can: skb: can_skb_set_owner(): fix ref counting if socket was closed before setting skb ownership

There are two ref count variables controlling the free()ing of a socket:
- struct sock::sk_refcnt - which is changed by sock_hold()/sock_put()
- struct sock::sk_wmem_alloc - which accounts the memory allocated by
  the skbs in the send path.

In case there are still TX skbs on the fly and the socket() is closed,
the struct sock::sk_refcnt reaches 0. In the TX-path the CAN stack
clones an "echo" skb, calls sock_hold() on the original socket and
references it. This produces the following back trace:

| WARNING: CPU: 0 PID: 280 at lib/refcount.c:25 refcount_warn_saturate+0x114/0x134
| refcount_t: addition on 0; use-after-free.
| Modules linked in: coda_vpu(E) v4l2_jpeg(E) videobuf2_vmalloc(E) imx_vdoa(E)
| CPU: 0 PID: 280 Comm: test_can.sh Tainted: G            E     5.11.0-04577-gf8ff6603c617 #203
| Hardware name: Freescale i.MX6 Quad/DualLite (Device Tree)
| Backtrace:
| [<80bafea4>] (dump_backtrace) from [<80bb0280>] (show_stack+0x20/0x24) r7:00000000 r6:600f0113 r5:00000000 r4:81441220
| [<80bb0260>] (show_stack) from [<80bb593c>] (dump_stack+0xa0/0xc8)
| [<80bb589c>] (dump_stack) from [<8012b268>] (__warn+0xd4/0x114) r9:00000019 r8:80f4a8c2 r7:83e4150c r6:00000000 r5:00000009 r4:80528f90
| [<8012b194>] (__warn) from [<80bb09c4>] (warn_slowpath_fmt+0x88/0xc8) r9:83f26400 r8:80f4a8d1 r7:00000009 r6:80528f90 r5:00000019 r4:80f4a8c2
| [<80bb0940>] (warn_slowpath_fmt) from [<80528f90>] (refcount_warn_saturate+0x114/0x134) r8:00000000 r7:00000000 r6:82b44000 r5:834e5600 r4:83f4d540
| [<80528e7c>] (refcount_warn_saturate) from [<8079a4c8>] (__refcount_add.constprop.0+0x4c/0x50)
| [<8079a47c>] (__refcount_add.constprop.0) from [<8079a57c>] (can_put_echo_skb+0xb0/0x13c)
| [<8079a4cc>] (can_put_echo_skb) from [<8079ba98>] (flexcan_start_xmit+0x1c4/0x230) r9:00000010 r8:83f48610 r7:0fdc0000 r6:0c080000 r5:82b44000 r4:834e5600
| [<8079b8d4>] (flexcan_start_xmit) from [<80969078>] (netdev_start_xmit+0x44/0x70) r9:814c0ba0 r8:80c8790c r7:00000000 r6:834e5600 r5:82b44000 r4:82ab1f00
| [<80969034>] (netdev_start_xmit) from [<809725a4>] (dev_hard_start_xmit+0x19c/0x318) r9:814c0ba0 r8:00000000 r7:82ab1f00 r6:82b44000 r5:00000000 r4:834e5600
| [<80972408>] (dev_hard_start_xmit) from [<809c6584>] (sch_direct_xmit+0xcc/0x264) r10:834e5600 r9:00000000 r8:00000000 r7:82b44000 r6:82ab1f00 r5:834e5600 r4:83f27400
| [<809c64b8>] (sch_direct_xmit) from [<809c6c0c>] (__qdisc_run+0x4f0/0x534)

To fix this problem, only set skb ownership to sockets which have still
a ref count > 0.

Fixes: 0ae89beb ("can: add destructor for self generated skbs")
Cc: Oliver Hartkopp <socketcan@hartkopp.net>
Cc: Andre Naujoks <nautsch2@gmail.com>
Link: https://lore.kernel.org/r/20210226092456.27126-1-o.rempel@pengutronix.deSuggested-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Oleksij Rempel <o.rempel@pengutronix.de>
Reviewed-by: Oliver Hartkopp <socketcan@hartkopp.net>
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>

include/linux/can/skb.h

@@ -65,8 +65,12 @@ static inline void can_skb_reserve(struct sk_buff *skb)
 
 static inline void can_skb_set_owner(struct sk_buff *skb, struct sock *sk)
 {
-	if (sk) {
-		sock_hold(sk);
+	/* If the socket has already been closed by user space, the
+	 * refcount may already be 0 (and the socket will be freed
+	 * after the last TX skb has been freed). So only increase
+	 * socket refcount if the refcount is > 0.
+	 */
+	if (sk && refcount_inc_not_zero(&sk->sk_refcnt)) {
 		skb->destructor = sock_efree;
 		skb->sk = sk;
 	}