Re: KMSAN: uninit-value in can_receive

From: Oliver Hartkopp
Date: Tue Nov 19 2019 - 02:38:58 EST

On 18/11/2019 22.15, Marc Kleine-Budde wrote:
On 11/18/19 9:49 PM, Oliver Hartkopp wrote:

On 18/11/2019 21.29, Marc Kleine-Budde wrote:
On 11/18/19 9:25 PM, Oliver Hartkopp wrote:

IMPORTANT: if you fix the bug, please add the following tag to the commit:
Reported-by: syzbot+b02ff0707a97e4e79ebb@xxxxxxxxxxxxxxxxxxxxxxxxx

BUG: KMSAN: uninit-value in can_receive+0x23c/0x5e0 net/can/af_can.c:649
CPU: 1 PID: 3490 Comm: syz-executor.2 Not tainted 5.4.0-rc5+ #0

In line 649 of 5.4.0-rc5+ we can find a while() statement:

while (!(can_skb_prv(skb)->skbcnt))
can_skb_prv(skb)->skbcnt = atomic_inc_return(&skbcounter);

In linux/include/linux/can/skb.h we see:

static inline struct can_skb_priv *can_skb_prv(struct sk_buff *skb)
return (struct can_skb_priv *)(skb->head);

IMO accessing can_skb_prv(skb)->skbcnt at this point is a valid
operation which has no uninitialized value.

Can this probably be a false positive of KMSAN?

The packet is injected via the packet socket into the kernel. Where does
skb->head point to in this case? When the skb is a proper
kernel-generated skb containing a CAN-2.0 or CAN-FD frame skb->head is
maybe properly initialized?

The packet is either received via vcan or vxcan which checks via
can_dropped_invalid_skb() if we have a valid ETH_P_CAN type skb.

According to the call stack it's injected into the kernel via a packet
socket and not via v(x)can.

See ioctl$ifreq

23:11:34 executing program 2:
r0 = socket(0x200000000000011, 0x3, 0x0)
ioctl$ifreq_SIOCGIFINDEX_vcan(r0, 0x8933, &(0x7f0000000040)={'vxcan1\x00', <r1=>0x0})
bind$packet(r0, &(0x7f0000000300)={0x11, 0xc, r1}, 0x14)
sendmmsg(r0, &(0x7f0000000d00), 0x400004e, 0x0)

We only can receive skbs from (v(x))can devices.
No matter if someone wrote to them via PF_CAN or PF_PACKET.
We check for ETH_P_CAN(FD) type and ARPHRD_CAN dev type at rx time.

We additionally might think about introducing a check whether we have a
can_skb_reserve() created skbuff.

But even if someone forged a skbuff without this reserved space the
access to can_skb_prv(skb)->skbcnt would point into some CAN frame
content - which is still no access to uninitialized content, right?

So this question remains still valid whether we have a false positive from KMSAN here.