Skip to contentSkip to conversation

f2fs: move f2fs_balance_fs to correct place in unlink

PATCH In review Namjae Jeon 2 participants 5 messages Mar 2, 2013 – Mar 8, 2013

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 3/5 f2fs: move f2fs_balance_fs to correct place in unlink

From: Namjae Jeon <[email protected]>

Actual dirty of pages will occur in f2fs_delete_entry so move the f2fs_balance_fs just before deletion.

---
 fs/f2fs/namei.c |    4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
fs/f2fs/namei.c +2 -2
index 1a49b88..eaa86f5 100644
@@ -223,8 +223,6 @@ static int f2fs_unlink(struct inode *dir, struct dentry *dentry)
 	struct page *page;
 	int err = -ENOENT;
 
-	f2fs_balance_fs(sbi);
-
 	de = f2fs_find_entry(dir, &dentry->d_name, &page);
 	if (!de)
 		goto fail;
@@ -236,6 +234,8 @@ static int f2fs_unlink(struct inode *dir, struct dentry *dentry)
 		goto fail;
 	}
 
+	f2fs_balance_fs(sbi);
+
 	f2fs_delete_entry(de, page, inode);
 
 	/* In order to evict this inode,  we set it dirty */
Signature · 1 line
1.7.9.5
#1 lore raw

2013-03-02 (토), 12:41 +0900, Namjae Jeon:

Quoted 25 lines

From: Namjae Jeon <[email protected]>

Actual dirty of pages will occur in f2fs_delete_entry so move the f2fs_balance_fs just before deletion.

---
 fs/f2fs/namei.c |    4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
fs/f2fs/namei.c +0 -2
index 1a49b88..eaa86f5 100644
@@ -223,8 +223,6 @@ static int f2fs_unlink(struct inode *dir, struct dentry *dentry)
 	struct page *page;
 	int err = -ENOENT;
 
-	f2fs_balance_fs(sbi);
-
 	de = f2fs_find_entry(dir, &dentry->d_name, &page);
 	if (!de)
 		goto fail;

fs/f2fs/namei.c static int f2fs_unlink(struct inode *dir, struct dentry *dentry) · line 238

Quoted 5 lines
fs/f2fs/namei.c +2 -0
@@ -236,6 +234,8 @@ static int f2fs_unlink(struct inode *dir, struct dentry *dentry)
 		goto fail;
 	}
 
+	f2fs_balance_fs(sbi);
+

I think we don't need to do this because of no issues on performance and reliability.
In addition, it would be better to call f2fs_balance_fs without any dentry page.

Quoted 3 lines
fs/f2fs/namei.c +0 -0
 	f2fs_delete_entry(de, page, inode);
 
 	/* In order to evict this inode,  we set it dirty */
Signature · 2 lines
Jaegeuk Kim
Samsung
#2 lore raw

2013/3/3, Jaegeuk Kim <[email protected]>:

Quoted 31 lines

2013-03-02 (토), 12:41 +0900, Namjae Jeon:

Quoted 30 lines

From: Namjae Jeon <[email protected]>

Actual dirty of pages will occur in f2fs_delete_entry so move the f2fs_balance_fs just before deletion.

---
 fs/f2fs/namei.c |    4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
fs/f2fs/namei.c +0 -2
index 1a49b88..eaa86f5 100644
@@ -223,8 +223,6 @@ static int f2fs_unlink(struct inode *dir, struct
 dentry *dentry)
 	struct page *page;
 	int err = -ENOENT;
 
-	f2fs_balance_fs(sbi);
-
 	de = f2fs_find_entry(dir, &dentry->d_name, &page);
 	if (!de)
fs/f2fs/namei.c +0 -0
 		goto fail;
@@ -236,6 +234,8 @@ static int f2fs_unlink(struct inode *dir, struct
 dentry *dentry)
 		goto fail;
 	}
 
Quoted 7 lines

fs/f2fs/namei.c static int f2fs_unlink(struct inode *dir, struct dentry *dentry) · line 238

Quoted 2 lines
fs/f2fs/namei.c +2 -0
+	f2fs_balance_fs(sbi);
+

I think we don't need to do this because of no issues on performance and reliability.
In addition, it would be better to call f2fs_balance_fs without any dentry page.

Regarding moving “f2fs_balance_fs” in unlink part– we considered one scenario. Suppose – when the disk is full and it really needed to trigger the Garbage collection. But in this we considered one scenario. Let’s say the ‘name’ being passed is for invalid file i.e., the file does not exist. So, primarily in this case – I think it should return immediately.
In such cases it actually results in wrong timing results for the non-existence files.
For this observation we thought that f2fs_balance_fs be instead called at a proper place i.e., after there is no lookup-failure.

Let me know your opinion.
Thanks.

Quoted 9 lines
Quoted 4 lines
fs/f2fs/namei.c +0 -0
 	f2fs_delete_entry(de, page, inode);
 

/* In order to evict this inode, we set it dirty */

--
Jaegeuk Kim
Samsung

#3 lore raw

2013-03-04 (월), 15:10 +0900, Namjae Jeon:

Quoted 39 lines

2013/3/3, Jaegeuk Kim <[email protected]>:

Quoted 31 lines

2013-03-02 (토), 12:41 +0900, Namjae Jeon:

Quoted 30 lines

From: Namjae Jeon <[email protected]>

Actual dirty of pages will occur in f2fs_delete_entry so move the f2fs_balance_fs just before deletion.

---
 fs/f2fs/namei.c |    4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
fs/f2fs/namei.c +0 -2
index 1a49b88..eaa86f5 100644
@@ -223,8 +223,6 @@ static int f2fs_unlink(struct inode *dir, struct
 dentry *dentry)
 	struct page *page;
 	int err = -ENOENT;
 
-	f2fs_balance_fs(sbi);
-
 	de = f2fs_find_entry(dir, &dentry->d_name, &page);
 	if (!de)
fs/f2fs/namei.c +0 -0
 		goto fail;
@@ -236,6 +234,8 @@ static int f2fs_unlink(struct inode *dir, struct
 dentry *dentry)
 		goto fail;
 	}
 
Quoted 7 lines

fs/f2fs/namei.c static int f2fs_unlink(struct inode *dir, struct dentry *dentry) · line 238

Quoted 2 lines
fs/f2fs/namei.c +2 -0
+	f2fs_balance_fs(sbi);
+

I think we don't need to do this because of no issues on performance and reliability.
In addition, it would be better to call f2fs_balance_fs without any dentry page.

Quoted 10 lines

Regarding moving “f2fs_balance_fs” in unlink part– we considered one scenario. Suppose – when the disk is full and it really needed to trigger the Garbage collection. But in this we considered one scenario. Let’s say the ‘name’ being passed is for invalid file i.e., the file does not exist. So, primarily in this case – I think it should return immediately.
In such cases it actually results in wrong timing results for the non-existence files.
For this observation we thought that f2fs_balance_fs be instead called at a proper place i.e., after there is no lookup-failure.

If so, could you check all the locations of f2fs_balance_fs and write one patch for whole things?
I suspect that many directory operations are exposed to this issue. Thanks,

Quoted 17 lines

Let me know your opinion.
Thanks.

Quoted 9 lines
Quoted 4 lines
fs/f2fs/namei.c +0 -0
 	f2fs_delete_entry(de, page, inode);
 

/* In order to evict this inode, we set it dirty */

--
Jaegeuk Kim
Samsung

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

Signature · 2 lines
Jaegeuk Kim
Samsung
#4 lore raw

2013/3/8, Jaegeuk Kim <[email protected]>:

Quoted 53 lines

2013-03-04 (월), 15:10 +0900, Namjae Jeon:

Quoted 40 lines

2013/3/3, Jaegeuk Kim <[email protected]>:

Quoted 31 lines

2013-03-02 (토), 12:41 +0900, Namjae Jeon:

Quoted 30 lines

From: Namjae Jeon <[email protected]>

Actual dirty of pages will occur in f2fs_delete_entry so move the f2fs_balance_fs just before deletion.

---
 fs/f2fs/namei.c |    4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
fs/f2fs/namei.c +0 -2
index 1a49b88..eaa86f5 100644
@@ -223,8 +223,6 @@ static int f2fs_unlink(struct inode *dir, struct
 dentry *dentry)
 	struct page *page;
 	int err = -ENOENT;
 
-	f2fs_balance_fs(sbi);
-
 	de = f2fs_find_entry(dir, &dentry->d_name, &page);
 	if (!de)
fs/f2fs/namei.c +0 -0
 		goto fail;
@@ -236,6 +234,8 @@ static int f2fs_unlink(struct inode *dir, struct
 dentry *dentry)
 		goto fail;
 	}
 
Quoted 8 lines

fs/f2fs/namei.c static int f2fs_unlink(struct inode *dir, struct dentry *dentry) · line 238

Quoted 2 lines
fs/f2fs/namei.c +2 -0
+	f2fs_balance_fs(sbi);
+

I think we don't need to do this because of no issues on performance and
reliability.
In addition, it would be better to call f2fs_balance_fs without any dentry page.

Quoted 11 lines

Regarding moving “f2fs_balance_fs” in unlink part– we considered one scenario.
Suppose – when the disk is full and it really needed to trigger the Garbage collection. But in this we considered one scenario. Let’s say the ‘name’ being passed is for invalid file i.e., the file does not exist. So, primarily in this case – I think it should return immediately.
In such cases it actually results in wrong timing results for the non-existence files.
For this observation we thought that f2fs_balance_fs be instead called at a proper place i.e., after there is no lookup-failure.

Quoted 4 lines

If so, could you check all the locations of f2fs_balance_fs and write one patch for whole things?
I suspect that many directory operations are exposed to this issue. Thanks,

Sure, I will check all places to use f2fs_balance_fs. And will post updated patch again.
Thanks Jaegeuk!

Quoted 24 lines
Quoted 13 lines

Let me know your opinion.
Thanks.

Quoted 9 lines
Quoted 4 lines
fs/f2fs/namei.c +0 -0
 	f2fs_delete_entry(de, page, inode);
 

/* In order to evict this inode, we set it dirty */

--
Jaegeuk Kim
Samsung

Quoted 6 lines

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

--
Jaegeuk Kim
Samsung