[PATCH] ext4: Fix refcount leak bug in __ext4_new_inode()

After successfully creating handle by __ext4_journal_start_sb(),
the function forgets to decrease the refcount of handle in several
paths, causing refcount leak.

Fix this issue by recording a flag when successfully getting handle
by __ext4_journal_start_sb(), and decrease the refcount of handle
when exiting this function if holding the flag.

Signed-off-by: Chenyuan Mi <redacted>
Signed-off-by: Xiyu Yang <redacted>
Signed-off-by: Xin Tan <redacted>

---
 fs/ext4/ialloc.c | 7 ++++++-
 1 file changed, 6 insertions(+), 1 deletion(-)

diff --git a/fs/ext4/ialloc.c b/fs/ext4/ialloc.c
index f73e5eb43eae..7563b892c64f 100644
--- a/fs/ext4/ialloc.c
+++ b/fs/ext4/ialloc.c

@@ -944,6 +944,7 @@ struct inode *__ext4_new_inode(struct user_namespace *mnt_userns,
 	ext4_group_t flex_group;
 	struct ext4_group_info *grp = NULL;
 	bool encrypt = false;
+	bool create_handle = false;
 
 	/* Cannot create files in a deleted directory */
 	if (!dir || !dir->i_nlink)

@@ -1085,6 +1086,7 @@ struct inode *__ext4_new_inode(struct user_namespace *mnt_userns,
 				ext4_std_error(sb, err);
 				goto out;
 			}
+			create_handle = true;
 		}
 		BUFFER_TRACE(inode_bitmap_bh, "get_write_access");
 		err = ext4_journal_get_write_access(handle, sb, inode_bitmap_bh,

@@ -1345,7 +1347,8 @@ struct inode *__ext4_new_inode(struct user_namespace *mnt_userns,
 		ext4_std_error(sb, err);
 		goto fail_free_drop;
 	}
-
+	if (create_handle)
+		ext4_journal_stop(handle);
 	ext4_debug("allocating inode %lu\n", inode->i_ino);
 	trace_ext4_allocate_inode(inode, dir, mode);
 	brelse(inode_bitmap_bh);

@@ -1357,6 +1360,8 @@ struct inode *__ext4_new_inode(struct user_namespace *mnt_userns,
 	clear_nlink(inode);
 	unlock_new_inode(inode);
 out:
+	if (create_handle)
+		ext4_journal_stop(handle);
 	dquot_drop(inode);
 	inode->i_flags |= S_NOQUOTA;
 	iput(inode);

-- 
2.17.1

On Wed 08-09-21 10:46:05, Chenyuan Mi wrote:

After successfully creating handle by __ext4_journal_start_sb(),
the function forgets to decrease the refcount of handle in several
paths, causing refcount leak.

Fix this issue by recording a flag when successfully getting handle
by __ext4_journal_start_sb(), and decrease the refcount of handle
when exiting this function if holding the flag.

Signed-off-by: Chenyuan Mi <redacted>
Signed-off-by: Xiyu Yang <redacted>
Signed-off-by: Xin Tan <redacted>

I'm sorry but this patch is wrong. We are expected to return with handle
running from __ext4_new_inode(). Please have a look into callers of
ext4_new_inode() and ext4_new_inode_start_handle() how they use and then
stop the handle. And similarly as for the second patch I'd note that even a
small amount of testing would show you that this patch does not work.

                                                                Honza

-- 
Jan Kara [off-list ref]
SUSE Labs, CR