[PATCH v3 2/2] cgroup: Limit frequency of reading cgroup.stat for unprivileged users

[Waiman Long]

Since cgroup_mutex is being acquired when reading from a world readable
cgroup.stat control file, it is possible that repeated reading of
cgroup.stat may be used as a denial of service (DoS) attack vector
by unprivileged users to greatly hinder cgroup related operations in
a system.

To prevent this, we are limiting the reading of cgroup.stat file from
unprivileged users to at most 8 times per second.

Signed-off-by: Waiman Long <longman@xxxxxxxxxx>
---
 kernel/cgroup/cgroup.c | 10 ++++++++++
 1 file changed, 10 insertions(+)

diff --git a/kernel/cgroup/cgroup.c b/kernel/cgroup/cgroup.c
index 18c982a06446..56ac9f14d100 100644
--- a/kernel/cgroup/cgroup.c
+++ b/kernel/cgroup/cgroup.c
@@ -3668,12 +3668,22 @@ static int cgroup_events_show(struct seq_file *seq, void *v)
 
 static int cgroup_stat_show(struct seq_file *seq, void *v)
 {
+	static unsigned long unpriv_timestamp;
 	struct cgroup *cgroup = seq_css(seq)->cgroup;
 	struct cgroup_subsys_state *css;
 	int ssid;
 
+	/*
+	 * Unprivileged users can only read cgroup.stat up to 8 times
+	 * per second to avoid potential DoS attack.
+	 */
+	if (!capable(CAP_SYS_ADMIN) &&
+	    time_before(jiffies, unpriv_timestamp + (HZ >> 3)))
+		msleep(HZ >> 3);
+
 	/* cgroup_mutex required for for_each_css() */
 	cgroup_lock();
+	unpriv_timestamp = jiffies;
 	seq_printf(seq, "nr_descendants %d\n",
 		   cgroup->nr_descendants);
 	seq_printf(seq, "nr_dying_descendants %d\n",
-- 
2.39.3