Re: [PATCH v2 5/6] iomap: don't mark blocks uptodate after partial zeroing

[Zhang Yi <yi.zhang@huaweicloud.com>]

On 2024/8/13 0:49, Darrick J. Wong wrote:
> On Mon, Aug 12, 2024 at 08:11:58PM +0800, Zhang Yi wrote:
>> From: Zhang Yi <yi.zhang@huawei.com>
>>
>> In __iomap_write_begin(), if we unaligned buffered write data to a hole
>> of a regular file, we only zero out the place where aligned to block
>> size that we don't want to write, but mark the whole range uptodate if
>> block size < folio size. This is wrong since the not zeroed part will
>> contains stale data and can be accessed by a concurrent buffered read
>> easily (on the filesystem may not hold inode->i_rwsem) once we mark the
>> range uptodate. Fix this by drop iomap_set_range_uptodate() in the
>> zeroing out branch.
>>
>> Fixes: 9dc55f1389f9 ("iomap: add support for sub-pagesize buffered I/O without buffer heads")
>> Reported-by: Matthew Wilcox <willy@infradead.org>
>> Closes: https://lore.kernel.org/all/ZqsN5ouQTEc1KAzV@casper.infradead.org/
>> Signed-off-by: Zhang Yi <yi.zhang@huawei.com>
>> ---
>>  fs/iomap/buffered-io.c | 2 +-
>>  1 file changed, 1 insertion(+), 1 deletion(-)
>>
>> diff --git a/fs/iomap/buffered-io.c b/fs/iomap/buffered-io.c
>> index ac762de9a27f..96600405dbb5 100644
>> --- a/fs/iomap/buffered-io.c
>> +++ b/fs/iomap/buffered-io.c
>> @@ -744,8 +744,8 @@ static int __iomap_write_begin(const struct iomap_iter *iter, loff_t pos,
>>  					poff, plen, srcmap);
>>  			if (status)
>>  				return status;
>> +			iomap_set_range_uptodate(folio, poff, plen);
>>  		}
>> -		iomap_set_range_uptodate(folio, poff, plen);
> 
> Don't we need to iomap_set_range_uptodate for the bytes that we zeroed
> with folio_zero_segments?
> 

We must do partial block zeroing here, hence we don't need to set update
bit.

Thanks,
Yi.

> --D
> 
>>  	} while ((block_start += plen) < block_end);
>>  
>>  	return 0;
>> -- 
>> 2.39.2
>>
>>