linux-f2fs-devel.lists.sourceforge.net archive mirror
 help / color / mirror / Atom feed
* [PATCH v2] f2fs: avoid dead loop in function find_fsync_dnodes
@ 2017-12-11  4:23 Yunlei He
  2017-12-11 13:46 ` Chao Yu
  0 siblings, 1 reply; 2+ messages in thread
From: Yunlei He @ 2017-12-11  4:23 UTC (permalink / raw)
  To: jaegeuk, yuchao0, linux-f2fs-devel; +Cc: ning.jia, heyunlei

Came across a dead loop in recovery like this:

......
[   24.680480s][pid:320,cpu0,init]find_fsync_dnodes: blkaddr =13597696
[   24.698394s][pid:320,cpu0,init]find_fsync_dnodes: blkaddr =13597697
[   24.724334s][pid:320,cpu0,init]find_fsync_dnodes: blkaddr =13597698
[   24.724334s][pid:320,cpu0,init]find_fsync_dnodes: blkaddr =13597698
[   24.724365s][pid:320,cpu0,init]find_fsync_dnodes: blkaddr =13597698
[   24.724365s][pid:320,cpu0,init]find_fsync_dnodes: blkaddr =13597698
[   24.724365s][pid:320,cpu0,init]find_fsync_dnodes: blkaddr =13597698
[   24.724395s][pid:320,cpu0,init]find_fsync_dnodes: blkaddr =13597698
[   24.724395s][pid:320,cpu0,init]find_fsync_dnodes: blkaddr =13597698
[   24.724395s][pid:320,cpu0,init]find_fsync_dnodes: blkaddr =13597698
[   24.724395s][pid:320,cpu0,init]find_fsync_dnodes: blkaddr =13597698
[   24.724426s][pid:320,cpu0,init]find_fsync_dnodes: blkaddr =13597698
......

Mount process will block in dead loop and fsck can do nothing with this
error, This patch abandon recovery if node chain is cyclical.

Signed-off-by: Yunlei He <heyunlei@huawei.com>
---
 fs/f2fs/f2fs.h     |  6 +++++
 fs/f2fs/recovery.c | 71 ++++++++++++++++++++++++++++++++++++++++++++++--------
 2 files changed, 67 insertions(+), 10 deletions(-)

diff --git a/fs/f2fs/f2fs.h b/fs/f2fs/f2fs.h
index 82f1dc3..4649663 100644
--- a/fs/f2fs/f2fs.h
+++ b/fs/f2fs/f2fs.h
@@ -289,6 +289,12 @@ struct fsync_inode_entry {
 	block_t last_dentry;	/* block address locating the last dentry */
 };
 
+/* for the block of node chain, used only during recovery */
+struct fsync_node_entry {
+	struct list_head list;	/* list head */
+	block_t blkaddr;	/* locating the node block in recovery chain */
+};
+
 #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/recovery.c b/fs/f2fs/recovery.c
index 7d63faf..9958abb 100644
--- a/fs/f2fs/recovery.c
+++ b/fs/f2fs/recovery.c
@@ -46,6 +46,7 @@
  */
 
 static struct kmem_cache *fsync_entry_slab;
+static struct kmem_cache *fsync_node_chain_slab;
 
 bool space_for_roll_forward(struct f2fs_sb_info *sbi)
 {
@@ -220,14 +221,55 @@ static void recover_inode(struct inode *inode, struct page *page)
 			ino_of_node(page), name);
 }
 
+static void destroy_node_blocks(struct list_head *head)
+{
+	struct fsync_node_entry *entry, *tmp;
+
+	list_for_each_entry_safe(entry, tmp, head, list) {
+		list_del(&entry->list);
+		kmem_cache_free(fsync_node_chain_slab, entry);
+	}
+}
+
+static bool blkaddr_already_in_list(struct list_head *head, block_t blkaddr)
+{
+	struct fsync_node_entry *entry;
+
+	list_for_each_entry(entry, head, list)
+		if (entry->blkaddr == blkaddr)
+			return true;
+
+	return false;
+}
+
+static void add_fsync_node(struct list_head *head, block_t blkaddr)
+{
+	struct fsync_node_entry *entry;
+
+	entry = f2fs_kmem_cache_alloc(fsync_node_chain_slab, GFP_F2FS_ZERO);
+	entry->blkaddr = blkaddr;
+	list_add_tail(&entry->list, head);
+}
+
+static void destroy_fsync_dnodes(struct list_head *head)
+{
+	struct fsync_inode_entry *entry, *tmp;
+
+	list_for_each_entry_safe(entry, tmp, head, list)
+		del_fsync_inode(entry);
+}
+
 static int find_fsync_dnodes(struct f2fs_sb_info *sbi, struct list_head *head,
 				bool check_only)
 {
 	struct curseg_info *curseg;
 	struct page *page = NULL;
+	struct list_head node_block_list;
 	block_t blkaddr;
 	int err = 0;
 
+	INIT_LIST_HEAD(&node_block_list);
+
 	/* get node pages in the current segment */
 	curseg = CURSEG_I(sbi, CURSEG_WARM_NODE);
 	blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
@@ -239,7 +281,13 @@ static int find_fsync_dnodes(struct f2fs_sb_info *sbi, struct list_head *head,
 			return 0;
 
 		page = get_tmp_page(sbi, blkaddr);
+		if (unlikely(blkaddr_already_in_list(&node_block_list, blkaddr))) {
+			f2fs_msg(sbi->sb, KERN_ERR, "Abandon dead loop node block list");
+			destroy_fsync_dnodes(head);
+			break;
+		}
 
+		add_fsync_node(&node_block_list, blkaddr);
 		if (!is_recoverable_dnode(page))
 			break;
 
@@ -284,18 +332,12 @@ static int find_fsync_dnodes(struct f2fs_sb_info *sbi, struct list_head *head,
 
 		ra_meta_pages_cond(sbi, blkaddr);
 	}
+
+	destroy_node_blocks(&node_block_list);
 	f2fs_put_page(page, 1);
 	return err;
 }
 
-static void destroy_fsync_dnodes(struct list_head *head)
-{
-	struct fsync_inode_entry *entry, *tmp;
-
-	list_for_each_entry_safe(entry, tmp, head, list)
-		del_fsync_inode(entry);
-}
-
 static int check_index_in_prev_nodes(struct f2fs_sb_info *sbi,
 			block_t blkaddr, struct dnode_of_data *dn)
 {
@@ -614,7 +656,14 @@ int recover_fsync_data(struct f2fs_sb_info *sbi, bool check_only)
 			sizeof(struct fsync_inode_entry));
 	if (!fsync_entry_slab) {
 		err = -ENOMEM;
-		goto out;
+		goto free_inode_entry;
+	}
+
+	fsync_node_chain_slab = f2fs_kmem_cache_create("f2fs_fsync_node_entry",
+			sizeof(struct fsync_node_entry));
+	if (!fsync_node_chain_slab) {
+		err = -ENOMEM;
+		goto free_node_entry;
 	}
 
 	INIT_LIST_HEAD(&inode_list);
@@ -664,8 +713,10 @@ int recover_fsync_data(struct f2fs_sb_info *sbi, bool check_only)
 		err = write_checkpoint(sbi, &cpc);
 	}
 
+	kmem_cache_destroy(fsync_node_chain_slab);
+free_node_entry:
 	kmem_cache_destroy(fsync_entry_slab);
-out:
+free_inode_entry:
 #ifdef CONFIG_QUOTA
 	/* Turn quotas off */
 	if (quota_enabled)
-- 
1.9.1


------------------------------------------------------------------------------
Check out the vibrant tech community on one of the world's most
engaging tech sites, Slashdot.org! http://sdm.link/slashdot

^ permalink raw reply related	[flat|nested] 2+ messages in thread

* Re: [PATCH v2] f2fs: avoid dead loop in function find_fsync_dnodes
  2017-12-11  4:23 [PATCH v2] f2fs: avoid dead loop in function find_fsync_dnodes Yunlei He
@ 2017-12-11 13:46 ` Chao Yu
  0 siblings, 0 replies; 2+ messages in thread
From: Chao Yu @ 2017-12-11 13:46 UTC (permalink / raw)
  To: Yunlei He, jaegeuk, yuchao0, linux-f2fs-devel; +Cc: ning.jia

On 2017/12/11 12:23, Yunlei He wrote:
> Came across a dead loop in recovery like this:
> 
> ......
> [   24.680480s][pid:320,cpu0,init]find_fsync_dnodes: blkaddr =13597696
> [   24.698394s][pid:320,cpu0,init]find_fsync_dnodes: blkaddr =13597697
> [   24.724334s][pid:320,cpu0,init]find_fsync_dnodes: blkaddr =13597698
> [   24.724334s][pid:320,cpu0,init]find_fsync_dnodes: blkaddr =13597698
> [   24.724365s][pid:320,cpu0,init]find_fsync_dnodes: blkaddr =13597698
> [   24.724365s][pid:320,cpu0,init]find_fsync_dnodes: blkaddr =13597698
> [   24.724365s][pid:320,cpu0,init]find_fsync_dnodes: blkaddr =13597698
> [   24.724395s][pid:320,cpu0,init]find_fsync_dnodes: blkaddr =13597698
> [   24.724395s][pid:320,cpu0,init]find_fsync_dnodes: blkaddr =13597698
> [   24.724395s][pid:320,cpu0,init]find_fsync_dnodes: blkaddr =13597698
> [   24.724395s][pid:320,cpu0,init]find_fsync_dnodes: blkaddr =13597698
> [   24.724426s][pid:320,cpu0,init]find_fsync_dnodes: blkaddr =13597698
> ......
> 
> Mount process will block in dead loop and fsck can do nothing with this
> error, This patch abandon recovery if node chain is cyclical.

What about tagging page with PG_checked flag to indicate current meta page
has been traversed?

Thanks,

> 
> Signed-off-by: Yunlei He <heyunlei@huawei.com>
> ---
>  fs/f2fs/f2fs.h     |  6 +++++
>  fs/f2fs/recovery.c | 71 ++++++++++++++++++++++++++++++++++++++++++++++--------
>  2 files changed, 67 insertions(+), 10 deletions(-)
> 
> diff --git a/fs/f2fs/f2fs.h b/fs/f2fs/f2fs.h
> index 82f1dc3..4649663 100644
> --- a/fs/f2fs/f2fs.h
> +++ b/fs/f2fs/f2fs.h
> @@ -289,6 +289,12 @@ struct fsync_inode_entry {
>  	block_t last_dentry;	/* block address locating the last dentry */
>  };
>  
> +/* for the block of node chain, used only during recovery */
> +struct fsync_node_entry {
> +	struct list_head list;	/* list head */
> +	block_t blkaddr;	/* locating the node block in recovery chain */
> +};
> +
>  #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/recovery.c b/fs/f2fs/recovery.c
> index 7d63faf..9958abb 100644
> --- a/fs/f2fs/recovery.c
> +++ b/fs/f2fs/recovery.c
> @@ -46,6 +46,7 @@
>   */
>  
>  static struct kmem_cache *fsync_entry_slab;
> +static struct kmem_cache *fsync_node_chain_slab;
>  
>  bool space_for_roll_forward(struct f2fs_sb_info *sbi)
>  {
> @@ -220,14 +221,55 @@ static void recover_inode(struct inode *inode, struct page *page)
>  			ino_of_node(page), name);
>  }
>  
> +static void destroy_node_blocks(struct list_head *head)
> +{
> +	struct fsync_node_entry *entry, *tmp;
> +
> +	list_for_each_entry_safe(entry, tmp, head, list) {
> +		list_del(&entry->list);
> +		kmem_cache_free(fsync_node_chain_slab, entry);
> +	}
> +}
> +
> +static bool blkaddr_already_in_list(struct list_head *head, block_t blkaddr)
> +{
> +	struct fsync_node_entry *entry;
> +
> +	list_for_each_entry(entry, head, list)
> +		if (entry->blkaddr == blkaddr)
> +			return true;
> +
> +	return false;
> +}
> +
> +static void add_fsync_node(struct list_head *head, block_t blkaddr)
> +{
> +	struct fsync_node_entry *entry;
> +
> +	entry = f2fs_kmem_cache_alloc(fsync_node_chain_slab, GFP_F2FS_ZERO);
> +	entry->blkaddr = blkaddr;
> +	list_add_tail(&entry->list, head);
> +}
> +
> +static void destroy_fsync_dnodes(struct list_head *head)
> +{
> +	struct fsync_inode_entry *entry, *tmp;
> +
> +	list_for_each_entry_safe(entry, tmp, head, list)
> +		del_fsync_inode(entry);
> +}
> +
>  static int find_fsync_dnodes(struct f2fs_sb_info *sbi, struct list_head *head,
>  				bool check_only)
>  {
>  	struct curseg_info *curseg;
>  	struct page *page = NULL;
> +	struct list_head node_block_list;
>  	block_t blkaddr;
>  	int err = 0;
>  
> +	INIT_LIST_HEAD(&node_block_list);
> +
>  	/* get node pages in the current segment */
>  	curseg = CURSEG_I(sbi, CURSEG_WARM_NODE);
>  	blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
> @@ -239,7 +281,13 @@ static int find_fsync_dnodes(struct f2fs_sb_info *sbi, struct list_head *head,
>  			return 0;
>  
>  		page = get_tmp_page(sbi, blkaddr);
> +		if (unlikely(blkaddr_already_in_list(&node_block_list, blkaddr))) {
> +			f2fs_msg(sbi->sb, KERN_ERR, "Abandon dead loop node block list");
> +			destroy_fsync_dnodes(head);
> +			break;
> +		}
>  
> +		add_fsync_node(&node_block_list, blkaddr);
>  		if (!is_recoverable_dnode(page))
>  			break;
>  
> @@ -284,18 +332,12 @@ static int find_fsync_dnodes(struct f2fs_sb_info *sbi, struct list_head *head,
>  
>  		ra_meta_pages_cond(sbi, blkaddr);
>  	}
> +
> +	destroy_node_blocks(&node_block_list);
>  	f2fs_put_page(page, 1);
>  	return err;
>  }
>  
> -static void destroy_fsync_dnodes(struct list_head *head)
> -{
> -	struct fsync_inode_entry *entry, *tmp;
> -
> -	list_for_each_entry_safe(entry, tmp, head, list)
> -		del_fsync_inode(entry);
> -}
> -
>  static int check_index_in_prev_nodes(struct f2fs_sb_info *sbi,
>  			block_t blkaddr, struct dnode_of_data *dn)
>  {
> @@ -614,7 +656,14 @@ int recover_fsync_data(struct f2fs_sb_info *sbi, bool check_only)
>  			sizeof(struct fsync_inode_entry));
>  	if (!fsync_entry_slab) {
>  		err = -ENOMEM;
> -		goto out;
> +		goto free_inode_entry;
> +	}
> +
> +	fsync_node_chain_slab = f2fs_kmem_cache_create("f2fs_fsync_node_entry",
> +			sizeof(struct fsync_node_entry));
> +	if (!fsync_node_chain_slab) {
> +		err = -ENOMEM;
> +		goto free_node_entry;
>  	}
>  
>  	INIT_LIST_HEAD(&inode_list);
> @@ -664,8 +713,10 @@ int recover_fsync_data(struct f2fs_sb_info *sbi, bool check_only)
>  		err = write_checkpoint(sbi, &cpc);
>  	}
>  
> +	kmem_cache_destroy(fsync_node_chain_slab);
> +free_node_entry:
>  	kmem_cache_destroy(fsync_entry_slab);
> -out:
> +free_inode_entry:
>  #ifdef CONFIG_QUOTA
>  	/* Turn quotas off */
>  	if (quota_enabled)
> 

------------------------------------------------------------------------------
Check out the vibrant tech community on one of the world's most
engaging tech sites, Slashdot.org! http://sdm.link/slashdot

^ permalink raw reply	[flat|nested] 2+ messages in thread

end of thread, other threads:[~2017-12-11 13:46 UTC | newest]

Thread overview: 2+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2017-12-11  4:23 [PATCH v2] f2fs: avoid dead loop in function find_fsync_dnodes Yunlei He
2017-12-11 13:46 ` Chao Yu

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).