l2tp: fix races with tunnel socket close

The tunnel socket tunnel->sock (struct sock) is accessed when
preparing a new ppp session on a tunnel at pppol2tp_session_init. If
the socket is closed by a thread while another is creating a new
session, the threads race. In pppol2tp_connect, the tunnel object may
be created if the pppol2tp socket is associated with the special
session_id 0 and the tunnel socket is looked up using the provided
fd. When handling this, pppol2tp_connect cannot sock_hold the tunnel
socket to prevent it being destroyed during pppol2tp_connect since
this may itself may race with the socket being destroyed. Doing
sockfd_lookup in pppol2tp_connect isn't sufficient to prevent
tunnel->sock going away either because a given tunnel socket fd may be
reused between calls to pppol2tp_connect. Instead, have
l2tp_tunnel_create sock_hold the tunnel socket before it does
sockfd_put. This ensures that the tunnel's socket is always extant
while the tunnel object exists. Hold a ref on the socket until the
tunnel is destroyed and ensure that all tunnel destroy paths go
through a common function (l2tp_tunnel_delete) since this will do the
final sock_put to release the tunnel socket.

Since the tunnel's socket is now guaranteed to exist if the tunnel
exists, we no longer need to use sockfd_lookup via l2tp_sock_to_tunnel
to derive the tunnel from the socket since this is always
sk_user_data.

Also, sessions no longer sock_hold the tunnel socket since sessions
already hold a tunnel ref and the tunnel sock will not be freed until
the tunnel is freed. Removing these sock_holds in
l2tp_session_register avoids a possible sock leak in the
pppol2tp_connect error path if l2tp_session_register succeeds but
attaching a ppp channel fails. The pppol2tp_connect error path could
have been fixed instead and have the sock ref dropped when the session
is freed, but doing a sock_put of the tunnel socket when the session
is freed would require a new session_free callback. It is simpler to
just remove the sock_hold of the tunnel socket in
l2tp_session_register, now that the tunnel socket lifetime is
guaranteed.

Finally, some init code in l2tp_tunnel_create is reordered to ensure
that the new tunnel object's refcount is set and the tunnel socket ref
is taken before the tunnel socket destructor callbacks are set.

kasan: CONFIG_KASAN_INLINE enabled
kasan: GPF could be caused by NULL-ptr deref or user memory access
general protection fault: 0000 [#1] SMP KASAN
Modules linked in:
CPU: 0 PID: 4360 Comm: syzbot_19c09769 Not tainted 4.16.0-rc2+ #34
Hardware name: innotek GmbH VirtualBox/VirtualBox, BIOS VirtualBox 12/01/2006
RIP: 0010:pppol2tp_session_init+0x1d6/0x500
RSP: 0018:ffff88001377fb40 EFLAGS: 00010212
RAX: dffffc0000000000 RBX: ffff88001636a940 RCX: ffffffff84836c1d
RDX: 0000000000000045 RSI: 0000000055976744 RDI: 0000000000000228
RBP: ffff88001377fb60 R08: ffffffff84836bc8 R09: 0000000000000002
R10: ffff88001377fab8 R11: 0000000000000001 R12: 0000000000000000
R13: ffff88001636aac8 R14: ffff8800160f81c0 R15: 1ffff100026eff76
FS:  00007ffb3ea66700(0000) GS:ffff88001a400000(0000) knlGS:0000000000000000
CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 0000000020e77000 CR3: 0000000016261000 CR4: 00000000000006f0
Call Trace:
 pppol2tp_connect+0xd18/0x13c0
 ? pppol2tp_session_create+0x170/0x170
 ? __might_fault+0x115/0x1d0
 ? lock_downgrade+0x860/0x860
 ? __might_fault+0xe5/0x1d0
 ? security_socket_connect+0x8e/0xc0
 SYSC_connect+0x1b6/0x310
 ? SYSC_bind+0x280/0x280
 ? __do_page_fault+0x5d1/0xca0
 ? up_read+0x1f/0x40
 ? __do_page_fault+0x3c8/0xca0
 SyS_connect+0x29/0x30
 ? SyS_accept+0x40/0x40
 do_syscall_64+0x1e0/0x730
 ? trace_hardirqs_off_thunk+0x1a/0x1c
 entry_SYSCALL_64_after_hwframe+0x42/0xb7
RIP: 0033:0x7ffb3e376259
RSP: 002b:00007ffeda4f6508 EFLAGS: 00000202 ORIG_RAX: 000000000000002a
RAX: ffffffffffffffda RBX: 0000000020e77012 RCX: 00007ffb3e376259
RDX: 000000000000002e RSI: 0000000020e77000 RDI: 0000000000000004
RBP: 00007ffeda4f6540 R08: 0000000000000000 R09: 0000000000000000
R10: 0000000000000000 R11: 0000000000000202 R12: 0000000000400b60
R13: 00007ffeda4f6660 R14: 0000000000000000 R15: 0000000000000000
Code: 80 3d b0 ff 06 02 00 0f 84 07 02 00 00 e8 13 d6 db fc 49 8d bc 24 28 02 00 00 48 b8 00 00 00 00 00 fc ff df 48 89 f
a 48 c1 ea 03 <80> 3c 02 00 0f 85 ed 02 00 00 4d 8b a4 24 28 02 00 00 e8 13 16

Fixes: 80d84ef3ff1dd ("l2tp: prevent l2tp_tunnel_delete racing with userspace close")
Signed-off-by: James Chapman <jchapman@katalix.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

1 files changed, 3 insertions, 7 deletions

diff --git a/net/l2tp/l2tp_ip.c b/net/l2tp/l2tp_ip.c
index ff61124fdf59..3428fba6f2b7 100644
--- a/net/l2tp/l2tp_ip.c
+++ b/net/l2tp/l2tp_ip.c
@@ -234,17 +234,13 @@ 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 = sk->sk_user_data;
 
 	while ((skb = __skb_dequeue_tail(&sk->sk_write_queue)) != NULL)
 		kfree_skb(skb);
 
-	if (tunnel) {
-		l2tp_tunnel_closeall(tunnel);
-		sock_put(sk);
-	}
-
-	sk_refcnt_debug_dec(sk);
+	if (tunnel)
+		l2tp_tunnel_delete(tunnel);
 }
 
 static int l2tp_ip_bind(struct sock *sk, struct sockaddr *uaddr, int addr_len)