[net] rxrpc: Fix possible NULL pointer access in ICMP handling
diff mbox series

Message ID 157071915431.29197.5055122258964729288.stgit@warthog.procyon.org.uk
State Accepted
Delegated to: David Miller
Headers show
Series
  • [net] rxrpc: Fix possible NULL pointer access in ICMP handling
Related show

Commit Message

David Howells Oct. 10, 2019, 2:52 p.m. UTC
If an ICMP packet comes in on the UDP socket backing an AF_RXRPC socket as
the UDP socket is being shut down, rxrpc_error_report() may get called to
deal with it after sk_user_data on the UDP socket has been cleared, leading
to a NULL pointer access when this local endpoint record gets accessed.

Fix this by just returning immediately if sk_user_data was NULL.

The oops looks like the following:

#PF: supervisor read access in kernel mode
#PF: error_code(0x0000) - not-present page
...
RIP: 0010:rxrpc_error_report+0x1bd/0x6a9
...
Call Trace:
 ? sock_queue_err_skb+0xbd/0xde
 ? __udp4_lib_err+0x313/0x34d
 __udp4_lib_err+0x313/0x34d
 icmp_unreach+0x1ee/0x207
 icmp_rcv+0x25b/0x28f
 ip_protocol_deliver_rcu+0x95/0x10e
 ip_local_deliver+0xe9/0x148
 __netif_receive_skb_one_core+0x52/0x6e
 process_backlog+0xdc/0x177
 net_rx_action+0xf9/0x270
 __do_softirq+0x1b6/0x39a
 ? smpboot_register_percpu_thread+0xce/0xce
 run_ksoftirqd+0x1d/0x42
 smpboot_thread_fn+0x19e/0x1b3
 kthread+0xf1/0xf6
 ? kthread_delayed_work_timer_fn+0x83/0x83
 ret_from_fork+0x24/0x30

Fixes: 17926a79320a ("[AF_RXRPC]: Provide secure RxRPC sockets for use by userspace and kernel both")
Reported-by: syzbot+611164843bd48cc2190c@syzkaller.appspotmail.com
Signed-off-by: David Howells <dhowells@redhat.com>
---

 net/rxrpc/peer_event.c |    3 +++
 1 file changed, 3 insertions(+)

Comments

David Miller Oct. 12, 2019, 4:21 a.m. UTC | #1
From: David Howells <dhowells@redhat.com>
Date: Thu, 10 Oct 2019 15:52:34 +0100

> If an ICMP packet comes in on the UDP socket backing an AF_RXRPC socket as
> the UDP socket is being shut down, rxrpc_error_report() may get called to
> deal with it after sk_user_data on the UDP socket has been cleared, leading
> to a NULL pointer access when this local endpoint record gets accessed.
> 
> Fix this by just returning immediately if sk_user_data was NULL.
> 
> The oops looks like the following:
 ...
> Fixes: 17926a79320a ("[AF_RXRPC]: Provide secure RxRPC sockets for use by userspace and kernel both")
> Reported-by: syzbot+611164843bd48cc2190c@syzkaller.appspotmail.com
> Signed-off-by: David Howells <dhowells@redhat.com>

Applied and queued up for -stable, thanks.
Eric Dumazet Oct. 12, 2019, 10:49 a.m. UTC | #2
On 10/10/19 7:52 AM, David Howells wrote:
> If an ICMP packet comes in on the UDP socket backing an AF_RXRPC socket as
> the UDP socket is being shut down, rxrpc_error_report() may get called to
> deal with it after sk_user_data on the UDP socket has been cleared, leading
> to a NULL pointer access when this local endpoint record gets accessed.
> 
> Fix this by just returning immediately if sk_user_data was NULL.
> 
> The oops looks like the following:
> 
> #PF: supervisor read access in kernel mode
> #PF: error_code(0x0000) - not-present page
> ...
> RIP: 0010:rxrpc_error_report+0x1bd/0x6a9
> ...
> Call Trace:
>  ? sock_queue_err_skb+0xbd/0xde
>  ? __udp4_lib_err+0x313/0x34d
>  __udp4_lib_err+0x313/0x34d
>  icmp_unreach+0x1ee/0x207
>  icmp_rcv+0x25b/0x28f
>  ip_protocol_deliver_rcu+0x95/0x10e
>  ip_local_deliver+0xe9/0x148
>  __netif_receive_skb_one_core+0x52/0x6e
>  process_backlog+0xdc/0x177
>  net_rx_action+0xf9/0x270
>  __do_softirq+0x1b6/0x39a
>  ? smpboot_register_percpu_thread+0xce/0xce
>  run_ksoftirqd+0x1d/0x42
>  smpboot_thread_fn+0x19e/0x1b3
>  kthread+0xf1/0xf6
>  ? kthread_delayed_work_timer_fn+0x83/0x83
>  ret_from_fork+0x24/0x30
> 
> Fixes: 17926a79320a ("[AF_RXRPC]: Provide secure RxRPC sockets for use by userspace and kernel both")
> Reported-by: syzbot+611164843bd48cc2190c@syzkaller.appspotmail.com
> Signed-off-by: David Howells <dhowells@redhat.com>
> ---
> 
>  net/rxrpc/peer_event.c |    3 +++
>  1 file changed, 3 insertions(+)
> 
> diff --git a/net/rxrpc/peer_event.c b/net/rxrpc/peer_event.c
> index c97ebdc043e4..61451281d74a 100644
> --- a/net/rxrpc/peer_event.c
> +++ b/net/rxrpc/peer_event.c
> @@ -151,6 +151,9 @@ void rxrpc_error_report(struct sock *sk)
>  	struct rxrpc_peer *peer;
>  	struct sk_buff *skb;
>  
> +	if (unlikely(!local))
> +		return;
> +
>  	_enter("%p{%d}", sk, local->debug_id);
>  
>  	/* Clear the outstanding error value on the socket so that it doesn't
> 

Okay, but we also need this.

diff --git a/net/rxrpc/peer_event.c b/net/rxrpc/peer_event.c
index c97ebdc043e44525eaecdd54bc447c1895bdca74..38db10e61f7a5cb50f9ee036b5e16ec284e723ac 100644
--- a/net/rxrpc/peer_event.c
+++ b/net/rxrpc/peer_event.c
@@ -145,9 +145,9 @@ static void rxrpc_adjust_mtu(struct rxrpc_peer *peer, struct sock_exterr_skb *se
  */
 void rxrpc_error_report(struct sock *sk)
 {
+       struct rxrpc_local *local = rcu_dereference_sk_user_data(sk);
        struct sock_exterr_skb *serr;
        struct sockaddr_rxrpc srx;
-       struct rxrpc_local *local = sk->sk_user_data;
        struct rxrpc_peer *peer;
        struct sk_buff *skb;
Eric Dumazet Oct. 12, 2019, 11:06 a.m. UTC | #3
On 10/12/19 3:49 AM, Eric Dumazet wrote:

> 
> Okay, but we also need this.
> 
> diff --git a/net/rxrpc/peer_event.c b/net/rxrpc/peer_event.c
> index c97ebdc043e44525eaecdd54bc447c1895bdca74..38db10e61f7a5cb50f9ee036b5e16ec284e723ac 100644
> --- a/net/rxrpc/peer_event.c
> +++ b/net/rxrpc/peer_event.c
> @@ -145,9 +145,9 @@ static void rxrpc_adjust_mtu(struct rxrpc_peer *peer, struct sock_exterr_skb *se
>   */
>  void rxrpc_error_report(struct sock *sk)
>  {
> +       struct rxrpc_local *local = rcu_dereference_sk_user_data(sk);
>         struct sock_exterr_skb *serr;
>         struct sockaddr_rxrpc srx;
> -       struct rxrpc_local *local = sk->sk_user_data;
>         struct rxrpc_peer *peer;
>         struct sk_buff *skb;
> 

I will psubmit the patch later once David pushes his net tree, since I do not know yet
the sha1 of your patch (to provide a proper Fixes: tag)
David Howells Oct. 15, 2019, 3:50 p.m. UTC | #4
Eric Dumazet <eric.dumazet@gmail.com> wrote:

>  void rxrpc_error_report(struct sock *sk)
>  {
> +       struct rxrpc_local *local = rcu_dereference_sk_user_data(sk);

Acked-by: David Howells <dhowells@redhat.com>

Patch
diff mbox series

diff --git a/net/rxrpc/peer_event.c b/net/rxrpc/peer_event.c
index c97ebdc043e4..61451281d74a 100644
--- a/net/rxrpc/peer_event.c
+++ b/net/rxrpc/peer_event.c
@@ -151,6 +151,9 @@  void rxrpc_error_report(struct sock *sk)
 	struct rxrpc_peer *peer;
 	struct sk_buff *skb;
 
+	if (unlikely(!local))
+		return;
+
 	_enter("%p{%d}", sk, local->debug_id);
 
 	/* Clear the outstanding error value on the socket so that it doesn't