Re: [PATCH 1/4] f2fs: support F2FS_IOC_PRECACHE_EXTENTS

[Jaegeuk Kim <jaegeuk@kernel.org>]

On 01/08, Chao Yu wrote:
> This patch introduces a new ioctl F2FS_IOC_PRECACHE_EXTENTS to precache
> extent info, in order to gain better performance during data/meta
> accessing.
> 
> Signed-off-by: Chao Yu <yuchao0@huawei.com>
> ---
>  fs/f2fs/data.c | 27 +++++++++++++++++++++++++++
>  fs/f2fs/f2fs.h |  2 ++
>  fs/f2fs/file.c | 35 +++++++++++++++++++++++++++++++++++
>  3 files changed, 64 insertions(+)
> 
> diff --git a/fs/f2fs/data.c b/fs/f2fs/data.c
> index 429764118604..f38e660bc889 100644
> --- a/fs/f2fs/data.c
> +++ b/fs/f2fs/data.c
> @@ -920,6 +920,7 @@ int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map,
>  	blkcnt_t prealloc;
>  	struct extent_info ei = {0,0,0};
>  	block_t blkaddr;
> +	unsigned int start_pgofs;
>  
>  	if (!maxblocks)
>  		return 0;
> @@ -935,6 +936,8 @@ int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map,
>  		map->m_pblk = ei.blk + pgofs - ei.fofs;
>  		map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgofs);
>  		map->m_flags = F2FS_MAP_MAPPED;
> +		if (map->m_next_pgofs)
> +			*map->m_next_pgofs = pgofs + map->m_len;
>  		goto out;
>  	}
>  
> @@ -957,6 +960,7 @@ int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map,
>  		goto unlock_out;
>  	}
>  
> +	start_pgofs = pgofs;
>  	prealloc = 0;
>  	last_ofs_in_node = ofs_in_node = dn.ofs_in_node;
>  	end_offset = ADDRS_PER_PAGE(dn.node_page, inode);
> @@ -990,6 +994,8 @@ int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map,
>  				map->m_pblk = 0;
>  				goto sync_out;
>  			}
> +			if (flag == F2FS_GET_BLOCK_PRECACHE)
> +				goto sync_out;
>  			if (flag == F2FS_GET_BLOCK_FIEMAP &&
>  						blkaddr == NULL_ADDR) {
>  				if (map->m_next_pgofs)
> @@ -1048,6 +1054,16 @@ int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map,
>  	else if (dn.ofs_in_node < end_offset)
>  		goto next_block;
>  
> +	if (flag == F2FS_GET_BLOCK_PRECACHE) {
> +		if (map->m_flags & F2FS_MAP_MAPPED) {
> +			unsigned int ofs = start_pgofs - map->m_lblk;
> +
> +			f2fs_update_extent_cache_range(&dn,
> +				start_pgofs, map->m_pblk + ofs,
> +				map->m_len - ofs);
> +		}
> +	}
> +
>  	f2fs_put_dnode(&dn);
>  
>  	if (create) {
> @@ -1057,6 +1073,17 @@ int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map,
>  	goto next_dnode;
>  
>  sync_out:
> +	if (flag == F2FS_GET_BLOCK_PRECACHE) {
> +		if (map->m_flags & F2FS_MAP_MAPPED) {
> +			unsigned int ofs = start_pgofs - map->m_lblk;
> +
> +			f2fs_update_extent_cache_range(&dn,
> +				start_pgofs, map->m_pblk + ofs,
> +				map->m_len - ofs);
> +		}
> +		if (map->m_next_pgofs)
> +			*map->m_next_pgofs = pgofs + 1;
> +	}
>  	f2fs_put_dnode(&dn);
>  unlock_out:
>  	if (create) {
> diff --git a/fs/f2fs/f2fs.h b/fs/f2fs/f2fs.h
> index 2f097139b8f7..14177cebc3e7 100644
> --- a/fs/f2fs/f2fs.h
> +++ b/fs/f2fs/f2fs.h
> @@ -353,6 +353,7 @@ static inline bool __has_cursum_space(struct f2fs_journal *journal,
>  						struct f2fs_gc_range)
>  #define F2FS_IOC_GET_FEATURES		_IOR(F2FS_IOCTL_MAGIC, 12, __u32)
>  #define F2FS_IOC_SET_PIN_FILE		_IOW(F2FS_IOCTL_MAGIC, 13, __u32)
> +#define F2FS_IOC_PRECACHE_EXTENTS	_IO(F2FS_IOCTL_MAGIC, 14)
>  
>  #define F2FS_IOC_SET_ENCRYPTION_POLICY	FS_IOC_SET_ENCRYPTION_POLICY
>  #define F2FS_IOC_GET_ENCRYPTION_POLICY	FS_IOC_GET_ENCRYPTION_POLICY
> @@ -558,6 +559,7 @@ enum {
>  	F2FS_GET_BLOCK_BMAP,
>  	F2FS_GET_BLOCK_PRE_DIO,
>  	F2FS_GET_BLOCK_PRE_AIO,
> +	F2FS_GET_BLOCK_PRECACHE,
>  };
>  
>  /*
> diff --git a/fs/f2fs/file.c b/fs/f2fs/file.c
> index e4078b32856c..14e75871acf3 100644
> --- a/fs/f2fs/file.c
> +++ b/fs/f2fs/file.c
> @@ -2734,6 +2734,38 @@ static int f2fs_ioc_set_pin_file(struct file *filp, unsigned long arg)
>  	return ret;
>  }
>  
> +static int f2fs_ioc_precache_extents(struct file *filp, unsigned long arg)
> +{
> +	struct inode *inode = file_inode(filp);
> +	struct f2fs_inode_info *fi = F2FS_I(inode);
> +	struct f2fs_map_blocks map;
> +	pgoff_t next_pgofs;
> +	loff_t end;
> +	int err;
> +
> +	if (is_inode_flag_set(inode, FI_NO_EXTENT))
> +		return -EOPNOTSUPP;
> +
> +	map.m_lblk = 0;
> +	map.m_next_pgofs = &next_pgofs;
> +	map.m_seg_type = NO_CHECK_TYPE;
> +	end = F2FS_I_SB(inode)->max_file_blocks;
> +
> +	while (map.m_lblk < end) {
> +		map.m_len = end - map.m_lblk;
> +
> +		down_write(&fi->dio_rwsem[WRITE]);
> +		err = f2fs_map_blocks(inode, &map, 0, F2FS_GET_BLOCK_PRECACHE);
> +		up_write(&fi->dio_rwsem[WRITE]);
> +		if (err)
> +			return err;
> +
> +		map.m_lblk = next_pgofs;
> +	}
> +
> +	return 0;
> +}
> +
>  long f2fs_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
>  {
>  	if (unlikely(f2fs_cp_error(F2FS_I_SB(file_inode(filp)))))
> @@ -2786,6 +2818,8 @@ long f2fs_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
>  		return f2fs_ioc_fssetxattr(filp, arg);
>  	case F2FS_IOC_SET_PIN_FILE:
>  		return f2fs_ioc_set_pin_file(filp, arg);
> +	case F2FS_IOC_PRECACHE_EXTENTS:

It looks like F2FS_IOC_CACHE_EXTENTS, and would be better to have an offset
range to look up. BTW, if this is to prepare coming data reads, how about doing
this through fadvise(POSIX_FADV_WILLNEED) instead of additional ioctl?
It seems we can cache extents if the page is marked as PageReadahead().

Thanks,

> +		return f2fs_ioc_precache_extents(filp, arg);
>  	default:
>  		return -ENOTTY;
>  	}
> @@ -2862,6 +2896,7 @@ long f2fs_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
>  	case F2FS_IOC_FSGETXATTR:
>  	case F2FS_IOC_FSSETXATTR:
>  	case F2FS_IOC_SET_PIN_FILE:
> +	case F2FS_IOC_PRECACHE_EXTENTS:
>  		break;
>  	default:
>  		return -ENOIOCTLCMD;
> -- 
> 2.15.0.55.gc2ece9dc4de6