Re: [PATCH] bpf, sockmap: Use sk_rmem_schedule in bpf_tcp_ingress

[Michal Kubiak <michal.kubiak@intel.com>]

On Thu, Jul 04, 2024 at 12:39:01PM +0800, zhengguoyong wrote:
> In sockmap redirect mode, when msg send to redir sk,
> we use sk_wrmem_schedule to check memory is enough,
> 
>     tcp_bpf_sendmsg
>         tcp_bpf_send_verdict
>             bpf_tcp_ingress
>                 sk_wmem_schedule
> 
> but in bpf_tcp_ingress, the parameter sk means receiver,
> so use sk_rmem_schedule here is more suitability.
> 
> Signed-off-by: GuoYong Zheng <zhenggy@chinatelecom.cn>
> ---
>  net/ipv4/tcp_bpf.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/net/ipv4/tcp_bpf.c b/net/ipv4/tcp_bpf.c
> index 53b0d62..88c58b5 100644
> --- a/net/ipv4/tcp_bpf.c
> +++ b/net/ipv4/tcp_bpf.c
> @@ -49,7 +49,7 @@ static int bpf_tcp_ingress(struct sock *sk, struct sk_psock *psock,
>  		sge = sk_msg_elem(msg, i);
>  		size = (apply && apply_bytes < sge->length) ?
>  			apply_bytes : sge->length;
> -		if (!sk_wmem_schedule(sk, size)) {
> +		if (!sk_rmem_schedule(sk, size)) {
>  			if (!copied)
>  				ret = -ENOMEM;
>  			break;
> -- 
> 1.8.3.1
> 
> 

From the commit message I'm not really sure about the intention of this
patch, however it seems the existing kernel implementation is correct.
Changing sk_wmem_schedule -> sk_rmem_schedule breaks the kernel compilation
because those 2 functions even have different input parameters.

Please see the Patchwork results for details:
https://patchwork.kernel.org/project/netdevbpf/patch/ae2569fa-f34a-40d6-9a03-33a455fbb9ea@chinatelecom.cn/

Thanks,
Michal

Nacked-by: Michal Kubiak <michal.kubiak@intel.com>