Re: [PATCH] f2fs: separate IPU policy for fdatasync from F2FS_IPU_FSYNC

From: Chao Yu
Date: Wed Nov 09 2022 - 08:40:07 EST

On 2022/11/9 20:56, qixiaoyu wrote:
On Tue, Nov 08, 2022 at 10:30:13PM +0800, Chao Yu wrote:
On 2022/11/8 20:32, qixiaoyu wrote:
On Sun, Nov 06, 2022 at 09:54:59PM +0800, Chao Yu wrote:
On 2022/11/2 20:25, qixiaoyu wrote:
Hi Chao,

fdatasync do in-place-update to avoid additional node writes, but currently
it only do that with F2FS_IPU_FSYNC as:

f2fs_do_sync_file:
if (datasync || get_dirty_pages(inode) <= SM_I(sbi)->min_fsync_blocks)
set_inode_flag(inode, FI_NEED_IPU);

check_inplace_update_policy:
/* this is only set during fdatasync */
if (policy & (0x1 << F2FS_IPU_FSYNC) &&
is_inode_flag_set(inode, FI_NEED_IPU))
return true;

So this patch separate in-place-update of fdatasync from F2FS_IPU_FSYNC to
apply it to all IPU policy.

BTW, we found small performance improvement with this patch on AndroBench app
using F2FS_IPU_SSR_UTIL on our product:

How this patch affects performance when F2FS_IPU_SSR_UTIL is on?

Thanks,


SQLite test in AndroBench app use fdatasync to sync file to the disk.
When switch to F2FS_IPU_SSR_UTIL ipu_policy, it will use out-of-place-update
even though SQLite calls fdatasync, which will introduce extra meta data write.

Why not using F2FS_IPU_SSR_UTIL | F2FS_IPU_FSYNC, I guess these two flags
cover different scenarios, F2FS_IPU_SSR_UTIL for ssr case, and F2FS_IPU_FSYNC
for f{data,}sync case.

Thanks,


As fsync(2) says:
fdatasync() is similar to fsync(), but does not flush modified metadata unless that
metadata is needed in order to allow a subsequent data retrieval to be correctly handled.

I guess it says it allows fdatasync to flush metatdata in order to recovery data in SPO
case.


I think fdatasync should try to perform in-place-update to avoid unnecessary metadata
update whatever the ipu_policy is, and F2FS_IPU_FSYNC is used for fsync independently.

IMO, FSYNC key word in F2FS_IPU_FSYNC means fsync path or interface name as below:

int (*fsync) (struct file *, loff_t, loff_t, int datasync);

And by default, f2fs enables F2FS_IPU_FSYNC, I didn't get why we need to disable it.

To Jaegeuk, any comments?

Thanks,


Thanks


Thanks.


F2FS_IPU_FSYNC F2FS_IPU_SSR_UTIL F2FS_IPU_SSR_UTIL(with patch)
SQLite Insert(QPS) 6818.08 6327.09(-7.20%) 6757.72
SQLite Update(QPS) 6528.81 6336.57(-2.94%) 6490.77
SQLite Delete(QPS) 9724.68 9378.37(-3.56%) 9622.27

Thanks

On Tue, Nov 01, 2022 at 11:14:55PM +0800, Chao Yu wrote:
On 2022/10/21 10:31, qixiaoyu1 wrote:
Currently IPU policy for fdatasync is coupled with F2FS_IPU_FSYNC.
Fix to apply it to all IPU policy.

Xiaoyu,

Sorry for the delay.

I didn't get the point, can you please explain more about the
issue?

Thanks,


Signed-off-by: qixiaoyu1 <qixiaoyu1@xxxxxxxxxx>
---
fs/f2fs/data.c | 8 +++-----
fs/f2fs/file.c | 4 +++-
2 files changed, 6 insertions(+), 6 deletions(-)

diff --git a/fs/f2fs/data.c b/fs/f2fs/data.c
index a71e818cd67b..fec8e15fe820 100644
--- a/fs/f2fs/data.c
+++ b/fs/f2fs/data.c
@@ -2518,6 +2518,9 @@ static inline bool check_inplace_update_policy(struct inode *inode,
if (policy & (0x1 << F2FS_IPU_HONOR_OPU_WRITE) &&
is_inode_flag_set(inode, FI_OPU_WRITE))
return false;
+ /* this is set by fdatasync or F2FS_IPU_FSYNC policy */
+ if (is_inode_flag_set(inode, FI_NEED_IPU))
+ return true;
if (policy & (0x1 << F2FS_IPU_FORCE))
return true;
if (policy & (0x1 << F2FS_IPU_SSR) && f2fs_need_SSR(sbi))
@@ -2538,11 +2541,6 @@ static inline bool check_inplace_update_policy(struct inode *inode,
!IS_ENCRYPTED(inode))
return true;
- /* this is only set during fdatasync */
- if (policy & (0x1 << F2FS_IPU_FSYNC) &&
- is_inode_flag_set(inode, FI_NEED_IPU))
- return true;
-
if (unlikely(fio && is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
!f2fs_is_checkpointed_data(sbi, fio->old_blkaddr)))
return true;
diff --git a/fs/f2fs/file.c b/fs/f2fs/file.c
index 82cda1258227..08091550cdf2 100644
--- a/fs/f2fs/file.c
+++ b/fs/f2fs/file.c
@@ -270,8 +270,10 @@ static int f2fs_do_sync_file(struct file *file, loff_t start, loff_t end,
goto go_write;
/* if fdatasync is triggered, let's do in-place-update */
- if (datasync || get_dirty_pages(inode) <= SM_I(sbi)->min_fsync_blocks)
+ if (datasync || (SM_I(sbi)->ipu_policy & (0x1 << F2FS_IPU_FSYNC) &&
+ get_dirty_pages(inode) <= SM_I(sbi)->min_fsync_blocks))
set_inode_flag(inode, FI_NEED_IPU);
+
ret = file_write_and_wait_range(file, start, end);
clear_inode_flag(inode, FI_NEED_IPU);