gtp: fix use-after-free in gtp_encap_destroy()
authorTaehee Yoo <ap420073@gmail.com>
Tue, 2 Jul 2019 15:22:25 +0000 (00:22 +0900)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Wed, 31 Jul 2019 05:28:33 +0000 (07:28 +0200)
[ Upstream commit 1788b8569f5de27da09087fa3f6580d2aa04cc75 ]

gtp_encap_destroy() is called twice.
1. When interface is deleted.
2. When udp socket is destroyed.
either gtp->sk0 or gtp->sk1u could be freed by sock_put() in
gtp_encap_destroy(). so, when gtp_encap_destroy() is called again,
it would uses freed sk pointer.

patch makes gtp_encap_destroy() to set either gtp->sk0 or gtp->sk1u to
null. in addition, both gtp->sk0 and gtp->sk1u pointer are protected
by rtnl_lock. so, rtnl_lock() is added.

Test command:
   gtp-link add gtp1 &
   killall gtp-link
   ip link del gtp1

Splat looks like:
[   83.182767] BUG: KASAN: use-after-free in __lock_acquire+0x3a20/0x46a0
[   83.184128] Read of size 8 at addr ffff8880cc7d5360 by task ip/1008
[   83.185567] CPU: 1 PID: 1008 Comm: ip Not tainted 5.2.0-rc6+ #50
[   83.188469] Call Trace:
[ ... ]
[   83.200126]  lock_acquire+0x141/0x380
[   83.200575]  ? lock_sock_nested+0x3a/0xf0
[   83.201069]  _raw_spin_lock_bh+0x38/0x70
[   83.201551]  ? lock_sock_nested+0x3a/0xf0
[   83.202044]  lock_sock_nested+0x3a/0xf0
[   83.202520]  gtp_encap_destroy+0x18/0xe0 [gtp]
[   83.203065]  gtp_encap_disable.isra.14+0x13/0x50 [gtp]
[   83.203687]  gtp_dellink+0x56/0x170 [gtp]
[   83.204190]  rtnl_delete_link+0xb4/0x100
[ ... ]
[   83.236513] Allocated by task 976:
[   83.236925]  save_stack+0x19/0x80
[   83.237332]  __kasan_kmalloc.constprop.3+0xa0/0xd0
[   83.237894]  kmem_cache_alloc+0xd8/0x280
[   83.238360]  sk_prot_alloc.isra.42+0x50/0x200
[   83.238874]  sk_alloc+0x32/0x940
[   83.239264]  inet_create+0x283/0xc20
[   83.239684]  __sock_create+0x2dd/0x540
[   83.240136]  __sys_socket+0xca/0x1a0
[   83.240550]  __x64_sys_socket+0x6f/0xb0
[   83.240998]  do_syscall_64+0x9c/0x450
[   83.241466]  entry_SYSCALL_64_after_hwframe+0x49/0xbe
[   83.242061]
[   83.242249] Freed by task 0:
[   83.242616]  save_stack+0x19/0x80
[   83.243013]  __kasan_slab_free+0x111/0x150
[   83.243498]  kmem_cache_free+0x89/0x250
[   83.244444]  __sk_destruct+0x38f/0x5a0
[   83.245366]  rcu_core+0x7e9/0x1c20
[   83.245766]  __do_softirq+0x213/0x8fa

Fixes: 1e3a3abd8b28 ("gtp: make GTP sockets in gtp_newlink optional")
Signed-off-by: Taehee Yoo <ap420073@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Sasha Levin <sashal@kernel.org>
drivers/net/gtp.c

index c751f87c935e2f8bdabe17e510f95b01c693d72f..53fd66534e3a6ae17a1c5e4910cfef0575aaa3fa 100644 (file)
@@ -289,13 +289,17 @@ static int gtp1u_udp_encap_recv(struct gtp_dev *gtp, struct sk_buff *skb)
        return gtp_rx(pctx, skb, hdrlen, gtp->role);
 }
 
-static void gtp_encap_destroy(struct sock *sk)
+static void __gtp_encap_destroy(struct sock *sk)
 {
        struct gtp_dev *gtp;
 
        lock_sock(sk);
        gtp = sk->sk_user_data;
        if (gtp) {
+               if (gtp->sk0 == sk)
+                       gtp->sk0 = NULL;
+               else
+                       gtp->sk1u = NULL;
                udp_sk(sk)->encap_type = 0;
                rcu_assign_sk_user_data(sk, NULL);
                sock_put(sk);
@@ -303,12 +307,19 @@ static void gtp_encap_destroy(struct sock *sk)
        release_sock(sk);
 }
 
+static void gtp_encap_destroy(struct sock *sk)
+{
+       rtnl_lock();
+       __gtp_encap_destroy(sk);
+       rtnl_unlock();
+}
+
 static void gtp_encap_disable_sock(struct sock *sk)
 {
        if (!sk)
                return;
 
-       gtp_encap_destroy(sk);
+       __gtp_encap_destroy(sk);
 }
 
 static void gtp_encap_disable(struct gtp_dev *gtp)
@@ -1045,6 +1056,7 @@ static int gtp_genl_new_pdp(struct sk_buff *skb, struct genl_info *info)
                return -EINVAL;
        }
 
+       rtnl_lock();
        rcu_read_lock();
 
        gtp = gtp_find_dev(sock_net(skb->sk), info->attrs);
@@ -1069,6 +1081,7 @@ static int gtp_genl_new_pdp(struct sk_buff *skb, struct genl_info *info)
 
 out_unlock:
        rcu_read_unlock();
+       rtnl_unlock();
        return err;
 }