Skip to contentSkip to conversation

f2fs: Add f2fs_balance_fs for direct IO

PATCH Huang Ying 3 participants 5 messages Jul 7, 2014 – Jul 9, 2014 F2FS file system

lore
Share & tools
lore

Huang Ying

author
#0 lore raw

Otherwise, if a large amount of direct IO writes were done, the segment allocation may be failed because no enough segments are gced.

---
 fs/f2fs/data.c |    6 +++++-
 1 file changed, 5 insertions(+), 1 deletion(-)
fs/f2fs/data.c +5 -1
@@ -1038,6 +1038,7 @@ static ssize_t f2fs_direct_IO(int rw, st
 {
 	struct file *file = iocb->ki_filp;
 	struct inode *inode = file->f_mapping->host;
+	struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 
 	/* Let buffer I/O handle the inline data case. */
 	if (f2fs_has_inline_data(inode))
@@ -1046,8 +1047,11 @@ static ssize_t f2fs_direct_IO(int rw, st
 	if (check_direct_IO(inode, rw, iter, offset))
 		return 0;
 
+	if (rw == WRITE)
+		f2fs_balance_fs(sbi);
+
 	/* clear fsync mark to recover these blocks */
-	fsync_mark_clear(F2FS_SB(inode->i_sb), inode->i_ino);
+	fsync_mark_clear(sbi, inode->i_ino);
 
 	return blockdev_direct_IO(rw, iocb, inode, iter, offset,
 				  get_data_block);
maintainer
#1 lore raw

Hi Huang,

It occurs a merge conflict.
Please check the patch.

Anyway, IMO, it needs to place f2fs_balance_fs somewhere in get_data_block.

Thanks,

Quoted 31 lines · Huang Ying

Otherwise, if a large amount of direct IO writes were done, the segment allocation may be failed because no enough segments are gced.

---
 fs/f2fs/data.c |    6 +++++-
 1 file changed, 5 insertions(+), 1 deletion(-)
fs/f2fs/data.c +5 -1
@@ -1038,6 +1038,7 @@ static ssize_t f2fs_direct_IO(int rw, st
 {
 	struct file *file = iocb->ki_filp;
 	struct inode *inode = file->f_mapping->host;
+	struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 
 	/* Let buffer I/O handle the inline data case. */
 	if (f2fs_has_inline_data(inode))
@@ -1046,8 +1047,11 @@ static ssize_t f2fs_direct_IO(int rw, st
 	if (check_direct_IO(inode, rw, iter, offset))
 		return 0;
 
+	if (rw == WRITE)
+		f2fs_balance_fs(sbi);
+
 	/* clear fsync mark to recover these blocks */
-	fsync_mark_clear(F2FS_SB(inode->i_sb), inode->i_ino);
+	fsync_mark_clear(sbi, inode->i_ino);
 
 	return blockdev_direct_IO(rw, iocb, inode, iter, offset,
 				  get_data_block);
Signature · 1 line
Jaegeuk Kim
author
#2 lore raw

Hi, Jaegeuk,

Quoted 4 lines · Jaegeuk Kim

Hi Huang,

It occurs a merge conflict.
Please check the patch.

Sorry, my fault. I think I need to base my patch on f2fs tree instead of latest Linus' tree?

Quoted 1 line

Anyway, IMO, it needs to place f2fs_balance_fs somewhere in get_data_block.

Yes. I think so too. So you prefer to add f2fs_balance_fs in both f2fs_direct_IO and get_data_block, or just add f2fs_blance_fs in get_data_block only?

Best Regards,
Huang, Ying

Quoted 32 lines

Thanks,

Quoted 29 lines · Huang Ying

Otherwise, if a large amount of direct IO writes were done, the segment allocation may be failed because no enough segments are gced.

---
 fs/f2fs/data.c |    6 +++++-
 1 file changed, 5 insertions(+), 1 deletion(-)
fs/f2fs/data.c +5 -1
@@ -1038,6 +1038,7 @@ static ssize_t f2fs_direct_IO(int rw, st
 {
 	struct file *file = iocb->ki_filp;
 	struct inode *inode = file->f_mapping->host;
+	struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 
 	/* Let buffer I/O handle the inline data case. */
 	if (f2fs_has_inline_data(inode))
@@ -1046,8 +1047,11 @@ static ssize_t f2fs_direct_IO(int rw, st
 	if (check_direct_IO(inode, rw, iter, offset))
 		return 0;
 
+	if (rw == WRITE)
+		f2fs_balance_fs(sbi);
+
 	/* clear fsync mark to recover these blocks */
-	fsync_mark_clear(F2FS_SB(inode->i_sb), inode->i_ino);
+	fsync_mark_clear(sbi, inode->i_ino);
 
Quoted 3 lines

fs/f2fs/data.c static ssize_t f2fs_direct_IO(int rw, st · line 1057

Quoted 2 lines
fs/f2fs/data.c +0 -0
 	return blockdev_direct_IO(rw, iocb, inode, iter, offset,
 				  get_data_block);

------------------------------------------------------------------------------
Open source business process management suite built on Java and Eclipse Turn processes into business applications with Bonita BPM Community Edition Quickly connect people, data, and systems into organized workflows Winner of BOSSIE, CODIE, OW2 and Gartner awards
http://p.sf.net/sfu/Bonitasoft

#3 lore raw

Hi Huang,

Quoted 9 lines

-----Original Message-----
From: Huang Ying [mailto:[email protected]]
Sent: Tuesday, July 08, 2014 4:17 PM
To: Jaegeuk Kim
Cc: [email protected]; [email protected]
Subject: Re: [f2fs-dev] f2fs: Add f2fs_balance_fs for direct IO

Hi, Jaegeuk,

Quoted 8 lines
Quoted 4 lines · Jaegeuk Kim

Hi Huang,

It occurs a merge conflict.
Please check the patch.

Sorry, my fault. I think I need to base my patch on f2fs tree instead of latest Linus' tree?

I think you patch is not conflict to both latest linux's tree and f2fs dev tree. but conflict to this patch
"f2fs: introduce f2fs_write_failed to handle error case when write" which maybe in Jaegeuk's local tree, not update to f2fs dev tree.

If possible, you could wait for Jaegeuk to update f2fs tree, or merge this patch, then apply your new patch.

Thanks,
Yu

Quoted 58 lines
Quoted 2 lines

Anyway, IMO, it needs to place f2fs_balance_fs somewhere in get_data_block.

Yes. I think so too. So you prefer to add f2fs_balance_fs in both f2fs_direct_IO and get_data_block, or just add f2fs_blance_fs in get_data_block only?

Best Regards,
Huang, Ying

Quoted 32 lines

Thanks,

Quoted 29 lines · Huang Ying

Otherwise, if a large amount of direct IO writes were done, the segment allocation may be failed because no enough segments are gced.

---
 fs/f2fs/data.c |    6 +++++-
 1 file changed, 5 insertions(+), 1 deletion(-)
fs/f2fs/data.c +5 -1
@@ -1038,6 +1038,7 @@ static ssize_t f2fs_direct_IO(int rw, st
 {
 	struct file *file = iocb->ki_filp;
 	struct inode *inode = file->f_mapping->host;
+	struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 
 	/* Let buffer I/O handle the inline data case. */
 	if (f2fs_has_inline_data(inode))
@@ -1046,8 +1047,11 @@ static ssize_t f2fs_direct_IO(int rw, st
 	if (check_direct_IO(inode, rw, iter, offset))
 		return 0;
 
+	if (rw == WRITE)
+		f2fs_balance_fs(sbi);
+
 	/* clear fsync mark to recover these blocks */
-	fsync_mark_clear(F2FS_SB(inode->i_sb), inode->i_ino);
+	fsync_mark_clear(sbi, inode->i_ino);
 
Quoted 3 lines

fs/f2fs/data.c static ssize_t f2fs_direct_IO(int rw, st · line 1057

Quoted 2 lines
fs/f2fs/data.c +0 -0
 	return blockdev_direct_IO(rw, iocb, inode, iter, offset,
 				  get_data_block);

------------------------------------------------------------------------------
Open source business process management suite built on Java and Eclipse Turn processes into business applications with Bonita BPM Community Edition Quickly connect people, data, and systems into organized workflows Winner of BOSSIE, CODIE, OW2 and Gartner awards
http://p.sf.net/sfu/Bonitasoft
_______________________________________________
Linux-f2fs-devel mailing list [email protected]
https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel

maintainer
#4 lore raw
In reply to Huang Ying
Quoted 10 lines · Huang Ying

Hi, Jaegeuk,

Quoted 4 lines · Jaegeuk Kim

Hi Huang,

It occurs a merge conflict.
Please check the patch.

Sorry, my fault. I think I need to base my patch on f2fs tree instead of latest Linus' tree?

My apologies.
As Chao said, it was due to the other patch.
I'll merge them soon.

Quoted 6 lines
Quoted 2 lines

Anyway, IMO, it needs to place f2fs_balance_fs somewhere in get_data_block.

Yes. I think so too. So you prefer to add f2fs_balance_fs in both f2fs_direct_IO and get_data_block, or just add f2fs_blance_fs in get_data_block only?

I think it should be placed on get_data_block only.
But, at a glance, we need to figure out some conditions when it should be called.
Since we don't or shouldn't call that in the look-up paths (e.g., read).

Thanks,

Quoted 39 lines

Best Regards,
Huang, Ying

Quoted 35 lines

Thanks,

Quoted 31 lines · Huang Ying

Otherwise, if a large amount of direct IO writes were done, the segment allocation may be failed because no enough segments are gced.

---
 fs/f2fs/data.c |    6 +++++-
 1 file changed, 5 insertions(+), 1 deletion(-)
fs/f2fs/data.c +5 -1
@@ -1038,6 +1038,7 @@ static ssize_t f2fs_direct_IO(int rw, st
 {
 	struct file *file = iocb->ki_filp;
 	struct inode *inode = file->f_mapping->host;
+	struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 
 	/* Let buffer I/O handle the inline data case. */
 	if (f2fs_has_inline_data(inode))
@@ -1046,8 +1047,11 @@ static ssize_t f2fs_direct_IO(int rw, st
 	if (check_direct_IO(inode, rw, iter, offset))
 		return 0;
 
+	if (rw == WRITE)
+		f2fs_balance_fs(sbi);
+
 	/* clear fsync mark to recover these blocks */
-	fsync_mark_clear(F2FS_SB(inode->i_sb), inode->i_ino);
+	fsync_mark_clear(sbi, inode->i_ino);
 
 	return blockdev_direct_IO(rw, iocb, inode, iter, offset,
 				  get_data_block);
Signature · 1 line
Jaegeuk Kim