]> git.proxmox.com Git - mirror_ubuntu-kernels.git/commitdiff
l2tp: Don't sleep and disable BH under writer-side sk_callback_lock
authorJakub Sitnicki <jakub@cloudflare.com>
Mon, 21 Nov 2022 08:54:26 +0000 (09:54 +0100)
committerDavid S. Miller <davem@davemloft.net>
Wed, 23 Nov 2022 12:45:19 +0000 (12:45 +0000)
When holding a reader-writer spin lock we cannot sleep. Calling
setup_udp_tunnel_sock() with write lock held violates this rule, because we
end up calling percpu_down_read(), which might sleep, as syzbot reports
[1]:

 __might_resched.cold+0x222/0x26b kernel/sched/core.c:9890
 percpu_down_read include/linux/percpu-rwsem.h:49 [inline]
 cpus_read_lock+0x1b/0x140 kernel/cpu.c:310
 static_key_slow_inc+0x12/0x20 kernel/jump_label.c:158
 udp_tunnel_encap_enable include/net/udp_tunnel.h:187 [inline]
 setup_udp_tunnel_sock+0x43d/0x550 net/ipv4/udp_tunnel_core.c:81
 l2tp_tunnel_register+0xc51/0x1210 net/l2tp/l2tp_core.c:1509
 pppol2tp_connect+0xcdc/0x1a10 net/l2tp/l2tp_ppp.c:723

Trim the writer-side critical section for sk_callback_lock down to the
minimum, so that it covers only operations on sk_user_data.

Also, when grabbing the sk_callback_lock, we always need to disable BH, as
Eric points out. Failing to do so leads to deadlocks because we acquire
sk_callback_lock in softirq context, which can get stuck waiting on us if:

1) it runs on the same CPU, or

       CPU0
       ----
  lock(clock-AF_INET6);
  <Interrupt>
    lock(clock-AF_INET6);

2) lock ordering leads to priority inversion

       CPU0                    CPU1
       ----                    ----
  lock(clock-AF_INET6);
                               local_irq_disable();
                               lock(&tcp_hashinfo.bhash[i].lock);
                               lock(clock-AF_INET6);
  <Interrupt>
    lock(&tcp_hashinfo.bhash[i].lock);

... as syzbot reports [2,3]. Use the _bh variants for write_(un)lock.

[1] https://lore.kernel.org/netdev/0000000000004e78ec05eda79749@google.com/
[2] https://lore.kernel.org/netdev/000000000000e38b6605eda76f98@google.com/
[3] https://lore.kernel.org/netdev/000000000000dfa31e05eda76f75@google.com/

v2:
- Check and set sk_user_data while holding sk_callback_lock for both
  L2TP encapsulation types (IP and UDP) (Tetsuo)

Cc: Tom Parkin <tparkin@katalix.com>
Cc: Tetsuo Handa <penguin-kernel@i-love.sakura.ne.jp>
Fixes: b68777d54fac ("l2tp: Serialize access to sk_user_data with sk_callback_lock")
Reported-by: Eric Dumazet <edumazet@google.com>
Reported-by: syzbot+703d9e154b3b58277261@syzkaller.appspotmail.com
Reported-by: syzbot+50680ced9e98a61f7698@syzkaller.appspotmail.com
Reported-by: syzbot+de987172bb74a381879b@syzkaller.appspotmail.com
Signed-off-by: Jakub Sitnicki <jakub@cloudflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
net/l2tp/l2tp_core.c

index 754fdda8a5f52e4e8e2c0f47331c3b22765033d0..9a1415fe3fa78bf419efc86c4a6ccfc7e0f0c78b 100644 (file)
@@ -1474,11 +1474,12 @@ int l2tp_tunnel_register(struct l2tp_tunnel *tunnel, struct net *net,
        }
 
        sk = sock->sk;
-       write_lock(&sk->sk_callback_lock);
-
+       write_lock_bh(&sk->sk_callback_lock);
        ret = l2tp_validate_socket(sk, net, tunnel->encap);
        if (ret < 0)
-               goto err_sock;
+               goto err_inval_sock;
+       rcu_assign_sk_user_data(sk, tunnel);
+       write_unlock_bh(&sk->sk_callback_lock);
 
        tunnel->l2tp_net = net;
        pn = l2tp_pernet(net);
@@ -1507,8 +1508,6 @@ int l2tp_tunnel_register(struct l2tp_tunnel *tunnel, struct net *net,
                };
 
                setup_udp_tunnel_sock(net, sock, &udp_cfg);
-       } else {
-               rcu_assign_sk_user_data(sk, tunnel);
        }
 
        tunnel->old_sk_destruct = sk->sk_destruct;
@@ -1522,16 +1521,18 @@ int l2tp_tunnel_register(struct l2tp_tunnel *tunnel, struct net *net,
        if (tunnel->fd >= 0)
                sockfd_put(sock);
 
-       write_unlock(&sk->sk_callback_lock);
        return 0;
 
 err_sock:
+       write_lock_bh(&sk->sk_callback_lock);
+       rcu_assign_sk_user_data(sk, NULL);
+err_inval_sock:
+       write_unlock_bh(&sk->sk_callback_lock);
+
        if (tunnel->fd < 0)
                sock_release(sock);
        else
                sockfd_put(sock);
-
-       write_unlock(&sk->sk_callback_lock);
 err:
        return ret;
 }