Skip to contentSkip to conversation

ocfs2: fix a refcount condition checking

PATCH New Wengang Wang 3 participants 7 messages Feb 4, 2010 – Feb 9, 2010 OCFS2

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 ocfs2: fix a refcount condition checking

Hi Joel/Tao,

I don't know the reflink very well, so please ignore this patch if I am wrong.

I think in ocfs2_prepare_inode_for_write(), we disable DIO write if the inode has reflink.
If am right, the way we determine if the inode has reflink is wrong in case (!has_refcount && direct_io).

---
 fs/ocfs2/file.c |    5 +++--
 1 files changed, 3 insertions(+), 2 deletions(-)
fs/ocfs2/file.c +3 -2
index 06ccf6a..77ebb6e 100644
@@ -1775,7 +1775,7 @@ static int ocfs2_prepare_inode_for_write(struct dentry *dentry,
 					 int *direct_io,
 					 int *has_refcount)
 {
-	int ret = 0, meta_level = 0;
+	int ret = 0, meta_level = 0, refcount = 0;
 	struct inode *inode = dentry->d_inode;
 	loff_t saved_pos, end;
 
@@ -1834,6 +1834,7 @@ static int ocfs2_prepare_inode_for_write(struct dentry *dentry,
 							       saved_pos,
 							       count,
 							       &meta_level);
+			refcount = 1;
 			if (has_refcount)
 				*has_refcount = 1;
 		}
@@ -1859,7 +1860,7 @@ static int ocfs2_prepare_inode_for_write(struct dentry *dentry,
 			break;
 		}
 
-		if (has_refcount && *has_refcount == 1) {
+		if (refcount) {
 			*direct_io = 0;
 			break;
 		}
Signature · 1 line
1.6.6
#1 lore raw

PATCH 1/1 ocfs2: fix a refcount condition checking

Hi wengang,

Quoted 8 lines · Wengang Wang

Hi Joel/Tao,

I don't know the reflink very well, so please ignore this patch if I am wrong.

I think in ocfs2_prepare_inode_for_write(), we disable DIO write if the inode has reflink.
If am right, the way we determine if the inode has reflink is wrong in case (!has_refcount && direct_io).

I just check the caller, all these 2 parameters are either set or NULL simultaneously. You patch only make sense in (!has_refcount && direct_io), but currently we don't have such a case. So why bother adding redundant code for a not-exist case?

Regards,
Tao

Quoted 36 lines
---
 fs/ocfs2/file.c |    5 +++--
 1 files changed, 3 insertions(+), 2 deletions(-)
fs/ocfs2/file.c +3 -2
index 06ccf6a..77ebb6e 100644
@@ -1775,7 +1775,7 @@ static int ocfs2_prepare_inode_for_write(struct dentry *dentry,
 					 int *direct_io,
 					 int *has_refcount)
 {
-	int ret = 0, meta_level = 0;
+	int ret = 0, meta_level = 0, refcount = 0;
 	struct inode *inode = dentry->d_inode;
 	loff_t saved_pos, end;
 
@@ -1834,6 +1834,7 @@ static int ocfs2_prepare_inode_for_write(struct dentry *dentry,
 							       saved_pos,
 							       count,
 							       &meta_level);
+			refcount = 1;
 			if (has_refcount)
 				*has_refcount = 1;
 		}
@@ -1859,7 +1860,7 @@ static int ocfs2_prepare_inode_for_write(struct dentry *dentry,
 			break;
 		}
 
-		if (has_refcount && *has_refcount == 1) {
+		if (refcount) {
 			*direct_io = 0;
 			break;
 		}
#2 lore raw

PATCH 1/1 ocfs2: fix a refcount condition checking

Hi Tao,

Quoted 7 lines · Tao Ma

Hi wengang,

Quoted 4 lines · Wengang Wang

Hi Joel/Tao,

I don't know the reflink very well, so please ignore this patch if I am wrong.

Quoted 8 lines
Quoted 4 lines

I think in ocfs2_prepare_inode_for_write(), we disable DIO write if the inode has reflink.
If am right, the way we determine if the inode has reflink is wrong in case (!has_refcount && direct_io).

I just check the caller, all these 2 parameters are either set or NULL simultaneously. You patch only make sense in (!has_refcount && direct_io), but currently we don't have such a case. So why bother adding redundant code for a not-exist case?

Yes that current calling has no problem. But such interface has potential danger for callers in future.
If you don't like change code, I think it's better to add comment that has_refcount and direct_io must be both NULL or both non-NULL.

regards,
wengang.

#3 lore raw

PATCH 1/1 ocfs2: fix a refcount condition checking

Quoted 19 lines · Wengang Wang

Hi Tao,

Quoted 7 lines · Tao Ma

Hi wengang,

Quoted 4 lines · Wengang Wang

Hi Joel/Tao,

I don't know the reflink very well, so please ignore this patch if I am wrong.

Quoted 8 lines
Quoted 4 lines

I think in ocfs2_prepare_inode_for_write(), we disable DIO write if the inode has reflink.
If am right, the way we determine if the inode has reflink is wrong in case (!has_refcount && direct_io).

I just check the caller, all these 2 parameters are either set or NULL simultaneously. You patch only make sense in (!has_refcount && direct_io), but currently we don't have such a case. So why bother adding redundant code for a not-exist case?

Quoted 4 lines

Yes that current calling has no problem. But such interface has potential danger for callers in future.
If you don't like change code, I think it's better to add comment that has_refcount and direct_io must be both NULL or both non-NULL.

Add it please as you wish.

Regards,
Tao

#4 lore raw

PATCH 1/1 ocfs2: fix a refcount condition checking

Quoted 22 lines · Tao Ma
Quoted 19 lines · Wengang Wang

Hi Tao,

Quoted 15 lines · Tao Ma

Hi wengang,

Quoted 8 lines · Wengang Wang

Hi Joel/Tao,

I don't know the reflink very well, so please ignore this patch if I am wrong.

I think in ocfs2_prepare_inode_for_write(), we disable DIO write if the inode has reflink.
If am right, the way we determine if the inode has reflink is wrong in case (!has_refcount && direct_io).

I just check the caller, all these 2 parameters are either set or NULL simultaneously. You patch only make sense in (!has_refcount && direct_io), but currently we don't have such a case. So why bother adding redundant code for a not-exist case?

Quoted 5 lines
Quoted 4 lines

Yes that current calling has no problem. But such interface has potential danger for callers in future.
If you don't like change code, I think it's better to add comment that has_refcount and direct_io must be both NULL or both non-NULL.

Add it please as you wish.

It's not the best way I want to do. I persist what I have post :).

regards,
wengang.

#5 lore raw

PATCH 1/1 ocfs2: fix a refcount condition checking

In reply to Wengang Wang
Quoted 13 lines · Wengang Wang

index 06ccf6a..77ebb6e 100644

fs/ocfs2/file.c +1 -1
@@ -1775,7 +1775,7 @@ static int ocfs2_prepare_inode_for_write(struct dentry *dentry,
 					 int *direct_io,
 					 int *has_refcount)
 {
-	int ret = 0, meta_level = 0;
+	int ret = 0, meta_level = 0, refcount = 0;
 	struct inode *inode = dentry->d_inode;
 	loff_t saved_pos, end;
 

fs/ocfs2/file.c static int ocfs2_prepare_inode_for_write(struct dentry *dentry, · line 1840

Quoted 7 lines
fs/ocfs2/file.c +1 -0
@@ -1834,6 +1834,7 @@ static int ocfs2_prepare_inode_for_write(struct dentry *dentry,
 							       saved_pos,
 							       count,
 							       &meta_level);
+			refcount = 1;
 			if (has_refcount)
 				*has_refcount = 1;
 		}

Tao is right that this is redundant - we're setting two refcount
bits, which is pointless.
Here's the thing: we know that refcounted extents mean no direct
io. Instead of adding your 'refcount' bit or Tao's comment, why not just clear direct_io right here?

@@ -1834,6 +1834,8 @@ static int ocfs2_prepare_inode_for_write(struct dentry *dentry,
saved_pos, count, &meta_level);
if (has_refcount)
*has_refcount = 1;
+ if (direct_io)
+ *direct_io = 0;
}

It's safe to clear if it exists - whether it was zero before, it must be zero now.

Joel

Signature · 7 lines
"War doesn't determine who's right; war determines who's left."

Joel Becker
Principal Software Developer
Oracle
E-mail: joel.becker at oracle.com
Phone: (650) 506-8127
#6 lore raw

PATCH 1/1 ocfs2: fix a refcount condition checking

Quoted 38 lines · Joel Becker
Quoted 13 lines · Wengang Wang

index 06ccf6a..77ebb6e 100644

fs/ocfs2/file.c +1 -1
@@ -1775,7 +1775,7 @@ static int ocfs2_prepare_inode_for_write(struct dentry *dentry,
 					 int *direct_io,
 					 int *has_refcount)
 {
-	int ret = 0, meta_level = 0;
+	int ret = 0, meta_level = 0, refcount = 0;
 	struct inode *inode = dentry->d_inode;
 	loff_t saved_pos, end;
 

fs/ocfs2/file.c static int ocfs2_prepare_inode_for_write(struct dentry *dentry, · line 1840

Quoted 7 lines
fs/ocfs2/file.c +1 -0
@@ -1834,6 +1834,7 @@ static int ocfs2_prepare_inode_for_write(struct dentry *dentry,
 							       saved_pos,
 							       count,
 							       &meta_level);
+			refcount = 1;
 			if (has_refcount)
 				*has_refcount = 1;
 		}

Tao is right that this is redundant - we're setting two refcount
bits, which is pointless.
Here's the thing: we know that refcounted extents mean no direct
io. Instead of adding your 'refcount' bit or Tao's comment, why not just clear direct_io right here?

fs/ocfs2/file.c +2 -0
@@ -1834,6 +1834,8 @@ static int ocfs2_prepare_inode_for_write(struct dentry *dentry,
 							       saved_pos,
 							       count,
 							       &meta_level);
 			if (has_refcount)
 				*has_refcount = 1;
+			if (direct_io)
+				*direct_io = 0;
 		}
Quoted 2 lines

It's safe to clear if it exists - whether it was zero before, it must be zero now.

yeah, this is better.
btw, we also need to remove the latter lines.

- if (has_refcount && *has_refcount == 1) {
- *direct_io = 0;
- break;
-}

Regards,
Tao