[PATCH 2/2] jffs2: solving deadlock on sync function

[Wagner Popov dos Santos <wpopov@gmail.com>]

Correcting AB-BA deadlock in jffs2_fsync() involving alloc_sem
semaphore and inodes.

The function jffs2_fsync() can't lock the inode because some
process, or even the same process, that call the CG will acquire
alloc_sem semaphore and will try to acquire the inode if it is
inside the Erase Block that is marked to be processed.

Fixes: 02c24a82187d ("fs: push i_mutex and filemap_write_and_wait down into ->fsync() handlers")

Signed-off-by: Wagner Popov dos Santos <wpopov@gmail.com>
---
 fs/jffs2/file.c | 10 +++++++---
 1 file changed, 7 insertions(+), 3 deletions(-)

diff --git a/fs/jffs2/file.c b/fs/jffs2/file.c
index 7d8654a1472e..7f139704cb8d 100644
--- a/fs/jffs2/file.c
+++ b/fs/jffs2/file.c
@@ -39,10 +39,14 @@ int jffs2_fsync(struct file *filp, loff_t start, loff_t end, int datasync)
 	if (ret)
 		return ret;
 
-	inode_lock(inode);
-	/* Trigger GC to flush any pending writes for this inode */
+	/* Trigger GC to flush any pending writes for this inode
+	 *
+	 * We need to leave the inode unlocked to avoid a deadlock condition
+	 * because the function jffs2_garbage_collect_pass() can try to lock
+	 * the same inode if it is inside the erase block that GC is
+	 * processing.
+	 */
 	jffs2_flush_wbuf_gc(c, inode->i_ino);
-	inode_unlock(inode);
 
 	return 0;
 }
-- 
2.20.1