Hello, Glauber.

On Wed, Oct 31, 2012 at 05:21:29PM +0400, Glauber Costa wrote:

quoted

+
+	local_irq_disable();
+
+	/* block new css_tryget() by deactivating refcnt */
+	for_each_subsys(cgrp->root, ss) {
+		struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
+
+		WARN_ON(atomic_read(&css->refcnt) < 0);
+		atomic_add(CSS_DEACT_BIAS, &css->refcnt);
 	}
-	/* NO css_tryget() can success after here. */
+
+	/*
+	 * Set REMOVED.  All in-progress css_tryget() will be released.
+	 * Put all the base refs.  Each css holds an extra reference to the
+	 * cgroup's dentry and cgroup removal proceeds regardless of css
+	 * refs.  On the last put of each css, whenever that may be, the
+	 * extra dentry ref is put so that dentry destruction happens only
+	 * after all css's are released.
+	 */
+	for_each_subsys(cgrp->root, ss) {
+		struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
+
+		set_bit(CSS_REMOVED, &css->flags);
+		css_put(css);
+	}
+
+	local_irq_enable();
+
 	finish_wait(&cgroup_rmdir_waitq, &wait);

It seems to me that the main reason for doing this in two passes was to
derive a global "failed" state on the first pass. Now that we can't
fail, why can't you just loop through the subsystems just once ?

Later in the series, pre_destroy() happens inbetween.  The first loop
marks the CSSes and cgrp dead, ->pre_destroy()s are called, and the
later loop will put the final ref.  I'll mention it in the commit
message.

Thanks.

-- 
tejun