f2fs: fix livelock calling f2fs_iget during f2fs_evict_inode

In f2fs_evict_inode,
 commit_inmemory_pages
   f2fs_gc
     f2fs_iget
       iget_locked
         -> wait for inode free

Here, if the inode is same as the one to be evicted, f2fs should wait forever.
Actually, we should not call f2fs_balance_fs during f2fs_evict_inode to avoid
this.

But, the commit_inmem_pages calls f2fs_balance_fs by default, even if
f2fs_evict_inode wants to free inmemory pages only.

Hence, this patch adds to trigger f2fs_balance_fs only when there is something
to write.

Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>

diff --git a/fs/f2fs/segment.c b/fs/f2fs/segment.c
index 9de857f6899b805220db8ba21bb9705842ebaed9..9a33e34d26ce9dcd31f0af918a3df510d6d47c52 100644 (file)
--- a/fs/f2fs/segment.c
+++ b/fs/f2fs/segment.c
@@ -230,7 +230,16 @@ void commit_inmem_pages(struct inode *inode, bool abort)
                .rw = WRITE_SYNC,
        };
 
-       f2fs_balance_fs(sbi);
+       /*
+        * The abort is true only when f2fs_evict_inode is called.
+        * Basically, the f2fs_evict_inode doesn't produce any data writes, so
+        * that we don't need to call f2fs_balance_fs.
+        * Otherwise, f2fs_gc in f2fs_balance_fs can wait forever until this
+        * inode becomes free by iget_locked in f2fs_iget.
+        */
+       if (!abort)
+               f2fs_balance_fs(sbi);
+
        f2fs_lock_op(sbi);
 
        mutex_lock(&fi->inmem_lock);