* Re: [PATCH] ext4: Remove expensive flush on fast commit
2021-01-06 1:32 ` [PATCH] ext4: Remove expensive flush on fast commit Daejun Park
@ 2021-01-06 21:58 ` harshad shirwadkar
2021-01-08 2:26 ` Andreas Dilger
[not found] ` <CGME20210106013242epcms2p5b6b4ed8ca86f29456fdf56aa580e74b4@epcms2p3>
2021-01-06 22:05 ` Darrick J. Wong
2021-01-14 3:44 ` Theodore Ts'o
2 siblings, 2 replies; 6+ messages in thread
From: harshad shirwadkar @ 2021-01-06 21:58 UTC (permalink / raw)
To: daejun7.park; +Cc: tytso, adilger.kernel, linux-ext4, linux-kernel
Thanks Daejun! This looks good.
Reviewed-by: Harshad Shirwadkar <harshadshirwadkar@gmail.com>
On Tue, Jan 5, 2021 at 5:32 PM Daejun Park <daejun7.park@samsung.com> wrote:
>
> In the fast commit, it adds REQ_FUA and REQ_PREFLUSH on each fast commit
> block when barrier is enabled. However, in recovery phase, ext4 compares
> CRC value in the tail. So it is sufficient adds REQ_FUA and REQ_PREFLUSH
> on the block that has tail.
>
> Signed-off-by: Daejun Park <daejun7.park@samsung.com>
> ---
> fs/ext4/fast_commit.c | 10 +++++-----
> 1 file changed, 5 insertions(+), 5 deletions(-)
>
> diff --git a/fs/ext4/fast_commit.c b/fs/ext4/fast_commit.c
> index 4fcc21c25e79..e66507be334c 100644
> --- a/fs/ext4/fast_commit.c
> +++ b/fs/ext4/fast_commit.c
> @@ -604,13 +604,13 @@ void ext4_fc_track_range(handle_t *handle, struct inode *inode, ext4_lblk_t star
> trace_ext4_fc_track_range(inode, start, end, ret);
> }
>
> -static void ext4_fc_submit_bh(struct super_block *sb)
> +static void ext4_fc_submit_bh(struct super_block *sb, bool is_tail)
> {
> int write_flags = REQ_SYNC;
> struct buffer_head *bh = EXT4_SB(sb)->s_fc_bh;
>
> - /* TODO: REQ_FUA | REQ_PREFLUSH is unnecessarily expensive. */
> - if (test_opt(sb, BARRIER))
> + /* Add REQ_FUA | REQ_PREFLUSH only its tail */
> + if (test_opt(sb, BARRIER) && is_tail)
> write_flags |= REQ_FUA | REQ_PREFLUSH;
> lock_buffer(bh);
> set_buffer_dirty(bh);
> @@ -684,7 +684,7 @@ static u8 *ext4_fc_reserve_space(struct super_block *sb, int len, u32 *crc)
> *crc = ext4_chksum(sbi, *crc, tl, sizeof(*tl));
> if (pad_len > 0)
> ext4_fc_memzero(sb, tl + 1, pad_len, crc);
> - ext4_fc_submit_bh(sb);
> + ext4_fc_submit_bh(sb, false);
>
> ret = jbd2_fc_get_buf(EXT4_SB(sb)->s_journal, &bh);
> if (ret)
> @@ -741,7 +741,7 @@ static int ext4_fc_write_tail(struct super_block *sb, u32 crc)
> tail.fc_crc = cpu_to_le32(crc);
> ext4_fc_memcpy(sb, dst, &tail.fc_crc, sizeof(tail.fc_crc), NULL);
>
> - ext4_fc_submit_bh(sb);
> + ext4_fc_submit_bh(sb, true);
>
> return 0;
> }
> --
> 2.25.1
>
^ permalink raw reply [flat|nested] 6+ messages in thread* Re: [PATCH] ext4: Remove expensive flush on fast commit
2021-01-06 21:58 ` harshad shirwadkar
@ 2021-01-08 2:26 ` Andreas Dilger
[not found] ` <CGME20210106013242epcms2p5b6b4ed8ca86f29456fdf56aa580e74b4@epcms2p3>
1 sibling, 0 replies; 6+ messages in thread
From: Andreas Dilger @ 2021-01-08 2:26 UTC (permalink / raw)
To: daejun7.park, harshad shirwadkar; +Cc: tytso, linux-ext4, linux-kernel
[-- Attachment #1: Type: text/plain, Size: 2603 bytes --]
On Tue, Jan 5, 2021 at 5:32 PM Daejun Park <daejun7.park@samsung.com> wrote:
>
> In the fast commit, it adds REQ_FUA and REQ_PREFLUSH on each fast commit
> block when barrier is enabled. However, in recovery phase, ext4 compares
> CRC value in the tail. So it is sufficient adds REQ_FUA and REQ_PREFLUSH
> on the block that has tail.
Does the tail block *always* contain a CRC, or is that dependent on
EXT4_FEATURE_RO_COMPAT_METADATA_CSUM, JBD2_FEATURE_INCOMPAT_CSUM_V2,
or JBD2_FEATURE_INCOMPAT_CSUM_V3 being enabled?
If one of those features is *required* before the FAST_COMMIT feature
can be used, then this patch looks OK. Otherwise, the CSUM feature
should be checked before the FUA is skipped for non-tail blocks.
Cheers, Andreas
> Signed-off-by: Daejun Park <daejun7.park@samsung.com>
> ---
> fs/ext4/fast_commit.c | 10 +++++-----
> 1 file changed, 5 insertions(+), 5 deletions(-)
>
> diff --git a/fs/ext4/fast_commit.c b/fs/ext4/fast_commit.c
> index 4fcc21c25e79..e66507be334c 100644
> --- a/fs/ext4/fast_commit.c
> +++ b/fs/ext4/fast_commit.c
> @@ -604,13 +604,13 @@ void ext4_fc_track_range(handle_t *handle, struct inode *inode, ext4_lblk_t star
> trace_ext4_fc_track_range(inode, start, end, ret);
> }
>
> -static void ext4_fc_submit_bh(struct super_block *sb)
> +static void ext4_fc_submit_bh(struct super_block *sb, bool is_tail)
> {
> int write_flags = REQ_SYNC;
> struct buffer_head *bh = EXT4_SB(sb)->s_fc_bh;
>
> - /* TODO: REQ_FUA | REQ_PREFLUSH is unnecessarily expensive. */
> - if (test_opt(sb, BARRIER))
> + /* Add REQ_FUA | REQ_PREFLUSH only its tail */
> + if (test_opt(sb, BARRIER) && is_tail)
> write_flags |= REQ_FUA | REQ_PREFLUSH;
> lock_buffer(bh);
> set_buffer_dirty(bh);
> @@ -684,7 +684,7 @@ static u8 *ext4_fc_reserve_space(struct super_block *sb, int len, u32 *crc)
> *crc = ext4_chksum(sbi, *crc, tl, sizeof(*tl));
> if (pad_len > 0)
> ext4_fc_memzero(sb, tl + 1, pad_len, crc);
> - ext4_fc_submit_bh(sb);
> + ext4_fc_submit_bh(sb, false);
>
> ret = jbd2_fc_get_buf(EXT4_SB(sb)->s_journal, &bh);
> if (ret)
> @@ -741,7 +741,7 @@ static int ext4_fc_write_tail(struct super_block *sb, u32 crc)
> tail.fc_crc = cpu_to_le32(crc);
> ext4_fc_memcpy(sb, dst, &tail.fc_crc, sizeof(tail.fc_crc), NULL);
>
> - ext4_fc_submit_bh(sb);
> + ext4_fc_submit_bh(sb, true);
>
> return 0;
> }
> --
> 2.25.1
Cheers, Andreas
[-- Attachment #2: Message signed with OpenPGP --]
[-- Type: application/pgp-signature, Size: 873 bytes --]
^ permalink raw reply [flat|nested] 6+ messages in thread[parent not found: <CGME20210106013242epcms2p5b6b4ed8ca86f29456fdf56aa580e74b4@epcms2p3>]
* RE: Re: [PATCH] ext4: Remove expensive flush on fast commit
[not found] ` <CGME20210106013242epcms2p5b6b4ed8ca86f29456fdf56aa580e74b4@epcms2p3>
@ 2021-01-08 2:37 ` Daejun Park
0 siblings, 0 replies; 6+ messages in thread
From: Daejun Park @ 2021-01-08 2:37 UTC (permalink / raw)
To: Andreas Dilger, Daejun Park, harshad shirwadkar
Cc: tytso, linux-ext4, linux-kernel
> > In the fast commit, it adds REQ_FUA and REQ_PREFLUSH on each fast commit
> > block when barrier is enabled. However, in recovery phase, ext4 compares
> > CRC value in the tail. So it is sufficient adds REQ_FUA and REQ_PREFLUSH
> > on the block that has tail.
>
> Does the tail block *always* contain a CRC, or is that dependent on
> EXT4_FEATURE_RO_COMPAT_METADATA_CSUM, JBD2_FEATURE_INCOMPAT_CSUM_V2,
> or JBD2_FEATURE_INCOMPAT_CSUM_V3 being enabled?
In the fast commit, the tail block always contain a CRC.
> If one of those features is *required* before the FAST_COMMIT feature
> can be used, then this patch looks OK. Otherwise, the CSUM feature
> should be checked before the FUA is skipped for non-tail blocks.
So, I think it is OK without checking other CSUM feature.
Thanks,
Daejun
^ permalink raw reply [flat|nested] 6+ messages in thread
* Re: [PATCH] ext4: Remove expensive flush on fast commit
2021-01-06 1:32 ` [PATCH] ext4: Remove expensive flush on fast commit Daejun Park
2021-01-06 21:58 ` harshad shirwadkar
@ 2021-01-06 22:05 ` Darrick J. Wong
2021-01-14 3:44 ` Theodore Ts'o
2 siblings, 0 replies; 6+ messages in thread
From: Darrick J. Wong @ 2021-01-06 22:05 UTC (permalink / raw)
To: Daejun Park
Cc: tytso, adilger.kernel, harshadshirwadkar, linux-ext4, linux-kernel
On Wed, Jan 06, 2021 at 10:32:42AM +0900, Daejun Park wrote:
> In the fast commit, it adds REQ_FUA and REQ_PREFLUSH on each fast commit
> block when barrier is enabled. However, in recovery phase, ext4 compares
> CRC value in the tail. So it is sufficient adds REQ_FUA and REQ_PREFLUSH
> on the block that has tail.
Just out of curiosity, how does the new fastcommit code handle tests
like generic/475 that simulate sudden disk failure?
--D
> Signed-off-by: Daejun Park <daejun7.park@samsung.com>
> ---
> fs/ext4/fast_commit.c | 10 +++++-----
> 1 file changed, 5 insertions(+), 5 deletions(-)
>
> diff --git a/fs/ext4/fast_commit.c b/fs/ext4/fast_commit.c
> index 4fcc21c25e79..e66507be334c 100644
> --- a/fs/ext4/fast_commit.c
> +++ b/fs/ext4/fast_commit.c
> @@ -604,13 +604,13 @@ void ext4_fc_track_range(handle_t *handle, struct inode *inode, ext4_lblk_t star
> trace_ext4_fc_track_range(inode, start, end, ret);
> }
>
> -static void ext4_fc_submit_bh(struct super_block *sb)
> +static void ext4_fc_submit_bh(struct super_block *sb, bool is_tail)
> {
> int write_flags = REQ_SYNC;
> struct buffer_head *bh = EXT4_SB(sb)->s_fc_bh;
>
> - /* TODO: REQ_FUA | REQ_PREFLUSH is unnecessarily expensive. */
> - if (test_opt(sb, BARRIER))
> + /* Add REQ_FUA | REQ_PREFLUSH only its tail */
> + if (test_opt(sb, BARRIER) && is_tail)
> write_flags |= REQ_FUA | REQ_PREFLUSH;
> lock_buffer(bh);
> set_buffer_dirty(bh);
> @@ -684,7 +684,7 @@ static u8 *ext4_fc_reserve_space(struct super_block *sb, int len, u32 *crc)
> *crc = ext4_chksum(sbi, *crc, tl, sizeof(*tl));
> if (pad_len > 0)
> ext4_fc_memzero(sb, tl + 1, pad_len, crc);
> - ext4_fc_submit_bh(sb);
> + ext4_fc_submit_bh(sb, false);
>
> ret = jbd2_fc_get_buf(EXT4_SB(sb)->s_journal, &bh);
> if (ret)
> @@ -741,7 +741,7 @@ static int ext4_fc_write_tail(struct super_block *sb, u32 crc)
> tail.fc_crc = cpu_to_le32(crc);
> ext4_fc_memcpy(sb, dst, &tail.fc_crc, sizeof(tail.fc_crc), NULL);
>
> - ext4_fc_submit_bh(sb);
> + ext4_fc_submit_bh(sb, true);
>
> return 0;
> }
> --
> 2.25.1
>
^ permalink raw reply [flat|nested] 6+ messages in thread* Re: [PATCH] ext4: Remove expensive flush on fast commit
2021-01-06 1:32 ` [PATCH] ext4: Remove expensive flush on fast commit Daejun Park
2021-01-06 21:58 ` harshad shirwadkar
2021-01-06 22:05 ` Darrick J. Wong
@ 2021-01-14 3:44 ` Theodore Ts'o
2 siblings, 0 replies; 6+ messages in thread
From: Theodore Ts'o @ 2021-01-14 3:44 UTC (permalink / raw)
To: Daejun Park; +Cc: adilger.kernel, harshadshirwadkar, linux-ext4, linux-kernel
On Wed, Jan 06, 2021 at 10:32:42AM +0900, Daejun Park wrote:
> In the fast commit, it adds REQ_FUA and REQ_PREFLUSH on each fast commit
> block when barrier is enabled. However, in recovery phase, ext4 compares
> CRC value in the tail. So it is sufficient adds REQ_FUA and REQ_PREFLUSH
> on the block that has tail.
>
> Signed-off-by: Daejun Park <daejun7.park@samsung.com>
Thanks, applied.
- Ted
^ permalink raw reply [flat|nested] 6+ messages in thread