Jaegeuk,

Let's consider this patch in next Linux version.

BTW, do you still encounter consistency problem w/o this version?

On 2020/6/8 20:01, Chao Yu wrote:
>     71.07%     0.01%  kworker/u256:1+  [kernel.kallsyms]  [k] wb_writeback
>             |
>              --71.06%--wb_writeback
>                        |
>                        |--68.96%--__writeback_inodes_wb
>                        |          |
>                        |           --68.95%--writeback_sb_inodes
>                        |                     |
>                        |                     
> |--65.08%--__writeback_single_inode
>                        |                     |          |
>                        |                     |           
> --64.35%--do_writepages
>                        |                     |                     |
>                        |                     |                     
> |--59.83%--f2fs_write_node_pages
>                        |                     |                     |          
> |
>                        |                     |                     |          
>  --59.74%--f2fs_sync_node_pages
>                        |                     |                     |          
>            |
>                        |                     |                     |          
>            |--27.91%--pagevec_lookup_range_tag
>                        |                     |                     |          
>            |          |
>                        |                     |                     |          
>            |           --27.90%--find_get_pages_range_tag
> 
> If filesystem was injected checkpoint errror, before umount, kworker will
> always hold one core in order to writeback a large number of node pages,
> that looks not reasonable, to avoid that, we can allow data/node write in
> such case, since we can force all data/node writes with OPU mode, and
> clear recovery flag on node, and checkpoint is not allowed as well, so we
> don't need to worry about writeback's effect on data/node in previous
> checkpoint, then with this way, it can decrease memory footprint cost by
> node/data pages and avoid looping into data/node writeback process.
> 
> Signed-off-by: Chao Yu <yuch...@huawei.com>
> ---
>  fs/f2fs/checkpoint.c |  7 +++++++
>  fs/f2fs/data.c       | 22 +++++++++++++++-------
>  fs/f2fs/node.c       |  7 +++++--
>  3 files changed, 27 insertions(+), 9 deletions(-)
> 
> diff --git a/fs/f2fs/checkpoint.c b/fs/f2fs/checkpoint.c
> index 236064930251..1bb8278a1c4a 100644
> --- a/fs/f2fs/checkpoint.c
> +++ b/fs/f2fs/checkpoint.c
> @@ -1238,6 +1238,13 @@ static int block_operations(struct f2fs_sb_info *sbi)
>               goto retry_flush_nodes;
>       }
>  
> +     if (unlikely(f2fs_cp_error(sbi))) {
> +             up_write(&sbi->node_write);
> +             up_write(&sbi->node_change);
> +             f2fs_unlock_all(sbi);
> +             return -EIO;
> +     }
> +
>       /*
>        * sbi->node_change is used only for AIO write_begin path which produces
>        * dirty node blocks and some checkpoint values by block allocation.
> diff --git a/fs/f2fs/data.c b/fs/f2fs/data.c
> index 267b5e76a02b..808e7734db19 100644
> --- a/fs/f2fs/data.c
> +++ b/fs/f2fs/data.c
> @@ -2508,6 +2508,8 @@ bool f2fs_should_update_outplace(struct inode *inode, 
> struct f2fs_io_info *fio)
>  {
>       struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
>  
> +     if (unlikely(f2fs_cp_error(sbi)))
> +             return true;
>       if (f2fs_lfs_mode(sbi))
>               return true;
>       if (S_ISDIR(inode->i_mode))
> @@ -2691,13 +2693,19 @@ int f2fs_write_single_data_page(struct page *page, 
> int *submitted,
>       /* we should bypass data pages to proceed the kworkder jobs */
>       if (unlikely(f2fs_cp_error(sbi))) {
>               mapping_set_error(page->mapping, -EIO);
> -             /*
> -              * don't drop any dirty dentry pages for keeping lastest
> -              * directory structure.
> -              */
> -             if (S_ISDIR(inode->i_mode))
> -                     goto redirty_out;
> -             goto out;
> +
> +             if (is_sbi_flag_set(sbi, SBI_IS_CLOSE))
> +                     goto out;
> +
> +             if (has_not_enough_free_secs(sbi, 0, 0)) {
> +                     /*
> +                      * don't drop any dirty dentry pages for keeping lastest
> +                      * directory structure.
> +                      */
> +                     if (S_ISDIR(inode->i_mode))
> +                             goto redirty_out;
> +                     goto out;
> +             }
>       }
>  
>       if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
> diff --git a/fs/f2fs/node.c b/fs/f2fs/node.c
> index 03e24df1c84f..372c04efad38 100644
> --- a/fs/f2fs/node.c
> +++ b/fs/f2fs/node.c
> @@ -1527,7 +1527,10 @@ static int __write_node_page(struct page *page, bool 
> atomic, bool *submitted,
>                       unlock_page(page);
>                       return 0;
>               }
> -             goto redirty_out;
> +             if (has_not_enough_free_secs(sbi, 0, 0))
> +                     goto redirty_out;
> +             set_fsync_mark(page, 0);
> +             set_dentry_mark(page, 0);
>       }
>  
>       if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
> @@ -1568,7 +1571,7 @@ static int __write_node_page(struct page *page, bool 
> atomic, bool *submitted,
>               goto redirty_out;
>       }
>  
> -     if (atomic && !test_opt(sbi, NOBARRIER))
> +     if (atomic && !test_opt(sbi, NOBARRIER) && !f2fs_cp_error(sbi))
>               fio.op_flags |= REQ_PREFLUSH | REQ_FUA;
>  
>       /* should add to global list before clearing PAGECACHE status */
> 

Reply via email to