Re: [f2fs-dev] [PATCH] f2fs: avoid f2fs_map_blocks() for consecutive holes in readpages

[Chao Yu <chao@kernel.org>]

On 1/15/2026 8:48 PM, Nanzhe Zhao wrote:
> Hi Chao:
> 
> On 2026/1/12 09:33, Chao Yu via Linux-f2fs-devel wrote:
>> For consecutive large hole mapping across {d,id,did}nodes , we don't
>> need to call f2fs_map_blocks() to check one hole block per one time,
>> instead, we can use map.m_next_pgofs as a hint of next potential valid
>> block, so that we can skip calling f2fs_map_blocks the range of
>> [cur_pgofs + 1, .m_next_pgofs).
>>
>> 1) regular case
>>
>> touch /mnt/f2fs/file
>> truncate -s $((1024*1024*1024)) /mnt/f2fs/file
>> time dd if=/mnt/f2fs/file of=/dev/null bs=1M count=1024
>>
>> Before:
>> real    0m0.706s
>> user    0m0.000s
>> sys     0m0.706s
>>
>> After:
>> real    0m0.620s
>> user    0m0.008s
>> sys     0m0.611s
>>
>> 2) large folio case
>>
>> touch /mnt/f2fs/file
>> truncate -s $((1024*1024*1024)) /mnt/f2fs/file
>> f2fs_io setflags immutable /mnt/f2fs/file
>> sync
>> echo 3 > /proc/sys/vm/drop_caches
>> time dd if=/mnt/f2fs/file of=/dev/null bs=1M count=1024
>>
>> Before:
>> real    0m0.438s
>> user    0m0.004s
>> sys     0m0.433s
>>
>> After:
>> real    0m0.368s
>> user    0m0.004s
>> sys     0m0.364s
>>
>> Signed-off-by: Chao Yu <chao@kernel.org>
>> ---
>>    fs/f2fs/data.c | 21 +++++++++++++++------
>>    1 file changed, 15 insertions(+), 6 deletions(-)
>>
>> diff --git a/fs/f2fs/data.c b/fs/f2fs/data.c
>> index a2c4769d0ae1..5b0642cd27ff 100644
>> --- a/fs/f2fs/data.c
>> +++ b/fs/f2fs/data.c
>> @@ -2176,10 +2176,13 @@ static int f2fs_read_single_page(struct inode *inode, struct folio *folio,
>>    	/*
>>    	 * Map blocks using the previous result first.
>>    	 */
>> -	if ((map->m_flags & F2FS_MAP_MAPPED) &&
>> -			block_in_file > map->m_lblk &&
>> +	if (map->m_flags & F2FS_MAP_MAPPED) {
>> +		if (block_in_file > map->m_lblk &&
>>    			block_in_file < (map->m_lblk + map->m_len))
>> +			goto got_it;
>> +	} else if (block_in_file < *map->m_next_pgofs) {
>>    		goto got_it;
>> +	}
>>    
>>    	/*
>>    	 * Then do more f2fs_map_blocks() calls until we are
>> @@ -2454,7 +2457,7 @@ static int f2fs_read_data_large_folio(struct inode *inode,
>>    	struct bio *bio = NULL;
>>    	sector_t last_block_in_bio = 0;
>>    	struct f2fs_map_blocks map = {0, };
>> -	pgoff_t index, offset;
>> +	pgoff_t index, offset, next_pgofs = 0;
>>    	unsigned max_nr_pages = rac ? readahead_count(rac) :
>>    				folio_nr_pages(folio);
>>    	unsigned nrpages;
>> @@ -2487,16 +2490,21 @@ static int f2fs_read_data_large_folio(struct inode *inode,
>>    		/*
>>    		 * Map blocks using the previous result first.
>>    		 */
>> -		if ((map.m_flags & F2FS_MAP_MAPPED) &&
>> -				index > map.m_lblk &&
>> +		if (map.m_flags & F2FS_MAP_MAPPED) {
>> +			if (index > map.m_lblk &&
>>    				index < (map.m_lblk + map.m_len))
>> +				goto got_it;
>> +		} else if (index < next_pgofs) {
>> +			/* hole case */
>>    			goto got_it;
>> +		}
>>    
>>    		/*
>>    		 * Then do more f2fs_map_blocks() calls until we are
>>    		 * done with this page.
>>    		 */
>>    		memset(&map, 0, sizeof(map));
>> +		map.m_next_pgofs = &next_pgofs;
>>    		map.m_seg_type = NO_CHECK_TYPE;
>>    		map.m_lblk = index;
>>    		map.m_len = max_nr_pages;
>> @@ -2617,6 +2625,7 @@ static int f2fs_mpage_readpages(struct inode *inode,
>>    	pgoff_t nc_cluster_idx = NULL_CLUSTER;
>>    	pgoff_t index;
>>    #endif
>> +	pgoff_t next_pgofs = 0;
>>    	unsigned nr_pages = rac ? readahead_count(rac) : 1;
>>    	struct address_space *mapping = rac ? rac->mapping : folio->mapping;
>>    	unsigned max_nr_pages = nr_pages;
>> @@ -2637,7 +2646,7 @@ static int f2fs_mpage_readpages(struct inode *inode,
>>    	map.m_lblk = 0;
>>    	map.m_len = 0;
>>    	map.m_flags = 0;
>> -	map.m_next_pgofs = NULL;
>> +	map.m_next_pgofs = &next_pgofs;
>>    	map.m_next_extent = NULL;
>>    	map.m_seg_type = NO_CHECK_TYPE;
>>    	map.m_may_create = false;
> 
> Do we have plans to also support reducing f2fs_map_blocks() calls for
> consectives holes in logical file postion with dnode have already been
> allocated in buffered large folio read?
> Such as consective NULL_ADDR or NEW_ADDR?

Nanzhe,

We have supported that for large folio read w/ this patch?

Thanks,

> 
> Thanks,
>