2017-03-13 13:05:49

by Chao Yu

Subject: [PATCH v2] f2fs: restrict write IO alignment condition

We should only align start offset of bio with defined IO_SIZE for below
conditions:
a. mode=lfs mount option
b. write IOs
c. Out-place-update
d. non-meta pages

Signed-off-by: Chao Yu <[email protected]>
---
fs/f2fs/data.c | 10 +++++++---
1 file changed, 7 insertions(+), 3 deletions(-)

diff --git a/fs/f2fs/data.c b/fs/f2fs/data.c
index 1375fef11146..53fe79c70406 100644
--- a/fs/f2fs/data.c
+++ b/fs/f2fs/data.c
@@ -190,6 +190,9 @@ static inline void __submit_bio(struct f2fs_sb_info *sbi,
current->plug && (type == DATA || type == NODE))
blk_finish_plug(current->plug);

+ if (!test_opt(sbi, LFS))
+ goto submit_io;
+
if (type != DATA && type != NODE)
goto submit_io;

@@ -395,11 +398,12 @@ int f2fs_submit_page_mbio(struct f2fs_io_info *fio)
__submit_merged_bio(io);
alloc_new:
if (io->bio == NULL) {
- if ((fio->type == DATA || fio->type == NODE) &&
+ if (test_opt(sbi, LFS) &&
+ !is_read && (fio->type == DATA || fio->type == NODE) &&
fio->new_blkaddr & F2FS_IO_SIZE_MASK(sbi)) {
err = -EAGAIN;
- if (!is_read)
- dec_page_count(sbi, WB_DATA_TYPE(bio_page));
+ f2fs_bug_on(sbi, fio->new_blkaddr == fio->old_blkaddr);
+ dec_page_count(sbi, WB_DATA_TYPE(bio_page));
goto out_fail;
}
io->bio = __bio_alloc(sbi, fio->new_blkaddr,
--
2.8.2.295.g3f1c1d0

2017-03-13 13:46:01

by Kinglong Mee

Subject: Re: [f2fs-dev] [PATCH v2] f2fs: restrict write IO alignment condition

On 3/13/2017 21:02, Chao Yu wrote:
> We should only align start offset of bio with defined IO_SIZE for below
> conditions:
> a. mode=lfs mount option
> b. write IOs
> c. Out-place-update
> d. non-meta pages
>
> Signed-off-by: Chao Yu <[email protected]>
> ---
> fs/f2fs/data.c | 10 +++++++---
> 1 file changed, 7 insertions(+), 3 deletions(-)
>
> diff --git a/fs/f2fs/data.c b/fs/f2fs/data.c
> index 1375fef11146..53fe79c70406 100644
> --- a/fs/f2fs/data.c
> +++ b/fs/f2fs/data.c
> @@ -190,6 +190,9 @@ static inline void __submit_bio(struct f2fs_sb_info *sbi,
> current->plug && (type == DATA || type == NODE))
> blk_finish_plug(current->plug);
>
> + if (!test_opt(sbi, LFS))
> + goto submit_io;
> +
> if (type != DATA && type != NODE)
> goto submit_io;

Is the position of checking LFS here right?

I have a plan of change the code order as,
diff --git a/fs/f2fs/data.c b/fs/f2fs/data.c
index 2ab5ca0..698dbf3 100644
--- a/fs/f2fs/data.c
+++ b/fs/f2fs/data.c
@@ -186,13 +186,12 @@ static inline void __submit_bio(struct f2fs_sb_info *sbi,
if (!is_read_io(bio_op(bio))) {
unsigned int start;

- if (f2fs_sb_mounted_blkzoned(sbi->sb) &&
- current->plug && (type == DATA || type == NODE))
- blk_finish_plug(current->plug);
-
if (type != DATA && type != NODE)
goto submit_io;

+ if (f2fs_sb_mounted_blkzoned(sbi->sb) && current->plug)
+ blk_finish_plug(current->plug);
+
start = bio->bi_iter.bi_size >> F2FS_BLKSIZE_BITS;
start %= F2FS_IO_SIZE(sbi);

thanks,
Kinglong Mee

>
> @@ -395,11 +398,12 @@ int f2fs_submit_page_mbio(struct f2fs_io_info *fio)
> __submit_merged_bio(io);
> alloc_new:
> if (io->bio == NULL) {
> - if ((fio->type == DATA || fio->type == NODE) &&
> + if (test_opt(sbi, LFS) &&
> + !is_read && (fio->type == DATA || fio->type == NODE) &&
> fio->new_blkaddr & F2FS_IO_SIZE_MASK(sbi)) {
> err = -EAGAIN;
> - if (!is_read)
> - dec_page_count(sbi, WB_DATA_TYPE(bio_page));
> + f2fs_bug_on(sbi, fio->new_blkaddr == fio->old_blkaddr);
> + dec_page_count(sbi, WB_DATA_TYPE(bio_page));
> goto out_fail;
> }
> io->bio = __bio_alloc(sbi, fio->new_blkaddr,
>

2017-03-13 20:06:08

by Jaegeuk Kim

Subject: Re: [PATCH v2] f2fs: restrict write IO alignment condition

On 03/13, Chao Yu wrote:
> We should only align start offset of bio with defined IO_SIZE for below
> conditions:
> a. mode=lfs mount option
> b. write IOs
> c. Out-place-update
> d. non-meta pages
>
> Signed-off-by: Chao Yu <[email protected]>
> ---
> fs/f2fs/data.c | 10 +++++++---
> 1 file changed, 7 insertions(+), 3 deletions(-)
>
> diff --git a/fs/f2fs/data.c b/fs/f2fs/data.c
> index 1375fef11146..53fe79c70406 100644
> --- a/fs/f2fs/data.c
> +++ b/fs/f2fs/data.c
> @@ -190,6 +190,9 @@ static inline void __submit_bio(struct f2fs_sb_info *sbi,
> current->plug && (type == DATA || type == NODE))
> blk_finish_plug(current->plug);
>
> + if (!test_opt(sbi, LFS))
> + goto submit_io;

The parse_options in fill_super checks this.

> +
> if (type != DATA && type != NODE)
> goto submit_io;
>
> @@ -395,11 +398,12 @@ int f2fs_submit_page_mbio(struct f2fs_io_info *fio)
> __submit_merged_bio(io);
> alloc_new:
> if (io->bio == NULL) {
> - if ((fio->type == DATA || fio->type == NODE) &&
> + if (test_opt(sbi, LFS) &&
> + !is_read && (fio->type == DATA || fio->type == NODE) &&

The missing case is !is_read, here?
But, actually, DATA and NODE don't issue any read commands at all.

Thanks,

> fio->new_blkaddr & F2FS_IO_SIZE_MASK(sbi)) {
> err = -EAGAIN;
> - if (!is_read)
> - dec_page_count(sbi, WB_DATA_TYPE(bio_page));
> + f2fs_bug_on(sbi, fio->new_blkaddr == fio->old_blkaddr);
> + dec_page_count(sbi, WB_DATA_TYPE(bio_page));
> goto out_fail;
> }
> io->bio = __bio_alloc(sbi, fio->new_blkaddr,
> --
> 2.8.2.295.g3f1c1d0

2017-03-14 11:32:47

by Chao Yu

Subject: Re: [PATCH v2] f2fs: restrict write IO alignment condition

On 2017/3/14 4:05, Jaegeuk Kim wrote:
> On 03/13, Chao Yu wrote:
>> We should only align start offset of bio with defined IO_SIZE for below
>> conditions:
>> a. mode=lfs mount option
>> b. write IOs
>> c. Out-place-update
>> d. non-meta pages
>>
>> Signed-off-by: Chao Yu <[email protected]>
>> ---
>> fs/f2fs/data.c | 10 +++++++---
>> 1 file changed, 7 insertions(+), 3 deletions(-)
>>
>> diff --git a/fs/f2fs/data.c b/fs/f2fs/data.c
>> index 1375fef11146..53fe79c70406 100644
>> --- a/fs/f2fs/data.c
>> +++ b/fs/f2fs/data.c
>> @@ -190,6 +190,9 @@ static inline void __submit_bio(struct f2fs_sb_info *sbi,
>> current->plug && (type == DATA || type == NODE))
>> blk_finish_plug(current->plug);
>>
>> + if (!test_opt(sbi, LFS))
>> + goto submit_io;
>
> The parse_options in fill_super checks this.

You're right, I missed that one...

>
>> +
>> if (type != DATA && type != NODE)
>> goto submit_io;
>>
>> @@ -395,11 +398,12 @@ int f2fs_submit_page_mbio(struct f2fs_io_info *fio)
>> __submit_merged_bio(io);
>> alloc_new:
>> if (io->bio == NULL) {
>> - if ((fio->type == DATA || fio->type == NODE) &&
>> + if (test_opt(sbi, LFS) &&
>> + !is_read && (fio->type == DATA || fio->type == NODE) &&
>
> The missing case is !is_read, here?
> But, actually, DATA and NODE don't issue any read commands at all.

Correct, let me add BUG_ON here. :)

Thanks,

>
> Thanks,
>
>> fio->new_blkaddr & F2FS_IO_SIZE_MASK(sbi)) {
>> err = -EAGAIN;
>> - if (!is_read)
>> - dec_page_count(sbi, WB_DATA_TYPE(bio_page));
>> + f2fs_bug_on(sbi, fio->new_blkaddr == fio->old_blkaddr);
>> + dec_page_count(sbi, WB_DATA_TYPE(bio_page));
>> goto out_fail;
>> }
>> io->bio = __bio_alloc(sbi, fio->new_blkaddr,
>> --
>> 2.8.2.295.g3f1c1d0
>
> .
>