On 2018/2/12 14:55, Yunlei He wrote:
> This patch introduce a method to make nat journal more fresh:
> i.  sort set list using entry # plus cp version difference (last modify
> set cp # and current cp #)
> ii. if meet with cache hit, update last modify cp # to current cp #

As we discuss off-line, I think we can record weighted average of cpver for all
dirty nat entries into nat set.

Thanks,

> 
> Signed-off-by: Yunlei He <heyun...@huawei.com>
> ---
>  fs/f2fs/f2fs.h |  2 ++
>  fs/f2fs/node.c | 57 +++++++++++++++++++++++++++++++++++++++++----------------
>  fs/f2fs/node.h |  2 ++
>  3 files changed, 45 insertions(+), 16 deletions(-)
> 
> diff --git a/fs/f2fs/f2fs.h b/fs/f2fs/f2fs.h
> index b7ba496..bd1e775 100644
> --- a/fs/f2fs/f2fs.h
> +++ b/fs/f2fs/f2fs.h
> @@ -297,6 +297,8 @@ struct fsync_inode_entry {
>       block_t last_dentry;    /* block address locating the last dentry */
>  };
>  
> +#define DEF_NAT_FACTOR 10
> +
>  #define nats_in_cursum(jnl)          (le16_to_cpu((jnl)->n_nats))
>  #define sits_in_cursum(jnl)          (le16_to_cpu((jnl)->n_sits))
>  
> diff --git a/fs/f2fs/node.c b/fs/f2fs/node.c
> index ced986d..f3f89e3 100644
> --- a/fs/f2fs/node.c
> +++ b/fs/f2fs/node.c
> @@ -195,8 +195,8 @@ static void __del_from_nat_cache(struct f2fs_nm_info 
> *nm_i, struct nat_entry *e)
>       __free_nat_entry(e);
>  }
>  
> -static void __set_nat_cache_dirty(struct f2fs_nm_info *nm_i,
> -                                             struct nat_entry *ne)
> +static void __set_nat_cache_dirty(struct f2fs_sb_info *sbi, bool 
> from_journal,
> +                     struct f2fs_nm_info *nm_i, struct nat_entry *ne)
>  {
>       nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
>       struct nat_entry_set *head;
> @@ -209,9 +209,15 @@ static void __set_nat_cache_dirty(struct f2fs_nm_info 
> *nm_i,
>               INIT_LIST_HEAD(&head->set_list);
>               head->set = set;
>               head->entry_cnt = 0;
> +             head->to_journal = false;
> +             head->cp_ver = cur_cp_version(F2FS_CKPT(sbi));
>               f2fs_radix_tree_insert(&nm_i->nat_set_root, set, head);
>       }
>  
> +     /* journal hit case, try to locate set in journal */
> +     if (head->to_journal && !from_journal)
> +             head->cp_ver = cur_cp_version(F2FS_CKPT(sbi));
> +
>       if (get_nat_flag(ne, IS_DIRTY))
>               goto refresh_list;
>  
> @@ -359,7 +365,7 @@ static void set_node_addr(struct f2fs_sb_info *sbi, 
> struct node_info *ni,
>       nat_set_blkaddr(e, new_blkaddr);
>       if (new_blkaddr == NEW_ADDR || new_blkaddr == NULL_ADDR)
>               set_nat_flag(e, IS_CHECKPOINTED, false);
> -     __set_nat_cache_dirty(nm_i, e);
> +     __set_nat_cache_dirty(sbi, false, nm_i, e);
>  
>       /* update fsync_mark if its inode nat entry is still alive */
>       if (ni->nid != ni->ino)
> @@ -2397,14 +2403,30 @@ static void remove_nats_in_journal(struct 
> f2fs_sb_info *sbi)
>                       spin_unlock(&nm_i->nid_list_lock);
>               }
>  
> -             __set_nat_cache_dirty(nm_i, ne);
> +             __set_nat_cache_dirty(sbi, true, nm_i, ne);
>       }
>       update_nats_in_cursum(journal, -i);
>       up_write(&curseg->journal_rwsem);
>  }
>  
> -static void __adjust_nat_entry_set(struct nat_entry_set *nes,
> -                                             struct list_head *head, int max)
> +static bool cost_compare(struct f2fs_sb_info *sbi,
> +                     struct nat_entry_set *nes, struct nat_entry_set * cur)
> +{
> +     struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
> +     __u64 cost1 = nes->entry_cnt * DEF_NAT_FACTOR +
> +                     (nes->cp_ver <= cur_cp_version(ckpt)) ?
> +                     cur_cp_version(ckpt) - nes->cp_ver :
> +                     ULONG_MAX - nes->cp_ver + cur_cp_version(ckpt);
> +     __u64 cost2 = cur->entry_cnt * DEF_NAT_FACTOR +
> +                     (cur->cp_ver <= cur_cp_version(ckpt)) ?
> +                     cur_cp_version(ckpt) - cur->cp_ver :
> +                     ULONG_MAX - cur->cp_ver + cur_cp_version(ckpt);
> +
> +     return cost1 <= cost2;
> +}
> +
> +static void __adjust_nat_entry_set(struct f2fs_sb_info *sbi,
> +                     struct nat_entry_set *nes, struct list_head *head, int 
> max)
>  {
>       struct nat_entry_set *cur;
>  
> @@ -2412,7 +2434,7 @@ static void __adjust_nat_entry_set(struct nat_entry_set 
> *nes,
>               goto add_out;
>  
>       list_for_each_entry(cur, head, set_list) {
> -             if (cur->entry_cnt >= nes->entry_cnt) {
> +             if (cost_compare(sbi, nes, cur)) {
>                       list_add(&nes->set_list, cur->set_list.prev);
>                       return;
>               }
> @@ -2460,7 +2482,6 @@ static void __flush_nat_entry_set(struct f2fs_sb_info 
> *sbi,
>       struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
>       struct f2fs_journal *journal = curseg->journal;
>       nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
> -     bool to_journal = true;
>       struct f2fs_nat_block *nat_blk;
>       struct nat_entry *ne, *cur;
>       struct page *page = NULL;
> @@ -2470,11 +2491,13 @@ static void __flush_nat_entry_set(struct f2fs_sb_info 
> *sbi,
>        * #1, flush nat entries to journal in current hot data summary block.
>        * #2, flush nat entries to nat page.
>        */
> +     set->to_journal = true;
> +
>       if (enabled_nat_bits(sbi, cpc) ||
>               !__has_cursum_space(journal, set->entry_cnt, NAT_JOURNAL))
> -             to_journal = false;
> +             set->to_journal = false;
>  
> -     if (to_journal) {
> +     if (set->to_journal) {
>               down_write(&curseg->journal_rwsem);
>       } else {
>               page = get_next_nat_page(sbi, start_nid);
> @@ -2490,7 +2513,7 @@ static void __flush_nat_entry_set(struct f2fs_sb_info 
> *sbi,
>  
>               f2fs_bug_on(sbi, nat_get_blkaddr(ne) == NEW_ADDR);
>  
> -             if (to_journal) {
> +             if (set->to_journal) {
>                       offset = lookup_journal_in_cursum(journal,
>                                                       NAT_JOURNAL, nid, 1);
>                       f2fs_bug_on(sbi, offset < 0);
> @@ -2511,17 +2534,19 @@ static void __flush_nat_entry_set(struct f2fs_sb_info 
> *sbi,
>               }
>       }
>  
> -     if (to_journal) {
> +     if (set->to_journal) {
>               up_write(&curseg->journal_rwsem);
>       } else {
>               __update_nat_bits(sbi, start_nid, page);
>               f2fs_put_page(page, 1);
>       }
>  
> -     /* Allow dirty nats by node block allocation in write_begin */
> +     if (!set->to_journal || cpc->reason & CP_UMOUNT) {
> +             radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
> +             kmem_cache_free(nat_entry_set_slab, set);
> +     }
> +
>       f2fs_bug_on(sbi, set->entry_cnt);
> -     radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
> -     kmem_cache_free(nat_entry_set_slab, set);
>  }
>  
>  /*
> @@ -2557,7 +2582,7 @@ void flush_nat_entries(struct f2fs_sb_info *sbi, struct 
> cp_control *cpc)
>               unsigned idx;
>               set_idx = setvec[found - 1]->set + 1;
>               for (idx = 0; idx < found; idx++)
> -                     __adjust_nat_entry_set(setvec[idx], &sets,
> +                     __adjust_nat_entry_set(sbi, setvec[idx], &sets,
>                                               MAX_NAT_JENTRIES(journal));
>       }
>  
> diff --git a/fs/f2fs/node.h b/fs/f2fs/node.h
> index 081ef0d..aba5ada 100644
> --- a/fs/f2fs/node.h
> +++ b/fs/f2fs/node.h
> @@ -149,6 +149,8 @@ struct nat_entry_set {
>       struct list_head entry_list;    /* link with dirty nat entries */
>       nid_t set;                      /* set number*/
>       unsigned int entry_cnt;         /* the # of nat entries in set */
> +     __u64 cp_ver;                   /* cp version of this set modify */
> +     bool to_journal;                /* set flush to journal */
>  };
>  
>  struct free_nid {
> 


------------------------------------------------------------------------------
Check out the vibrant tech community on one of the world's most
engaging tech sites, Slashdot.org! http://sdm.link/slashdot
_______________________________________________
Linux-f2fs-devel mailing list
Linux-f2fs-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel

Reply via email to