Re: [PATCH ipsec v1 1/2] xfrm: Update offload configuration during SA updates

[Simon Horman <horms@kernel.org>]

On Wed, Jan 22, 2025 at 12:09:40PM +0000, Chiachang Wang wrote:
> The offload setting is set to HW when the ipsec session is
> initialized but cannot be changed until the session is torn
> down. The session administrator should be able to update
> the SA via netlink message.
> 
> This patch ensures that when a SA is updated, the associated
> offload configuration is also updated. This is necessary to
> maintain consistency between the SA and the offload device,
> especially when the device is configured for IPSec offload.
> 
> Any offload changes to the SA are reflected in the kernel
> and offload device.
> 
> Test: Enable both in/out crypto offload, and verify with
>       Android device on WiFi/cellular network, including
>       1. WiFi + crypto offload -> WiFi + no offload
>       2. WiFi + no offload -> WiFi + crypto offload
>       3. Cellular + crypto offload -> Cellular + no offload
>       4. Cellular + no offload -> Cellular + crypto offload
> Signed-off-by: Chiachang Wang <chiachangwang@google.com>
> ---
>  net/xfrm/xfrm_state.c | 24 +++++++++++++++++++++++-
>  1 file changed, 23 insertions(+), 1 deletion(-)
> 
> diff --git a/net/xfrm/xfrm_state.c b/net/xfrm/xfrm_state.c

...

> @@ -2124,7 +2125,28 @@ int xfrm_state_update(struct xfrm_state *x)
>  			__xfrm_state_bump_genids(x1);
>  			spin_unlock_bh(&net->xfrm.xfrm_state_lock);
>  		}
> +#ifdef CONFIG_XFRM_OFFLOAD
> +	x1->type_offload = x->type_offload;
> +
> +	if (memcmp(&x1->xso, &x->xso, sizeof(x1->xso))) {
> +		old_dev = x1->xso.dev;
> +		memcpy(&x1->xso, &x->xso, sizeof(x1->xso));
> +
> +		if (old_dev)
> +			old_dev->xfrmdev_ops->xdo_dev_state_delete(x1);
> +
> +		if (x1->xso.dev) {
> +			xso = &x1->xso;
> +			netdev_hold(xso->dev, &xso->dev_tracker, GFP_ATOMIC);
> +			err = xso->dev->xfrmdev_ops->xdo_dev_state_add(x1, NULL);
>  
> +			if (err) {
> +				netdev_put(xso->dev, &xso->dev_tracker);
> +				goto fail;
> +			}
> +		}
> +	}
> +#endif

For consistency, it looks like all of the code above should be indented by
one more tabstop.

>  		err = 0;
>  		x->km.state = XFRM_STATE_DEAD;
>  		__xfrm_state_put(x);