Re: [PATCH] f2fs: fix to account total free nid correctly

[Jaegeuk Kim]

On Mon, Nov 14, 2016 at 07:24:56PM +0800, Chao Yu wrote:
> Thread A		Thread B		Thread C
> - f2fs_create
>  - f2fs_new_inode
>   - f2fs_lock_op
>    - alloc_nid
>     alloc last nid
>   - f2fs_unlock_op
> 			- f2fs_create
> 			 - f2fs_new_inode
> 			  - f2fs_lock_op
> 			   - alloc_nid
> 			    as node count still not
> 			    be increased, we will
> 			    loop in alloc_nid
> 						- f2fs_write_node_pages
> 						 - f2fs_balance_fs_bg
> 						  - f2fs_sync_fs
> 						   - write_checkpoint
> 						    - block_operations
> 						     - f2fs_lock_all
>  - f2fs_lock_op
> 
> While creating new inode, we do not allocate and account nid atomically,
> so that when there is almost no free nids left, we may encounter deadloop
> like above stack.
> 
> In order to avoid that, add nm_i::free_nid_cnt for accounting free nids
> and do nid allocation atomically during node creation.

How about using nm_i::avaiable_nids for this?
It seems that we don't need both of variables at the same time.

Thanks,

> 
> Signed-off-by: Chao Yu <yuchao0@xxxxxxxxxx>
> ---
>  fs/f2fs/f2fs.h |  1 +
>  fs/f2fs/node.c | 19 +++++++++++++++----
>  2 files changed, 16 insertions(+), 4 deletions(-)
> 
> diff --git a/fs/f2fs/f2fs.h b/fs/f2fs/f2fs.h
> index 6de1fbf..9de6f20 100644
> --- a/fs/f2fs/f2fs.h
> +++ b/fs/f2fs/f2fs.h
> @@ -551,6 +551,7 @@ struct f2fs_nm_info {
>  	struct radix_tree_root free_nid_root;/* root of the free_nid cache */
>  	struct list_head nid_list[MAX_NID_LIST];/* lists for free nids */
>  	unsigned int nid_cnt[MAX_NID_LIST];	/* the number of free node id */
> +	unsigned int free_nid_cnt;	/* the number of total free nid */
>  	spinlock_t nid_list_lock;	/* protect nid lists ops */
>  	struct mutex build_lock;	/* lock for build free nids */
>  
> diff --git a/fs/f2fs/node.c b/fs/f2fs/node.c
> index d58438f..e412d0e 100644
> --- a/fs/f2fs/node.c
> +++ b/fs/f2fs/node.c
> @@ -1885,11 +1885,13 @@ bool alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
>  		return false;
>  	}
>  #endif
> -	if (unlikely(sbi->total_valid_node_count + 1 > nm_i->available_nids))
> -		return false;
> -
>  	spin_lock(&nm_i->nid_list_lock);
>  
> +	if (unlikely(nm_i->free_nid_cnt == 0)) {
> +		spin_unlock(&nm_i->nid_list_lock);
> +		return false;
> +	}
> +
>  	/* We should not use stale free nids created by build_free_nids */
>  	if (nm_i->nid_cnt[FREE_NID_LIST] && !on_build_free_nids(nm_i)) {
>  		f2fs_bug_on(sbi, list_empty(&nm_i->nid_list[FREE_NID_LIST]));
> @@ -1900,6 +1902,7 @@ bool alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
>  		__remove_nid_from_list(sbi, i, FREE_NID_LIST, true);
>  		i->state = NID_ALLOC;
>  		__insert_nid_to_list(sbi, i, ALLOC_NID_LIST, false);
> +		nm_i->free_nid_cnt--;
>  		spin_unlock(&nm_i->nid_list_lock);
>  		return true;
>  	}
> @@ -1951,6 +1954,9 @@ void alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
>  		i->state = NID_NEW;
>  		__insert_nid_to_list(sbi, i, FREE_NID_LIST, false);
>  	}
> +
> +	nm_i->free_nid_cnt++;
> +
>  	spin_unlock(&nm_i->nid_list_lock);
>  
>  	if (need_free)
> @@ -2222,8 +2228,12 @@ static void __flush_nat_entry_set(struct f2fs_sb_info *sbi,
>  		raw_nat_from_node_info(raw_ne, &ne->ni);
>  		nat_reset_flag(ne);
>  		__clear_nat_cache_dirty(NM_I(sbi), ne);
> -		if (nat_get_blkaddr(ne) == NULL_ADDR)
> +		if (nat_get_blkaddr(ne) == NULL_ADDR) {
>  			add_free_nid(sbi, nid, false);
> +			spin_lock(&NM_I(sbi)->nid_list_lock);
> +			NM_I(sbi)->free_nid_cnt++;
> +			spin_unlock(&NM_I(sbi)->nid_list_lock);
> +		}
>  	}
>  
>  	if (to_journal)
> @@ -2302,6 +2312,7 @@ static int init_node_manager(struct f2fs_sb_info *sbi)
>  	nm_i->nid_cnt[FREE_NID_LIST] = 0;
>  	nm_i->nid_cnt[ALLOC_NID_LIST] = 0;
>  	nm_i->nat_cnt = 0;
> +	nm_i->free_nid_cnt = nm_i->available_nids - sbi->total_valid_node_count;
>  	nm_i->ram_thresh = DEF_RAM_THRESHOLD;
>  	nm_i->ra_nid_pages = DEF_RA_NID_PAGES;
>  	nm_i->dirty_nats_ratio = DEF_DIRTY_NAT_RATIO_THRESHOLD;
> -- 
> 2.8.2.311.gee88674