Re: [f2fs-dev] [PATCH RESEND 1/5] f2fs: fix extent merge failure when largest extent is not in rb-tree

On 6/15/26 7:26 PM, Chao Yu via Linux-f2fs-devel wrote:
> On 6/12/26 19:58, Yongpeng Yang wrote:
>> From: Yongpeng Yang <[email protected]>
>>
>> The following scenario can cause fiemap to report incorrect extents:
>>
>> $ mkfs.f2fs /dev/vdb -f
>> $ mount -o mode=lfs /dev/vdb /mnt/f2fs/
>> $ dd if=/dev/urandom of=data bs=4K count=874 conv=notrunc
>> $ f2fs_io fiemap 0 1000000 data 1
>> $ shrink all extent
>> $ dd if=/dev/urandom of=data bs=4K count=150 seek=874 conv=notrunc
>> $ f2fs_io fiemap 0 1000000 data 1
>> Fiemap: offset = 0 len = 1000000
>>          logical addr.    physical addr.   length           flags
>> 0       0000000000000000 00000002868d4000 000000000036a000 00001000
>> 1       000000000036a000 0000000286c3e000 0000000000096000 00001001
>>
>> The root cause is that when the largest extent is not in the extent
>> tree, mergeable extents are not merged, causing f2fs_map_blocks to
>> misjudge and output an incorrect extent list.
>>
>> Fix this by allowing the extent being inserted to merge with the largest
>> extent. When updating the extent tree range, if the new extent can be
>> front-merged or back-merged with the largest extent and the largest
>> extent is not in the rb-tree, merge them before the normal lookup.
>>
>> Fixes: 429511cdf8b3 ("f2fs: add core functions for rb-tree extent cache")
>> Signed-off-by: Yongpeng Yang <[email protected]>
>> ---
>>   fs/f2fs/extent_cache.c | 21 +++++++++++++++++++++
>>   1 file changed, 21 insertions(+)
>>
>> diff --git a/fs/f2fs/extent_cache.c b/fs/f2fs/extent_cache.c
>> index 61f6b9714366..aa368a01b035 100644
>> --- a/fs/f2fs/extent_cache.c
>> +++ b/fs/f2fs/extent_cache.c
>> @@ -702,6 +702,27 @@ static void __update_extent_tree_range(struct inode *inode,
>>   		__drop_largest_extent(et, fofs, len);
>>   	}
>>   
> 
> This can only be used for EX_READ?

Yes, we only handle the case where the newly inserted mapping of EX_READ
can be merged with the largest extent.

> 
>> +	if (et->largest.len != 0 &&
>> +			(__is_front_mergeable(tei, &et->largest, type) ||
>> +			 __is_back_mergeable(tei, &et->largest, type))) {
>> +		/* 0. try to merge with largest extent. */
>> +		en = __lookup_extent_node_ret(&et->root,
>> +				et->cached_en, et->largest.fofs,
>> +				&prev_en, &next_en,
>> +				&insert_p, &insert_parent,
>> +				&leftmost);
>> +		if (!en) {
>> +			if (__is_back_mergeable(tei, &et->largest, type)) {
> 
> Can we reuse result of previous __is_back_mergeable()?

OK, I'll change it in v2 patch.

Thanks
Yongpeng,

> 
> Thanks,
> 
>> +				tei->fofs = et->largest.fofs;
>> +				tei->blk = et->largest.blk;
>> +				fofs = tei->fofs;
>> +			}
>> +			tei->len += et->largest.len;
>> +			len = tei->len;
>> +			end = fofs + len;
>> +		}
>> +	}
>> +
>>   	/* 1. lookup first extent node in range [fofs, fofs + len - 1] */
>>   	en = __lookup_extent_node_ret(&et->root,
>>   					et->cached_en, fofs,
> 
> 
> 
> _______________________________________________
> Linux-f2fs-devel mailing list
> [email protected]
> https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel



_______________________________________________
Linux-f2fs-devel mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel