> I guess erofs doesn't call the above functions for non-compressed
> files?

mpage are obsoleted interfaces (of course EROFS could use
them instead, see my backport to centos 7 [1]), and iomap
is used for recent unencoded I/O use cases.

It would be better to add support for iomap too, but I guess
PG_mappedtodisk has very few users in the upstream kernel,
so they might ask for further use cases tho ;-)

Thanks,
Gao Xiang

[1] https://github.com/erofs/kmod-erofs/blob/main/src/data.c#L249

> 
>>
>> Thanks,
>> Gao Xiang
> 
> Barry