Skip to contentSkip to conversation

f2fs: correct i_size change for atomic writes

PATCH Superseded Daeho Jeong 2 participants 4 messages Oct 4, 2022 F2FS file system

v3 of 6 Superseded Latest is v6 →

lore
Share & tools
lore
#0 lore raw

PATCH v3 1/2 f2fs: correct i_size change for atomic writes

From: Daeho Jeong <[email protected]>

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

---
v3: make sure inode is clean while atomic writing

---
 fs/f2fs/f2fs.h    |  1 +
 fs/f2fs/file.c    | 18 +++++++++++-------
 fs/f2fs/inode.c   |  3 +++
 fs/f2fs/segment.c |  4 +++-
 4 files changed, 18 insertions(+), 8 deletions(-)
fs/f2fs/f2fs.h +1 -0
index dee7b67a17a6..539da7f12cfc 100644
@@ -821,6 +821,7 @@ 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,
fs/f2fs/file.c +11 -7
index 5efe0e4a725a..ce2336d2f688 100644
@@ -1989,6 +1989,7 @@ 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))
@@ -2047,7 +2048,12 @@ 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);
 
 	spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
 	sbi->atomic_files++;
@@ -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;
fs/f2fs/inode.c +3 -0
index cde0a3dc80c3..64d7772b4cd9 100644
@@ -30,6 +30,9 @@ void f2fs_mark_inode_dirty_sync(struct inode *inode, bool sync)
 	if (f2fs_inode_dirtied(inode, sync))
 		return;
 
+	if (f2fs_is_atomic_file(inode))
+		return;
+
 	mark_inode_dirty_sync(inode);
 }
 
fs/f2fs/segment.c +3 -1
index 460048f3c850..143b7ea0fb8e 100644
@@ -193,8 +193,10 @@ void f2fs_abort_atomic_write(struct inode *inode, bool clean)
 	if (!f2fs_is_atomic_file(inode))
 		return;
 
-	if (clean)
+	if (clean) {
 		truncate_inode_pages_final(inode->i_mapping);
+		f2fs_i_size_write(inode, fi->original_i_size);
+	}
 	clear_inode_flag(fi->cow_inode, FI_COW_FILE);
 	iput(fi->cow_inode);
 	fi->cow_inode = NULL;
Signature · 1 line
2.38.0.rc1.362.ged0d419d3c-goog
#1 lore raw

PATCH v3 2/2 f2fs: introduce F2FS_IOC_START_ATOMIC_REPLACE

From: Daeho Jeong <[email protected]>

introduce a new ioctl to replace the whole content of a file atomically, which means it induces truncate and content update at the same time. We can start it with F2FS_IOC_START_ATOMIC_REPLACE and complete it with F2FS_IOC_COMMIT_ATOMIC_WRITE. Or abort it with F2FS_IOC_ABORT_ATOMIC_WRITE.

---
v3: move i_size change after setting atomic write flag
v2: add undefined ioctl number reported by <[email protected]>

---
 fs/f2fs/data.c            |  3 +++
 fs/f2fs/f2fs.h            |  1 +
 fs/f2fs/file.c            | 20 ++++++++++++++------
 fs/f2fs/segment.c         | 14 +++++++++++++-
 include/uapi/linux/f2fs.h |  1 +
 5 files changed, 32 insertions(+), 7 deletions(-)
fs/f2fs/data.c +3 -0
index 6cd29a575105..d3d32db3a25d 100644
@@ -3438,6 +3438,9 @@ static int prepare_atomic_write_begin(struct f2fs_sb_info *sbi,
 	else if (*blk_addr != NULL_ADDR)
 		return 0;
 
+	if (is_inode_flag_set(inode, FI_ATOMIC_REPLACE))
+		goto reserve_block;
+
 	/* Look for the block in the original inode */
 	err = __find_data_block(inode, index, &ori_blk_addr);
 	if (err)
fs/f2fs/f2fs.h +1 -0
index 539da7f12cfc..2c49da12d6d8 100644
@@ -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_REPLACE,	/* indicate atomic replace */
 	FI_MAX,			/* max flag, never be used */
 };
 
fs/f2fs/file.c +14 -6
index ce2336d2f688..66d62ea42c73 100644
@@ -1982,7 +1982,7 @@ static int f2fs_ioc_getversion(struct file *filp, unsigned long arg)
 	return put_user(inode->i_generation, (int __user *)arg);
 }
 
-static int f2fs_ioc_start_atomic_write(struct file *filp)
+static int f2fs_ioc_start_atomic_write(struct file *filp, bool truncate)
 {
 	struct inode *inode = file_inode(filp);
 	struct user_namespace *mnt_userns = file_mnt_user_ns(filp);
@@ -2051,10 +2051,6 @@ static int f2fs_ioc_start_atomic_write(struct file *filp)
 
 	f2fs_write_inode(inode, NULL);
 
-	isize = i_size_read(inode);
-	fi->original_i_size = isize;
-	f2fs_i_size_write(fi->cow_inode, isize);
-
 	spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
 	sbi->atomic_files++;
 	spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
@@ -2064,6 +2060,16 @@ static int f2fs_ioc_start_atomic_write(struct file *filp)
 	clear_inode_flag(fi->cow_inode, FI_INLINE_DATA);
 	f2fs_up_write(&fi->i_gc_rwsem[WRITE]);
 
+	isize = i_size_read(inode);
+	fi->original_i_size = isize;
+	if (truncate) {
+		set_inode_flag(inode, FI_ATOMIC_REPLACE);
+		truncate_inode_pages_final(inode->i_mapping);
+		f2fs_i_size_write(inode, 0);
+		isize = 0;
+	}
+	f2fs_i_size_write(fi->cow_inode, isize);
+
 	f2fs_update_time(sbi, REQ_TIME);
 	fi->atomic_write_task = current;
 	stat_update_max_atomic_write(inode);
@@ -4082,7 +4088,9 @@ static long __f2fs_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
 	case FS_IOC_GETVERSION:
 		return f2fs_ioc_getversion(filp, arg);
 	case F2FS_IOC_START_ATOMIC_WRITE:
-		return f2fs_ioc_start_atomic_write(filp);
+		return f2fs_ioc_start_atomic_write(filp, false);
+	case F2FS_IOC_START_ATOMIC_REPLACE:
+		return f2fs_ioc_start_atomic_write(filp, true);
 	case F2FS_IOC_COMMIT_ATOMIC_WRITE:
 		return f2fs_ioc_commit_atomic_write(filp);
 	case F2FS_IOC_ABORT_ATOMIC_WRITE:
fs/f2fs/segment.c +13 -1
index 143b7ea0fb8e..c524538a9013 100644
@@ -263,14 +263,26 @@ static void __complete_revoke_list(struct inode *inode, struct list_head *head,
 					bool revoke)
 {
 	struct revoke_entry *cur, *tmp;
+	pgoff_t start_index = 0;
+	bool truncate = is_inode_flag_set(inode, FI_ATOMIC_REPLACE);
 
 	list_for_each_entry_safe(cur, tmp, head, list) {
-		if (revoke)
+		if (revoke) {
 			__replace_atomic_write_block(inode, cur->index,
 						cur->old_addr, NULL, true);
+		} else if (truncate) {
+			f2fs_truncate_hole(inode, start_index, cur->index);
+			start_index = cur->index + 1;
+		}
+
 		list_del(&cur->list);
 		kmem_cache_free(revoke_entry_slab, cur);
 	}
+
+	if (!revoke && truncate) {
+		f2fs_do_truncate_blocks(inode, start_index * PAGE_SIZE, false);
+		clear_inode_flag(inode, FI_ATOMIC_REPLACE);
+	}
 }
 
 static int __f2fs_commit_atomic_write(struct inode *inode)
include/uapi/linux/f2fs.h +1 -0
index 3121d127d5aa..955d440be104 100644
@@ -42,6 +42,7 @@
 						struct f2fs_comp_option)
 #define F2FS_IOC_DECOMPRESS_FILE	_IO(F2FS_IOCTL_MAGIC, 23)
 #define F2FS_IOC_COMPRESS_FILE		_IO(F2FS_IOCTL_MAGIC, 24)
+#define F2FS_IOC_START_ATOMIC_REPLACE	_IO(F2FS_IOCTL_MAGIC, 25)
 
 /*
  * should be same as XFS_IOC_GOINGDOWN.
Signature · 1 line
2.38.0.rc1.362.ged0d419d3c-goog
maintainer
#2 lore raw
In reply to Daeho Jeong
Quoted 93 lines · Daeho Jeong

From: Daeho Jeong <[email protected]>

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

---
v3: make sure inode is clean while atomic writing

---
  fs/f2fs/f2fs.h    |  1 +
  fs/f2fs/file.c    | 18 +++++++++++-------
  fs/f2fs/inode.c   |  3 +++
  fs/f2fs/segment.c |  4 +++-
  4 files changed, 18 insertions(+), 8 deletions(-)
fs/f2fs/f2fs.h +1 -0
index dee7b67a17a6..539da7f12cfc 100644
@@ -821,6 +821,7 @@ 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,
fs/f2fs/file.c +11 -7
index 5efe0e4a725a..ce2336d2f688 100644
@@ -1989,6 +1989,7 @@ 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))
@@ -2047,7 +2048,12 @@ 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);
 
  	spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
  	sbi->atomic_files++;
@@ -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;
fs/f2fs/inode.c +3 -0
index cde0a3dc80c3..64d7772b4cd9 100644
@@ -30,6 +30,9 @@ void f2fs_mark_inode_dirty_sync(struct inode *inode, bool sync)
  	if (f2fs_inode_dirtied(inode, sync))
  		return;
 
+	if (f2fs_is_atomic_file(inode))
+		return;
+
  	mark_inode_dirty_sync(inode);
  }
 
fs/f2fs/segment.c +0 -0
index 460048f3c850..143b7ea0fb8e 100644

fs/f2fs/segment.c void f2fs_abort_atomic_write(struct inode *inode, bool clean) · line 198

Quoted 7 lines
fs/f2fs/segment.c +2 -1
@@ -193,8 +193,10 @@ void f2fs_abort_atomic_write(struct inode *inode, bool clean)
  	if (!f2fs_is_atomic_file(inode))
  		return;
 
-	if (clean)
+	if (clean) {
  		truncate_inode_pages_final(inode->i_mapping);
+		f2fs_i_size_write(inode, fi->original_i_size);

inode has atomic_write flag, it can not be set dirtied here, how about writing i_size ater clear_inode_flag(inode, FI_ATOMIC_FILE)?

Thanks,

Quoted 4 lines
fs/f2fs/segment.c +1 -0
+	}
  	clear_inode_flag(fi->cow_inode, FI_COW_FILE);
  	iput(fi->cow_inode);
  	fi->cow_inode = NULL;
#3 lore raw
Quoted 98 lines · Chao Yu
Quoted 96 lines · Daeho Jeong

From: Daeho Jeong <[email protected]>

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

---
v3: make sure inode is clean while atomic writing

---
  fs/f2fs/f2fs.h    |  1 +
  fs/f2fs/file.c    | 18 +++++++++++-------
  fs/f2fs/inode.c   |  3 +++
  fs/f2fs/segment.c |  4 +++-
  4 files changed, 18 insertions(+), 8 deletions(-)
fs/f2fs/f2fs.h +1 -0
index dee7b67a17a6..539da7f12cfc 100644
@@ -821,6 +821,7 @@ 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,
fs/f2fs/file.c +11 -7
index 5efe0e4a725a..ce2336d2f688 100644
@@ -1989,6 +1989,7 @@ 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))
@@ -2047,7 +2048,12 @@ 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);
 
      spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
      sbi->atomic_files++;
@@ -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;
fs/f2fs/inode.c +3 -0
index cde0a3dc80c3..64d7772b4cd9 100644
@@ -30,6 +30,9 @@ void f2fs_mark_inode_dirty_sync(struct inode *inode, bool sync)
      if (f2fs_inode_dirtied(inode, sync))
              return;
 
+     if (f2fs_is_atomic_file(inode))
+             return;
+
      mark_inode_dirty_sync(inode);
  }
 
fs/f2fs/segment.c +0 -0
index 460048f3c850..143b7ea0fb8e 100644
@@ -193,8 +193,10 @@ void f2fs_abort_atomic_write(struct inode *inode, bool clean)
      if (!f2fs_is_atomic_file(inode))
              return;
 
Quoted 7 lines

fs/f2fs/segment.c void f2fs_abort_atomic_write(struct inode *inode, bool clean) · line 198

Quoted 4 lines
fs/f2fs/segment.c +2 -1
-     if (clean)
+     if (clean) {
              truncate_inode_pages_final(inode->i_mapping);
+             f2fs_i_size_write(inode, fi->original_i_size);

inode has atomic_write flag, it can not be set dirtied here, how about writing i_size ater clear_inode_flag(inode, FI_ATOMIC_FILE)?

Oh, right.
Thanks~!

Quoted 7 lines

Thanks,

Quoted 4 lines
fs/f2fs/segment.c +1 -0
+     }
      clear_inode_flag(fi->cow_inode, FI_COW_FILE);
      iput(fi->cow_inode);
      fi->cow_inode = NULL;