Re: [PATCH 3/9] f2fs: rework write preallocations

[Chao Yu <chao@kernel.org>]

On 2021/7/27 23:33, Darrick J. Wong wrote:
> On Tue, Jul 27, 2021 at 04:30:16PM +0800, Chao Yu wrote:
>> On 2021/7/27 15:38, Eric Biggers wrote:
>>> That's somewhat helpful, but I've been doing some more investigation and now I'm
>>> even more confused.  How can f2fs support non-overwrite DIO writes at all
>>> (meaning DIO writes in LFS mode as well as DIO writes to holes in non-LFS mode),
>>> given that it has no support for unwritten extents?  AFAICS, as-is users can
>>
>> I'm trying to pick up DAX support patch created by Qiuyang from huawei, and it
>> looks it faces the same issue, so it tries to fix this by calling sb_issue_zeroout()
>> in f2fs_map_blocks() before it returns.
> 
> I really hope you don't, because zeroing the region before memcpy'ing it
> is absurd.  I don't know if f2fs can do that (xfs can't really) without
> pinning resources during a potentially lengthy memcpy operation, but you
> /could/ allocate the space in ->iomap_begin, attach some record of that
> to iomap->private, and only commit the mapping update in ->iomap_end.

Thanks for the suggestion, let me check this a little bit later, since now I
just try to stabilize the codes...

Thanks,

> 
> --D
> 
>>> easily leak uninitialized disk contents on f2fs by issuing a DIO write that
>>> won't complete fully (or might not complete fully), then reading back the blocks
>>> that got allocated but not written to.
>>>
>>> I think that f2fs will have to take the ext2 approach of not allowing
>>> non-overwrite DIO writes at all...
>> Yes,
>>
>> Another option is to enhance f2fs metadata's scalability which needs to update layout
>> of dnode block or SSA block, after that we can record the status of unwritten data block
>> there... it's a big change though...
>>
>> Thanks,
>>
>>>
>>> - Eric
>>>