Re: [PATCH 4/5] btrfs: prepare extent_io.c for future larger folio support

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

在 2025/2/21 22:42, Johannes Thumshirn 写道:
> On 20.02.25 10:24, Qu Wenruo wrote:
>> @@ -2468,8 +2468,8 @@ void extent_write_locked_range(struct inode *inode, const struct folio *locked_f
>>    	ASSERT(IS_ALIGNED(start, sectorsize) && IS_ALIGNED(end + 1, sectorsize));
>>
>>    	while (cur <= end) {
>> -		u64 cur_end = min(round_down(cur, PAGE_SIZE) + PAGE_SIZE - 1, end);
>> -		u32 cur_len = cur_end + 1 - cur;
>> +		u64 cur_end;
>> +		u32 cur_len;
>>    		struct folio *folio;
>>
>>    		folio = filemap_get_folio(mapping, cur >> PAGE_SHIFT);
>> @@ -2479,13 +2479,18 @@ void extent_write_locked_range(struct inode *inode, const struct folio *locked_f
>>    		 * code is just in case, but shouldn't actually be run.
>>    		 */
>>    		if (IS_ERR(folio)) {
>> +			cur_end = min(round_down(cur, PAGE_SIZE) + PAGE_SIZE - 1, end);
>> +			cur_len = cur_end + 1 - cur;
>
> Why is it still using PAGE_SIZE here?

This is because we are at a failure path where there is no folio.

But we still want to skip to the next slot (may or may not exist
though), so we have to increase the bytenr by the minimal unit of
filemap, which is still a page.

Thanks,
Qu

>
>>    			btrfs_mark_ordered_io_finished(BTRFS_I(inode), NULL,
>>    						       cur, cur_len, false);
>>    			mapping_set_error(mapping, PTR_ERR(folio));
>> -			cur = cur_end + 1;
>> +			cur = cur_end;
>>    			continue;
>>    		}
>>
>> +		cur_end = min(folio_pos(folio) + folio_size(folio) - 1, end);
>> +		cur_len = cur_end + 1 - cur;
>> +
>>    		ASSERT(folio_test_locked(folio));
>>    		if (pages_dirty && folio != locked_folio)
>>    			ASSERT(folio_test_dirty(folio));