Re: [PATCH] f2fs:modify the entering condition for f2fs_migrate_blocks()

[Chao Yu]

On 2024/5/15 16:24, Liao Yuanhong wrote:
> Currently, when we allocating a swap file on zone UFS, this file will
> created on conventional UFS. If the swap file size is not aligned with the
> zone size, the last extent will enter f2fs_migrate_blocks(), resulting in
> significant additional I/O overhead and prolonged lock occupancy. In most
> cases, this is unnecessary, because on Conventional UFS, as long as the
> start block of the swap file is aligned with zone, it is sequentially
> aligned.To circumvent this issue, we have altered the conditions for
> entering f2fs_migrate_blocks(). Now, if the start block of the last extent
> is aligned with the start of zone, we avoids entering
> f2fs_migrate_blocks().

Hi,

Is it possible that we can pin swapfile, and fallocate on it aligned to
zone size, then mkswap and swapon?

Thanks,

> Signed-off-by: Liao Yuanhong <liaoyuanhong@xxxxxxxx>
> Signed-off-by: Wu Bo <bo.wu@xxxxxxxx>
> ---
>   fs/f2fs/data.c | 23 +++++++++++++++++++++--
>   1 file changed, 21 insertions(+), 2 deletions(-)
>
> diff --git a/fs/f2fs/data.c b/fs/f2fs/data.c
> index 50ceb25b3..4d58fb6c2 100644
> --- a/fs/f2fs/data.c
> +++ b/fs/f2fs/data.c
> @@ -3925,10 +3925,12 @@ static int check_swap_activate(struct swap_info_struct *sis,
>          block_t pblock;
>          block_t lowest_pblock = -1;
>          block_t highest_pblock = 0;
> +       block_t blk_start;
>          int nr_extents = 0;
>          unsigned int nr_pblocks;
>          unsigned int blks_per_sec = BLKS_PER_SEC(sbi);
>          unsigned int not_aligned = 0;
> +       unsigned int cur_sec;
>          int ret = 0;
>
>          /*
> @@ -3965,23 +3967,39 @@ static int check_swap_activate(struct swap_info_struct *sis,
>                  pblock = map.m_pblk;
>                  nr_pblocks = map.m_len;
>
> -               if ((pblock - SM_I(sbi)->main_blkaddr) % blks_per_sec ||
> +               blk_start = pblock - SM_I(sbi)->main_blkaddr;
> +
> +               if (blk_start % blks_per_sec ||
>                                  nr_pblocks % blks_per_sec ||
>                                  !f2fs_valid_pinned_area(sbi, pblock)) {
>                          bool last_extent = false;
>
>                          not_aligned++;
>
> +                       cur_sec = (blk_start + nr_pblocks) / BLKS_PER_SEC(sbi);
>                          nr_pblocks = roundup(nr_pblocks, blks_per_sec);
> -                       if (cur_lblock + nr_pblocks > sis->max)
> +                       if (cur_lblock + nr_pblocks > sis->max) {
>                                  nr_pblocks -= blks_per_sec;
>
> +                               /* the start address is aligned to section */
> +                               if (!(blk_start % blks_per_sec))
> +                                       last_extent = true;
> +                       }
> +
>                          /* this extent is last one */
>                          if (!nr_pblocks) {
>                                  nr_pblocks = last_lblock - cur_lblock;
>                                  last_extent = true;
>                          }
>
> +                       /*
> +                        * the last extent which located on conventional UFS doesn't
> +                        * need migrate
> +                        */
> +                       if (last_extent && f2fs_sb_has_blkzoned(sbi) &&
> +                               cur_sec < GET_SEC_FROM_SEG(sbi, first_zoned_segno(sbi)))
> +                               goto next;
> +
>                          ret = f2fs_migrate_blocks(inode, cur_lblock,
>                                                          nr_pblocks);
>                          if (ret) {
> @@ -3994,6 +4012,7 @@ static int check_swap_activate(struct swap_info_struct *sis,
>                                  goto retry;
>                  }
>
> +next:
>                  if (cur_lblock + nr_pblocks >= sis->max)
>                          nr_pblocks = sis->max - cur_lblock;
>
> --
> 2.25.1