tipc: pass tunnel dev as NULL to udp_tunnel(6)_xmit_skb
authorXin Long <lucien.xin@gmail.com>
Mon, 17 Jun 2019 13:34:15 +0000 (21:34 +0800)
committerDavid S. Miller <davem@davemloft.net>
Wed, 19 Jun 2019 00:48:45 +0000 (20:48 -0400)
udp_tunnel(6)_xmit_skb() called by tipc_udp_xmit() expects a tunnel device
to count packets on dev->tstats, a perpcu variable. However, TIPC is using
udp tunnel with no tunnel device, and pass the lower dev, like veth device
that only initializes dev->lstats(a perpcu variable) when creating it.

Later iptunnel_xmit_stats() called by ip(6)tunnel_xmit() thinks the dev as
a tunnel device, and uses dev->tstats instead of dev->lstats. tstats' each
pointer points to a bigger struct than lstats, so when tstats->tx_bytes is
increased, other percpu variable's members could be overwritten.

syzbot has reported quite a few crashes due to fib_nh_common percpu member
'nhc_pcpu_rth_output' overwritten, call traces are like:

  BUG: KASAN: slab-out-of-bounds in rt_cache_valid+0x158/0x190
  net/ipv4/route.c:1556
    rt_cache_valid+0x158/0x190 net/ipv4/route.c:1556
    __mkroute_output net/ipv4/route.c:2332 [inline]
    ip_route_output_key_hash_rcu+0x819/0x2d50 net/ipv4/route.c:2564
    ip_route_output_key_hash+0x1ef/0x360 net/ipv4/route.c:2393
    __ip_route_output_key include/net/route.h:125 [inline]
    ip_route_output_flow+0x28/0xc0 net/ipv4/route.c:2651
    ip_route_output_key include/net/route.h:135 [inline]
  ...

or:

  kasan: GPF could be caused by NULL-ptr deref or user memory access
  RIP: 0010:dst_dev_put+0x24/0x290 net/core/dst.c:168
    <IRQ>
    rt_fibinfo_free_cpus net/ipv4/fib_semantics.c:200 [inline]
    free_fib_info_rcu+0x2e1/0x490 net/ipv4/fib_semantics.c:217
    __rcu_reclaim kernel/rcu/rcu.h:240 [inline]
    rcu_do_batch kernel/rcu/tree.c:2437 [inline]
    invoke_rcu_callbacks kernel/rcu/tree.c:2716 [inline]
    rcu_process_callbacks+0x100a/0x1ac0 kernel/rcu/tree.c:2697
  ...

The issue exists since tunnel stats update is moved to iptunnel_xmit by
Commit 039f50629b7f ("ip_tunnel: Move stats update to iptunnel_xmit()"),
and here to fix it by passing a NULL tunnel dev to udp_tunnel(6)_xmit_skb
so that the packets counting won't happen on dev->tstats.

Reported-by: syzbot+9d4c12bfd45a58738d0a@syzkaller.appspotmail.com
Reported-by: syzbot+a9e23ea2aa21044c2798@syzkaller.appspotmail.com
Reported-by: syzbot+c4c4b2bb358bb936ad7e@syzkaller.appspotmail.com
Reported-by: syzbot+0290d2290a607e035ba1@syzkaller.appspotmail.com
Reported-by: syzbot+a43d8d4e7e8a7a9e149e@syzkaller.appspotmail.com
Reported-by: syzbot+a47c5f4c6c00fc1ed16e@syzkaller.appspotmail.com
Fixes: 039f50629b7f ("ip_tunnel: Move stats update to iptunnel_xmit()")
Signed-off-by: Xin Long <lucien.xin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
net/tipc/udp_media.c

index 7fc02d84c4f1087a08cffae68812a793ab199bd2..1405ccc9101c1599c53d0881d42f93d8de231df3 100644 (file)
@@ -176,7 +176,6 @@ static int tipc_udp_xmit(struct net *net, struct sk_buff *skb,
                        goto tx_error;
                }
 
-               skb->dev = rt->dst.dev;
                ttl = ip4_dst_hoplimit(&rt->dst);
                udp_tunnel_xmit_skb(rt, ub->ubsock->sk, skb, src->ipv4.s_addr,
                                    dst->ipv4.s_addr, 0, ttl, 0, src->port,
@@ -195,10 +194,9 @@ static int tipc_udp_xmit(struct net *net, struct sk_buff *skb,
                if (err)
                        goto tx_error;
                ttl = ip6_dst_hoplimit(ndst);
-               err = udp_tunnel6_xmit_skb(ndst, ub->ubsock->sk, skb,
-                                          ndst->dev, &src->ipv6,
-                                          &dst->ipv6, 0, ttl, 0, src->port,
-                                          dst->port, false);
+               err = udp_tunnel6_xmit_skb(ndst, ub->ubsock->sk, skb, NULL,
+                                          &src->ipv6, &dst->ipv6, 0, ttl, 0,
+                                          src->port, dst->port, false);
 #endif
        }
        return err;