net: sched: fix memory leak in tcindex_partial_destroy_work

Syzbot reported memory leak in tcindex_set_parms(). The problem was in
non-freed perfect hash in tcindex_partial_destroy_work().

In tcindex_set_parms() new tcindex_data is allocated and some fields from
old one are copied to new one, but not the perfect hash. Since
tcindex_partial_destroy_work() is the destroy function for old
tcindex_data, we need to free perfect hash to avoid memory leak.

Reported-and-tested-by: syzbot+f0bbb2287b8993d4fa74@syzkaller.appspotmail.com
Fixes: a4b67f07dc6f ("net: sched: RCU cls_tcindex")
Signed-off-by: Pavel Skripkin <paskripkin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

diff --git a/net/sched/cls_tcindex.c b/net/sched/cls_tcindex.c
index 5b274534264c2de4c1325efa3f388099f32cd7ee..e9a8a2c86bbdd033b3dca556fcdb1e79b835b5cd 100644 (file)
--- a/net/sched/cls_tcindex.c
+++ b/net/sched/cls_tcindex.c
@@ -278,6 +278,8 @@ static int tcindex_filter_result_init(struct tcindex_filter_result *r,
                             TCA_TCINDEX_POLICE);
 }
 
+static void tcindex_free_perfect_hash(struct tcindex_data *cp);
+
 static void tcindex_partial_destroy_work(struct work_struct *work)
 {
        struct tcindex_data *p = container_of(to_rcu_work(work),
@@ -285,7 +287,8 @@ static void tcindex_partial_destroy_work(struct work_struct *work)
                                              rwork);
 
        rtnl_lock();
-       kfree(p->perfect);
+       if (p->perfect)
+               tcindex_free_perfect_hash(p);
        kfree(p);
        rtnl_unlock();
 }