vfs: Move waiting for inode writeback from end_writeback() to evict_inode()

Currently, I_SYNC can never be set when evict_inode() (and thus
end_writeback()) is called because flusher thread holds inode reference while
inode is under writeback. As a result inode_sync_wait() in those places
currently does nothing. However that is going to change and unveils problems
with calling inode_sync_wait() from end_writeback(). Several filesystems call
end_writeback() after they have deleted the inode (btrfs, gfs2, ...) and other
filesystems (ext3, ext4, reiserfs, ...) can deadlock when waiting for I_SYNC
because they call end_writeback() from within a transaction.

To avoid these issues, we move inode_sync_wait() into evict_inode() before
calling ->evict_inode(). That way we preserve the current property that
->evict_inode() and writeback never run in parallel and all filesystems are
safe.

Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Fengguang Wu <fengguang.wu@intel.com>

diff --git a/fs/inode.c b/fs/inode.c
index 9f4f5fecc0963c12db1f806be61999a26a9ba7ea..501fc5daf6f44f8847825efc778164db34bf7612 100644 (file)
--- a/fs/inode.c
+++ b/fs/inode.c
@@ -500,7 +500,6 @@ void end_writeback(struct inode *inode)
        BUG_ON(!list_empty(&inode->i_data.private_list));
        BUG_ON(!(inode->i_state & I_FREEING));
        BUG_ON(inode->i_state & I_CLEAR);
-       inode_sync_wait(inode);
        /* don't need i_lock here, no concurrent mods to i_state */
        inode->i_state = I_FREEING | I_CLEAR;
 }
@@ -531,6 +530,8 @@ static void evict(struct inode *inode)
 
        inode_sb_list_del(inode);
 
+       inode_sync_wait(inode);
+
        if (op->evict_inode) {
                op->evict_inode(inode);
        } else {