[v5,1/2] f2fs: correct i_size change for atomic writes

Message ID 20221018170017.672824-1-daeho43@gmail.com
State New
Headers
Series [v5,1/2] f2fs: correct i_size change for atomic writes |

Commit Message

Daeho Jeong Oct. 18, 2022, 5 p.m. UTC
  From: Daeho Jeong <daehojeong@google.com>

We need to make sure i_size doesn't change until atomic write commit is
successful and restore it when commit is failed.

Signed-off-by: Daeho Jeong <daehojeong@google.com>
---
v5: prevent only i_size update for atomic files when dirtying inode
v4: move i_size update after clearing atomic file flag in
    f2fs_abort_atomic_write()
v3: make sure inode is clean while atomic writing
---
 fs/f2fs/f2fs.h    |  8 ++++++++
 fs/f2fs/file.c    | 18 +++++++++++-------
 fs/f2fs/inode.c   |  5 ++++-
 fs/f2fs/segment.c | 14 ++++++++++----
 4 files changed, 33 insertions(+), 12 deletions(-)
  

Comments

Chao Yu Oct. 29, 2022, 2:48 a.m. UTC | #1
On 2022/10/19 1:00, Daeho Jeong wrote:
> From: Daeho Jeong <daehojeong@google.com>
> 
> We need to make sure i_size doesn't change until atomic write commit is
> successful and restore it when commit is failed.
> 
> Signed-off-by: Daeho Jeong <daehojeong@google.com>
> ---
> v5: prevent only i_size update for atomic files when dirtying inode
> v4: move i_size update after clearing atomic file flag in
>      f2fs_abort_atomic_write()
> v3: make sure inode is clean while atomic writing
> ---
>   fs/f2fs/f2fs.h    |  8 ++++++++
>   fs/f2fs/file.c    | 18 +++++++++++-------
>   fs/f2fs/inode.c   |  5 ++++-
>   fs/f2fs/segment.c | 14 ++++++++++----
>   4 files changed, 33 insertions(+), 12 deletions(-)
> 
> diff --git a/fs/f2fs/f2fs.h b/fs/f2fs/f2fs.h
> index e7e750e6b332..68d2a3383c38 100644
> --- a/fs/f2fs/f2fs.h
> +++ b/fs/f2fs/f2fs.h
> @@ -764,6 +764,7 @@ enum {
>   	FI_COMPRESS_RELEASED,	/* compressed blocks were released */
>   	FI_ALIGNED_WRITE,	/* enable aligned write */
>   	FI_COW_FILE,		/* indicate COW file */
> +	FI_ATOMIC_COMMIT,	/* indicate atomic commit completed except disk sync */

How about FI_ATOMIC_WRITTEN or FI_ATOMIC_COMMITTED? which may be
more readable?

>   	FI_MAX,			/* max flag, never be used */
>   };
>   
> @@ -822,6 +823,7 @@ struct f2fs_inode_info {
>   	unsigned int i_cluster_size;		/* cluster size */
>   
>   	unsigned int atomic_write_cnt;
> +	loff_t original_i_size;		/* original i_size before atomic write */
>   };
>   
>   static inline void get_extent_info(struct extent_info *ext,
> @@ -3072,6 +3074,8 @@ static inline void f2fs_i_blocks_write(struct inode *inode,
>   		set_inode_flag(inode, FI_AUTO_RECOVER);
>   }
>   
> +static inline bool f2fs_is_atomic_file(struct inode *inode);
> +
>   static inline void f2fs_i_size_write(struct inode *inode, loff_t i_size)
>   {
>   	bool clean = !is_inode_flag_set(inode, FI_DIRTY_INODE);
> @@ -3081,6 +3085,10 @@ static inline void f2fs_i_size_write(struct inode *inode, loff_t i_size)
>   		return;
>   
>   	i_size_write(inode, i_size);
> +
> +	if (f2fs_is_atomic_file(inode))
> +		return;
> +
>   	f2fs_mark_inode_dirty_sync(inode, true);
>   	if (clean || recover)
>   		set_inode_flag(inode, FI_AUTO_RECOVER);
> diff --git a/fs/f2fs/file.c b/fs/f2fs/file.c
> index ec9ee0f6d502..7ce629c95f4a 100644
> --- a/fs/f2fs/file.c
> +++ b/fs/f2fs/file.c
> @@ -1991,6 +1991,7 @@ static int f2fs_ioc_start_atomic_write(struct file *filp)
>   	struct f2fs_inode_info *fi = F2FS_I(inode);
>   	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
>   	struct inode *pinode;
> +	loff_t isize;
>   	int ret;
>   
>   	if (!inode_owner_or_capable(mnt_userns, inode))
> @@ -2049,7 +2050,12 @@ static int f2fs_ioc_start_atomic_write(struct file *filp)
>   		f2fs_up_write(&fi->i_gc_rwsem[WRITE]);
>   		goto out;
>   	}
> -	f2fs_i_size_write(fi->cow_inode, i_size_read(inode));
> +
> +	f2fs_write_inode(inode, NULL);
> +
> +	isize = i_size_read(inode);
> +	fi->original_i_size = isize;
> +	f2fs_i_size_write(fi->cow_inode, isize);
>   
>   	stat_inc_atomic_inode(inode);
>   
> @@ -2087,16 +2093,14 @@ static int f2fs_ioc_commit_atomic_write(struct file *filp)
>   
>   	if (f2fs_is_atomic_file(inode)) {
>   		ret = f2fs_commit_atomic_write(inode);
> -		if (ret)
> -			goto unlock_out;
> -
> -		ret = f2fs_do_sync_file(filp, 0, LLONG_MAX, 0, true);
>   		if (!ret)
> -			f2fs_abort_atomic_write(inode, false);
> +			ret = f2fs_do_sync_file(filp, 0, LLONG_MAX, 0, true);
> +
> +		f2fs_abort_atomic_write(inode, ret);
>   	} else {
>   		ret = f2fs_do_sync_file(filp, 0, LLONG_MAX, 1, false);
>   	}
> -unlock_out:
> +
>   	inode_unlock(inode);
>   	mnt_drop_write_file(filp);
>   	return ret;
> diff --git a/fs/f2fs/inode.c b/fs/f2fs/inode.c
> index 9f0d3864d9f1..044f132b87bc 100644
> --- a/fs/f2fs/inode.c
> +++ b/fs/f2fs/inode.c
> @@ -621,9 +621,12 @@ void f2fs_update_inode(struct inode *inode, struct page *node_page)
>   	ri->i_uid = cpu_to_le32(i_uid_read(inode));
>   	ri->i_gid = cpu_to_le32(i_gid_read(inode));
>   	ri->i_links = cpu_to_le32(inode->i_nlink);
> -	ri->i_size = cpu_to_le64(i_size_read(inode));
>   	ri->i_blocks = cpu_to_le64(SECTOR_TO_BLOCK(inode->i_blocks) + 1);
>   
> +	if (!f2fs_is_atomic_file(inode) ||
> +			is_inode_flag_set(inode, FI_ATOMIC_COMMIT))
> +		ri->i_size = cpu_to_le64(i_size_read(inode));

So we expect that below flow can update i_size correctly, right?

- f2fs_ioc_commit_atomic_write
  - f2fs_do_sync_file
   - f2fs_skip_inode_update
   - f2fs_write_inode
    - f2fs_update_inode_page
     - f2fs_update_inode

Thanks,

> +
>   	if (et) {
>   		read_lock(&et->lock);
>   		set_raw_extent(&et->largest, &ri->i_ext);
> diff --git a/fs/f2fs/segment.c b/fs/f2fs/segment.c
> index 289bcb7ca300..264b4e352319 100644
> --- a/fs/f2fs/segment.c
> +++ b/fs/f2fs/segment.c
> @@ -192,14 +192,18 @@ void f2fs_abort_atomic_write(struct inode *inode, bool clean)
>   	if (!f2fs_is_atomic_file(inode))
>   		return;
>   
> -	if (clean)
> -		truncate_inode_pages_final(inode->i_mapping);
>   	clear_inode_flag(fi->cow_inode, FI_COW_FILE);
>   	iput(fi->cow_inode);
>   	fi->cow_inode = NULL;
>   	release_atomic_write_cnt(inode);
> +	clear_inode_flag(inode, FI_ATOMIC_COMMIT);
>   	clear_inode_flag(inode, FI_ATOMIC_FILE);
>   	stat_dec_atomic_inode(inode);
> +
> +	if (clean) {
> +		truncate_inode_pages_final(inode->i_mapping);
> +		f2fs_i_size_write(inode, fi->original_i_size);
> +	}
>   }
>   
>   static int __replace_atomic_write_block(struct inode *inode, pgoff_t index,
> @@ -335,10 +339,12 @@ static int __f2fs_commit_atomic_write(struct inode *inode)
>   	}
>   
>   out:
> -	if (ret)
> +	if (ret) {
>   		sbi->revoked_atomic_block += fi->atomic_write_cnt;
> -	else
> +	} else {
>   		sbi->committed_atomic_block += fi->atomic_write_cnt;
> +		set_inode_flag(inode, FI_ATOMIC_COMMIT);
> +	}
>   
>   	__complete_revoke_list(inode, &revoke_list, ret ? true : false);
>
  
Daeho Jeong Oct. 31, 2022, 6:37 p.m. UTC | #2
On Fri, Oct 28, 2022 at 7:48 PM Chao Yu <chao@kernel.org> wrote:
>
> On 2022/10/19 1:00, Daeho Jeong wrote:
> > From: Daeho Jeong <daehojeong@google.com>
> >
> > We need to make sure i_size doesn't change until atomic write commit is
> > successful and restore it when commit is failed.
> >
> > Signed-off-by: Daeho Jeong <daehojeong@google.com>
> > ---
> > v5: prevent only i_size update for atomic files when dirtying inode
> > v4: move i_size update after clearing atomic file flag in
> >      f2fs_abort_atomic_write()
> > v3: make sure inode is clean while atomic writing
> > ---
> >   fs/f2fs/f2fs.h    |  8 ++++++++
> >   fs/f2fs/file.c    | 18 +++++++++++-------
> >   fs/f2fs/inode.c   |  5 ++++-
> >   fs/f2fs/segment.c | 14 ++++++++++----
> >   4 files changed, 33 insertions(+), 12 deletions(-)
> >
> > diff --git a/fs/f2fs/f2fs.h b/fs/f2fs/f2fs.h
> > index e7e750e6b332..68d2a3383c38 100644
> > --- a/fs/f2fs/f2fs.h
> > +++ b/fs/f2fs/f2fs.h
> > @@ -764,6 +764,7 @@ enum {
> >       FI_COMPRESS_RELEASED,   /* compressed blocks were released */
> >       FI_ALIGNED_WRITE,       /* enable aligned write */
> >       FI_COW_FILE,            /* indicate COW file */
> > +     FI_ATOMIC_COMMIT,       /* indicate atomic commit completed except disk sync */
>
> How about FI_ATOMIC_WRITTEN or FI_ATOMIC_COMMITTED? which may be
> more readable?

Sounds reasonable. I will take FI_ATOMIC_COMMITTED.

>
> >       FI_MAX,                 /* max flag, never be used */
> >   };
> >
> > @@ -822,6 +823,7 @@ struct f2fs_inode_info {
> >       unsigned int i_cluster_size;            /* cluster size */
> >
> >       unsigned int atomic_write_cnt;
> > +     loff_t original_i_size;         /* original i_size before atomic write */
> >   };
> >
> >   static inline void get_extent_info(struct extent_info *ext,
> > @@ -3072,6 +3074,8 @@ static inline void f2fs_i_blocks_write(struct inode *inode,
> >               set_inode_flag(inode, FI_AUTO_RECOVER);
> >   }
> >
> > +static inline bool f2fs_is_atomic_file(struct inode *inode);
> > +
> >   static inline void f2fs_i_size_write(struct inode *inode, loff_t i_size)
> >   {
> >       bool clean = !is_inode_flag_set(inode, FI_DIRTY_INODE);
> > @@ -3081,6 +3085,10 @@ static inline void f2fs_i_size_write(struct inode *inode, loff_t i_size)
> >               return;
> >
> >       i_size_write(inode, i_size);
> > +
> > +     if (f2fs_is_atomic_file(inode))
> > +             return;
> > +
> >       f2fs_mark_inode_dirty_sync(inode, true);
> >       if (clean || recover)
> >               set_inode_flag(inode, FI_AUTO_RECOVER);
> > diff --git a/fs/f2fs/file.c b/fs/f2fs/file.c
> > index ec9ee0f6d502..7ce629c95f4a 100644
> > --- a/fs/f2fs/file.c
> > +++ b/fs/f2fs/file.c
> > @@ -1991,6 +1991,7 @@ static int f2fs_ioc_start_atomic_write(struct file *filp)
> >       struct f2fs_inode_info *fi = F2FS_I(inode);
> >       struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
> >       struct inode *pinode;
> > +     loff_t isize;
> >       int ret;
> >
> >       if (!inode_owner_or_capable(mnt_userns, inode))
> > @@ -2049,7 +2050,12 @@ static int f2fs_ioc_start_atomic_write(struct file *filp)
> >               f2fs_up_write(&fi->i_gc_rwsem[WRITE]);
> >               goto out;
> >       }
> > -     f2fs_i_size_write(fi->cow_inode, i_size_read(inode));
> > +
> > +     f2fs_write_inode(inode, NULL);
> > +
> > +     isize = i_size_read(inode);
> > +     fi->original_i_size = isize;
> > +     f2fs_i_size_write(fi->cow_inode, isize);
> >
> >       stat_inc_atomic_inode(inode);
> >
> > @@ -2087,16 +2093,14 @@ static int f2fs_ioc_commit_atomic_write(struct file *filp)
> >
> >       if (f2fs_is_atomic_file(inode)) {
> >               ret = f2fs_commit_atomic_write(inode);
> > -             if (ret)
> > -                     goto unlock_out;
> > -
> > -             ret = f2fs_do_sync_file(filp, 0, LLONG_MAX, 0, true);
> >               if (!ret)
> > -                     f2fs_abort_atomic_write(inode, false);
> > +                     ret = f2fs_do_sync_file(filp, 0, LLONG_MAX, 0, true);
> > +
> > +             f2fs_abort_atomic_write(inode, ret);
> >       } else {
> >               ret = f2fs_do_sync_file(filp, 0, LLONG_MAX, 1, false);
> >       }
> > -unlock_out:
> > +
> >       inode_unlock(inode);
> >       mnt_drop_write_file(filp);
> >       return ret;
> > diff --git a/fs/f2fs/inode.c b/fs/f2fs/inode.c
> > index 9f0d3864d9f1..044f132b87bc 100644
> > --- a/fs/f2fs/inode.c
> > +++ b/fs/f2fs/inode.c
> > @@ -621,9 +621,12 @@ void f2fs_update_inode(struct inode *inode, struct page *node_page)
> >       ri->i_uid = cpu_to_le32(i_uid_read(inode));
> >       ri->i_gid = cpu_to_le32(i_gid_read(inode));
> >       ri->i_links = cpu_to_le32(inode->i_nlink);
> > -     ri->i_size = cpu_to_le64(i_size_read(inode));
> >       ri->i_blocks = cpu_to_le64(SECTOR_TO_BLOCK(inode->i_blocks) + 1);
> >
> > +     if (!f2fs_is_atomic_file(inode) ||
> > +                     is_inode_flag_set(inode, FI_ATOMIC_COMMIT))
> > +             ri->i_size = cpu_to_le64(i_size_read(inode));
>
> So we expect that below flow can update i_size correctly, right?

Yes, right.

Thanks,

>
> - f2fs_ioc_commit_atomic_write
>   - f2fs_do_sync_file
>    - f2fs_skip_inode_update
>    - f2fs_write_inode
>     - f2fs_update_inode_page
>      - f2fs_update_inode
>
> Thanks,
>
> > +
> >       if (et) {
> >               read_lock(&et->lock);
> >               set_raw_extent(&et->largest, &ri->i_ext);
> > diff --git a/fs/f2fs/segment.c b/fs/f2fs/segment.c
> > index 289bcb7ca300..264b4e352319 100644
> > --- a/fs/f2fs/segment.c
> > +++ b/fs/f2fs/segment.c
> > @@ -192,14 +192,18 @@ void f2fs_abort_atomic_write(struct inode *inode, bool clean)
> >       if (!f2fs_is_atomic_file(inode))
> >               return;
> >
> > -     if (clean)
> > -             truncate_inode_pages_final(inode->i_mapping);
> >       clear_inode_flag(fi->cow_inode, FI_COW_FILE);
> >       iput(fi->cow_inode);
> >       fi->cow_inode = NULL;
> >       release_atomic_write_cnt(inode);
> > +     clear_inode_flag(inode, FI_ATOMIC_COMMIT);
> >       clear_inode_flag(inode, FI_ATOMIC_FILE);
> >       stat_dec_atomic_inode(inode);
> > +
> > +     if (clean) {
> > +             truncate_inode_pages_final(inode->i_mapping);
> > +             f2fs_i_size_write(inode, fi->original_i_size);
> > +     }
> >   }
> >
> >   static int __replace_atomic_write_block(struct inode *inode, pgoff_t index,
> > @@ -335,10 +339,12 @@ static int __f2fs_commit_atomic_write(struct inode *inode)
> >       }
> >
> >   out:
> > -     if (ret)
> > +     if (ret) {
> >               sbi->revoked_atomic_block += fi->atomic_write_cnt;
> > -     else
> > +     } else {
> >               sbi->committed_atomic_block += fi->atomic_write_cnt;
> > +             set_inode_flag(inode, FI_ATOMIC_COMMIT);
> > +     }
> >
> >       __complete_revoke_list(inode, &revoke_list, ret ? true : false);
> >
  

Patch

diff --git a/fs/f2fs/f2fs.h b/fs/f2fs/f2fs.h
index e7e750e6b332..68d2a3383c38 100644
--- a/fs/f2fs/f2fs.h
+++ b/fs/f2fs/f2fs.h
@@ -764,6 +764,7 @@  enum {
 	FI_COMPRESS_RELEASED,	/* compressed blocks were released */
 	FI_ALIGNED_WRITE,	/* enable aligned write */
 	FI_COW_FILE,		/* indicate COW file */
+	FI_ATOMIC_COMMIT,	/* indicate atomic commit completed except disk sync */
 	FI_MAX,			/* max flag, never be used */
 };
 
@@ -822,6 +823,7 @@  struct f2fs_inode_info {
 	unsigned int i_cluster_size;		/* cluster size */
 
 	unsigned int atomic_write_cnt;
+	loff_t original_i_size;		/* original i_size before atomic write */
 };
 
 static inline void get_extent_info(struct extent_info *ext,
@@ -3072,6 +3074,8 @@  static inline void f2fs_i_blocks_write(struct inode *inode,
 		set_inode_flag(inode, FI_AUTO_RECOVER);
 }
 
+static inline bool f2fs_is_atomic_file(struct inode *inode);
+
 static inline void f2fs_i_size_write(struct inode *inode, loff_t i_size)
 {
 	bool clean = !is_inode_flag_set(inode, FI_DIRTY_INODE);
@@ -3081,6 +3085,10 @@  static inline void f2fs_i_size_write(struct inode *inode, loff_t i_size)
 		return;
 
 	i_size_write(inode, i_size);
+
+	if (f2fs_is_atomic_file(inode))
+		return;
+
 	f2fs_mark_inode_dirty_sync(inode, true);
 	if (clean || recover)
 		set_inode_flag(inode, FI_AUTO_RECOVER);
diff --git a/fs/f2fs/file.c b/fs/f2fs/file.c
index ec9ee0f6d502..7ce629c95f4a 100644
--- a/fs/f2fs/file.c
+++ b/fs/f2fs/file.c
@@ -1991,6 +1991,7 @@  static int f2fs_ioc_start_atomic_write(struct file *filp)
 	struct f2fs_inode_info *fi = F2FS_I(inode);
 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 	struct inode *pinode;
+	loff_t isize;
 	int ret;
 
 	if (!inode_owner_or_capable(mnt_userns, inode))
@@ -2049,7 +2050,12 @@  static int f2fs_ioc_start_atomic_write(struct file *filp)
 		f2fs_up_write(&fi->i_gc_rwsem[WRITE]);
 		goto out;
 	}
-	f2fs_i_size_write(fi->cow_inode, i_size_read(inode));
+
+	f2fs_write_inode(inode, NULL);
+
+	isize = i_size_read(inode);
+	fi->original_i_size = isize;
+	f2fs_i_size_write(fi->cow_inode, isize);
 
 	stat_inc_atomic_inode(inode);
 
@@ -2087,16 +2093,14 @@  static int f2fs_ioc_commit_atomic_write(struct file *filp)
 
 	if (f2fs_is_atomic_file(inode)) {
 		ret = f2fs_commit_atomic_write(inode);
-		if (ret)
-			goto unlock_out;
-
-		ret = f2fs_do_sync_file(filp, 0, LLONG_MAX, 0, true);
 		if (!ret)
-			f2fs_abort_atomic_write(inode, false);
+			ret = f2fs_do_sync_file(filp, 0, LLONG_MAX, 0, true);
+
+		f2fs_abort_atomic_write(inode, ret);
 	} else {
 		ret = f2fs_do_sync_file(filp, 0, LLONG_MAX, 1, false);
 	}
-unlock_out:
+
 	inode_unlock(inode);
 	mnt_drop_write_file(filp);
 	return ret;
diff --git a/fs/f2fs/inode.c b/fs/f2fs/inode.c
index 9f0d3864d9f1..044f132b87bc 100644
--- a/fs/f2fs/inode.c
+++ b/fs/f2fs/inode.c
@@ -621,9 +621,12 @@  void f2fs_update_inode(struct inode *inode, struct page *node_page)
 	ri->i_uid = cpu_to_le32(i_uid_read(inode));
 	ri->i_gid = cpu_to_le32(i_gid_read(inode));
 	ri->i_links = cpu_to_le32(inode->i_nlink);
-	ri->i_size = cpu_to_le64(i_size_read(inode));
 	ri->i_blocks = cpu_to_le64(SECTOR_TO_BLOCK(inode->i_blocks) + 1);
 
+	if (!f2fs_is_atomic_file(inode) ||
+			is_inode_flag_set(inode, FI_ATOMIC_COMMIT))
+		ri->i_size = cpu_to_le64(i_size_read(inode));
+
 	if (et) {
 		read_lock(&et->lock);
 		set_raw_extent(&et->largest, &ri->i_ext);
diff --git a/fs/f2fs/segment.c b/fs/f2fs/segment.c
index 289bcb7ca300..264b4e352319 100644
--- a/fs/f2fs/segment.c
+++ b/fs/f2fs/segment.c
@@ -192,14 +192,18 @@  void f2fs_abort_atomic_write(struct inode *inode, bool clean)
 	if (!f2fs_is_atomic_file(inode))
 		return;
 
-	if (clean)
-		truncate_inode_pages_final(inode->i_mapping);
 	clear_inode_flag(fi->cow_inode, FI_COW_FILE);
 	iput(fi->cow_inode);
 	fi->cow_inode = NULL;
 	release_atomic_write_cnt(inode);
+	clear_inode_flag(inode, FI_ATOMIC_COMMIT);
 	clear_inode_flag(inode, FI_ATOMIC_FILE);
 	stat_dec_atomic_inode(inode);
+
+	if (clean) {
+		truncate_inode_pages_final(inode->i_mapping);
+		f2fs_i_size_write(inode, fi->original_i_size);
+	}
 }
 
 static int __replace_atomic_write_block(struct inode *inode, pgoff_t index,
@@ -335,10 +339,12 @@  static int __f2fs_commit_atomic_write(struct inode *inode)
 	}
 
 out:
-	if (ret)
+	if (ret) {
 		sbi->revoked_atomic_block += fi->atomic_write_cnt;
-	else
+	} else {
 		sbi->committed_atomic_block += fi->atomic_write_cnt;
+		set_inode_flag(inode, FI_ATOMIC_COMMIT);
+	}
 
 	__complete_revoke_list(inode, &revoke_list, ret ? true : false);