[PATCH v4] f2fs: avoid dead loop in function find_fsync_dnodes

[PATCH v4] f2fs: avoid dead loop in function find_fsync_dnodes

[Yunlei He]

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/recovery.c | 27 +++++++++++++++++++--------
 1 file changed, 19 insertions(+), 8 deletions(-)

diff --git a/fs/f2fs/recovery.c b/fs/f2fs/recovery.c
index 7d63faf..f7fc2ed 100644
--- a/fs/f2fs/recovery.c
+++ b/fs/f2fs/recovery.c
@@ -106,6 +106,14 @@ static void del_fsync_inode(struct fsync_inode_entry *entry)
 	kmem_cache_free(fsync_entry_slab, entry);
 }
 
+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 recover_dentry(struct inode *inode, struct page *ipage,
 						struct list_head *dir_list)
 {
@@ -239,6 +247,17 @@ static int find_fsync_dnodes(struct f2fs_sb_info *sbi, struct list_head *head,
 			return 0;
 
 		page = get_tmp_page(sbi, blkaddr);
+		if (PageChecked(page)) {
+			f2fs_msg(sbi->sb, KERN_ERR, "Abandon looped node block list");
+			destroy_fsync_dnodes(head);
+			break;
+		}
+
+		/*
+		 * it's not needed to clear PG_checked flag in temp page since we
+		 * will truncate all those pages in the end of recovery.
+		 */
+		SetPageChecked(page);
 
 		if (!is_recoverable_dnode(page))
 			break;
@@ -288,14 +307,6 @@ static int find_fsync_dnodes(struct f2fs_sb_info *sbi, struct list_head *head,
 	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)
 {
-- 
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

Re: [PATCH v4] f2fs: avoid dead loop in function find_fsync_dnodes

[Chao Yu]

On 2017/12/15 15:56, 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.
> 
> Signed-off-by: Yunlei He <heyunlei@huawei.com>

Reviewed-by: Chao Yu <yuchao0@huawei.com>

Thanks,


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