Skip to contentSkip to conversation

f2fs: remove race condition in releasing cblocks

PATCH In review Daeho Jeong 2 participants 5 messages May 8, 2020 F2FS file system

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 f2fs: remove race condition in releasing cblocks

From: Daeho Jeong <[email protected]>

Now, if writing pages and releasing compress blocks occur simultaneously, and releasing cblocks is executed more than one time to a file, then total block count of filesystem and block count of the file could be incorrect and damaged.

We have to execute releasing compress blocks only one time for a file without being interfered by writepages path.

---
 fs/f2fs/file.c | 31 ++++++++++++++++++++++++-------
 1 file changed, 24 insertions(+), 7 deletions(-)
fs/f2fs/file.c +24 -7
index 4aab4b42d8ba..a92bc51b9b28 100644
@@ -3488,6 +3488,7 @@ static int f2fs_release_compress_blocks(struct file *filp, unsigned long arg)
 	pgoff_t page_idx = 0, last_idx;
 	unsigned int released_blocks = 0;
 	int ret;
+	int writecount;
 
 	if (!f2fs_sb_has_compression(F2FS_I_SB(inode)))
 		return -EOPNOTSUPP;
@@ -3509,13 +3510,29 @@ static int f2fs_release_compress_blocks(struct file *filp, unsigned long arg)
 
 	inode_lock(inode);
 
-	if (!IS_IMMUTABLE(inode)) {
-		F2FS_I(inode)->i_flags |= F2FS_IMMUTABLE_FL;
-		f2fs_set_inode_flags(inode);
-		inode->i_ctime = current_time(inode);
-		f2fs_mark_inode_dirty_sync(inode, true);
+	writecount = atomic_read(&inode->i_writecount);
+	if ((filp->f_mode & FMODE_WRITE && writecount != 1) || writecount) {
+		ret = -EBUSY;
+		goto out;
 	}
 
+	if (IS_IMMUTABLE(inode)) {
+		ret = -EINVAL;
+		goto out;
+	}
+
+	ret = filemap_write_and_wait_range(inode->i_mapping, 0, LLONG_MAX);
+	if (ret)
+		goto out;
+
+	if (!F2FS_I(inode)->i_compr_blocks)
+		goto out;
+
+	F2FS_I(inode)->i_flags |= F2FS_IMMUTABLE_FL;
+	f2fs_set_inode_flags(inode);
+	inode->i_ctime = current_time(inode);
+	f2fs_mark_inode_dirty_sync(inode, true);
+
 	down_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
 	down_write(&F2FS_I(inode)->i_mmap_sem);
 
@@ -3554,9 +3571,9 @@ static int f2fs_release_compress_blocks(struct file *filp, unsigned long arg)
 
 	up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
 	up_write(&F2FS_I(inode)->i_mmap_sem);
-
-	inode_unlock(inode);
 out:
+	inode_unlock(inode);
+
 	mnt_drop_write_file(filp);
 
 	if (ret >= 0) {
Signature · 1 line
2.26.2.526.g744177e7f7-goog
#1 lore raw
Quoted 20 lines · Daeho Jeong

From: Daeho Jeong <[email protected]>

Now, if writing pages and releasing compress blocks occur simultaneously, and releasing cblocks is executed more than one time to a file, then total block count of filesystem and block count of the file could be incorrect and damaged.

We have to execute releasing compress blocks only one time for a file without being interfered by writepages path.

---
 fs/f2fs/file.c | 31 ++++++++++++++++++++++++-------
 1 file changed, 24 insertions(+), 7 deletions(-)
fs/f2fs/file.c +0 -0
index 4aab4b42d8ba..a92bc51b9b28 100644

fs/f2fs/file.c static int f2fs_release_compress_blocks(struct file *filp, unsigned long arg) · line 3494

Quoted 7 lines
fs/f2fs/file.c +1 -0
@@ -3488,6 +3488,7 @@ static int f2fs_release_compress_blocks(struct file *filp, unsigned long arg)
 	pgoff_t page_idx = 0, last_idx;
 	unsigned int released_blocks = 0;
 	int ret;
+	int writecount;
 
 	if (!f2fs_sb_has_compression(F2FS_I_SB(inode)))
 		return -EOPNOTSUPP;

Before inode_lock(), there is one case we may jump to out label, in this case, we may unlock inode incorrectly.

	if (!F2FS_I(inode)->i_compr_blocks)
		goto out;

fs/f2fs/file.c static int f2fs_release_compress_blocks(struct file *filp, unsigned long arg) · line 3577

Quoted 6 lines
fs/f2fs/file.c +2 -2
-
-	inode_unlock(inode);
 out:
+	inode_unlock(inode);
+
 	mnt_drop_write_file(filp);

Thanks,

#2 lore raw

I moved checking i_compr_blocks phrase after calling inode_lock() already, because we should check this after writing pages.

So, if it fails to check i_compr_blocks, we need to call inode_unlock().

Am I missing something?

2020년 5월 8일 (금) 오후 3:50, Chao Yu <[email protected]>님이 작성:

Quoted 44 lines · Daeho Jeong
Quoted 20 lines

From: Daeho Jeong <[email protected]>

Now, if writing pages and releasing compress blocks occur simultaneously, and releasing cblocks is executed more than one time to a file, then total block count of filesystem and block count of the file could be incorrect and damaged.

We have to execute releasing compress blocks only one time for a file without being interfered by writepages path.

---
 fs/f2fs/file.c | 31 ++++++++++++++++++++++++-------
 1 file changed, 24 insertions(+), 7 deletions(-)
fs/f2fs/file.c +0 -0
index 4aab4b42d8ba..a92bc51b9b28 100644

fs/f2fs/file.c static int f2fs_release_compress_blocks(struct file *filp, unsigned long arg) · line 3494

Quoted 7 lines
fs/f2fs/file.c +1 -0
@@ -3488,6 +3488,7 @@ static int f2fs_release_compress_blocks(struct file *filp, unsigned long arg)
      pgoff_t page_idx = 0, last_idx;
      unsigned int released_blocks = 0;
      int ret;
+     int writecount;
 
      if (!f2fs_sb_has_compression(F2FS_I_SB(inode)))
              return -EOPNOTSUPP;

Before inode_lock(), there is one case we may jump to out label, in this case, we may unlock inode incorrectly.

        if (!F2FS_I(inode)->i_compr_blocks)
                goto out;

fs/f2fs/file.c static int f2fs_release_compress_blocks(struct file *filp, unsigned long arg) · line 3577

Quoted 6 lines
fs/f2fs/file.c +2 -2
-
-     inode_unlock(inode);
 out:
+     inode_unlock(inode);
+
      mnt_drop_write_file(filp);

Thanks,

#3 lore raw
Quoted 6 lines · Daeho Jeong

I moved checking i_compr_blocks phrase after calling inode_lock() already, because we should check this after writing pages.

So, if it fails to check i_compr_blocks, we need to call inode_unlock().

Am I missing something?

After applying this patch, I get this:

	ret = mnt_want_write_file(filp);
	if (ret)
		return ret;

	if (!F2FS_I(inode)->i_compr_blocks)
		goto out;

	f2fs_balance_fs(F2FS_I_SB(inode), true);

	inode_lock(inode);
Quoted 48 lines

2020년 5월 8일 (금) 오후 3:50, Chao Yu <[email protected]>님이 작성:

Quoted 36 lines · Daeho Jeong
Quoted 20 lines

From: Daeho Jeong <[email protected]>

Now, if writing pages and releasing compress blocks occur simultaneously, and releasing cblocks is executed more than one time to a file, then total block count of filesystem and block count of the file could be incorrect and damaged.

We have to execute releasing compress blocks only one time for a file without being interfered by writepages path.

---
 fs/f2fs/file.c | 31 ++++++++++++++++++++++++-------
 1 file changed, 24 insertions(+), 7 deletions(-)
fs/f2fs/file.c +0 -0
index 4aab4b42d8ba..a92bc51b9b28 100644

fs/f2fs/file.c static int f2fs_release_compress_blocks(struct file *filp, unsigned long arg) · line 3494

Quoted 7 lines
fs/f2fs/file.c +1 -0
@@ -3488,6 +3488,7 @@ static int f2fs_release_compress_blocks(struct file *filp, unsigned long arg)
      pgoff_t page_idx = 0, last_idx;
      unsigned int released_blocks = 0;
      int ret;
+     int writecount;
 
      if (!f2fs_sb_has_compression(F2FS_I_SB(inode)))
              return -EOPNOTSUPP;

Before inode_lock(), there is one case we may jump to out label, in this case, we may unlock inode incorrectly.

        if (!F2FS_I(inode)->i_compr_blocks)
                goto out;
Quoted 8 lines

fs/f2fs/file.c static int f2fs_release_compress_blocks(struct file *filp, unsigned long arg) · line 3577

Quoted 6 lines
fs/f2fs/file.c +2 -2
-
-     inode_unlock(inode);
 out:
+     inode_unlock(inode);
+
      mnt_drop_write_file(filp);

Thanks,

.

#4 lore raw

Oops,

I will re-check it.

Thanks,

2020년 5월 8일 (금) 오후 4:09, Chao Yu <[email protected]>님이 작성:

Quoted 70 lines · Daeho Jeong
Quoted 6 lines

I moved checking i_compr_blocks phrase after calling inode_lock() already, because we should check this after writing pages.

So, if it fails to check i_compr_blocks, we need to call inode_unlock().

Am I missing something?

After applying this patch, I get this:

        ret = mnt_want_write_file(filp);
        if (ret)
                return ret;

        if (!F2FS_I(inode)->i_compr_blocks)
                goto out;

        f2fs_balance_fs(F2FS_I_SB(inode), true);

        inode_lock(inode);
Quoted 48 lines

2020년 5월 8일 (금) 오후 3:50, Chao Yu <[email protected]>님이 작성:

Quoted 36 lines · Daeho Jeong
Quoted 20 lines

From: Daeho Jeong <[email protected]>

Now, if writing pages and releasing compress blocks occur simultaneously, and releasing cblocks is executed more than one time to a file, then total block count of filesystem and block count of the file could be incorrect and damaged.

We have to execute releasing compress blocks only one time for a file without being interfered by writepages path.

---
 fs/f2fs/file.c | 31 ++++++++++++++++++++++++-------
 1 file changed, 24 insertions(+), 7 deletions(-)
fs/f2fs/file.c +0 -0
index 4aab4b42d8ba..a92bc51b9b28 100644

fs/f2fs/file.c static int f2fs_release_compress_blocks(struct file *filp, unsigned long arg) · line 3494

Quoted 7 lines
fs/f2fs/file.c +1 -0
@@ -3488,6 +3488,7 @@ static int f2fs_release_compress_blocks(struct file *filp, unsigned long arg)
      pgoff_t page_idx = 0, last_idx;
      unsigned int released_blocks = 0;
      int ret;
+     int writecount;
 
      if (!f2fs_sb_has_compression(F2FS_I_SB(inode)))
              return -EOPNOTSUPP;

Before inode_lock(), there is one case we may jump to out label, in this case, we may unlock inode incorrectly.

        if (!F2FS_I(inode)->i_compr_blocks)
                goto out;
Quoted 8 lines

fs/f2fs/file.c static int f2fs_release_compress_blocks(struct file *filp, unsigned long arg) · line 3577

Quoted 6 lines
fs/f2fs/file.c +2 -2
-
-     inode_unlock(inode);
 out:
+     inode_unlock(inode);
+
      mnt_drop_write_file(filp);

Thanks,

.