On Mon, Feb 12, 2018 at 10:11:06AM +0000, James Chapman wrote:
> If an L2TPIP socket is closed, add RCU protection when we deref
> sk_user_data to prevent races with another thread closing the same
> tunnel.
>
> Fixes: 0d76751fad ("l2tp: Add L2TPv3 IP encapsulation (no UDP) support")
>
> refcount_t: increment on 0; use-after-free.
> WARNING: CPU: 2 PID: 2892 at lib/refcount.c:153 refcount_inc+0x2b/0x30
> Modules linked in:
> CPU: 2 PID: 2892 Comm: pppol2tp_chaos Not tainted 4.15.0-rc9+ #1
> Hardware name: innotek GmbH VirtualBox/VirtualBox, BIOS VirtualBox 12/01/2006
> RIP: 0010:refcount_inc+0x2b/0x30
> RSP: 0018:ffff880014147b50 EFLAGS: 00010282
> RAX: 000000000000002b RBX: ffff8800194785c0 RCX: 0000000000000000
> RDX: 0000000000000001 RSI: ffff88001ad1f758 RDI: ffff88001ad1f758
> RBP: ffff880014147b50 R08: 0000000000000000 R09: 0000000000000000
> R10: 0000000000000001 R11: 0000000000000001 R12: ffff8800194785c8
> R13: ffff8800194785c0 R14: ffff88001a2c6c20 R15: ffff880013a9d580
> FS: 0000000000000000(0000) GS:ffff88001ad00000(0000) knlGS:0000000000000000
> CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
> CR2: 00007fbc17ea7000 CR3: 0000000003022000 CR4: 00000000000006e0
> Call Trace:
> l2tp_tunnel_delete+0x34/0x60
> l2tp_ip_destroy_sock+0x6d/0x80
> sk_common_release+0x19/0xd0
> l2tp_ip_close+0x89/0xa0
> inet_release+0x37/0x60
> sock_release+0x1a/0x70
> sock_close+0xd/0x20
> __fput+0xed/0x1f0
> ____fput+0x9/0x10
> task_work_run+0x77/0xb0
> do_exit+0x311/0xcf0
> do_group_exit+0x47/0xc0
> get_signal+0x343/0x8e0
> do_signal+0x23/0x780
> ? find_held_lock+0x39/0xb0
> exit_to_usermode_loop+0x4a/0x93
> syscall_return_slowpath+0x102/0x150
> entry_SYSCALL_64_fastpath+0x98/0x9a
> RIP: 0033:0x7fbc172d7fcf
> RSP: 002b:00007ffd524cd4d8 EFLAGS: 00000246 ORIG_RAX: 000000000000000e
> RAX: 0000000000000000 RBX: 0000000000000006 RCX: 00007fbc172d7fcf
> RDX: 0000000000000000 RSI: 00007ffd524cd460 RDI: 0000000000000002
> RBP: 0000000000404930 R08: 0000000000000000 R09: 00007ffd524cd460
> R10: 0000000000000008 R11: 0000000000000246 R12: 000000000000001e
> R13: 0000000000404a03 R14: 0000000000000000 R15: 0000000000000000
> Code: 55 48 89 e5 e8 87 ff ff ff 84 c0 74 02 5d c3 80 3d 97 87 bb 01 00 75
> f5 48 c7 c7 58 3e cc 82 c6 05 87 87 bb 01
> ---
> net/l2tp/l2tp_ip.c | 5 ++++-
> net/l2tp/l2tp_ip6.c | 5 ++++-
> 2 files changed, 8 insertions(+), 2 deletions(-)
>
> diff --git a/net/l2tp/l2tp_ip.c b/net/l2tp/l2tp_ip.c
> index ff61124fdf59..42f3c2f72bf4 100644
> --- a/net/l2tp/l2tp_ip.c
> +++ b/net/l2tp/l2tp_ip.c
> @@ -234,15 +234,18 @@ static void l2tp_ip_close(struct sock *sk, long timeout)
> static void l2tp_ip_destroy_sock(struct sock *sk)
> {
> struct sk_buff *skb;
> - struct l2tp_tunnel *tunnel = l2tp_sock_to_tunnel(sk);
> + struct l2tp_tunnel *tunnel;
>
> while ((skb = __skb_dequeue_tail(&sk->sk_write_queue)) != NULL)
> kfree_skb(skb);
>
> + rcu_read_lock();
> + tunnel = rcu_dereference_sk_user_data(sk);
Is this safe? We don't wait for a grace period after resetting ->sk_user_data.
> if (tunnel) {
> l2tp_tunnel_closeall(tunnel);
I'm pretty sure l2tp_tunnel_closeall() can sleep and can't be called in
RCU critical sections.
> sock_put(sk);
> }
> + rcu_read_unlock();
>
