[net] udp: must lock the socket in udp_disconnect()

From: Eric Dumazet <edumazet@google.com>

Baozeng Ding reported KASAN traces showing uses after free in
udp_lib_get_port() and other related UDP functions.

A CONFIG_DEBUG_PAGEALLOC=y kernel would eventually crash.

I could write a reproducer with two threads doing :

static int sock_fd;
static void *thr1(void *arg)
{
	for (;;) {
		connect(sock_fd, (const struct sockaddr *)arg,
			sizeof(struct sockaddr_in));
	}
}

static void *thr2(void *arg)
{
	struct sockaddr_in unspec;

	for (;;) {
		memset(&unspec, 0, sizeof(unspec));
	        connect(sock_fd, (const struct sockaddr *)&unspec,
			sizeof(unspec));
        }
}

Problem is that udp_disconnect() could run without holding socket lock,
and this was causing list corruptions.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Reported-by: Baozeng Ding <sploving1@gmail.com>
---
 include/net/udp.h   |    1 +
 net/ipv4/ping.c     |    2 +-
 net/ipv4/raw.c      |    2 +-
 net/ipv4/udp.c      |   13 +++++++++++--
 net/ipv6/ping.c     |    2 +-
 net/ipv6/raw.c      |    2 +-
 net/l2tp/l2tp_ip.c  |    2 +-
 net/l2tp/l2tp_ip6.c |    2 +-
 8 files changed, 18 insertions(+), 8 deletions(-)

From: Eric Dumazet <eric.dumazet@gmail.com>
Date: Thu, 20 Oct 2016 09:39:40 -0700

> From: Eric Dumazet <edumazet@google.com>
> 
> Baozeng Ding reported KASAN traces showing uses after free in
> udp_lib_get_port() and other related UDP functions.
> 
> A CONFIG_DEBUG_PAGEALLOC=y kernel would eventually crash.
> 
> I could write a reproducer with two threads doing :
> 
> static int sock_fd;
> static void *thr1(void *arg)
> {
> 	for (;;) {
> 		connect(sock_fd, (const struct sockaddr *)arg,
> 			sizeof(struct sockaddr_in));
> 	}
> }
> 
> static void *thr2(void *arg)
> {
> 	struct sockaddr_in unspec;
> 
> 	for (;;) {
> 		memset(&unspec, 0, sizeof(unspec));
> 	        connect(sock_fd, (const struct sockaddr *)&unspec,
> 			sizeof(unspec));
>         }
> }
> 
> Problem is that udp_disconnect() could run without holding socket lock,
> and this was causing list corruptions.
> 
> Signed-off-by: Eric Dumazet <edumazet@google.com>
> Reported-by: Baozeng Ding <sploving1@gmail.com>

Applied, sounds like I should queue this up for -stable too right?

On Thu, 2016-10-20 at 14:46 -0400, David Miller wrote:

> 
> Applied, sounds like I should queue this up for -stable too right?

Yes, I believe all stable versions have this bug.
Thanks.

On Thu, Oct 20, 2016 at 9:39 AM, Eric Dumazet <eric.dumazet@gmail.com> wrote:
> From: Eric Dumazet <edumazet@google.com>
>
> Baozeng Ding reported KASAN traces showing uses after free in
> udp_lib_get_port() and other related UDP functions.
>
> A CONFIG_DEBUG_PAGEALLOC=y kernel would eventually crash.
>
> I could write a reproducer with two threads doing :
>
> static int sock_fd;
> static void *thr1(void *arg)
> {
>         for (;;) {
>                 connect(sock_fd, (const struct sockaddr *)arg,
>                         sizeof(struct sockaddr_in));
>         }
> }
>
> static void *thr2(void *arg)
> {
>         struct sockaddr_in unspec;
>
>         for (;;) {
>                 memset(&unspec, 0, sizeof(unspec));
>                 connect(sock_fd, (const struct sockaddr *)&unspec,
>                         sizeof(unspec));
>         }
> }
>
> Problem is that udp_disconnect() could run without holding socket lock,
> and this was causing list corruptions.

If this is the cause of the hashlist corruption (I am still unsure about this),
then why only UDP? Don't all of those using ip4_datagram_connect()
as ->connect() and using udp_disconnect() as ->disconnect() need this fix?

For example, after your patch,

        .connect =      ip4_datagram_connect,
-       .disconnect =   udp_disconnect,
+       .disconnect =   __udp_disconnect,

Ping socket still doesn't have sock lock for ->disconnect() but has it for
->connect()? I must miss something...

On Thu, 2016-10-20 at 18:12 -0700, Cong Wang wrote:

> If this is the cause of the hashlist corruption (I am still unsure about this),
> then why only UDP? Don't all of those using ip4_datagram_connect()
> as ->connect() and using udp_disconnect() as ->disconnect() need this fix?
> 
> For example, after your patch,
> 
>         .connect =      ip4_datagram_connect,
> -       .disconnect =   udp_disconnect,
> +       .disconnect =   __udp_disconnect,
> 
> Ping socket still doesn't have sock lock for ->disconnect() but has it for
> ->connect()? I must miss something...

ping is less complex, it is protected by a single ping_table.lock

While UDP has to maintain two hash chains with may locks, and has to
handle rehash(), because autobind happens before full 4-tuple is setup.