Re: [PATCH] f2fs: enable atgc if atgc_age_threshold from user is less than elapsed_time

From: Zhiguo Niu
Date: Wed Jun 05 2024 - 02:00:03 EST

On Wed, Jun 5, 2024 at 11:48 AM Chao Yu <chao@xxxxxxxxxx> wrote:
>
> On 2024/6/3 17:05, Zhiguo Niu wrote:
> > On Mon, Jun 3, 2024 at 3:41 PM Chao Yu <chao@xxxxxxxxxx> wrote:
> >>
> >> On 2024/5/20 19:36, Zhiguo Niu wrote:
> >>> Now atgc can be enabled based on the following conditions:
> >>> -ATGC mount option is set
> >>> -elapsed_time is more than atgc_age_threshold already
> >>> but these conditions are check when umounted->mounted device again.
> >>> If the device has not be umounted->mounted for a long time, atgc can
> >>> not work even the above conditions met.
> >>>
> >>> It is better to enable atgc dynamiclly when user change atgc_age_threshold
> >>> meanwhile this vale is less than elapsed_time and ATGC mount option is on.
> >>>
> >>> Signed-off-by: Zhiguo Niu <zhiguo.niu@xxxxxxxxxx>
> >>> ---
> >>> fs/f2fs/f2fs.h | 1 +
> >>> fs/f2fs/segment.c | 9 ++++-----
> >>> fs/f2fs/sysfs.c | 16 ++++++++++++++++
> >>> 3 files changed, 21 insertions(+), 5 deletions(-)
> >>>
> >>> diff --git a/fs/f2fs/f2fs.h b/fs/f2fs/f2fs.h
> >>> index 1974b6a..e441d2d 100644
> >>> --- a/fs/f2fs/f2fs.h
> >>> +++ b/fs/f2fs/f2fs.h
> >>> @@ -3694,6 +3694,7 @@ void f2fs_clear_prefree_segments(struct f2fs_sb_info *sbi,
> >>> int f2fs_init_inmem_curseg(struct f2fs_sb_info *sbi);
> >>> void f2fs_save_inmem_curseg(struct f2fs_sb_info *sbi);
> >>> void f2fs_restore_inmem_curseg(struct f2fs_sb_info *sbi);
> >>> +int f2fs_init_atgc_curseg(struct f2fs_sb_info *sbi);
> >>> int f2fs_allocate_segment_for_resize(struct f2fs_sb_info *sbi, int type,
> >>> unsigned int start, unsigned int end);
> >>> int f2fs_allocate_new_section(struct f2fs_sb_info *sbi, int type, bool force);
> >>> diff --git a/fs/f2fs/segment.c b/fs/f2fs/segment.c
> >>> index 71dc8042..44923d4 100644
> >>> --- a/fs/f2fs/segment.c
> >>> +++ b/fs/f2fs/segment.c
> >>> @@ -2931,14 +2931,11 @@ static int get_atssr_segment(struct f2fs_sb_info *sbi, int type,
> >>> return ret;
> >>> }
> >>>
> >>> -static int __f2fs_init_atgc_curseg(struct f2fs_sb_info *sbi)
> >>> +int f2fs_init_atgc_curseg(struct f2fs_sb_info *sbi)
> >>> {
> >>> struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_ALL_DATA_ATGC);
> >>> int ret = 0;
> >>>
> >>> - if (!sbi->am.atgc_enabled)
> >>> - return 0;
> >>> -
> >>> f2fs_down_read(&SM_I(sbi)->curseg_lock);
> >>>
> >>> mutex_lock(&curseg->curseg_mutex);
> >>> @@ -2955,7 +2952,9 @@ static int __f2fs_init_atgc_curseg(struct f2fs_sb_info *sbi)
> >>> }
> >>> int f2fs_init_inmem_curseg(struct f2fs_sb_info *sbi)
> >>> {
> >>> - return __f2fs_init_atgc_curseg(sbi);
> >>> + if (!sbi->am.atgc_enabled)
> >>> + return 0;
> >>> + return f2fs_init_atgc_curseg(sbi);
> >>> }
> >>>
> >>> static void __f2fs_save_inmem_curseg(struct f2fs_sb_info *sbi, int type)
> >>> diff --git a/fs/f2fs/sysfs.c b/fs/f2fs/sysfs.c
> >>> index 09d3ecf..72676c5 100644
> >>> --- a/fs/f2fs/sysfs.c
> >>> +++ b/fs/f2fs/sysfs.c
> >>> @@ -673,6 +673,22 @@ static ssize_t __sbi_store(struct f2fs_attr *a,
> >>> return count;
> >>> }
> >>>
> >>> + if (!strcmp(a->attr.name, "atgc_age_threshold")) {
> >>> + if (t < 0)
> >>> + return -EINVAL;
> >>> + sbi->am.age_threshold = t;
> >>> + if (sbi->am.atgc_enabled)
> >>> + return count;
> >>> +
> >>> + if (test_opt(sbi, ATGC) &&
> >>> + le64_to_cpu(sbi->ckpt->elapsed_time) >= t) {
> >>
> > Hi Chao,
> >> Oh, I guess you want to fix this:
> > Yes, Sorry for not making it clear before.
> >>
> >> static void init_atgc_management(struct f2fs_sb_info *sbi)
> >> {
> >> ...
> >> if (test_opt(sbi, ATGC) &&
> >> SIT_I(sbi)->elapsed_time >= DEF_GC_THREAD_AGE_THRESHOLD)
> >> am->atgc_enabled = true;
> >>
> >> What about enabling atgc_enabled during checkpoint once elapsed time is
> >> satisfed w/ the condition? I guess this can give another chance whenever
> >> CP is been triggered, and it can avoid the racing condition as well.
> > 1. I'm not sure if this will increase checkpoint time consumption?
>
> Since it won't increase any IO time, I guess it's fine tolerate time consumed
> by initialization of atgc curseg.
>
> > 2. dynamically enabling atgc may have other problems. Is this confirmed?
>
> I think so, checkpoint has avoided most race cases.
>
> So, how do you think of below diff:
Hi Chao,
I think flow is ok, some details need to be discussed.
>
> ---
> fs/f2fs/checkpoint.c | 2 ++
> fs/f2fs/f2fs.h | 1 +
> fs/f2fs/segment.c | 26 +++++++++++++++++++++++---
> 3 files changed, 26 insertions(+), 3 deletions(-)
>
> diff --git a/fs/f2fs/checkpoint.c b/fs/f2fs/checkpoint.c
> index 55d444bec5c0..4a73bd481a25 100644
> --- a/fs/f2fs/checkpoint.c
> +++ b/fs/f2fs/checkpoint.c
> @@ -1718,6 +1718,8 @@ int f2fs_write_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
> }
>
> f2fs_restore_inmem_curseg(sbi);
> + f2fs_reinit_atgc_curseg(sbi);
> +
> stat_inc_cp_count(sbi);
> stop:
> unblock_operations(sbi);
> diff --git a/fs/f2fs/f2fs.h b/fs/f2fs/f2fs.h
> index 9688df332147..8d385a1c75ad 100644
> --- a/fs/f2fs/f2fs.h
> +++ b/fs/f2fs/f2fs.h
> @@ -3693,6 +3693,7 @@ void f2fs_release_discard_addrs(struct f2fs_sb_info *sbi);
> int f2fs_npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra);
> bool f2fs_segment_has_free_slot(struct f2fs_sb_info *sbi, int segno);
> int f2fs_init_inmem_curseg(struct f2fs_sb_info *sbi);
> +int f2fs_reinit_atgc_curseg(struct f2fs_sb_info *sbi);
> void f2fs_save_inmem_curseg(struct f2fs_sb_info *sbi);
> void f2fs_restore_inmem_curseg(struct f2fs_sb_info *sbi);
> int f2fs_allocate_segment_for_resize(struct f2fs_sb_info *sbi, int type,
> diff --git a/fs/f2fs/segment.c b/fs/f2fs/segment.c
> index bdc2247387e8..6d4273faf061 100644
> --- a/fs/f2fs/segment.c
> +++ b/fs/f2fs/segment.c
> @@ -2949,12 +2949,12 @@ static int get_atssr_segment(struct f2fs_sb_info *sbi, int type,
> return ret;
> }
>
> -static int __f2fs_init_atgc_curseg(struct f2fs_sb_info *sbi)
> +static int __f2fs_init_atgc_curseg(struct f2fs_sb_info *sbi, bool force)
> {
> struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_ALL_DATA_ATGC);
> int ret = 0;
>
> - if (!sbi->am.atgc_enabled)
> + if (!sbi->am.atgc_enabled && !force)
Is this parameter "force" unnecessary?
in mount flow, even atgc enable conditions are all met, it is not
enabled because of force=false.
Or am I missing something?
> return 0;
>
> f2fs_down_read(&SM_I(sbi)->curseg_lock);
> @@ -2971,9 +2971,29 @@ static int __f2fs_init_atgc_curseg(struct f2fs_sb_info *sbi)
> f2fs_up_read(&SM_I(sbi)->curseg_lock);
> return ret;
> }
> +
> int f2fs_init_inmem_curseg(struct f2fs_sb_info *sbi)
> {
> - return __f2fs_init_atgc_curseg(sbi);
> + return __f2fs_init_atgc_curseg(sbi, false);
> +}
> +
> +int f2fs_reinit_atgc_curseg(struct f2fs_sb_info *sbi)
> +{
> + int ret;
> +
> + if (!test_opt(sbi, ATGC))
> + return 0;
> + if (sbi->am.atgc_enabled)
> + return 0;
> + if (SIT_I(sbi)->elapsed_time < sbi->am.age_threshold)
SIT(sbi)->elapsed_time is just updated in mount flow, so
ckpt->elapsed_time is more suitable here?
thanks!
> + return 0;
> +
> + ret = __f2fs_init_atgc_curseg(sbi, true);
> + if (!ret) {
> + sbi->am.atgc_enabled = true;
> + f2fs_info(sbi, "reenabled age threshold GC");
> + }
> + return ret;
> }
>
> static void __f2fs_save_inmem_curseg(struct f2fs_sb_info *sbi, int type)
> --
> 2.40.1
>
> > thanks！
> >>
> >> Thanks,
> >>
> >>> + if (f2fs_init_atgc_curseg(sbi))
> >>> + return -EINVAL;
> >>> + sbi->am.atgc_enabled = true;
> >>> + }
> >>> + return count;
> >>> + }
> >>> +
> >>> if (!strcmp(a->attr.name, "gc_segment_mode")) {
> >>> if (t < MAX_GC_MODE)
> >>> sbi->gc_segment_mode = t;