Re: [PATCH] Bluetooth: hci_core: Fix slab-use-after-free in hci_cmd_work

[Hillf Danton]

On Thu, 25 Dec 2025 00:54:07 +0100 Szymon Wilczek wrote:
> syzbot reported a slab-use-after-free in hci_cmd_work.
> The issue is that hci_send_cmd_sync() consumes the skb reference
> (either by passing it to the driver which frees it, or by calling
> kfree_skb() on error), but the skb might be accessed after the call
> in certain configurations or due to race conditions with the freeing
> process (e.g. vhci_read).
> 
> Explicitly hold a reference to the skb using skb_get() before calling
> hci_send_cmd_sync() and release it with kfree_skb() afterwards. This
> ensures the skb object remains valid throughout the function call,
> regardless of how hci_send_cmd_sync() handles the original reference.
> 
> Reported-by: syzbot+4d6b203d625d2f57d4ca@xxxxxxxxxxxxxxxxxxxxxxxxx
> Closes: https://syzkaller.appspot.com/bug?extid=4d6b203d625d2f57d4ca
> Signed-off-by: Szymon Wilczek <swilczek.lx@xxxxxxxxx>
> ---
>  net/bluetooth/hci_core.c | 3 +++
>  1 file changed, 3 insertions(+)
> 
> diff --git a/net/bluetooth/hci_core.c b/net/bluetooth/hci_core.c
> index 8ccec73dce45..af4ef31295c4 100644
> --- a/net/bluetooth/hci_core.c
> +++ b/net/bluetooth/hci_core.c
> @@ -4149,7 +4149,10 @@ static void hci_cmd_work(struct work_struct *work)
>  		if (!skb)
>  			return;
>  
> +		skb_get(skb);
>  		err = hci_send_cmd_sync(hdev, skb);
> +		kfree_skb(skb);
> +
>  		if (err)
>  			return;
>  
> -- 
> 2.52.0
> 
Given skb_dequeue() in both vhci_read() and hci_cmd_work(), what is
missed is the root cause of the issue.