ax25: improve the incomplete fix to avoid UAF and NPD bugs

The previous commit 1ade48d0c27d ("ax25: NPD bug when detaching
AX25 device") introduce lock_sock() into ax25_kill_by_device to
prevent NPD bug. But the concurrency NPD or UAF bug will occur,
when lock_sock() or release_sock() dereferences the ax25_cb->sock.

The NULL pointer dereference bug can be shown as below:

ax25_kill_by_device()        | ax25_release()
                             |   ax25_destroy_socket()
                             |     ax25_cb_del()
  ...                        |     ...
                             |     ax25->sk=NULL;
  lock_sock(s->sk); //(1)    |
  s->ax25_dev = NULL;        |     ...
  release_sock(s->sk); //(2) |
  ...                        |

The root cause is that the sock is set to null before dereference
site (1) or (2). Therefore, this patch extracts the ax25_cb->sock
in advance, and uses ax25_list_lock to protect it, which can synchronize
with ax25_cb_del() and ensure the value of sock is not null before
dereference sites.

The concurrency UAF bug can be shown as below:

ax25_kill_by_device()        | ax25_release()
                             |   ax25_destroy_socket()
  ...                        |   ...
                             |   sock_put(sk); //FREE
  lock_sock(s->sk); //(1)    |
  s->ax25_dev = NULL;        |   ...
  release_sock(s->sk); //(2) |
  ...                        |

The root cause is that the sock is released before dereference
site (1) or (2). Therefore, this patch uses sock_hold() to increase
the refcount of sock and uses ax25_list_lock to protect it, which
can synchronize with ax25_cb_del() in ax25_destroy_socket() and
ensure the sock wil not be released before dereference sites.

Signed-off-by: Duoming Zhou <duoming@zju.edu.cn>
Signed-off-by: David S. Miller <davem@davemloft.net>

diff --git a/net/ax25/af_ax25.c b/net/ax25/af_ax25.c
index 02f43f3..44a8730 100644 (file)
--- a/net/ax25/af_ax25.c
+++ b/net/ax25/af_ax25.c
@@ -77,6 +77,7 @@ static void ax25_kill_by_device(struct net_device *dev)
 {
        ax25_dev *ax25_dev;
        ax25_cb *s;
+       struct sock *sk;
 
        if ((ax25_dev = ax25_dev_ax25dev(dev)) == NULL)
                return;
@@ -85,13 +86,15 @@ static void ax25_kill_by_device(struct net_device *dev)
 again:
        ax25_for_each(s, &ax25_list) {
                if (s->ax25_dev == ax25_dev) {
+                       sk = s->sk;
+                       sock_hold(sk);
                        spin_unlock_bh(&ax25_list_lock);
-                       lock_sock(s->sk);
+                       lock_sock(sk);
                        s->ax25_dev = NULL;
-                       release_sock(s->sk);
+                       release_sock(sk);
                        ax25_disconnect(s, ENETUNREACH);
                        spin_lock_bh(&ax25_list_lock);
-
+                       sock_put(sk);
                        /* The entry could have been deleted from the
                         * list meanwhile and thus the next pointer is
                         * no longer valid.  Play it safe and restart