Re: [f2fs-dev] [PATCH v5 1/1] f2fs: dax: implement direct access

[Chao Yu]

Hi Qiuyang,

On 2017/6/27 15:43, sunqiuyang wrote:
> From: Qiuyang Sun <sunqiuyang@xxxxxxxxxx>
> 
> This patch implements Direct Access (DAX) in F2FS, including:
> - a mount option to choose whether to enable DAX or not
> - read/write and mmap of regular files in the DAX way
> - zero-out of unaligned partial blocks in the DAX way
> - garbage collection of DAX files, by mapping both old and new physical
>   addresses of a data page into memory and copy data between them directly
> - incompatibility of DAX with inline data, atomic or volatile write, etc.
> 
> Signed-off-by: Qiuyang Sun <sunqiuyang@xxxxxxxxxx>
> ---
> Changlog v4 -> v5:
> 
> In DAX write, to avoid stale data, the newly allocated blocks should be
> zeroed out before writing to them. We introduce a new flag
> F2FS_GET_BLOCK_ZERO for f2fs_map_blocks(), for which a single struct
> f2fs_map_blocks can never contain old and new blocks at the same time, so
> that zeroing out new blocks would not affect old ones. This flag is used in
> the write path of f2fs_iomap_begin() for DAX I/O, replacing 
> F2FS_GET_BLOCK_PRE_DIO in earlier versions.
> 
> ---
>  fs/f2fs/data.c   | 115 +++++++++++++++++++++++++++++++++
>  fs/f2fs/f2fs.h   |   9 +++
>  fs/f2fs/file.c   | 192 ++++++++++++++++++++++++++++++++++++++++++++++++++++++-
>  fs/f2fs/gc.c     | 104 ++++++++++++++++++++++++++++--
>  fs/f2fs/inline.c |   4 ++
>  fs/f2fs/inode.c  |   8 ++-
>  fs/f2fs/namei.c  |   5 ++
>  fs/f2fs/super.c  |  15 +++++
>  8 files changed, 445 insertions(+), 7 deletions(-)
> 
> diff --git a/fs/f2fs/data.c b/fs/f2fs/data.c
> index 7d3af48..b28d97a 100644
> --- a/fs/f2fs/data.c
> +++ b/fs/f2fs/data.c
> @@ -908,6 +908,15 @@ int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map,
>  				err = -EIO;
>  				goto sync_out;
>  			}
> +			/*
> +			 * If newly allocated blocks are to be zeroed out later,
> +			 * a single f2fs_map_blocks must not contain both old
> +			 * and new blocks at the same time.
> +			 */
> +			if (flag == F2FS_GET_BLOCK_ZERO
> +					&& (map->m_flags & F2FS_MAP_MAPPED)
> +					&& !(map->m_flags & F2FS_MAP_NEW))
> +				goto sync_out;

IIUC, here we do the check for below case during lookuping mapping info:

pgidx		offsset			offset + 1
blkaddr		old mapped PBA		new allocated PBA

Shoudn't we cover following case as well?

pgidx		offsset			offset + 1
blkaddr		new allocated PBA	old mapped PBA

>  			if (flag == F2FS_GET_BLOCK_PRE_AIO) {
>  				if (blkaddr == NULL_ADDR) {
>  					prealloc++;
> @@ -994,6 +1003,12 @@ int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map,
>  	goto next_dnode;
>  
>  sync_out:
> +	if (flag == F2FS_GET_BLOCK_ZERO && map->m_flags & F2FS_MAP_NEW) {
> +		clean_bdev_aliases(inode->i_sb->s_bdev,
> +				map->m_pblk, map->m_len);
> +		sb_issue_zeroout(inode->i_sb, map->m_pblk,
> +				map->m_len, GFP_NOFS);

Need to check return value here.

> +	}
>  	f2fs_put_dnode(&dn);
>  unlock_out:
>  	if (create) {
> @@ -2257,3 +2272,103 @@ int f2fs_migrate_page(struct address_space *mapping,
>  	.migratepage    = f2fs_migrate_page,
>  #endif
>  };
> +
> +#ifdef CONFIG_FS_DAX
> +#include <linux/iomap.h>
> +#include <linux/dax.h>
> +
> +static int f2fs_iomap_begin(struct inode *inode, loff_t offset,
> +	loff_t length, unsigned int flags, struct iomap *iomap)
> +{
> +	struct block_device *bdev;
> +	unsigned long first_block = F2FS_BYTES_TO_BLK(offset);
> +	unsigned long last_block = F2FS_BYTES_TO_BLK(offset + length - 1);
> +	struct f2fs_map_blocks map;
> +	int ret;
> +
> +	if (WARN_ON_ONCE(f2fs_has_inline_data(inode)))
> +		return -ERANGE;
> +
> +	map.m_lblk = first_block;
> +	map.m_len = last_block - first_block + 1;
> +	map.m_next_pgofs = NULL;
> +
> +	if (!(flags & IOMAP_WRITE))
> +		ret = f2fs_map_blocks(inode, &map, 0, F2FS_GET_BLOCK_FIEMAP);
> +	else {
> +	/* i_size should be kept here and changed later in f2fs_iomap_end */
> +		loff_t original_i_size = i_size_read(inode);
> +
> +		ret = f2fs_map_blocks(inode, &map, 1, F2FS_GET_BLOCK_ZERO);
> +		if (i_size_read(inode) > original_i_size) {
> +			f2fs_i_size_write(inode, original_i_size);
> +			if (ret) {
> +				truncate_pagecache(inode, original_i_size);
> +				truncate_blocks(inode, original_i_size, true);

Should cover with i_mmap_sem?

> +			}
> +		}
> +	}
> +
> +	if (ret)
> +		return ret;
> +
> +	iomap->flags = 0;
> +	bdev = inode->i_sb->s_bdev;
> +	iomap->bdev = bdev;
> +	if (blk_queue_dax(bdev->bd_queue))
> +		iomap->dax_dev = dax_get_by_host(bdev->bd_disk->disk_name);
> +	else
> +		iomap->dax_dev = NULL;
> +	iomap->offset = F2FS_BLK_TO_BYTES((u64)first_block);
> +
> +	if (map.m_len == 0) {
> +		iomap->type = IOMAP_HOLE;
> +		iomap->blkno = IOMAP_NULL_BLOCK;
> +		iomap->length = F2FS_BLKSIZE;
> +	} else {
> +		if (map.m_flags & F2FS_MAP_UNWRITTEN)
> +			iomap->type = IOMAP_UNWRITTEN;
> +		else if (map.m_flags & F2FS_MAP_MAPPED)
> +			iomap->type = IOMAP_MAPPED;
> +		else {
> +			WARN_ON_ONCE(1);
> +			return -EIO;
> +		}
> +		iomap->blkno =
> +			(sector_t)map.m_pblk << F2FS_LOG_SECTORS_PER_BLOCK;
> +		iomap->length = F2FS_BLK_TO_BYTES((u64)map.m_len);
> +	}
> +
> +	if (map.m_flags & F2FS_MAP_NEW)
> +		iomap->flags |= IOMAP_F_NEW;
> +	return 0;
> +}
> +
> +static int f2fs_iomap_end(struct inode *inode, loff_t offset, loff_t length,
> +	ssize_t written, unsigned int flags, struct iomap *iomap)
> +{
> +	put_dax(iomap->dax_dev);
> +	if (!(flags & IOMAP_WRITE) || (flags & IOMAP_FAULT))
> +		return 0;
> +
> +	if (offset + written > i_size_read(inode))
> +		f2fs_i_size_write(inode, offset + written);
> +
> +	if (iomap->offset + iomap->length >
> +			ALIGN(i_size_read(inode), F2FS_BLKSIZE)) {
> +		block_t written_blk = F2FS_BYTES_TO_BLK(offset + written);
> +		block_t end_blk = F2FS_BYTES_TO_BLK(offset + length);
> +
> +		if (written_blk < end_blk)
> +			f2fs_write_failed(inode->i_mapping, offset + length);
> +	}
> +
> +	f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
> +	return 0;
> +}
> +
> +struct iomap_ops f2fs_iomap_ops = {
> +	.iomap_begin	= f2fs_iomap_begin,
> +	.iomap_end	= f2fs_iomap_end,
> +};
> +#endif
> diff --git a/fs/f2fs/f2fs.h b/fs/f2fs/f2fs.h
> index db5d7ae..f79d904 100644
> --- a/fs/f2fs/f2fs.h
> +++ b/fs/f2fs/f2fs.h
> @@ -88,6 +88,11 @@ struct f2fs_fault_info {
>  #define F2FS_MOUNT_FAULT_INJECTION	0x00010000
>  #define F2FS_MOUNT_ADAPTIVE		0x00020000
>  #define F2FS_MOUNT_LFS			0x00040000
> +#ifdef CONFIG_FS_DAX
> +#define F2FS_MOUNT_DAX			0x00080000 /* Direct Access */
> +#else
> +#define F2FS_MOUNT_DAX			0
> +#endif
>  
>  #define clear_opt(sbi, option)	((sbi)->mount_opt.opt &= ~F2FS_MOUNT_##option)
>  #define set_opt(sbi, option)	((sbi)->mount_opt.opt |= F2FS_MOUNT_##option)
> @@ -476,6 +481,7 @@ struct f2fs_map_blocks {
>  #define F2FS_GET_BLOCK_BMAP		3
>  #define F2FS_GET_BLOCK_PRE_DIO		4
>  #define F2FS_GET_BLOCK_PRE_AIO		5
> +#define F2FS_GET_BLOCK_ZERO		6
>  
>  /*
>   * i_advise uses FADVISE_XXX_BIT. We can add additional hints later.
> @@ -2402,6 +2408,9 @@ void f2fs_invalidate_page(struct page *page, unsigned int offset,
>  int f2fs_migrate_page(struct address_space *mapping, struct page *newpage,
>  			struct page *page, enum migrate_mode mode);
>  #endif
> +#ifdef CONFIG_FS_DAX
> +extern struct iomap_ops f2fs_iomap_ops;
> +#endif
>  
>  /*
>   * gc.c
> diff --git a/fs/f2fs/file.c b/fs/f2fs/file.c
> index 3f56b27..73aca18 100644
> --- a/fs/f2fs/file.c
> +++ b/fs/f2fs/file.c
> @@ -23,6 +23,10 @@
>  #include <linux/uio.h>
>  #include <linux/uuid.h>
>  #include <linux/file.h>
> +#ifdef CONFIG_FS_DAX
> +#include <linux/dax.h>
> +#include <linux/iomap.h>
> +#endif
>  
>  #include "f2fs.h"
>  #include "node.h"
> @@ -121,6 +125,64 @@ static int f2fs_vm_page_mkwrite(struct vm_fault *vmf)
>  	.page_mkwrite	= f2fs_vm_page_mkwrite,
>  };
>  
> +#ifdef CONFIG_FS_DAX
> +static int f2fs_dax_huge_fault(struct vm_fault *vmf,
> +	enum page_entry_size pe_size)
> +{
> +	int result;
> +	struct inode *inode = file_inode(vmf->vma->vm_file);
> +	struct super_block *sb = inode->i_sb;
> +	bool write = vmf->flags & FAULT_FLAG_WRITE;
> +
> +	if (write) {
> +		sb_start_pagefault(sb);
> +		file_update_time(vmf->vma->vm_file);
> +	}
> +	down_read(&F2FS_I(inode)->i_mmap_sem);
> +	result = dax_iomap_fault(vmf, pe_size, &f2fs_iomap_ops);
> +	up_read(&F2FS_I(inode)->i_mmap_sem);
> +	if (write)
> +		sb_end_pagefault(sb);
> +
> +	return result;
> +}
> +
> +static int f2fs_dax_fault(struct vm_fault *vmf)
> +{
> +	return f2fs_dax_huge_fault(vmf, PE_SIZE_PTE);
> +}
> +
> +static int f2fs_dax_pfn_mkwrite(struct vm_fault *vmf)
> +{
> +	struct inode *inode = file_inode(vmf->vma->vm_file);
> +	struct super_block *sb = inode->i_sb;
> +	loff_t size;
> +	int ret;
> +
> +	sb_start_pagefault(sb);
> +	file_update_time(vmf->vma->vm_file);
> +	down_read(&F2FS_I(inode)->i_mmap_sem);
> +	size = (i_size_read(inode) + PAGE_SIZE - 1) >> PAGE_SHIFT;
> +	if (vmf->pgoff >= size)
> +		ret = VM_FAULT_SIGBUS;
> +	else
> +		ret = dax_pfn_mkwrite(vmf);
> +	up_read(&F2FS_I(inode)->i_mmap_sem);
> +	sb_end_pagefault(sb);
> +
> +	return ret;
> +}
> +
> +static const struct vm_operations_struct f2fs_dax_vm_ops = {
> +	.fault		= f2fs_dax_fault,
> +	.huge_fault	= f2fs_dax_huge_fault,
> +	.page_mkwrite	= f2fs_dax_fault,
> +	.pfn_mkwrite	= f2fs_dax_pfn_mkwrite,
> +};
> +#else
> +#define f2fs_dax_vm_ops f2fs_file_vm_ops
> +#endif
> +
>  static int get_parent_ino(struct inode *inode, nid_t *pino)
>  {
>  	struct dentry *dentry;
> @@ -436,7 +498,13 @@ static int f2fs_file_mmap(struct file *file, struct vm_area_struct *vma)
>  		return err;
>  
>  	file_accessed(file);
> -	vma->vm_ops = &f2fs_file_vm_ops;
> +
> +	if (IS_DAX(inode)) {
> +		vma->vm_ops = &f2fs_dax_vm_ops;
> +		vma->vm_flags |= VM_MIXEDMAP | VM_HUGEPAGE;
> +	} else
> +		vma->vm_ops = &f2fs_file_vm_ops;
> +
>  	return 0;
>  }
>  
> @@ -520,6 +588,17 @@ static int truncate_partial_data_page(struct inode *inode, u64 from,
>  	if (!offset && !cache_only)
>  		return 0;
>  
> +#ifdef CONFIG_FS_DAX
> +	if (IS_DAX(inode)) {
> +		int ret;
> +
> +		down_read(&F2FS_I(inode)->dio_rwsem[WRITE]);
> +		ret = iomap_zero_range(inode, from, PAGE_SIZE - offset,
> +			NULL, &f2fs_iomap_ops);
> +		up_read(&F2FS_I(inode)->dio_rwsem[WRITE]);
> +		return ret;
> +	}
> +#endif
>  	if (cache_only) {
>  		page = find_lock_page(mapping, index);
>  		if (page && PageUptodate(page))
> @@ -786,6 +865,18 @@ static int fill_zero(struct inode *inode, pgoff_t index,
>  	if (!len)
>  		return 0;
>  
> +#ifdef CONFIG_FS_DAX
> +	if (IS_DAX(inode)) {
> +		int ret;
> +
> +		down_read(&F2FS_I(inode)->dio_rwsem[WRITE]);
> +		ret = iomap_zero_range(inode,
> +			F2FS_BLK_TO_BYTES((loff_t)index) + start,
> +			len, NULL, &f2fs_iomap_ops);
> +		up_read(&F2FS_I(inode)->dio_rwsem[WRITE]);
> +		return ret;
> +	}
> +#endif
>  	f2fs_balance_fs(sbi, true);
>  
>  	f2fs_lock_op(sbi);
> @@ -1108,6 +1199,11 @@ static int f2fs_collapse_range(struct inode *inode, loff_t offset, loff_t len)
>  	loff_t new_size;
>  	int ret;
>  
> +#ifdef CONFIG_FS_DAX
> +	/* The current implementation does not apply to DAX files. */
> +	if (IS_DAX(inode))
> +		return -EINVAL;
> +#endif
>  	if (offset + len >= i_size_read(inode))
>  		return -EINVAL;
>  
> @@ -1298,6 +1394,11 @@ static int f2fs_insert_range(struct inode *inode, loff_t offset, loff_t len)
>  	loff_t new_size;
>  	int ret = 0;
>  
> +#ifdef CONFIG_FS_DAX
> +	/* The current implementation does not apply to DAX files. */
> +	if (IS_DAX(inode))
> +		return -EINVAL;
> +#endif
>  	new_size = i_size_read(inode) + len;
>  	ret = inode_newsize_ok(inode, new_size);
>  	if (ret)
> @@ -1561,6 +1662,10 @@ static int f2fs_ioc_start_atomic_write(struct file *filp)
>  	struct inode *inode = file_inode(filp);
>  	int ret;
>  
> +#ifdef CONFIG_FS_DAX
> +	if (IS_DAX(inode))
> +		return -EINVAL;
> +#endif
>  	if (!inode_owner_or_capable(inode))
>  		return -EACCES;
>  
> @@ -1610,6 +1715,9 @@ static int f2fs_ioc_commit_atomic_write(struct file *filp)
>  	struct inode *inode = file_inode(filp);
>  	int ret;
>  
> +#ifdef CONFIG_FS_DAX
> +	f2fs_bug_on(F2FS_I_SB(inode), IS_DAX(inode));
> +#endif
>  	if (!inode_owner_or_capable(inode))
>  		return -EACCES;
>  
> @@ -1646,6 +1754,10 @@ static int f2fs_ioc_start_volatile_write(struct file *filp)
>  	struct inode *inode = file_inode(filp);
>  	int ret;
>  
> +#ifdef CONFIG_FS_DAX
> +	if (IS_DAX(inode))
> +		return -EINVAL;
> +#endif
>  	if (!inode_owner_or_capable(inode))
>  		return -EACCES;
>  
> @@ -1681,6 +1793,9 @@ static int f2fs_ioc_release_volatile_write(struct file *filp)
>  	struct inode *inode = file_inode(filp);
>  	int ret;
>  
> +#ifdef CONFIG_FS_DAX
> +	f2fs_bug_on(F2FS_I_SB(inode), IS_DAX(inode));
> +#endif
>  	if (!inode_owner_or_capable(inode))
>  		return -EACCES;
>  
> @@ -1710,6 +1825,9 @@ static int f2fs_ioc_abort_volatile_write(struct file *filp)
>  	struct inode *inode = file_inode(filp);
>  	int ret;
>  
> +#ifdef CONFIG_FS_DAX
> +	f2fs_bug_on(F2FS_I_SB(inode), IS_DAX(inode));
> +#endif
>  	if (!inode_owner_or_capable(inode))
>  		return -EACCES;
>  
> @@ -2124,6 +2242,10 @@ static int f2fs_ioc_defragment(struct file *filp, unsigned long arg)
>  	struct f2fs_defragment range;
>  	int err;
>  
> +#ifdef CONFIG_FS_DAX
> +	if (IS_DAX(inode))
> +		return -EINVAL;
> +#endif
>  	if (!capable(CAP_SYS_ADMIN))
>  		return -EPERM;
>  
> @@ -2173,6 +2295,10 @@ static int f2fs_move_file_range(struct file *file_in, loff_t pos_in,
>  	size_t dst_osize;
>  	int ret;
>  
> +#ifdef CONFIG_FS_DAX
> +	if (IS_DAX(src) || IS_DAX(dst))
> +		return -EINVAL;
> +#endif
>  	if (file_in->f_path.mnt != file_out->f_path.mnt ||
>  				src->i_sb != dst->i_sb)
>  		return -EXDEV;
> @@ -2414,6 +2540,61 @@ long f2fs_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
>  	}
>  }
>  
> +#ifdef CONFIG_FS_DAX
> +static ssize_t f2fs_dax_read_iter(struct kiocb *iocb, struct iov_iter *to)
> +{
> +	struct inode *inode = file_inode(iocb->ki_filp);
> +	ssize_t ret;
> +
> +	inode_lock_shared(inode);
> +
> +	if (!IS_DAX(inode)) {
> +		inode_unlock_shared(inode);
> +		return generic_file_read_iter(iocb, to);
> +	}
> +
> +	down_read(&F2FS_I(inode)->dio_rwsem[READ]);
> +	ret = dax_iomap_rw(iocb, to, &f2fs_iomap_ops);
> +	up_read(&F2FS_I(inode)->dio_rwsem[READ]);
> +	inode_unlock_shared(inode);
> +
> +	file_accessed(iocb->ki_filp);
> +	return ret;
> +}
> +
> +static ssize_t f2fs_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
> +{
> +	if (!iov_iter_count(to))
> +		return 0; /* skip atime */
> +
> +	if (IS_DAX(file_inode(iocb->ki_filp)))
> +		return f2fs_dax_read_iter(iocb, to);
> +
> +	return generic_file_read_iter(iocb, to);
> +}
> +
> +static ssize_t f2fs_dax_write_iter(struct kiocb *iocb, struct iov_iter *from)
> +{
> +	struct inode *inode = file_inode(iocb->ki_filp);
> +	ssize_t ret;
> +
> +	ret = file_remove_privs(iocb->ki_filp);
> +	if (ret)
> +		return ret;
> +	ret = file_update_time(iocb->ki_filp);
> +	if (ret)
> +		return ret;
> +
> +	down_read(&F2FS_I(inode)->dio_rwsem[WRITE]);
> +	ret = dax_iomap_rw(iocb, from, &f2fs_iomap_ops);
> +	up_read(&F2FS_I(inode)->dio_rwsem[WRITE]);
> +
> +	return ret;
> +}
> +#else
> +#define f2fs_dax_write_iter	__generic_file_write_iter
> +#endif
> +
>  static ssize_t f2fs_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
>  {
>  	struct file *file = iocb->ki_filp;
> @@ -2435,7 +2616,10 @@ static ssize_t f2fs_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
>  			return err;
>  		}
>  		blk_start_plug(&plug);
> -		ret = __generic_file_write_iter(iocb, from);
> +		if (IS_DAX(inode))
> +			ret = f2fs_dax_write_iter(iocb, from);
> +		else
> +			ret = __generic_file_write_iter(iocb, from);
>  		blk_finish_plug(&plug);
>  		clear_inode_flag(inode, FI_NO_PREALLOC);
>  	}
> @@ -2484,7 +2668,11 @@ long f2fs_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
>  
>  const struct file_operations f2fs_file_operations = {
>  	.llseek		= f2fs_llseek,
> +#ifdef CONFIG_FS_DAX
> +	.read_iter	= f2fs_file_read_iter,
> +#else
>  	.read_iter	= generic_file_read_iter,
> +#endif
>  	.write_iter	= f2fs_file_write_iter,
>  	.open		= f2fs_file_open,
>  	.release	= f2fs_release_file,
> diff --git a/fs/f2fs/gc.c b/fs/f2fs/gc.c
> index fa3d2e2..ade4f71 100644
> --- a/fs/f2fs/gc.c
> +++ b/fs/f2fs/gc.c
> @@ -700,6 +700,100 @@ static void move_encrypted_block(struct inode *inode, block_t bidx,
>  	f2fs_put_page(page, 1);
>  }
>  
> +#ifdef CONFIG_FS_DAX
> +#include <linux/dax.h>
> +
> +static void dax_move_data_page(struct inode *inode, block_t bidx,
> +				unsigned int segno, int off)
> +{
> +	struct block_device *bdev = inode->i_sb->s_bdev;
> +	struct dax_device *dax_dev;
> +	struct dnode_of_data dn;
> +	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
> +	struct f2fs_summary sum;
> +	struct node_info ni;
> +	block_t old_blkaddr, new_blkaddr;
> +	int err, id;
> +	long map_len;
> +	pgoff_t pgoff;
> +	void *kaddr_old, *kaddr_new;
> +	pfn_t pfn;
> +
> +	f2fs_bug_on(sbi, f2fs_is_atomic_file(inode));
> +
> +	if (blk_queue_dax(bdev->bd_queue))
> +		dax_dev = dax_get_by_host(bdev->bd_disk->disk_name);
> +	else
> +		return;
> +
> +	if (!check_valid_map(sbi, segno, off))
> +		return;

Need to release dax_dev before return.

BTW, how about check this before grabbing dax_dev?

> +
> +	if (!down_write_trylock(&F2FS_I(inode)->i_mmap_sem))
> +		return;

Need to release dax_dev before return.

Thanks,

> +
> +	unmap_mapping_range(inode->i_mapping, (loff_t)bidx << PAGE_SHIFT,
> +			PAGE_SIZE, 1);
> +	/* find the old block address */
> +	set_new_dnode(&dn, inode, NULL, NULL, 0);
> +	err = get_dnode_of_data(&dn, bidx, LOOKUP_NODE);
> +	if (err)
> +		goto out;
> +	old_blkaddr = dn.data_blkaddr;
> +	/* This page is already truncated */
> +	if (old_blkaddr == NULL_ADDR)
> +		goto put_dn;
> +
> +	/* allocate a new block address */
> +	get_node_info(sbi, dn.nid, &ni);
> +	set_summary(&sum, dn.nid, dn.ofs_in_node, ni.version);
> +	allocate_data_block(sbi, NULL, old_blkaddr, &new_blkaddr,
> +			&sum, CURSEG_COLD_DATA, NULL, false);
> +
> +	/* copy data page from old to new address in dax_bdev */
> +	id = dax_read_lock();
> +	err = bdev_dax_pgoff(bdev, SECTOR_FROM_BLOCK(old_blkaddr),
> +			PAGE_SIZE, &pgoff);
> +	if (err)
> +		goto recover;
> +	map_len = dax_direct_access(dax_dev, pgoff, 1, &kaddr_old, &pfn);
> +	if (map_len < 0)
> +		goto recover;
> +	err = bdev_dax_pgoff(bdev, SECTOR_FROM_BLOCK(new_blkaddr),
> +			PAGE_SIZE, &pgoff);
> +	if (err)
> +		goto recover;
> +	map_len = dax_direct_access(dax_dev, pgoff, 1, &kaddr_new, &pfn);
> +	if (map_len < 0)
> +		goto recover;
> +	copy_page((void __force *)kaddr_new, (void __force *)kaddr_old);
> +
> +	f2fs_update_data_blkaddr(&dn, new_blkaddr);
> +	set_inode_flag(inode, FI_APPEND_WRITE);
> +	if (bidx == 0)
> +		set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
> +
> +recover:
> +	if (err || map_len < 0)
> +		__f2fs_replace_block(sbi, &sum, new_blkaddr, old_blkaddr,
> +							true, true);
> +	dax_read_unlock(id);
> +put_dn:
> +	f2fs_put_dnode(&dn);
> +out:
> +	unmap_mapping_range(inode->i_mapping, (loff_t)bidx << PAGE_SHIFT,
> +			PAGE_SIZE, 1);
> +	up_write(&F2FS_I(inode)->i_mmap_sem);
> +	put_dax(dax_dev);
> +}
> +#else
> +static void dax_move_data_page(struct inode *inode, block_t bidx,
> +				unsigned int segno, int off)
> +{
> +	return;
> +}
> +#endif
> +
>  static void move_data_page(struct inode *inode, block_t bidx, int gc_type,
>  							unsigned int segno, int off)
>  {
> @@ -818,9 +912,9 @@ static void gc_data_segment(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
>  			if (IS_ERR(inode) || is_bad_inode(inode))
>  				continue;
>  
> -			/* if encrypted inode, let's go phase 3 */
> -			if (f2fs_encrypted_inode(inode) &&
> -						S_ISREG(inode->i_mode)) {
> +			/* if DAX or encrypted inode, let's go phase 3 */
> +			if (IS_DAX(inode) || (f2fs_encrypted_inode(inode) &&
> +						S_ISREG(inode->i_mode))) {
>  				add_gc_inode(gc_list, inode);
>  				continue;
>  			}
> @@ -858,7 +952,9 @@ static void gc_data_segment(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
>  
>  			start_bidx = start_bidx_of_node(nofs, inode)
>  								+ ofs_in_node;
> -			if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
> +			if (IS_DAX(inode))
> +				dax_move_data_page(inode, start_bidx, segno, off);
> +			else if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
>  				move_encrypted_block(inode, start_bidx, segno, off);
>  			else
>  				move_data_page(inode, start_bidx, gc_type, segno, off);
> diff --git a/fs/f2fs/inline.c b/fs/f2fs/inline.c
> index e0fd437..fd8b290 100644
> --- a/fs/f2fs/inline.c
> +++ b/fs/f2fs/inline.c
> @@ -28,6 +28,10 @@ bool f2fs_may_inline_data(struct inode *inode)
>  	if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
>  		return false;
>  
> +#ifdef CONFIG_FS_DAX
> +	if (IS_DAX(inode))
> +		return false;
> +#endif
>  	return true;
>  }
>  
> diff --git a/fs/f2fs/inode.c b/fs/f2fs/inode.c
> index 1ff5bd4..aa16c52 100644
> --- a/fs/f2fs/inode.c
> +++ b/fs/f2fs/inode.c
> @@ -43,8 +43,14 @@ void f2fs_set_inode_flags(struct inode *inode)
>  		new_fl |= S_NOATIME;
>  	if (flags & FS_DIRSYNC_FL)
>  		new_fl |= S_DIRSYNC;
> +#ifdef CONFIG_FS_DAX
> +	if (test_opt(F2FS_I_SB(inode), DAX) && S_ISREG(inode->i_mode) &&
> +		!f2fs_has_inline_data(inode) && !f2fs_encrypted_inode(inode) &&
> +		!f2fs_is_atomic_file(inode) && !f2fs_is_volatile_file(inode))
> +		new_fl |= S_DAX;
> +#endif
>  	inode_set_flags(inode, new_fl,
> -			S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
> +			S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_DAX);
>  }
>  
>  static void __get_inode_rdev(struct inode *inode, struct f2fs_inode *ri)
> diff --git a/fs/f2fs/namei.c b/fs/f2fs/namei.c
> index c31b40e..8feeb9a 100644
> --- a/fs/f2fs/namei.c
> +++ b/fs/f2fs/namei.c
> @@ -60,6 +60,11 @@ static struct inode *f2fs_new_inode(struct inode *dir, umode_t mode)
>  	if (f2fs_encrypted_inode(dir) && f2fs_may_encrypt(inode))
>  		f2fs_set_encrypted_inode(inode);
>  
> +#ifdef CONFIG_FS_DAX
> +	if (test_opt(sbi, DAX) && S_ISREG(inode->i_mode) &&
> +		!f2fs_has_inline_data(inode) && !f2fs_encrypted_inode(inode))
> +		inode->i_flags |= S_DAX;
> +#endif
>  	set_inode_flag(inode, FI_NEW_INODE);
>  
>  	if (test_opt(sbi, INLINE_XATTR))
> diff --git a/fs/f2fs/super.c b/fs/f2fs/super.c
> index 8e39b85..74277cd 100644
> --- a/fs/f2fs/super.c
> +++ b/fs/f2fs/super.c
> @@ -106,6 +106,7 @@ enum {
>  	Opt_fault_injection,
>  	Opt_lazytime,
>  	Opt_nolazytime,
> +	Opt_dax,
>  	Opt_err,
>  };
>  
> @@ -141,6 +142,7 @@ enum {
>  	{Opt_fault_injection, "fault_injection=%u"},
>  	{Opt_lazytime, "lazytime"},
>  	{Opt_nolazytime, "nolazytime"},
> +	{Opt_dax, "dax"},
>  	{Opt_err, NULL},
>  };
>  
> @@ -380,6 +382,15 @@ static int parse_options(struct super_block *sb, char *options)
>  		case Opt_nolazytime:
>  			sb->s_flags &= ~MS_LAZYTIME;
>  			break;
> +#ifdef CONFIG_FS_DAX
> +		case Opt_dax:
> +			set_opt(sbi, DAX);
> +			break;
> +#else
> +		case Opt_dax:
> +			f2fs_msg(sb, KERN_INFO, "dax option not supported");
> +			break;
> +#endif
>  		default:
>  			f2fs_msg(sb, KERN_ERR,
>  				"Unrecognized mount option \"%s\" or missing value",
> @@ -775,6 +786,10 @@ static int f2fs_show_options(struct seq_file *seq, struct dentry *root)
>  		seq_printf(seq, ",fault_injection=%u",
>  				sbi->fault_info.inject_rate);
>  #endif
> +#ifdef CONFIG_FS_DAX
> +	if (test_opt(sbi, DAX))
> +		seq_puts(seq, ",dax");
> +#endif
>  
>  	return 0;
>  }
>