From: Henry Martin <bsdhenryma@tencent.com>

vmci_transport_dgram_dequeue lack of buffer length validation before
accessing `vmci_datagram` header.

Trigger Path:
1. Attacker sends a datagram with length < sizeof(struct
   vmci_datagram).
2. `skb_recv_datagram()` returns the malformed sk_buff (skb->len <
   sizeof(struct vmci_datagram)).
3. Code casts skb->data to struct vmci_datagram *dg without verifying
   skb->len.
4. Accessing `dg->payload_size` (Line: `payload_len =
   dg->payload_size;`) reads out-of-bounds memory.

Fixes: d021c344051a ("VSOCK: Introduce VM Sockets")
Reported-by: TCS Robot <tcs_robot@tencent.com>
Signed-off-by: Henry Martin <bsdhenryma@tencent.com>
---
 net/vmw_vsock/vmci_transport.c | 5 +++++
 1 file changed, 5 insertions(+)

diff --git a/net/vmw_vsock/vmci_transport.c b/net/vmw_vsock/vmci_transport.c
index 7eccd6708d66..0be605e19b2e 100644
--- a/net/vmw_vsock/vmci_transport.c
+++ b/net/vmw_vsock/vmci_transport.c
@@ -1749,6 +1749,11 @@ static int vmci_transport_dgram_dequeue(struct vsock_sock *vsk,
 	if (!skb)
 		return err;
 
+	if (skb->len < sizeof(struct vmci_datagram)) {
+		err = -EINVAL;
+		goto out;
+	}
+
 	dg = (struct vmci_datagram *)skb->data;
 	if (!dg)
 		/* err is 0, meaning we read zero bytes. */
-- 
2.41.3

On Tue, Aug 05, 2025 at 02:20:41PM +0800, bsdhenrymartin@gmail.com wrote:
>From: Henry Martin <bsdhenryma@tencent.com>
>
>vmci_transport_dgram_dequeue lack of buffer length validation before
>accessing `vmci_datagram` header.
>
>Trigger Path:
>1. Attacker sends a datagram with length < sizeof(struct
>   vmci_datagram).

How?

>2. `skb_recv_datagram()` returns the malformed sk_buff (skb->len <
>   sizeof(struct vmci_datagram)).

The sk_buff is queued by vmci_transport_recv_dgram_cb() calling
sk_receive_skb(). And It is allocated with this code:

#define VMCI_DG_HEADERSIZE sizeof(struct vmci_datagram)
#define VMCI_DG_SIZE(_dg) (VMCI_DG_HEADERSIZE + (size_t)(_dg)->payload_size)

static int vmci_transport_recv_dgram_cb(void *data, struct vmci_datagram *dg)
{
	...
	size = VMCI_DG_SIZE(dg);

	/* Attach the packet to the socket's receive queue as an sk_buff. */
	skb = alloc_skb(size, GFP_ATOMIC);
	...
	skb_put(skb, size);
	...
}

So I don't understand what this patch is fixing...

>3. Code casts skb->data to struct vmci_datagram *dg without verifying
>   skb->len.
>4. Accessing `dg->payload_size` (Line: `payload_len =
>   dg->payload_size;`) reads out-of-bounds memory.
>
>Fixes: d021c344051a ("VSOCK: Introduce VM Sockets")
>Reported-by: TCS Robot <tcs_robot@tencent.com>

Please fix your robot and also check your patches.
This is the second no-sense patch from you I reviewed today, I'll start
to ignore if you continue.

Stefano

>Signed-off-by: Henry Martin <bsdhenryma@tencent.com>
>---
> net/vmw_vsock/vmci_transport.c | 5 +++++
> 1 file changed, 5 insertions(+)
>
>diff --git a/net/vmw_vsock/vmci_transport.c b/net/vmw_vsock/vmci_transport.c
>index 7eccd6708d66..0be605e19b2e 100644
>--- a/net/vmw_vsock/vmci_transport.c
>+++ b/net/vmw_vsock/vmci_transport.c
>@@ -1749,6 +1749,11 @@ static int vmci_transport_dgram_dequeue(struct vsock_sock *vsk,
> 	if (!skb)
> 		return err;
>
>+	if (skb->len < sizeof(struct vmci_datagram)) {
>+		err = -EINVAL;
>+		goto out;
>+	}
>+
> 	dg = (struct vmci_datagram *)skb->data;
> 	if (!dg)
> 		/* err is 0, meaning we read zero bytes. */
>-- 
>2.41.3
>